American Journal of Clinical Nutrition Volume 86, Issue 6 (Dec 1 2007),pp.1577-1809

Editorials G Harvey Anderson Much ado about high-fructose corn syrup in beverages: the meat of the matter Am J Clin Nutr 2007 86: 1577-1578. Jeri W Nieves and Robert Lindsay Calcium and fracture risk Am J Clin Nutr 2007 86: 1579-1580.

Commentary Conrad Wagner and Mark J Koury S-Adenosylhomocysteine—a better indicator of vascular disease than homocysteine? Am J Clin Nutr 2007 86: 1581-1585.

Obesity and eating disorders Stijn Soenen and Margriet S Westerterp-Plantenga No differences in satiety or energy intake after high-fructose corn syrup, sucrose, or milk preloads Am J Clin Nutr 2007 86: 1586-1594 Christine L Pelkman, Juan L Navia, Allison E Miller, and Rachael J Pohle Novel calcium-gelled, alginate-pectin beverage reduced energy intake in nondieting overweight and obese women: interactions with dietary restraint status Am J Clin Nutr 2007 86: 1595-1602. Lisa J Moran, Manny Noakes, Peter M Clifton, Gary A Wittert, Carel W Le Roux, Mohammed A Ghatei, Stephen R Bloom, and Robert J Norman Postprandial ghrelin, cholecystokinin, peptide YY, and appetite before and after weight loss in overweight women with and without polycystic ovary syndrome Am J Clin Nutr 2007 86: 1603-1610.

Lipids Lars Berglund, Michael Lefevre, Henry N Ginsberg, Penny M Kris-Etherton, Patricia J Elmer, Paul W Stewart, Abby Ershow, Thomas A Pearson, Barbara H Dennis, Paul S Roheim, Rajasekhar Ramakrishnan, Roberta Reed, Kent Stewart, Katherine M Phillips for the DELTA Investigators Comparison of monounsaturated fat with carbohydrates as a replacement for saturated fat in subjects with a high metabolic risk profile: studies in the fasting and postprandial states Am J Clin Nutr 2007 86: 1611-1620. William S Harris, James V Pottala, Scott A Sands, and Philip G Jones Comparison of the effects of fish and fish-oil capsules on the n–3 fatty acid content of blood cells and plasma phospholipids Am J Clin Nutr 2007 86: 1621-1625.

Cardiovascular disease risk Haidong Kan, June Stevens, Gerardo Heiss, Ronald Klein, Kathryn M Rose, and Stephanie J London Dietary fiber intake and retinal vascular caliber in the Atherosclerosis Risk in Communities Study Am J Clin Nutr 2007 86: 1626-1632. Marguerite Gastaldi, Sophie Dizière, Catherine Defoort, Henri Portugal, Denis Lairon, Michel Darmon, and Richard Planells Sex-specific association of fatty acid binding protein 2 and microsomal triacylglycerol transfer protein variants with response to dietary lipid changes in the 3-mo Medi-RIVAGE primary intervention study Am J Clin Nutr 2007 86: 1633-1641. M Luisa Trirogoff, Ayumi Shintani, Jonathan Himmelfarb, and T Alp Ikizler Body mass index and fat mass are the primary correlates of insulin resistance in nondiabetic stage 3–4 chronic kidney disease patients Am J Clin Nutr 2007 86: 1642-1648.

Nutritional status, dietary intake, and body composition Maggie L Zou, Paul J Moughan, Ajay Awati, and Geoffrey Livesey Accuracy of the Atwater factors and related food energy conversion factors with low-fat, high-fiber diets when energy intake is reduced spontaneously Am J Clin Nutr 2007 86: 1649-1656. Sonia A Talwar, John F Aloia, Simcha Pollack, and James K Yeh Dose response to vitamin D supplementation among postmenopausal African American women Am J Clin Nutr 2007 86: 1657-1662

Kaisu Keskitalo, Hely Tuorila, Tim D Spector, Lynn F Cherkas, Antti Knaapila, Karri Silventoinen, and Markus Perola Same genetic components underlie different measures of sweet taste preference Am J Clin Nutr 2007 86: 1663-1669. Morvarid Kabir, Geraldine Skurnik, Nadia Naour, Valeria Pechtner, Emmanuelle Meugnier, Sophie Rome, Annie Quignard-Boulangé, Hubert Vidal, Gérard Slama, Karine Clément, Michèle Guerre-Millo, and Salwa W Rizkalla Treatment for 2 mo with n–3 polyunsaturated fatty acids reduces adiposity and some atherogenic factors but does not improve insulin sensitivity in women with type 2 diabetes: a randomized controlled study Am J Clin Nutr 2007 86: 1670-1679.

Vitamins, minerals, and phytochemicals Maria Wijaya-Erhardt, Juergen G Erhardt, Juliawati Untoro, Elvina Karyadi, Lindawati Wibowo, and Rainer Gross Effect of daily or weekly multiple-micronutrient and iron foodlike tablets on body iron stores of Indonesian infants aged 6–12 mo: a double-blind, randomized, placebo-controlled trial Am J Clin Nutr 2007 86: 1680-1686. Marianne C Walsh, Lorraine Brennan, Estelle Pujos-Guillot, Jean-Louis Sébédio, Augustin Scalbert, Ailís Fagan, Desmond G Higgins, and Michael J Gibney Influence of acute phytochemical intake on human urinary metabolomic profiles Am J Clin Nutr 2007 86: 1687-1693. Alisha J Rovner, Virginia A Stallings, Joan I Schall, Mary B Leonard, and Babette S Zemel Vitamin D insufficiency in children, adolescents, and young adults with cystic fibrosis despite routine oral supplementation Am J Clin Nutr 2007 86: 1694-1699.

Growth, development, and pediatrics Nadina Karaolis-Danckert, Anke LB Günther, Anja Kroke, Claudia Hornberg, and Anette E Buyken How early dietary factors modify the effect of rapid weight gain in infancy on subsequent body-composition development in term children whose birth weight was appropriate for gestational age Am J Clin Nutr 2007 86: 1700-1708.

Jennifer O Fisher, Angeles Arreola, Leann L Birch, and Barbara J Rolls Portion size effects on daily energy intake in low-income Hispanic and African American children and their mothers Am J Clin Nutr 2007 86: 1709-1716. Michael S Kramer, Lidia Matush, Irina Vanilovich, Robert W Platt, Natalia Bogdanovich, Zinaida Sevkovskaya, Irina Dzikovich, Gyorgy Shishko, Jean-Paul Collet, Richard M Martin, George Davey Smith, Matthew W Gillman, Beverley Chalmers, Ellen Hodnett, Stanley Shapiro for the Promotion of Breastfeeding Intervention Trial (PROBIT) Study Group Effects of prolonged and exclusive breastfeeding on child height, weight, adiposity, and blood pressure at age 6.5 y: evidence from a large randomized trial Am J Clin Nutr 2007 86: 1717-1721.

Cancer Jolieke C van der Pols, Chris Bain, David Gunnell, George Davey Smith, Clare Frobisher, and Richard M Martin Childhood dairy intake and adult cancer risk: 65-y follow-up of the Boyd Orr cohort Am J Clin Nutr 2007 86: 1722-1729. Elisa V Bandera, Lawrence H Kushi, Dirk F Moore, Dina M Gifkins, and Marjorie L McCullough Association between dietary fiber and endometrial cancer: a dose-response meta-analysis Am J Clin Nutr 2007 86: 1730-1737.

Aging Mariano Malaguarnera, Lisa Cammalleri, Maria Pia Gargante, Marco Vacante, Valentina Colonna, and Massimo Motta L-Carnitine treatment reduces severity of physical and mental fatigue and increases cognitive functions in centenarians: a randomized and controlled clinical trial Am J Clin Nutr 2007 86: 1738-1744.

Nutritional epidemiology and public health PK Newby, Janice Maras, Peter Bakun, Denis Muller, Luigi Ferrucci, and Katherine L Tucker Intake of whole grains, refined grains, and cereal fiber measured with 7-d diet records and associations with risk factors for chronic disease Am J Clin Nutr 2007 86: 1745-1753.

Amy E Millen, Amy F Subar, Barry I Graubard, Ulrike Peters, Richard B Hayes, Joel L Weissfeld, Lance A Yokochi, Regina G Ziegler for the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial Project Team Fruit and vegetable intake and prevalence of colorectal adenoma in a cancer screening trial Am J Clin Nutr 2007 86: 1754-1764. Anke LB Günther, Thomas Remer, Anja Kroke, and Anette E Buyken Early protein intake and later obesity risk: which protein sources at which time points throughout infancy and childhood are important for body mass index and body fat percentage at 7 y of age? Am J Clin Nutr 2007 86: 1765-1772. Eoin P Quinlivan and Jesse F Gregory, III Reassessing folic acid consumption patterns in the United States (1999– 2004): potential effect on neural tube defects and overexposure to folate Am J Clin Nutr 2007 86: 1773-1779.

Bone metabolism Heike A Bischoff-Ferrari, Bess Dawson-Hughes, John A Baron, Peter Burckhardt, Ruifeng Li, Donna Spiegelman, Bonny Specker, John E Orav, John B Wong, Hannes B Staehelin, Eilis O'Reilly, Douglas P Kiel, and Walter C Willett Calcium intake and hip fracture risk in men and women: a meta-analysis of prospective cohort studies and randomized controlled trials Am J Clin Nutr 2007 86: 1780-1790. Esther J Waugh, Janet Polivy, Rowena Ridout, and Gillian A Hawker A prospective investigation of the relations among cognitive dietary restraint, subclinical ovulatory disturbances, physical activity, and bone mass in healthy young women Am J Clin Nutr 2007 86: 1791-1801.

Letters to the Editor Frits AJ Muskiet, Remko S Kuipers, Ella N Smit, and Joséphine CA Joordens The basis of recommendations for docosahexaenoic and arachidonic acids in infant formula: absolute or relative standards? Am J Clin Nutr 2007 86: 1802-1803. J Thomas Brenna, Behzad Varamini, Deborah A Diersen-Schade, Julia A Boettcher, and Linda M Arterburn Reply to FAJ Muskiet et al Am J Clin Nutr 2007 86: 1803-1804. Rohit P Ojha, Martha J Felini, and Lori A Fischbach Vitamin D for cancer prevention: valid assertion or premature anointment? Am J Clin Nutr 2007 86: 1804-1805.

Joan M Lappe and Robert P Heaney Reply to RP Ojha et al Am J Clin Nutr 2007 86: 1805-1806. Preston W Estep, III Many factors modify the physiological response to sugary liquids Am J Clin Nutr 2007 86: 1806-1808.

Robert G Moses, Megan Luebke, Peter Petocz, and Jennie C Brand-Miller Maternal diet and infant size 2 y after the completion of a study of a lowglycemic-index diet in pregnancy Am J Clin Nutr 2007 86: 1806. Adam Drewnowski and France Bellisle Reply to PW Estep III Am J Clin Nutr 2007 86: 1808.

Books Received Books Received Am J Clin Nutr 2007 86: 1809.

Editorial See corresponding article on page 1586.

Much ado about high-fructose corn syrup in beverages: the meat of the matter1,2 G Harvey Anderson Claudio: I pray you leave me. Benedick: Ho, now you strike like the blind man—’twas the boy that stole your meat, and you’ll beat the post (1).

Over the past 35 y the prevalence of obesity has risen concurrently with an increased availability of added sugars in the food supply. Food disappearance data, used as an indicator of trends in food consumption, have shown a 20% increase in the availability of caloric sweeteners (sugars) in the United States from 1970 – 1974 to 2000 (2). Obesity has been blamed on sugars and sugarsweetened beverages, but the debate has raged for many years with little resolution (3). More recently, the intensity of the debate was fueled by the hypothesis that introduction in the 1970s of high-fructose corn syrup (HFCS) as a caloric sweetener in beverages was specifically at fault (4). HFCS was proposed to lead to obesity because fructose bypasses food intake regulatory systems and favors lipogenesis. However, although the availability of sugars has increased in the US food supply over the past 4 decades, it has not increased disproportionately as a contributor to the increase in total energy availability. From 1970 –74 to 2000 the increase in per capita availability of total energy, sugars, carbohydrates, and fats was 25%, 22%, 26%, and 48%, respectively (2). Furthermore, in the United States, HFCS has primarily been used to substitute for sucrose as a caloric sweetener rather than to be used in addition to sucrose. Sucrose use has declined from 80% of total caloric sweetener availability in 1970 to 40% of caloric sweetener availability in 1997. This reduction in sucrose consumption was simply made up for by HFCS, which has increased from nearly 0% in 1970 to 40% of total caloric sweeteners in 1997. The availability of HFCS in the US food supply did not change from 1997 [60.4 pounds (27.4 kg) per capita annually] to 2004 [59.2 pounds (26.9 kg) per capita annually] (2), during a time of continued weight gain. There is no evidence that the ratio of fructose and glucose consumed from sugars has changed over the past 4 decades as a result of HFCS replacing sucrose in many applications. The term “high fructose corn syrup ” is not a good descriptor of its composition, but the term was mandated to distinguish the newly developed fructose-containing corn syrup from the traditional all-glucose corn syrup. HFCS is predominantly sold as HFCS-55 (55% fructose, 41% glucose, and 4% glucose polymers) or HFCS-42 (42% fructose, 53% glucose, and 5% glucose polymers) (5). In North America, the former is used in beverages and the latter in solid foods.

The significance of replacing sucrose with HFCS in soft drinks and other beverages is addressed in this issue of the Journal by Soenen and Westerterp-Plantenga (6) and merits emphasis because it challenges the argument of biologic plausibility that was proposed to support the hypothesis. When compared with a diet drink, subjective appetite and food intake were decreased by the consumption of solutions of sugars (86 g/800 mL water). The drink mixtures contained primarily HFCS or sucrose and were prepared by using syrups added to beverages consumed in Europe. They contained either sucrose or HFCS to which was added glucose syrup (90% glucose and 9% fructose). The final composition of the sucrose beverage was 64% glucose and 36% fructose, whereas the HFCS beverage contained 41% glucose and 59% fructose. Thus, the difference in fructose content of the test beverages was exaggerated relative to the differences in beverages containing only sucrose or HFCS. Yet, no difference in food intake was found at a meal consumed 50 min later or in postprandial blood concentrations of glucose, insulin, glucagonlike peptide 1, or ghrelin—all components of food intake regulatory mechanisms. Implicit in the argument that HFCS in soft drinks is different from sucrose is the notion that sucrose, as the disaccharide, may in some way stimulate metabolic satiety signals or increase fatty acid synthesis more than the monosaccharide mixtures. However, these possibilities were recently explored also (7). HFCS and sucrose, compared with solutions containing equal proportions of glucose and fructose, resulted in no differences in food intake 80 min after consumption. Sucrose and solutions of its monosaccharide equivalents resulted in no differences in postprandial blood glucose, insulin, and ghrelin concentrations. Thus, there is no evidence that sucrose—when consumed in its intact form—would confer any benefits over HFCS, which contains the 2 unbound monosaccharides. Furthermore, when added to the acidic environment of soft drinks, sucrose is most likely consumed as the monosaccharide because of hydrolysis. The article by Soenen and Westerterp-Platenga (6) also suggests that all beverages may be created equal. The results support the view that compensation for calories in beverages is incomplete. Neither food intake nor caloric compensation was different 1

From the Department of Nutritional Sciences, University of Toronto, Toronto, Canada. 2 Reprints not available. Address correspondence to GH Anderson, Department of Nutritional Sciences, University of Toronto, Toronto, Ontario M5S 3E2, Canada. E-mail: [email protected].

Am J Clin Nutr 2007;86:1577– 8. Printed in USA. © 2007 American Society for Nutrition

1577

1578

EDITORIAL

after subjects consumed HFCS, sucrose, or milk drinks (1.5 mol). However, caloric compensation, the reduction of intake at the test meal as a percentage of calories in the preload, averaged 쏝40% for women and 60% for men. Whether compensation would be higher for a solid food of similar composition remains to be determined (8). There are multidimensional determinants of obesity. It was previously noted that neither sugar nor carbohydrate consumption has been clearly delineated as a direct cause of obesity (9). However, it is clear that energy imbalance for most individuals is accounted for by energy intake exceeding expenditure. The lifestyle factors that lead to this problem are too little exercise and too much food, but the determinants of such vary greatly between individuals. A food solution to obesity remains elusive, but a reductionist approach that focuses on one food or one component of the food supply, in the presence of too much (9), is unlikely to succeed. Unfortunately the recent focus on HFCS has done little to resolve the role of sugars in contributing to energy imbalance. The hypothesis that the replacement of sucrose with HFCS in beverages plays a causative role in obesity is not supported on the basis of its composition, biologic actions, or short-term effects on food intake. Had the hypothesis been phrased in the converse, namely that replacing HFCS with sucrose in beverages would be a solution for the obesity epidemic, its merit would have been seen more clearly. Put simply, a proposal that a return to sucrosecontaining beverages would be a credible solution to the obesity epidemic would have been met with outright dismissal. In many countries where trade barriers have prevented the replacement of sucrose with HFCS, the prevalence of obesity is high. Therefore, what role HFCS in beverages plays in the etiology of obesity, as in Much Ado about Nothing (1), may simply be a play on words.

Claudio: Now you talk of a sheet of paper, I remember a pretty jest your daughter told us of. Leonato: O, when she had writ it and was reading it over, she found Benedick and Beatrice between the sheet (1). I thank David Jenkins for his comments during the preparation of this editorial. GHA serves as a science advisor to the Canadian Sugar Institute and to Archer Daniels Midland, a producer of HFCS, but has no equity or other financial interests in either industry. GHA has received unrestricted grant funding from the US Sugar Association and from the Canadian Sugar Institute.

REFERENCES 1. Shakespeare W. Much ado about nothing. Ed: Zitner SP. Oxford, United Kingdom: Oxford University Press, 1993:97-202. 2. USDA. U.S. per capita food consumption. Beltsville, MD: US Department of Agriculture, Economic Research Services, 2006. 3. Pereira MA. The possible role of sugar-sweetened beverages in obesity etiology: a review of the evidence. Int J Obes Relat Metab Disord 2006; 30(suppl):S28 –36. 4. Bray GA, Nielsen SJ, Popkin BM. Consumption of high-fructose corn syrup in beverages may play a role in the epidemic of obesity. Am J Clin Nutr 2004;79:537– 43. 5. Forshee RA, Storey MA, Allison DB, et al. A critical examination of the evidence relating high fructose corn syrup and weight gain. Crit Rev Food Sci Nutr 2007;47:561– 82. 6. Soenen S, Westerterp-Plantenga MS. No differences in satiety or energy intake after high-fructose corn syrup, sucrose, or milk preloads. Am J Clin Nutr 2007;86:1586 –94. 7. Akhavan T, Anderson GH. Effects of the glucose to fructose ratios in solutions on subjective satiety, food intake, and satiety hormones in young men. Am J Clin Nutr (in press). 8. Anderson GH. Sugars-containing beverages and post-prandial satiety and food intake Int J Obes Relat Metab Disord 2006;30(suppl):S52–9. 9. Jenkins DJ, Kendall CW, Marchie A, Augustin LS. Too much sugar, too much carbohydrate, or just too much? Am J Clin Nutr 2004;79:711–2.

Editorial See corresponding article on page 1780.

Calcium and fracture risk1,2 Jeri W Nieves and Robert Lindsay A meta-analysis of cohort studies and clinical trials evaluating the effects of calcium on hip fractures by Bischoff-Ferrari et al (1) appears in this issue of the Journal. The authors’ analysis concluded that calcium, either in food or as a supplement, has no effect on hip fracture risk. This conclusion differs from that of the fracture prevention trials of calcium and vitamin D, is at odds with the use of calcium (with or without vitamin D) as background therapy in clinical trials, and departs from the standard of care for osteoporosis prevention and treatment—where does that leave clinicians? Osteoporosis is a complex disease whose pathogenesis often involves multiple factors. Even in controlled clinical trials involving calcium and vitamin D, the fracture effects have been relatively small, although often statistically significant. Perhaps these data suggest that calcium supplementation, to be effective, requires the addition of vitamin D supplementation. Recently, there has been a move to increase the recommendations for vitamin D intake. The realization that 25-hydroxyvitamin D [25(OH)D] concentrations 쏝80 nmol/L are associated with markers of poor skeletal health (ie, increased parathyroid hormone concentrations, lower calcium absorption, and lower bone mineral density) has led several experts to recommend that vitamin D intake be increased to 1000 IU/d (from the current Dietary Reference Intake of 400 to 600 IU/d) with the goal of a 25(OH)D concentration 쏜80 nmol/L. Calcium nutrition is intimately linked to vitamin D status, and, if 80 nmol/L represents a vitamin D–replete status, then a large segment of the population—from 앒30% to 앒60%, depending on the population being studied—would be vitamin D insufficient (2). Studies of calcium without regard to vitamin D status may then lead to erroneous conclusions. In the meta-analysis of Bischoff-Ferrari et al, only one clinical trial reported baseline 25(OH)D concentrations 쏜80 nmol/L, and another study had summer but not winter values that were 쏜80 nmol/L. Those authors discussed the possibility that the efficacy of calcium intake may be enhanced by additional vitamin D but did not discuss the converse possibility—that the antifracture efficacy of vitamin D may be improved by calcium. This association is of particular importance because, as BischoffFerrari et al noted, recent meta-analyses have shown a decrease in hip fracture risk in persons taking both vitamin D and calcium (3, 4) It is also difficult to accurately assess calcium intake and impossible to determine whether that intake represents a state of sufficiency for any particular person. Because calcium is a threshold element, supplementation of persons who are already

calcium replete would be unlikely to provide a further effect on fractures. The variables that affect calcium sufficiency include intestinal absorption efficiency and renal calcium conservation. Calcium absorption averages 앒30 –35% of ingested load, but it can vary enormously, generally increasing with inadequacy of either calcium intake or vitamin D status. Renal conservation also has marked individual variations. Variability in absorption and renal conservation makes it difficult to define accurately the intake required for calcium sufficiency in any person. Above that threshold, when intake is sufficient for serum calcium maintenance, the feedback system will reduce absorption efficiency and increase renal clearance but will not increase the flow of calcium into the skeleton. At calcium intakes below that threshold, the continued need to supply calcium to the circulation will increase bone remodeling, which itself can increase fracture risk. The resulting bone loss would also contribute to fracture risk. Calcium supplementation would be expected to reduce turnover and reduce fracture risk, but only in those with baseline concentrations below the threshold. The proportion of such subjects in either cohort studies or clinical trials cannot be defined, and any calcium benefit would be overwhelmed by the inclusion of calcium-replete subjects. This conclusion is supported by the findings of the Women’s Health Initiative in which mean intakes were already at 1200 mg/d before supplementation, and further intake of calcium provided no benefit (5). It is also possible that, in cohort studies such as those evaluated in this meta-analysis, there may be insufficient variability in calcium intakes to allow detection of a fracture effect. In several of the cohort studies, mean baseline calcium intake was not reported; in those studies that noted mean calcium intakes, they were between 550 and 780 mg/d—an intake that is generally considered inadequate, although at least some persons will clearly be replete at even these intakes. Although BischoffFerrari et al tried to address this problem (see Figure 3 in 1), without knowledge of the sample size per intake level, their analysis was limited. Furthermore, a single calcium assessment may not reflect usual calcium intake over the observation period of the study. 1 From the Departments of Medicine (RL) and Epidemiology (JWN), Columbia University, New York, NY, and the Clinical Research Center, Helen Hayes Hospital, West Haverstraw, NY (JWN and RL). 2 Reprints not available. Address correspondence to R Lindsay, Clinical Research Center, Helen Hayes Hospital, Route 9W, West Haverstraw, NY 10993. E-mail: [email protected].

Am J Clin Nutr 2007;86:1579 – 80. Printed in USA. © 2007 American Society for Nutrition

1579

1580

EDITORIAL

As in any clinical trial, compliance and adherence also are issues of concern; compliance was noted to range from 42% to 77% in the few trials that evaluated these 2 factors. After compliance was taken into account, what was the actual intake in these populations? The adherence population analysis provided brought the pooled risk ratio for nonvertebral fracture in relation to calcium intake down to 0.83 (95% CI: 0.64, 1.09), which is similar to that seen in the combined calcium and vitamin D trials, although here it is clearly not statistically significant (3, 4). A generalization from the literature, without a formal metaanalysis, may be that we need adequate supplies of both vitamin D and calcium to obtain significant reductions in nonvertebral fractures (especially hip fractures), and that those effects may be seen only in those persons who have insufficient vitamin D or calcium (or both). In addition, persons need to consume an overall healthful diet that meets all nutrient requirements. Protein for fracture prevention and healing (6 –9) and plenty of fruit and vegetables for overall health (10, 11) are of particular importance. A well-rounded diet is important, and evaluation of one element or vitamin does not give the whole story. So where does that leave clinicians? The best public health recommendation would be that people should consume an overall healthful diet including adequate consumption of both calcium and vitamin D. This meta-analysis highlights the importance of not segmenting nutrition into heterogeneous populations and isolated nutrients. Bone is not just calcium, and calcium does not function in isolation. Neither of the authors had a personal or financial conflict of interest.

REFERENCES 1. Bischoff-Ferrari HA, Dawson-Hughes B, Baron JA, et al. Calcium intake and hip fracture risk in men and women: a meta-analysis of prospective cohort studies and randomized controlled trials. Am J Clin Nutr 2007;86:1780 –90. 2. Holick MF. High prevalence of vitamin D inadequacy and implications for health. Mayo Clin Proc 2006;81:353–73. 3. Avenell A, Gillespie WJ, Gillespie LD, O’Connell DL. Vitamin D and vitamin D analogues for preventing fractures associated with involutional and post-menopausal osteoporosis. Cochrane Database Syst Rev 2005;CD000227. 4. Boonen S, Lips P, Bouillon R, Bischoff-Ferrari HA, Vanderschueren D, Haentjens P. Need for additional calcium to reduce the risk of hip fracture with vitamin D supplementation: evidence from a comparative metaanalysis of randomized controlled trials. J Clin Endocrinol Metab 2007; 92:1415–23. 5. Jackson RD, LaCroix AZ, Gass M, et al. Calcium plus vitamin D supplementation and the risk of fractures. N Engl J Med 2006;354:669 – 83. 6. Bonjour JP. Dietary protein: an essential nutrient for bone health. J Am Coll Nutr 2005;24(suppl):526S–36S. 7. Avenell A, Handoll HH. Nutritional supplementation for hip fracture aftercare in older people. Cochrane Database Syst Rev 2006;CD001880. 8. Schurch MA, Rizzoli R, Slosman D, Vadas L, Vergnaud P, Bonjour JP. Protein supplements increase serum insulin-like growth factor-I levels and attenuate proximal femur bone loss in patients with recent hip fracture. A randomized, double-blind, placebo-controlled trial. Ann Intern Med 1998;128:801–9. 9. Eneroth M, Olsson UB, Thorngren KG. Nutritional supplementation decreases hip fracture-related complications. Clin Orthop Relat Res 2006;451:212–7. 10. New SA. Do vegetarians have a normal bone mass? Osteoporos Int 2004;15:679 – 88. 11. Prynne CJ, Mishra GD, O’Connell MA, et al. Fruit and vegetable intakes and bone mineral status: a cross-sectional study in 5 age and sex cohorts. Am J Clin Nutr 2006;83:1420 – 8.

Commentary

S-Adenosylhomocysteine—a better indicator of vascular disease than homocysteine?1–3 Conrad Wagner and Mark J Koury ABSTRACT It is widely accepted that elevated plasma total homocysteine is an independent risk factor for vascular disease. The relation is believed to be causal, but there is no generally accepted mechanism for the pathophysiology involved. The metabolic precursor of homocysteine in all tissues is S-adenosylhomocysteine (AdoHcy). AdoHcy is present in normal human plasma at concentrations approximately 1-500th of those of homocysteine, a fact that presents difficulties in measurement. The requirement for specialized equipment, complicated time-consuming methodology, or both is a reason that measurement of plasma AdoHcy has not generally been carried out in large studies. A recently published rapid immunoassay for AdoHcy in human plasma should make measurement of this important metabolite available for general use. Advantages of the measurement of plasma AdoHcy include 1) a smaller overlap of values between control subjects and patients, and thus the possibility of observing significant differences in fewer samples, 2) an accepted mechanism of metabolic activity as an inhibitor of all S-adenosylmethionine– mediated methyltransferases, and 3) evidence (from recent studies) that a higher plasma concentration of AdoHcy is a more sensitive indicator of vascular disease than is a higher plasma concentration of homocysteine. Am J Clin Nutr 2007;86: 1581–5. KEY WORDS S-Adenosylhomocysteine, S-adenosylmethionine, homocysteine, vascular disease, methionine, risk factors, plasma

INTRODUCTION

S-Adenosylhomocysteine (AdoHcy) is the immediate precursor of all of the homocysteine produced in the body. The reaction is catalyzed by S-adenosylhomocysteine hydrolase and is reversible with the equilibrium favoring formation of AdoHcy. In vivo, the reaction is driven in the direction of homocysteine formation by the action of the enzyme adenosine deaminase, which converts the second product of the S-adenosylhomocysteine hydrolase reaction, adenosine, to inosine (1). Homocysteine is a branch point in the metabolism of methionine. In one direction, it can be remethylated either by the vitamin B-12– dependent enzyme system, methionine synthase, or it can accept a methyl group from betaine to regenerate methionine. In a second direction, homocysteine can be degraded by the transsulfuration pathway by conversion to cystathionine with the use of the enzyme cystathionine-␤-synthase (2).

EVIDENCE FOR HOMOCYSTEINE’S INVOLVEMENT IN VASCULAR DISEASE

The initial study by McCully (3) showed that homocystinuria resulted in massive thromboses and generalized vascular damage. The associated elevation in plasma homocysteine accompanying the homocystinuria in such cases can range from 150 to 500 ␮mol/L (normal: 앒10 ␮mol/L). However, it was not until Wilcken and Wilcken (4) examined patients with and without cardiovascular disease (CVD) that it was suggested that a moderate increase in plasma homocysteine was associated with vascular disease. Since that time, thousands of journal articles have been published on the relation between plasma or serum concentrations of homocysteine and vascular disease. Wilcken and Wilcken showed that 앒28% of patients with coronary heart disease had an abnormal methionine load test, which indicated a lower ability to metabolize methionine. An oral methionine load stresses the systems metabolizing methionine and results in a higher and more prolonged increase in plasma methionine in patients with CVD than in control subjects. An abnormal methionine load test was shown by Clarke et al (5) to be an independent risk factor for coronary, peripheral, and cerebral vascular disease. An elevation of plasma homocysteine in patients with vascular disease was also observed in those without a methionine load (6). Many subsequent studies have provided support for this conclusion. The meta-analysis of 27 studies by Boushey et al (7) concluded that there was a strong association of elevated plasma homocysteine with coronary, cerebrovascular, and peripheral vascular disease and that as much as 10% of the coronary artery disease in the United States could be attributed to high plasma concentrations of homocysteine. A survey of articles published between 1966 and 1998 analyzed results from 30 prospective or retrospective studies (8) and found a stronger association in the retrospective than in the prospective studies. The common polymorphism in the methylenetrahydrofolate reductase gene (677C3 T) found in 앒10% of the population is associated with 1 From the Departments of Biochemistry (CW) and Medicine (MJK), Vanderbilt University School of Medicine, Nashville, TN, and the Veterans Affairs Medical Center, Nashville, TN (CW and MJK). 2 Supported by grant no. DK15289 from the National Institutes of Health (to CW) and by a Merit Revue Award from the Department of Veterans Affairs (to MJK). 3 Reprints not available. Address correspondence to Conrad Wagner, Department of Biochemistry, Vanderbilt University School of Medicine, Nashville, TN 37232. E-mail: [email protected]. Received March 23, 2007. Accepted for publication May 19, 2007.

Am J Clin Nutr 2007;86:1581–5. Printed in USA. © 2007 American Society for Nutrition

1581

1582

WAGNER AND KOURY

higher plasma homocysteine concentrations in persons with below-normal folate concentrations (9). Another meta-analysis of 72 studies investigated the effect of mutations in this gene on homocysteine concentrations and vascular disease (10). The authors of the meta-analysis concluded from these genetic studies as well as from prospective studies showing a highly significant association between plasma homocysteine and a variety of vascular diseases that this association was causal, at least with respect to CVD (10). There is also a dietary component. Many studies have provided evidence that, because folate, vitamin B-12, and vitamin B-6 are cofactors in the metabolic disposition of homocysteine, elevated plasma homocysteine is inversely correlated with plasma concentrations of these vitamins (11) (12). In the absence of vitamin B-12 deficiency, elevated plasma homocysteine concentrations are most responsive to folate supplementation and can be returned to normal (13) by that treatment.

WHAT IS THE MECHANISM OF THE VASCULAR DAMAGE?

There is no generally accepted mechanism for the pathophysiology of elevated plasma homocysteine as a cause of vascular disease. Various mechanisms for the toxic action of homocysteine include a change in the redox status of the tissues with production of reactive oxygen species (14); an inhibition of anticoagulation mechanisms mediated by the vascular endothelium (15); antiplatelet effects related to the reaction of elevated homocysteine with nitric oxide to form S-nitrosohomocysteine (16, 17); a direct effect of homocysteine on vascular endothelial (18) or smooth muscle cells (19); the formation of homocysteine thiolactone that modifies endothelial proteins (20); and the induction of programmed death of endothelial cells (21). In most cases, these actions have been indicated by effects caused by the addition of homocysteine to cells in culture. The principal problem with most of these studies has been the use of concentrations of homocysteine far higher (50 –1000 ␮mol/L) than those present in plasma to show these effects. Rarely have any effects been shown with concentrations of homocysteine as low as 10 ␮mol/L. Homocysteine has a free sulfhydryl group and is oxidized with a second homocysteine molecule to form the disulfide, homocystine, and also with cysteine to form a mixed disulfide. In human plasma, most homocysteine exists in disulfide linkage to cysteine in albumin. For this reason, it has been the standard practice to measure total homocysteine (tHcy) that is produced after the reduction of the bound homocysteine. The normal concentration of tHcy in human males is 앒10 ␮mol/L. However, as was pointed out by Jacobsen (22), the amount of free homocysteine in human plasma is 쏝1% (쏝 0.1 ␮mol/L). Therefore, although high plasma concentrations of homocysteine are associated with vascular disease, it has been difficult to show that they are the proximal cause of the damage.

COULD S-ADENOSYLHOMOCYSTEINE BE A CAUSE OF THE VASCULAR DAMAGE?

An alternative possible cause of the pathophysiology associated with hyperhomocysteinemia is AdoHcy. This compound is the precursor of all of the homocysteine in tissues. Except for methyl transfer from betaine and from methylcobalamin in the

methionine synthase reaction, AdoHcy is the product of all methylation reactions that involve S-adenosylmethionine (AdoMet) as the methyl donor. There are 앒50 reactions that carry out methyl transfer in cells. AdoHcy is well known as a potent inhibitor of most, if not all, methyltransferases (23). Increased concentrations of AdoHcy in tissues are usually accompanied by decreased concentrations of AdoMet. The use of the ratio of AdoMet to AdoHcy as an indicator of the methylating capacity of the cell was first suggested by Cantoni et al (24), and this ratio has been referred to as the “methylation index” (25). However, in certain situations, the elevation of AdoHcy appears to be a better indication of the inhibition of methylation than does the ratio of AdoMet to AdoHcy (26, 27). Methylation is significant in epigenetic regulation of protein expression via DNA and histone methylation. The inhibition of these AdoMet-mediated processes by AdoHcy is a proven mechanism for metabolic alteration. Because the conversion of AdoHcy to homocysteine is reversible, with the equilibrium favoring the formation of AdoHcy, increases in plasma homocysteine are accompanied by an elevation of AdoHcy in most cases. Measurement of plasma AdoHcy has not been carried out in most studies, mostly because the concentration of AdoHcy in plasma is 앒1-500th that of plasma tHcy, and complicated methods are needed to measure AdoHcy. Most of these methods are cumbersome and timeconsuming, or they involve specialized equipment (28, 29). IS THERE ANY ADVANTAGE TO THE MEASUREMENT OF S-ADENOSYLHOMOCYSTEINE RATHER THAN HOMOCYSTEINT?

Relatively few studies have directly compared plasma homocysteine and plasma AdoHcy as indicators of vascular disease, probably because of the complex methods involved in the measurement of plasma AdoHcy in large studies. Several small studies have shown that measurement of plasma AdoHcy is a better indicator of the risk of vascular disease than is measurement of plasma tHcy. Loehrer et al (30) first showed that, when compared with control subjects, patients with end-stage renal disease had 44-fold greater plasma AdoHcy but only 5-fold greater plasma homocysteine concentrations. Both measurements were significantly (P 쏝 0.001) different, but the authors drew no conclusions about which measurement was more sensitive. In a study published in 2001 comparing patients with proven CVD and matched controls, there was a significant difference in the plasma AdoHcy concentrations between the patients and controls but no significant difference in the homocysteine concentrations (31). This inability to discriminate between patients with CVD and controls by using homocysteine concentrations was probably due to the small numbers of patients (n ҃ 30) and control subjects (n ҃ 29) in the study. This insensitivity illustrates one of the major problems of using plasma homocysteine as an indicator of the risk of vascular disease. Because there is a large overlap in values between patients and control subjects, large numbers of subjects are needed to show a relation between high homocysteine concentrations and vascular disease. This makes it impossible to predict that any one person with moderately elevated plasma homocysteine is at greater risk than any other person with the same plasma homocysteine concentration. An association of high homocysteine concentrations with low renal function has been noted many times. There is evidence for significant metabolism of methionine by the kidney (32). Plasma homocysteine is

PLASMA S-ADENOSYLHOMOCYSTEINE VS HOMOCYSTEINE

highly elevated in patients with renal disease—to concentrations that are generally higher than those in patients with CVD. Contrary to its effect in other patients with hyperhomocysteinemia, supplementation with folic acid in those with renal disease lowers, but does not normalize, plasma homocysteine (33, 34). A study comparing adult renal disease patients with control subjects showed that plasma AdoHcy was a significantly more sensitive test of renal insufficiency than was homocysteine (35). In the studies comparing plasma homocysteine and plasma AdoHcy in patients who were selected because they had renal disease (35) and in patients who were selected because they had CVD (31), the values for plasma AdoHcy in both patients and control subjects overlapped much less than those for homocysteine. Two studies have noted that both plasma AdoHcy and homocysteine are elevated in patients with kidney disease (30, 35). Adults with kidney disease generally are older and have other diseases that are known be associated with elevated plasma homocysteine (eg, hypertension, diabetes, and CVD), which makes it difficult to determine the primary reason for the elevated homocysteine. To determine whether decreased renal function was the reason for the elevated homocysteine, a group of children who had only renal disease were studied (36). In that group of patients, there was no statistical correlation between glomerular filtration rate and plasma tHcy, but there was a strong correlation with plasma AdoHcy. The study suggested that a reduction in the ability to metabolize or excrete AdoHcy (or both) is a primary event in renal disease. This change is probably a function of the fact that AdoHcy is readily excreted in the urine (37), whereas homocysteine is not (38). A significant correlation between elevated plasma homocysteine and serum creatinine has been noted in many previous studies of the association of homocysteine with CVD (39), which raises the question of whether decreased renal function and kidney disease due to the involvement of renal vessels with the vascular component of CVD may have been the underlying reason for the elevated plasma homocysteine in some of those earlier studies. Although it may be expected that plasma AdoHcy and homocysteine values would tend to change in the same way, that is not always the case, as shown above for the children with renal disease only (36). In a particularly revealing study, Becker et al (40) showed that, in contrast to the plasma homocysteine concentration, the plasma AdoHcy concentration was not associated with folate concentrations. As pointed out by Becker et al, if AdoHcy is the actual toxic agent rather than homocysteine, then the use of folic acid supplementation to reduce plasma homocysteine concentrations will do nothing to reduce the incidence of vascular disease. This possibility is noteworthy in view of recent epidemiologic studies showing that folate supplementation did not reduce the risk of vascular disease, although plasma homocysteine was reduced (41, 42). It should be noted, however, that these studies were secondary prevention trials and that any effect of folate in reducing risk may have taken place at a time before supplementation was begun. Whether reduction of plasma homocysteine concentrations by B vitamin supplementation can reduce the incidence of vascular disease is under investigation in current clinical trials (41, 42). In a review of several large clinical trials, Clarke et al (43) carried out a meta-analysis of 4 trials that have been completed. It was concluded that there were no beneficial effects of B vitamin supplementation on either coronary

1583

heart disease or stroke. An additional 8 large studies are underway; together, these trials may have the statistical power to answer this question (43). MEASUREMENT OF PLASMA HOMOCYSTEINE AND S-ADENOSYLHOMOCYSTEINE

Many methods for the measurement of homocysteine in human plasma have been published. Because homocysteine contains a free thiol group and because it can form disulfide linkages with another molecule of homocysteine, with free cysteine, or with cysteine residues in proteins, only a small amount of the tHcy in plasma is free. Jacobsen has estimated that 쏝1% is the free thiol (22). For this reason, plasma or serum must first be treated with a reducing agent to obtain the total amount of homocysteine. The normal concentration of tHcy is 앒10 ␮mol/L, and analytic methods usually involve a reduction step that is followed first by derivatization to a form more easily detected and then by separation with the use of HPLC (39). An immunoassay was developed to detect the homocysteine in plasma after reduction (44). Measurement of AdoHcy in plasma presents a greater challenge, because its concentrations are 앒1/500th of those of homocysteine in normal plasma—앒20 nmol/L. Indeed, the existence of AdoHcy in plasma was unexpected until Lohrer et al (45) devised the first sensitive method for its measurement. This method depended on the formation of the fluorescent 1,N6etheno derivative of the adenosine portion of AdoHcy and then on separation by HPLC. This method measured AdoMet as well as AdoHcy in plasma; the reaction took a long time, although the results for measurement of AdoHcy compared favorably with those of other methods. More recently, Castro et al (46) were able to shorten the derivatization time from 8 to 4 h, and they could detect as little as 2.5 nmol AdoHcy/L in plasma. Their method uses a single HPLC column but requires the use of 1.0 mL plasma, an amount that may be difficult to obtain from small children. A method for measuring AdoHcy (and AdoMet) by using a very sensitive reaction with naphthalene dicarboxaldehyde to produce a fluorescent derivative was developed, but it too was cumbersome, requiring 2 HPLC separations in addition to the derivatization step (47). Several other methods have used highly specialized equipment such as tandem mass spectrometry (29, 37, 48) and coulometric electrochemical detection (28) to obtain greater sensitivity. In hindsight, it would seem useful to have measured plasma AdoHcy as well as homocysteine and the relation of AdoHcy concentrations in response to folate and other B vitamins. The reason for not having done so is that the existing methods were not suitable for epidemiologic studies. Recently, a simple, rapid immunoassay developed for the measurement of AdoHcy in human plasma promises to be useful in such studies (49). No significant change was seen in concentrations of AdoHcy in plasma or serum samples that had been frozen at Ҁ80 °C and then thawed and kept for 2 h at 4 °C (37). We have seen no change in values for plasma AdoHcy kept for 욷4 y at Ҁ80 °C (C Wagner, unpublished data, 2007). With regard to the stability of AdoHcy in freshly drawn plasma, we have noted, when using a method that measures both AdoMet and AdoHcy (47), that, when fresh plasma is kept at room temperature, there is little or no loss of either AdoMet or AdoHcy for 5 h. However, when frozen plasma is thawed and kept at room temperature for 5 h, there is a rapid loss of AdoMet

1584

WAGNER AND KOURY

but a very slight increase in AdoHcy. The amount of change varied from subject to subject. We ascribe these findings to some sort of activation of an enzyme in plasma, because it can be prevented by the addition of HgCL2 (C Wagner, unpublished data, 2005). We do not know whether the changes described above are due to the conversion of AdoMet to AdoHcy by a methyltransferase that is activated in frozen plasma. We believe that such changes are unlikely if the samples are kept on ice while being thawed and before analysis. The results described by Capdevila et al (49) for normal values obtained by this immunoassay are comparable to normal values published for several other, more complicated methods. If S-adenosylhomocysteine hydrolase is present and active in human plasma, there is a possibility that free AdoHcy in plasma could react with plasma adenosine to change AdoHcy concentrations; however, we are unaware of any reports of such activity in human plasma. Elevated homocysteine has been implicated as a risk factor in numerous neurologic disorders (50 –52), and a recent study showed that concentrations of homocysteine, AdoHcy, and AdoMet in plasma and cerebrospinal fluid are correlated (53). It would seem useful to determine whether cerebrospinal fluid measurements of AdoHcy are more informative than those of homocysteine in neurologic disorders such as dementia and Alzheimer disease. The authors’ responsibilities were as follows—CW and MJK: contributed equally to the writing of this article. Neither author had a personal or financial conflict of interest.

REFERENCES 1. Cantoni GL, Chiang PK. The role of S-adenosylhomocysteine hydrolase in the control of biological methylations. In: Cavallini D, Gaull GE, Zappia V, eds. Natural sulfur compounds: novel biochemical and structural aspects. New York, NY: Plenum Press, 1980:67– 80. 2. Finkelstein JD. Pathways and regulation of homocysteine metabolism in mammals. Semin Thromb Haemost 2000;26:219 –25. 3. McCully KS. Vascular pathology of homocysteinemia: implications for the pathogenesis of arteriosclerosis. Am J Pathol 1969;56:111–28. 4. Wilcken DE, Wilcken B. The pathogenesis of coronary artery disease. A possible role for methionine metabolism. J Clin Invest 1976;57:1079 – 82. 5. Clarke R, Daly L, Robinson K, et al. Hyperhomocysteinemia: an independent risk factor for vascular disease. N Engl J Med 1991;324:1149 – 55. 6. Brattstrom LE, Hardebo JE, Hultberg BL. Moderate homocysteinemia—a possible risk factor for arteriosclerotic cerebrovascular disease. Stroke 1984;15:1012– 6. 7. Boushey CJ, Beresford SA, Omenn GS, Motulsky AG. A quantitative assessment of plasma homocysteine as a risk factor for vascular disease. Probable benefits of increasing folic acid intakes. JAMA 1995;274: 1049 –57. 8. Homocysteine Studies Collaboration. Homocysteine and risk of ischemic heart disease and stroke: a meta-analysis. JAMA 2002;288:2015– 22. 9. Jacques PF, Bostom AG, Williams RR, et al. Relation between folate status, a common mutation in methylenetetrahydrofolate reductase, and plasma homocysteine concentrations. Circulation 1996;93:7–9. 10. Wald DS, Law M, Morris JK. Homocysteine and cardiovascular disease: evidence on causality from a meta-analysis. BMJ 2002;325:1202– 8. 11. Ueland PM, Refsum H. Plasma homocysteine, a risk factor for vascular disease: plasma levels in health, disease, and drug therapy. J Lab Clin Med 1989;114:473–501. 12. Selhub J, Jacques PF, Bostom AG, et al. Relationship between plasma homocysteine, vitamin status and extracranial carotid-artery stenosis in the Framingham Study population. J Nutr 1996;126(suppl):1258S– 65S. 13. Jacques PF, Selhub J, Bostom AG, Wilson PW, Rosenberg IH. The effect of folic acid fortification on plasma folate and total homocysteine concentrations. N Engl J Med 1999;340:1449 –54.

14. Heinecke JW. Unique aspects of sulfur chemistry: homocysteine and lipid oxidation. In: Carmel R, Jacobsen DW, eds. Homocysteine in health and disease. Cambridge, United Kingdom: Cambridge University Press, 2001:32– 8. 15. Ling Q, Hajjar KA. Inhibition of endothelial cell thromboresistance by homocysteine. J Nutr 2000;130(suppl):373S– 6S. 16. Stamler JS, Osborne JA, Jaraki O, et al. Adverse vascular effects of homocysteine are modulated by endothelium-derived relaxing factor and related oxides of nitrogen. J Clin Invest 1993;91:308 –18. 17. Gow AJ, Cobb F, Stamler JS. Homocysteine, nitric oxide and nitrosothiols. In: Carmel R, Jacobsen DW, eds. Homocysteine in health and disease. Cambridge, United Kingdom: Cambridge University Press, 2001:39 – 45. 18. Wang H, Yoshizumi M, Lai K, et al. Inhibition of growth and p21ras methylation in vascular endothelial cells by homocysteine but not cysteine. J Biol Chem 1997;272:25380 –5. 19. Tsai JC, Perrella MA, Yoshizumi M, et al. Promotion of vascular smooth muscle cell growth by homocysteine: a link to atherosclerosis. Proc Natl Acad Sci U S A 1994;91:6369 –73. 20. Jakubowski H. Metabolism of homocysteine thiolactone in human cell cultures. Possible mechanism for pathological consequences of elevated homocysteine levels. J Biol Chem 1997;272:1935– 42. 21. Hossain GS, Van Thienen JV, Werstuck GH, et al. TDAG51 is induced by homocysteine, promotes detachment-mediated programmed cell death, and contributes to the development of atherosclerosis in hyperhomocysteinemia. J Biol Chem 2003;278:30317–27. 22. Jacobsen DW. Practical chemistry of homocysteine and other thios. In: Carmel R, Jacobsen DW, eds. Homocysteine in health and disease. Cambridge, United Kingdom: Cambridge University Press, 2001:9 –20. 23. Clarke S, Banfield K. S-adenosylmethionine-dependent methyltransferases. In: Carmel R, Jacobsen DW, eds. Homocysteine in health and disease. Cambridge, United Kingdm: Cambridge University Press, 2001:63–78. 24. Cantoni GL, Richards HH, Chiang PK. Inhibitors of S-adenosylhomocysteine hydrolase and their role in the regulation of biological methylation. In: Usdin E, Borchardt RT, Creveling CR, eds. Transmethylation. New York, NY: Elsevier North Holland, 1978:155– 64. 25. Hershfield MS, Kredich NM, Koller CA, et al. S-adenosylhomocysteine catabolism and basis for acquired resistance during treatment of T-cell acute lymphoblastic leukemia with 2'-deoxycoformycin alone and in combination with 9-beta-D-arabinofuranosyladenine. Cancer Res 1983; 43:3451– 8. 26. Caudill MA, Wang JC, Melnyk S, et al. Intracellular S-adenosylhomocysteine concentrations predict global DNA hypomethylation in tissues of methyl-deficient cystathionine beta-synthase heterozygous mice. J Nutr 2001;131:2811– 8. 27. Capdevila A, Decha-Umphai W, Song KH, Borchardt RT, Wagner C. Pancreatic exocrine secretion is blocked by inhibitors of methylation. Arch Biochem Biophys 1997;345:47–55. 28. Melnyk S, Pogribna M, Pogribny IP, Yi P, James SJ. Measurement of plasma and intracellular S-adenosylmethionine and S-adenosylhomocysteine utilizing coulometric electrochemical detection: alterations with plasma homocysteine and pyridoxal 5'-phosphate concentrations. Clin Chem 2000 46:265-72. 29. Struys EA, Jansen EE, de Meer K, Jakobs C. Determination of S-adenosylmethionine and S-adenosylhomocysteine in plasma and cerebrospinal fluid by stable-isotope dilution tandem mass spectrometry. Clin Chem 2000;46:1650 – 6. 30. Loehrer FM, Angst CP, Brunner FP, Haefeli WE, Fowler B. Evidence for disturbed S-adenosylmethionine:S-adenosylhomocysteine ratio in patients with end-stage renal failure: a cause for disturbed methylation reactions? Nephrol Dial Transplant 1998;13:656 – 61. 31. Kerins DM, Koury MJ, Capdevila A, Rana S, Wagner C. Plasma S-adenosylhomocysteine is a more sensitive indicator of cardiovascular disease than plasma homocysteine. Am J Clin Nutr 2001;74:723–9. 32. House JD, Brosnan ME, Brosnan JT. Characterization of homocysteine metabolism in the rat kidney. Biochem J 1997;328:287–92. 33. van Guldener C, Robinson K. Homocysteine and renal disease. Semin Thromb Hemost 2000;26:313–24. 34. De Vriese AS, Verbeke F, Schrijvers BF, Lameire NH. Is folate a promising agent in the prevention and treatment of cardiovascular disease in patients with renal failure? Kidney Int 2002;61:1199 –209. 35. Wagner C, Stone WJ, Koury MJ, Dupont WD, Kerins DM.

PLASMA S-ADENOSYLHOMOCYSTEINE VS HOMOCYSTEINE

36. 37.

38. 39.

40.

41. 42. 43. 44.

S-adenosylhomocysteine is a more sensitive indicator of renal insufficiency than homocysteine. Nutr Res 2004;24:487–94. Jabs K, Koury MJ, Dupont WD, Wagner C. Relationship between plasma S-adenosylhomocysteine concentration and glomerular filtration rate in children. Metabolism 2006;55:252–7. Stabler SP, Allen RH. Quantification of serum and urinary S-adenosylmethionine and S-adenosylhomocysteine by stable-isotopedilution liquid chromatography-mass spectrometry. Clin Chem 2004; 50:365–72. van Guldener C, Donker AJ, Jakobs C, Teerlink T, de Meer K, Stehouwer CD. No net renal extraction of homocysteine in fasting humans. Kidney Int 1998;54:166 –9. Ueland P, Refsum H, Brattstrom L. Plasma Homocysteine and cardiovascular disease. In: Francis RB, ed. Atherosclerotic cardiovascular disease, hemostasis, and endothelial function. New York, NY: Marcel Dekker Inc, 1992:183–235. Becker A, Smulders YM, Teerlink T, et al. S-adenosylhomocysteine and the ratio of S-adenosylmethionine to S-adenosylhomocysteine are not related to folate, cobalamin and vitamin B6 concentrations. Eur J Clin Invest 2003;33:17–25. Lonn E, Yusuf S, Arnold MJ, et al. Homocysteine lowering with folic acid and B vitamins in vascular disease. N Engl J Med 2006;354:1567– 77. Bonaa KH, Njolstad I, Ueland PM, et al. Homocysteine lowering and cardiovascular events after acute myocardial infarction. N Engl J Med 2006;354:1578 – 88. Clarke R, Lewington S, Sherliker P, Armitage J. Effects of B-vitamins on plasma homocysteine concentrations and on risk of cardiovascular disease and dementia. Curr Opin Clin Nutr Metab Care 2007;10:32–9. Frantzen F, Faaren AL, Alfheim I, Nordhei AK. Enzyme conversion immunoassay for determining total homocysteine in plasma or serum. Clin Chem 1998;44:311– 6.

1585

45. Loehrer FM, Haefeli WE, Angst CP, Browne G, Frick G, Fowler B. Effect of methionine loading on 5-methyltetrahydrofolate, S-adenosylmethionine and S-adenosylhomocysteine in plasma of healthy humans. Clin Sci 1996;91:79 – 86. 46. Castro R, Rivera I, Struys EA, et al. Increased homocysteine and S-adenosylhomocysteine concentrations and DNA hypomethylation in vascular disease. Clin Chem 2003;49:1292– 6. 47. Capdevila A, Wagner C. Measurement of plasma S-adenosylmethionine and S-adenosylhomocysteine as their fluorescent isoindoles. Anal Biochem 1998;264:180 – 4. 48. Gellekink H, van Oppenraaij-Emmerzaal D, van Rooij A, Struys EA, den Heijer M, Blom HJ. Stable-isotope dilution liquid chromatographyelectrospray injection tandem mass spectrometry method for fast, selective measurement of S-adenosylmethionine and S-adenosylhomocysteine in plasma. Clin Chem 2005;51:1487–92. 49. Capdevila A, Burk RF, Freedman J, Wagner C. A simple rapid immunoassay for S-adenosylhomocysteine in plasma. J Nutr Biochem 2007 May 15 (Epub ahead of print). 50. Bottiglieri T, Hyland K. S-adenosylmethionine levels in psychiatric and neurological disorders: a review. Acta Neurol Scand. Suppl 1994;154: 19 –26. 51. Seshadri S, Beiser A, Selhub J, et al. Plasma homocysteine as a risk factor for dementia and Alzheimer’s disease. N Engl J Med 2002;346:476 – 83. 52. McIlroy SP, Dynan KB, Lawson JT, Patterson CC, Passmore AP. Moderately elevated plasma homocysteine, methylenetetrahydrofolate reductase genotype, and risk for stroke, vascular dementia, and Alzheimer disease in Northern Ireland. Stroke 2002;33:2351– 6. 53. Obeid R, Kostopoulos P, Knapp JP, et al. Biomarkers of folate and vitamin B12 are related in blood and cerebrospinal fluid. Clin Chem 2007;53:326 –33.

Original Research Communications

See corresponding editorial on page 1577.

No differences in satiety or energy intake after high-fructose corn syrup, sucrose, or milk preloads1–3 Stijn Soenen and Margriet S Westerterp-Plantenga ABSTRACT Background: It is unclear whether energy-containing drinks, especially those sweetened with high-fructose corn syrup (HFCS), promote positive energy balance and thereby play a role in the development of obesity. Objective: The objective was to examine the satiating effects of HFCS and sucrose in comparison with milk and a diet drink. Design: The effects of 4800-mL drinks containing no energy or 1.5 MJ from sucrose, HFCS, or milk on satiety were assessed, first in 15 men and 15 women with a mean (앐SD) body mass index (BMI; in kg/m2) of 22.1 앐 1.9 according to visual analogue scales (VAS) and blood variables and second in 20 men and 20 women (BMI: 22.4 앐 2.1) according to ingestion of a standardized ad libitum meal (granola cereal ѿ yogurt, 10.1 kJ/g). Results: Fifty minutes after consumption of the 1.5-MJ preload drinks containing sucrose, HFCS, or milk, 170%-mm VAS changes in satiety were observed. Glucagon-like peptide 1 (GLP-1) (P 쏝 0.001) and ghrelin (P 쏝 0.05) concentrations changed accordingly. Compensatory energy intake did not differ significantly between the 3 preloads and ranged from 30% to 45%. Energy intake compensations were related to satiety (r ҃ 0.35, P 쏝 0.05). No differences were observed between the effects of the sucrose- and HFCScontaining drinks on changes in VAS and on insulin, glucose, GLP-1, and ghrelin concentrations. Changes in appetite VAS ratings were a function of changes in GLP-1, ghrelin, insulin, and glucose concentrations. Conclusion: Energy balance consequences of HFCS-sweetened soft drinks are not different from those of other isoenergetic drinks, eg, a sucrose-drink or milk. Am J Clin Nutr 2007;86:1586 –94. KEY WORDS Glucagon-like peptide 1, ghrelin, insulin, glucose, energy intake

decades in the United States (8 –10). In the 1970s, the food industry in the United States introduced high-fructose corn syrup (HFCS) sweetener as a substitute for sucrose (11). It has been suggested that the obesity epidemic may have been aggravated by the increase in HFCS consumption (12). Drinking HFCS-sweetened soda was reported to increase energy intake and body weight (13). However, several studies have reported that fructose, when consumed alone, reduced subsequent energy intake equally in some (14 –16) or significantly more in other studies (17–19) compared with a monosaccharide glucose preload. Yet, it should be noted that the principal sweetener in soft drinks in the United States, HFCS, is not pure fructose but a mixture of fructose (55%) and glucose (45%). Factors that may account for the different effects of fructose alone or a mix of fructose and glucose are its gastrointestinal effects and absorption characteristics (20, 21). In addition to the composition of ingested carbohydrates, the physical state of intake may be important in influencing subsequent energy intake compensation. Compensatory dietary responses to energy-containing beverages have been found to be less precise than those to isoenergetic solid loads (22, 23). Thus, fluid carbohydrates such as soft drinks could increase the risk of excess total energy intake. An effect of soft drink consumption, eg, of sucrose compared with artificial sweeteners, on weight gain and obesity has been found in children (24 –26), adolescents (27), and adults (28, 29). On the basis of these studies, it is suggested that carbohydrates in liquid form promote a positive energy balance and therefore contribute to the development of obesity. Compensation for energy intake from drinks by a change in energy intake at the subsequent meal depends on the moment in 1

INTRODUCTION

Trends in overweight are consistent with increased energy intake over recent decades (1). The upward shift in energy intake may partly consist of the consumption of soft drinks (2–5). Increased soft drink consumption has coincided with the increase in prevalence of overweight and obesity (6, 7) over the past 3

1586

From the Department of Human Biology, Maastricht University, Maastricht, Netherlands. 2 Supported by Suikerstichting Nederland (Baarn, Netherlands). 3 Address reprint requests and correspondence to S Soenen, Department of Human Biology, Maastricht University, PO Box 616, 6200 MD Maastricht, Netherlands. E-mail: [email protected]. Received January 18, 2007. Accepted for publication June 1, 2007.

Am J Clin Nutr 2007;86:1586 –94. Printed in USA. © 2007 American Society for Nutrition

1587

SOFT DRINK INTAKE, SATIETY, AND COMPENSATION TABLE 1 Subject characteristics Study 1

Age (y) Weight (kg) Height (cm) BMI (kg/m2) Systolic blood pressure (mm Hg) Diastolic blood pressure (mm Hg) F1, cognitive restraint3 F2, disinhibition4 F3, hunger5 Insulin (mU/L)6 Glucose (mmol/L)6

Study 2

Women (n ҃ 15)

Men (n ҃ 15)

Women (n ҃ 20)

Men (n ҃ 20)

21.1 앐 1.5 (7)2 63.7 앐 7.3 (12) 171.4 앐 5.6 (3) 21.6 앐 1.9 (9) 123 앐 14 (11) 73 앐 9 (12) 5.1 앐 2.9 (57) 4.9 앐 1.9 (39) 4.5 앐 2.9 (64) 12.9 앐 4.1 (32) 4.9 앐 0.3 (6)

21.5 앐 1.8 (8) 75.8 앐 9.5 (13) 183.3 앐 8.0 (4) 22.5 앐 1.8 (8) 131 앐 11 (8) 78 앐 10 (13) 3.3 앐 2.2 (67) 3.1 앐 1.1 (35) 3.3 앐 1.8 (55) 12.7 앐 3.8 (30) 5.3 앐 0.4 (8)

21.2 앐 2.2 (10) 65.0 앐 7.7 (12) 171.6 앐 4.6 (3) 22.0 앐 2.1 (10) 123 앐 11 (9) 74 앐 8 (11) 5.5 앐 3.0 (55) 5.0 앐 2.0 (40) 4.4 앐 2.9 (66)

22.3 앐 4.5 (20) 76.2 앐 6.0 (8) 183.0 앐 7.2 (4) 22.8 앐 2.0 (9) 130 앐 10 (8) 77 앐 7 (9) 3.3 앐 2.1 (64) 4.0 앐 2.0 (50) 5.1 앐 3.4 (67)

P (ANOVA)1 쏝0.001 쏝0.001 쏝0.05 쏝0.005 쏝0.01 쏝0.001

Represents differences between men and women; all subjects participated in either study 1 or 2 (n ҃ 57; 13 subjects participated in both studies). x៮ 앐 SD; CV in parentheses (all such values). 3 A measure of cognitive restraint with the Three-Factor Eating Questionnaire (TFEQ); minimum score ҃ 0, maximum score ҃ 21; cutoff point for the Dutch population was 9. Values 쏜9 indicate cognitive restraint eating. 4 A measure of disinhibition or emotional eating with the TFEQ; minimum score ҃ 0, maximum score ҃ 14. 5 A general feeling of hunger with the TFEQ; minimum score ҃ 0, maximum score ҃ 14. 6 Average plasma concentrations over the 4 test days after the subjects fasted overnight. 1 2

time of preload ingestion. Time delay between preload and test meal interferes with the outcome of preload studies (30 –32). The objective of the present study was to examine whether there is a difference in response between a HFCS-sweetened and a sucrose-sweetened isoenergetic, isovolumetric orangeflavored preload and a no-energy control. A milk preload was used to compare the soft drinks with another type of liquid preload. In the first study, the responses were measured as the appetite profile using visual analogue scales (VAS) and as a possible change in the satiety hormones: glucagon-like peptide 1 (GLP-1), insulin, ghrelin, and glucose. Moreover, the latest time point after ingestion when relevant differences in satiety scores or satiety hormone concentrations were still present was determined as the moment in time for the subsequent test meal. In the second study, possible compensation in energy intake during an ad libitum subsequent meal was determined. The studies were conducted in Europe, so subjects had a negligible history of consuming HFCS-containing products. SUBJECTS AND METHODS

Subjects Subjects were recruited by means of an advertisement in local newspapers and on notice boards at Maastricht University. Subjects who were willing to participate in the study were subsequently screened by means of a detailed medical history and a physical examination. All subjects were in good health, were normotensive, were nonsmokers, were nonrestrained eaters, were regular breakfast consumers, were at most moderate alcohol users, had a stable body weight (a change of 쏝2 kg over at least the past 2 mo) and did not use prescription medication. Excluded from the study were athletes, defined as those who trained 쏜10 h/week. Thirty subjects (equal numbers of men and women) participated in the first study, 40 in the second study. Subject characteristics are given in Table 1. Subjects were re-

quested to maintain their customary level of physical activity and normal dietary habits and not to gain or lose weight for the duration of the study. All subjects gave written informed consent, and the experimental protocol was approved by the local Medical Ethics Committee of the University of Maastricht, Maastricht, Netherlands. Study design A within-subjects design was used, with each subject returning for 4 separate test days 욷1 wk apart. The preloads were offered blindly and in randomized order to avoid the order-oftreatment effect. To analyze possible differences in the appetite profile, VAS ratings and blood samples for the measurement of GLP-1, ghrelin, insulin, and glucose concentrations were collected before and after preload consumption in the first study. The last moment in time at which relevant differences in satiety were present was determined to decide on the timing of the test meal in the second study. The second study consisted of the same preload consumptions as in the first study, with VAS ratings of the appetite profile before and after the preload and a test meal at the relevant moment in time, as defined by the first study. Anthropometric measures Body weight was determined during screening and on each test day with a digital balance (weighing accuracy of 0.02 kg; ChyoMW-150K; Chyo, Japan) while the subjects were wearing underwear and in a fasted state and after they had emptied their bladders. Height was measured to the nearest 0.1 cm with a wall-mounted stadiometer (model 220; Seca, Hamburg, Germany). Body mass index (BMI) was calculated by dividing body weight (kg) by height squared (m2). Systolic and diastolic blood pressures were recorded during screening with an automatic blood pressure monitor (OSZ 5 easy; Spreidel & Keller GmBH and Co, KG, Jungingen, Germany).

1588

SOENEN ET AL

TABLE 2 Energy and macronutrient composition of the 4 preloads and the meal1 Preload

Carbohydrate [kJ (%)] Glucose [kJ (%)] Fructose [kJ (%)] Lactose [kJ (%)] Protein [kJ (%)] Fat [kJ (%)] Energy (kJ) Volume (mL) Energy density (kJ/g)

Sucrose-containing preload2

HFCS-containing preload3

Milk preload

1500 960 (64) 540 (36) 0 0 0 1500 800 1.9

1500 615 (41) 885 (59) 0 0 0 1500 800 1.9

632 0 0 632 (42) 442 (30) 426 (28) 1500 800 1.9

Diet preload

Meal4

0 0 0

554 (55)

2 0 2 800 0

80 (8) 378 (37) 1012 10.1

1

HFCS, high-fructose corn syrup. Values reported as percentages represent the percentage of energy of the energy-containing macronutrients. 66% sucrose and 34% glucose syrup (91% glucose and 9% fructose). 3 55% fructose and 45% glucose syrup (91% glucose and 9% fructose). 4 The test meal consisted of a granola cereal and yogurt; values are expressed per 100 g. 2

Preloads The 4 beverages were as follows: a beverage containing sucrose, one containing HFCS, one containing milk, and a diet drink. The energy content and macronutrient composition of the 4 beverages are specified in Table 2. All 4 drinks were isovolumetric and had a volume of 800 mL. The energy drinks were isoenergetic and provided 1.5 MJ. The diet drink had an energy content of 0.2 MJ. The drinks containing sucrose or HFCS and the diet drink were orangeflavored custom-made beverages and were equally sweet. The sucrose-containing preload had the same consistency as a commercially available sucrose-sweetened drink containing 450 g sucrose and 236 g glucose syrup (91% glucose and 9% fructose). The HFCScontaining preload had the consistency of a commercially available HFCS-sweetened drink containing 55% fructose and 45% glucose syrup (91% glucose and 9% fructose). The diet preload consisted of the sweeteners aspartame, acesulfame-K, and sodium cyclamate. Additionally, all 3 preloads contained water, citric acid, orange flavoring, coloring E160, preservative E202, and antioxidant E300. Drinks were prepared by diluting 133 mL syrup with 667 mL water. All 4 beverages were served chilled at 8 °C. Test meal The test meal that was served in the second study consisted of a granola cereal with yogurt. The nutrient composition of the test meal is shown in Table 2. Subjects were requested to continue eating until they felt comfortably full. All foods were preweighed at the time of serving, and plate waste was collected and weighed.

in study 1 and at 7 time points in study 2. The scale ranged from “not at all” on the left to “extremely” on the right. Subjects were instructed to mark, with a single vertical line, a point where the length of the line matched their subjective sensation. All VASs were provided on a separate form at each time point and were collected immediately after they had been completed. Taste perception and hedonics Subjects rated their taste perception and hedonics for the 4 test drinks on anchored 100-mm VAS during screening and at the first and last sip of the beverage consumed during each test day (Table 3). The following scales had to be completed: how sweet, sour, bitter, or salty the drink was; how rich, creamy, and fresh the flavor of the drink was; and how pleasant the drink was in the mouth. Blood samples Venous blood samples were taken at 5 time points: one fasting sample at baseline before and 4 samples 15, 30, 60, and 120 min after preload consumption. After each blood collection, the intravenous cannula was rinsed with 0.9% sterile sodium chloride solution containing 1% heparin. Blood samples were taken to determine concentrations of plasma GLP-1, ghrelin, insulin, and glucose. The TABLE 3 Perception of taste characteristics1

Attitude toward eating

Sucrosecontaining preload

HFCScontaining preload

Milk preload

Diet preload

66 앐 14a 20 앐 20a,b 12 앐 12a 8 앐 10a 44 앐 23a 13 앐 16a 67 앐 14a 70 앐 14a 51 앐 19a

70 앐 17a 29 앐 23a,c 15 앐 17a,c 8 앐 12a 50 앐 24a 14 앐 14a 66 앐 14a 68 앐 15a 59 앐 21a

25 앐 22b 13 앐 16b 12 앐 17a,b 10 앐 15a 68 앐 16b 77 앐 13b 41 앐 20b 52 앐 21b 50 앐 26a

52 앐 22c 37 앐 21c 23 앐 21c 14 앐 17a 41 앐 25a 13 앐 15a 61 앐 19a 50 앐 23b 48 앐 22a

Appetite profile

Sweetness Sourness Bitterness Saltiness Richness Creaminess Refreshing Pleasantness Intenseness

The subjects’ feelings of hunger, satiety, fullness, prospective food and drink consumption, and desire to eat and drink were scored on anchored 100-mm VAS at 6 different 0.5-h time points

All values are x៮ 앐 SD. HFCS, high-fructose corn syrup. Means in a row with different superscript letters are significantly different, P 쏝 0.05 (ANOVA).

The subjects’ attitude toward eating was determined during screening with the use of a validated Dutch translation of the Three-Factor Eating Questionnaire (TFEQ) (33, 34). The scores on cognitive restrained and unrestrained eating behavior (F1), emotional eating and disinhibition of control (F2), and subjective feeling of hunger (F3) are shown in Table 1.

1

SOFT DRINK INTAKE, SATIETY, AND COMPENSATION

blood samples were collected in tubes containing EDTA to prevent clotting. Blood samples for GLP-1 analysis were collected in icechilled syringes containing 20 ␮L dipeptidyl peptidase-IV (DPPIV) inhibitor (Linco Research Inc, St Charles, MO) to prevent degradation. Plasma was obtained by centrifugation (1500 ҂ g, 10 min, 4 °C), frozen in liquid nitrogen, and stored at Ҁ80 °C until analyzed. Plasma ghrelin samples were mixed with hydrochloric acid, methanol, and phenylmethanesulfonyl fluoride (Sigma-Aldrich, Zwijndrecht, Netherlands). Plasma concentrations of active ghrelin were measured by radioimmunoassay (Linco Research Inc) and those of active GLP-1 by enzyme-linked immunosorbent assay (EGLP35K; Linco Research Inc). Insulin samples were analyzed with a radioimmunoassay kit (Linco Research Inc), and glucose samples were measured by using a hexokinase method (ABX Diagnostics, Montpellier, France). Test day procedure After fasting overnight, the subjects arrived at the laboratory at 0815. The subjects were asked to consume their habitual evening meals, to refrain from alcohol or strenuous exercise, and to refrain from eating and drinking after 2300 on the day before each test. Body weight was measured, and an intravenous Venflon cannula (Baxter BV, Utrecht, Netherlands) was inserted in the antecubital vein to enable blood sampling (study 1). The subjects remained seated in comfortable chairs separated by large room dividers with minimal disturbance from the investigators throughout the experimental session. During each test day, the subjects were isolated from time cues to eliminate as much as possible habitual (time-determined) meal patterns; no watches, clocks, or radios were present in the test room, and the research refrained from making time-related statements. The subjects were allowed to stretch their legs, use the bathroom, read, listen to music, or watch movies, but not while drinking the preload or eating the meal (study 2). At 0900, after collection of the baseline appetite profile and blood sample, the subjects received 1 of the 4 liquid preloads. The preloads had to be consumed entirely within 10 min. The preloads were accompanied by a VAS of taste perception and hedonics at the first and last sips of the beverage. Blood sampling in study 1 was repeated 15, 30, 60, and 120 min after preload consumption and the appetite profile 20, 50, 80, 110, and 140 (last time point only in study 2) min after preload consumption. The catheter was removed after the last blood sample had been taken. The meal in study 2 was served 50 min after preload consumption based on the VAS ratings or differences in increases of satiety hormones in the first study. Statistical analysis Data are presented as means 앐 SDs or SEs. VAS ratings were measured in millimeters from the left end of the scale. The changes in concentrations of the hormones from baseline and changes from baseline in VAS ratings of the appetite profile were compared by analysis of variance (ANOVA), repeated-measures ANOVA (analysis of change score), and analysis of covariance (ANCOVA) with the baseline values as covariates. Because the experiment was fully randomized with a 1 wk washout between the tests, because there was no significant difference between the baseline scores, and because the washout period was longer than the actual experiment, it is more appropriate to use the analysis of change score from baseline with an n-factor repeated-measures ANOVA instead of ANCOVA (35). An ANCOVA may give bias

1589

because of the “weight” of the baseline values (35). Post hoc analysis was carried out with a Fisher’s protected leastsignificant difference test, Sheffe’s F test, or a Tukey’s test. Taste perception and energy intake after the preloads were compared by ANOVA. Differences in responses between the drinks containing sucrose and HFCS were compared with a 2-tailed paired Student’s t test. Sex differences were assessed by using ANOVA. Time-by-sex interactions were assessed by using repeatedmeasures ANOVA, and time-by-treatment-by sex interactions were assessed by using multivariate ANOVA with preload condition and sex as fixed factors. Changes in the desire to eat from baseline were analyzed as a function of changes in concentrations of hormones and glucose from baseline by regression analysis. Compensation was calculated as the difference between energy intake after the diet preload and energy intake after any of the energy preloads as a percentage of the energy content of these preloads. Overconsumption was calculated as a difference between total energy intake after any of the energy preloads and total energy intake after the diet preload as a percentage of energy intake after the diet preload. All analyses were performed with the Statistical Package for the Social Sciences (SPSS) version 11.0.3 for Macintosh OS X (SPSS Inc, Chicago, IL). Differences were regarded as significant if P 쏝 0.05.

RESULTS

Perception of taste characteristics Drinks containing sucrose or HFCS (800 mL, 1.5 MJ) did not differ in taste perception or palatability. The milk preload (800 mL, 1.5 MJ) was perceived as less sweet, sour, refreshing, and pleasant (P 쏝 0.01) and more rich and creamy than the preloads containing sucrose or HFCS (P 쏝 0.005). The diet preload (800 mL, 2 kJ) was perceived as less pleasant and less sweet than preloads containing sucrose or HFCS (P 쏝 0.001) (Table 3). Taste perception did not differ between sexes. Perceptions of thirst after the preloads did not differ between the preloads. Thirst was significantly more reduced in women than in men [change in area under the curve (AUC) from baseline: Ҁ18 앐 9 compared with Ҁ31 앐 16 mm VAS/min respectively; P 쏝 0.05]. Determination of the moment in time to serve the test meal in study 2 In study 1 we determined the moment in time to serve the meal in study 2. The right moment was determined by identifying the moment in time when the mean difference in responses to the preloads containing sucrose or HFCS was statistically significant. This moment appeared to be 50 min after the preload consumption. This moment in time was underscored by the following. Although preloads containing sucrose or HFCS did not differ in satiety and hunger ratings in the total group (Figure 1), the reduction in hunger relative to baseline after a preload differed significantly between men and women (P 쏝 0.05). Men had a significantly greater reduction in hunger after the preload containing HFCS than after the preload containing sucrose at the 50-min time point (Ҁ8 앐 14 compared with Ҁ17 앐 15 mm VAS, respectively; P 쏝 0.05), whereas women showed the opposite. Women had a significantly greater reduction in hunger ratings at the 50-, 80-, and 110-min time points, with the maximal difference occurring 50 min (Ҁ24 앐 18 compared with Ҁ7 앐 19 mm VAS; P 쏝 0.05) after consumption of

1590

SOENEN ET AL

FIGURE 1. Mean (앐SEM) change (⌬) in hunger of the total group (n ҃ 30) and in men (n ҃ 15) and women (n ҃ 15) separately as a function of preload condition (study 1). There was a significant time-by-treatment-bysex interaction, P 쏝 0.05 (multivariate ANOVA). ‡Significant difference between the diet preload and the sucrose- and high-fructose corn syrup (HFCS)– containing preloads and the milk preload, P 쏝 0.05 (repeatedmeasures ANOVA). †Significant difference between the diet preload and the sucrose- and HFCS-containing preloads, P 쏝 0.005 (ANOVA). *Significant difference between the sucrose- and HFCS-containing preloads, P 쏝 0.05 (2-tailed paired Student’s t test).

the preload containing sucrose compared with the preload containing HFCS. Thus, the adequate moment in time to serve the test meal in study 2 was 50 min, as underscored by the significant treatmentby-sex interaction at 50 min (P 쏝 0.05). Differences in VAS ratings between treatments differed by sex. This moment in time was not supported by differences in concentrations in GLP-1, ghrelin, insulin, or glucose relative to baseline, as illustrated in Figure 2. However, changes in VAS ratings relative to baseline were a function of changes in concentrations of the hormones GLP-1, ghrelin, and insulin relative to baseline values (Table 4). Stepwise multiple linear regression analysis of VAS appetite ratings showed that change in GLP-1 (r ҃ Ҁ0.242, P ҃ 0.014) and insulin (r ҃ Ҁ0.239, P ҃ 0.029) independently predicted changes in satiety. Moreover, glucose and insulin concentrations were related after preload consumption, as expected, and GLP-1 and ghrelin concentrations were related to insulin concentrations. GLP-1 and ghrelin concentrations were not related to each other (Table 4). Furthermore, the determination of the adequate moment in time to serve the meal in study 2 was underscored by the decrease in glucose concentrations (Figure 2). Energy-containing preloads compared with the diet preload Meal size and energy intake were significantly lower after consumption of preloads containing sucrose or HFCS or the milk preload than after the diet preload (Table 5). This finding was supported by the significantly higher GLP-1 and insulin concentrations (Figure 2; P 쏝 0.001) and the significantly lower ghrelin concentrations (Figure 2; P 쏝 0.05) and hunger (Figure 1; P 쏝 0.05) after the energy-containing preloads than after the diet preload. Thus, less energy was consumed after consumption of an energy drink than after a drink designed to not deliver energy. Total energy intake (preload ѿ meal) with the energy-containing preloads was significantly higher than total energy intake with the diet preload (Table 5). Therefore, during the meal, energy

intake was only partly compensated for. Compensation for energy intake from the preloads containing sucrose, HFCS, or milk did not differ significantly (Table 5) and ranged from 30% to 45%. Energy consumed after preloads, compensation, and overconsumption differed significantly between men and women (P 쏝 0.01). This sex difference was supported by the significant time-by-sex interactions for glucose and GLP-1 concentrations (P 쏝 0.01). Compared with women, men had lower GLP-1 concentrations at baseline (P 쏝 0.05) and a smaller change in GLP-1 concentration from baseline after preload consumption (P 쏝 0.01). Appetite ratings after drink consumption decreased significantly more in women than in men (P 쏝 0.05). Decreases in hunger scores were not different between the 4 conditions after ingestion of the meals. Compensation after the energy-containing preloads was a function of the magnitude of change in satiety scores from baseline (r ҃ 0.350, P ҃ 0.023). In the men, overconsumption after the preload containing sucrose (r ҃ Ҁ0.934, P ҃ 0.020) or milk (r ҃ Ҁ0.999, P 쏝 0.001) was a function of the magnitude of change in satiety scores from baseline; after the preload containing HFCS, this relation was not observed. Hunger ratings were significantly more suppressed at each time point after the milk preload than after the diet preload (P 쏝 0.05). The change from baseline in GLP-1 concentrations was significantly larger (P 쏝 0.05) 30 min after the milk preload (3.6 앐 3.4 pmol/L) than after the preloads containing sucrose (2.1 앐 2.3 pmol/L) or HFCS (2.1 앐 3.3 pmol/L). In men, this difference was observed at each time point (P 쏝 0.05). Furthermore, compensation and satiety (r ҃ 0.412, P 쏝 0.05) were positively related to change in pleasantness of taste after the preload containing sucrose (the greater the suppression in pleasantness of taste, the larger the satiety and compensation), as shown in Figure 3. Accordingly, plasma glucose concentrations were significantly higher over time after the drinks containing sucrose or HFCS than after the milk or diet preloads (P 쏝 0.001). Moreover, plasma glucose concentrations were linearly related to the content of glucose of the preloads (r ҃ 0.581, P 쏝 0.001).

DISCUSSION

Do the satiation effects of isocaloric isovolumetric sucrose- or HFCS-containing preloads differ from those of milk as measured on the basis of VAS (in mm) or GLP-1 or ghrelin responses? The increase in satiety from baseline as AUC did not differ significantly between the sucrose, HFCS, or milk preload. Furthermore, satiety was expressed as compensation or overconsumption during the next meal; no significant differences between the different preloads were observed. From these observations we concluded that there are no differences in the satiety or energy balance effects of isovolumetric sucrose- or HFCS-containing preloads or milk. Subsequently the mechanisms underscoring the increases in satiety were revealed. Although no differences in satiety were observed, the mechanisms underlying satiety due to sucrose- or HFCS-containing drinks or milk were different and were related to evoking different increases in satiety hormone concentrations. No significant differences in energy intakes or in total energy consumed were observed 50 min after consumption of the 1.5-MJ (800 mL) drinks containing sucrose or HFCS. Also, energy intake after the isoenergetic isovolumetric milk preload did not differ from that after the sucrose or HFCS drinks. Similarly

SOFT DRINK INTAKE, SATIETY, AND COMPENSATION

1591

FIGURE 2. Mean (앐SEM) change (⌬) in glucagon-like peptide 1 (GLP-1), ghrelin, insulin, and glucose of men (n ҃ 15) and women (n ҃ 15) as a function of preload condition. A: ‡Significant difference between the diet preload and the sucrose- and high-fructose corn syrup (HFCS)– containing preloads and the milk preload, P 쏝 0.005 (ANOVA). *Significant difference between the milk preload and the sucrose- and HFCS-containing preloads, P 쏝 0.05 (ANOVA); significant time-by-treatment interaction between the diet preload and the sucrose- and HFCS-containing preloads and the milk preload, P 쏝 0.001 (repeatedmeasures ANOVA); significant time-by-sex interaction, P 쏝 0.01 (repeated-measures ANOVA). B: †Significant difference between the diet preload and the sucrose- and HFCS-containing preloads, P 쏝 0.05 (ANOVA); significant time-by-treatment interaction between the diet preload and the sucrose- and HFCS-containing preloads, P 쏝 0.05 (repeated-measures ANOVA); the time-by-sex interaction was not significant (repeated-measures ANOVA). C: ‡ Significant difference between the diet preload and the sucrose- and HFCS-containing preloads and the milk preload, P 쏝 0.001 (ANOVA). *Significant difference between the milk preload and the sucrose- and HFCS-containing preloads, P 쏝 0.05 (ANOVA); significant time-by-treatment interaction between the diet preload and the sucrose- and HFCS-containing preloads and the milk preload, P 쏝 0.001 (repeated-measures ANOVA); the time-by-sex interaction was not significant (repeated-measures ANOVA). D: ‡Significant difference between the diet and the milk preloads, P 쏝 0.05 (ANOVA). *Significant difference between the milk preload and the sucrose- and HFCS-containing preloads, P 쏝 0.001 (ANOVA). †Significant difference between the diet preload and the sucrose- and HFCS-containing preloads, P 쏝 0.001 (ANOVA); significant time-by-treatment interaction between the diet preload and the sucrose- and HFCS-containing preload, P 쏝 0.001 (repeated-measures ANOVA); significant time-by-treatment interaction between the milk preload and the sucrose- and HFCS-containing preload, P 쏝 0.001 (repeated-measures ANOVA); significant time-by-sex interaction, P 쏝 0.01 (repeated-measures ANOVA).

to our observations, a previous study found no significant differences between the effects of cola or chocolate-milk consumption (0.9 MJ, 500 mL) with ad libitum intake 30 min later, despite significantly greater satiety 30 min after the chocolate milk (36) or in subsequent meal compensation 135 min after preloads (1.036 MJ, 590 mL) of cola, orange juice, and milk relative to sparkling water (37). As usual, energy intake including the energy-containing preloads was higher than total energy intake including the diet preload, despite the smaller consumption during the subsequent meal. Thus, subsequent energy intake only partly compensated for the energy delivered by the preloads; ie, for 45% with the sucrose-containing preload, for 42% with the HFCS-containing preload, and for 30% with the milk preload, all compared with energy intake after the diet preload. So, consumption of an energy-containing preload followed by a meal at 50

min led to overconsumption compared with a diet preload and subsequent meal. Previously, consumption of a 1.26-MJ highfructose-glucose mixture (80 –20%) was compensated with 12% of the meal consumed 60 min after preload, which was not different from that of an equisweet sucrose drink with 42% compensation (38). In conclusion, on the basis of these studies, subsequent energy intake did not differ significantly 30 –135 min after a 0.9 –1.5-MJ preload containing sucrose or HFCS or a milk preload. Therefore, in general, the effects of energy balance are positive, yet not different between different energy containing drinks. A sex effect was observed in VAS ratings, energy intake, and energy compensation and overconsumption. A possible explanations for these sex differences was the different responses in GLP-1 and glucose when preloads of the same size were offered. Previous studies support these higher concentrations in women

1592

SOENEN ET AL

TABLE 4 Satiety [change in area under the curve (⌬AUC)] as a function of the hormones glucagon-like peptide 1 (GLP-1) ghrelin, and insulin (⌬AUC) and correlations between GLP-1, ghrelin, and insulin (⌬AUC) and glucose (⌬AUC)1 GLP-1 Preload

Ghrelin

Insulin

r

P

r

P

r

P

0.253 0.382 0.429 — —

쏝0.01 쏝0.05 쏝0.05 — —

— — Ҁ0.407 0.423 —

— — 쏝0.05 쏝0.05 —

0.241 0.370 — — —

쏝0.05 쏝0.05 — — —

— — 0.36 —

— — 쏝0.001 —

— — Ҁ0.19 Ҁ0.22

— — 쏝0.05 쏝0.05

— — — 0.49

— — — 쏝0.001

2

All Satiety Sucrose3 HFCS3 Milk3 Diet3 All2 GLP-1 Ghrelin Insulin Glucose

FIGURE 3. Compensation after the sucrose-containing preload in men and women (n ҃ 14) as a function of the area under the curve (AUC; 0 –50 min) of pleasantness of taste. Compensation ҃ energy intake from diet Ҁ energy intake after any preload as a percentage of the preload. VAS, visual analogue scale.

1

HFCS, high-fructose corn syrup. n for all preloads was 120. 3 n for each preload was 30. 2

(39, 40). Obviously, the preloads that were consumed by the men represented a smaller part of energy requirement than the preloads consumed by the women. Moreover, sex differences in water turnover may play a role (41) because it has been suggested that the increased energy intake after drinks may have been derived from physiologic mechanisms giving priority to quenching thirst (42). The preloads suppressed thirst equally, significantly more in women than in men however. Are different mechanisms responsible for the satiety achieved after sucrose- or HFCS-containing preloads or a milk preload? Consumption of the preloads containing sucrose or HFCS caused similar changes in plasma concentrations of the hormones GLP-1, ghrelin, and insulin and of glucose. Also, leptin concentrations did not differ after consumption of either sucrose or

HFCS (43). The increase in satiety was underscored by the increase in GLP-1 with the sucrose- or HFCS-containing preloads, but not with the milk preload. Because satiety did not differ between energy-containing preloads, it may well be that other satiety hormones such as peptide YY3-36 and cholecystokinin, which were not measured, supported the milk-induced satiety. Satiety after the sucrose-containing preload was also underscored by the increase in insulin and satiety after the HFCScontaining preload by the decrease in ghrelin. The changes in VAS ratings of the appetite profile were supported by the changes in the concentrations of the hormones GLP-1, ghrelin, and insulin and glucose. Stepwise regression showed that satiety was primarily related to increases in GLP-1 concentrations and secondarily to insulin concentrations. Thus, sucrose and HFCS likely trigger GLP-1 release, which may have triggered insulin release and a related increase in satiety.

TABLE 5 Energy intake from the meal and from the meal ѿ preload, energy compensation, and energy overconsumption1

Sucrose-containing preload Women Men HFCS-containing preload Women Men Milk preload Women Men Diet preload Women Men

Meal size2

Total energy intake (preload ѿ meal)3

Compensation4

Overconsumption5

kJ

kJ

%

%

1742 앐 7306 2372 앐 794

3215 앐 7306 3845 앐 794

37 앐 376 53 앐 47

53 앐 536 29 앐 30

1873 앐 8686 2335 앐 786

3347 앐 8686 3808 앐 786

28 앐 426 55 앐 54

57 앐 506 29 앐 34

1945 앐 7566 2626 앐 880

3441 앐 7566 4122 앐 880

24 앐 426 36 앐 55

64 앐 606 37 앐 34

2290 앐 7736 3148 앐 984

2292 앐 7736 3150 앐 984

All values are x៮ 앐 SD; n ҃ 40. HFCS, high-fructose corn syrup. The treatment-by-sex interaction was not significant (multivariate ANOVA). Significant difference between the diet preload and the other 3 preloads (ANOVA): 2 P 쏝 0.05, 3 P 쏝 0.001. 4 Compensation ҃ energy intake from the diet Ҁ energy intake after any preload as a percentage of the preload. 5 Overconsumption ҃ total energy intake from the diet Ҁ total energy intake after any preload as a percentage of the preload. 6 Significantly different from men, P 쏝 0.05 (ANOVA). 1

2,3

SOFT DRINK INTAKE, SATIETY, AND COMPENSATION

On the other hand, satiety and compensation after the preload containing sucrose correlated with change in pleasantness of taste. Individuals do not eat solely based on hunger, taste is another reason for eating a specific food, and a decrease in pleasantness of taste is often given as a reason for terminating or reducing food intake. Therefore, the less sweet, refreshing, and pleasant milk preload may have contributed to incomplete compensation at the subsequent meal. Furthermore, high glycemic carbohydrates have been shown to be associated with a reduced appetite and food intake in the very short term (eg, 1 h), whereas lower glycemic carbohydrates showed a more delayed effect on the perception of satiety (eg, 2–3 h) (44, 45). We found a linear relation between the glucose content of the preloads and AUC plasma glucose concentrations. The glycemic indexes (GIs) of the monosaccharides glucose, fructose, and lactose are 99, 19, and 46, respectively (46). The GI of sucrose is 68 (46) and of HFCS is 73 (47) and 68 (48). The glucose concentrations peaked at 30 min and dropped below baseline at 60 min after the carbohydrate preloads and remained low until the end of the experiment. The same pattern of an initial steep increase in plasma glucose and insulin concentrations followed by a rebound effect, which stimulates hunger and food intake, has been found in several studies (16, 17, 32, 49 –56). Thus, a rapid rise in blood glucose and a large insulin response stimulates peripheral glucose uptake to such an extent that the blood glucose concentration falls below the fasting concentration. Therefore, the lower GI of milk, full-fat milk (GI: 27), and skim milk (GI: 32) (46), may have contributed to its satiety effect. Is satiety after sucrose- or HFCS-containing preloads influenced by its biochemical properties? The carbohydrate sucrose is a disaccharide and consists of one molecule of glucose and one molecule of fructose, which are not available for absorption until sucrose is hydrolyzed by intestinal brush-border enzymes. HFCS, on the other hand, contains glucose and fructose in their monosaccharide forms, which gives the solution a higher osmotic pressure. In soft drinks, however, a proportion of the sucrose is hydrolyzed into glucose and fructose by the acidic pH before the drinks are consumed. Fructose is passively absorbed in the duodenum and jejunum by a GLUT 5 transporter, which has a smaller absorption capacity than does the actively sodiumdependent hexose transporter, which absorbs glucose in the duodenum (57–59). However, there is a more complete and faster transport accompanied by a decrease in malabsorption when fructose is consumed in combination with other carbohydrates (20, 21). Both the differences in duration in the intestines and in the osmotic pressure of glucose and fructose could influence satiety differently. Furthermore, glucose triggers glucose sensors in the central nervous system involved in the regulation of food intake (60). Fructose, however, does not cross the blood-brain barrier (61). Fructose could trigger satiety by its oxidation (62), greater thermogenic response (63– 65), and rapid metabolism in the liver (61). The liver is sensitive to its own metabolism and signals to the brain via the vagus nerve to inhibit the central control for meal initiation (61). Thus, glucose and fructose in sucrose- or HFCS-sweetened drinks contribute to satiety through different biochemical mechanisms. In summary, a 1.5-MJ preload containing sucrose or HFCS or a milk preload did not affect energy intake differently 50 min later. Differences in satiety were absent despite different mechanisms underlying satiety due to sucrose- or HFCS-containing drinks or milk. Sucrose and HFCS triggered GLP-1 release,

1593

which triggered insulin release and a related increase in satiety. The different responses in GLP-1, glucose, and thirst when preloads of the same sizes were offered could explain the sex effect that was observed in VAS ratings, energy intake, and energy compensation and overconsumption. Obviously, the preloads that were consumed represented a smaller part of the energy requirement in men than in women. On the basis of partial compensation for and overconsumption due to the energy-containing preloads, a long-term study to assess the effect on body weight regulation would be a necessary follow-up. The question remains whether, in the long-term, this partial overconsumption of 앒40 –50% of the meal, amounting to 1 MJ, will accumulate. If no other long-term compensating mechanisms occurred, an increase in body weight over time of 앒1 kg over 1 mo would occur. Here, an additional 30 MJ accounts for a gain in body weight of 1 kg (66). To confirm this hypothetical approach or to find long-term compensating mechanisms, a well-controlled long-term study would be necessary. In conclusion, despite differences in the biochemical properties of preloads containing sucrose, HFCS, or milk and differences in the mechanisms underlying satiety in relation to GLP-1 release and ghrelin release, no differences in satiety, compensation, or overconsumption were observed. We gratefully acknowledge Marijke Prins, Jos Stegen, and Wendy Sluijsmans for their assistance. The study was supported by Suikerstichting Nederland (Baarn, Netherlands). The drinks were made by United Soft Drinks Ltd, Utrecht, Netherlands. SS designed the experiment, collected the data, analyzed the data, and wrote the manuscript. MSW-P designed the experiment, helped analyze the data and write the manuscript, and supervised the project. None of the authors had any financial or personal interest in any company or organization sponsoring the research.

REFERENCES 1. Harnack LJ, Jeffery RW, Boutelle KN. Temporal trends in energy intake in the United States: an ecologic perspective. Am J Clin Nutr 2000;71:1478–84. 2. Krebs-Smith SM. Choose beverages and foods to moderate your intake of sugars: measurement requires quantification. J Nutr 2001;131(suppl): 527S–35S. 3. Nielsen SJ, Siega-Riz AM, Popkin BM. Trends in energy intake in U.S. between 1977 and 1996: similar shifts seen across age groups. Obes Res 2002;10:370 – 8. 4. French SA, Lin BH, Guthrie JF. National trends in soft drink consumption among children and adolescents age 6 to 17 years: prevalence, amounts, and sources, 1977/1978 to 1994/1998. J Am Diet Assoc 2003; 103:1326 –31. 5. Pereira MA. The possible role of sugar-sweetened beverages in obesity etiology: a review of the evidence. Int J Obes (Lond) 2006;30(suppl):S28–36. 6. Flegal KM, Carroll MD, Ogden CL, Johnson CL. Prevalence and trends in obesity among US adults, 1999 –2000. JAMA 2002;288:1723–7. 7. Ogden CL, Carroll MD, Curtin LR, McDowell MA, Tabak CJ, Flegal KM. Prevalence of overweight and obesity in the United States, 1999 – 2004. JAMA 2006;295:1549 –55. 8. Guthrie JF, Morton JF, Schulze MB, et al. Sugar-sweetened beverages, weight gain, and incidence of type 2 diabetes in young and middle-aged women. J Am Diet Assoc 2000;100:43–51, quiz 49 –50. 9. Nielsen SJ, Popkin BM. Patterns and trends in food portion sizes, 1977– 1998. JAMA 2003;289:450 –3. 10. Nielsen SJ, Popkin BM. Changes in beverage intake between 1977 and 2001. Am J Prev Med 2004;27:205–10. 11. Cordain L, Eaton SB, Sebastian A, et al. Origins and evolution of the Western diet: health implications for the 21st century. Am J Clin Nutr 2005;81:341–54. 12. Bray GA, Nielsen SJ, Popkin BM. Consumption of high-fructose corn syrup in beverages may play a role in the epidemic of obesity. Am J Clin Nutr 2004;79:537– 43.

1594

SOENEN ET AL

13. Tordoff MG, Alleva AM. Effect of drinking soda sweetened with aspartame or high-fructose corn syrup on food intake and body weight. Am J Clin Nutr 1990;51:963–9. 14. Guss JL, Kissileff HR, Pi-Sunyer FX. Effects of glucose and fructose solutions on food intake and gastric emptying in nonobese women. Am J Physiol 1994;267:R1537– 44. 15. Kong MF, Chapman I, Goble E, et al. Effects of oral fructose and glucose on plasma GLP-1 and appetite in normal subjects. Peptides 1999;20:545–51. 16. Vozzo R, Baker B, Wittert GA, et al. Glycemic, hormone, and appetite responses to monosaccharide ingestion in patients with type 2 diabetes. Metabolism 2002;51:949 –57. 17. Spitzer L, Rodin J. Effects of fructose and glucose preloads on subsequent food intake. Appetite 1987;8:135– 45. 18. Rodin J, Reed D, Jamner L. Metabolic effects of fructose and glucose: implications for food intake. Am J Clin Nutr 1988;47:683–9. 19. Rodin J. Comparative effects of fructose, aspartame, glucose, and water preloads on calorie and macronutrient intake. Am J Clin Nutr 1990;51:428–35. 20. Rumessen JJ, Gudmand-Hoyer E. Absorption capacity of fructose in healthy adults. Comparison with sucrose and its constituent monosaccharides. Gut 1986;27:1161– 8. 21. Riby JE, Fujisawa T, Kretchmer N. Fructose absorption. Am J Clin Nutr 1993;58(suppl):748S–53S. 22. Mattes RD. Dietary compensation by humans for supplemental energy provided as ethanol or carbohydrate in fluids. Physiol Behav 1996;59: 179 – 87. 23. DiMeglio DP, Mattes RD. Liquid versus solid carbohydrate: effects on food intake and body weight. Int J Obes Relat Metab Disord 2000;24:794 – 800. 24. Harnack L, Stang J, Story M. Soft drink consumption among US children and adolescents: nutritional consequences. J Am Diet Assoc 1999;99:436–41. 25. Ludwig DS, Peterson KE, Gortmaker SL. Relation between consumption of sugar-sweetened drinks and childhood obesity: a prospective, observational analysis. Lancet 2001;357:505– 8. 26. Mrdjenovic G, Levitsky DA. Nutritional and energetic consequences of sweetened drink consumption in 6- to 13-year-old children. J Pediatr 2003;142:604 –10. 27. Berkey CS, Rockett HR, Field AE, Gillman MW, Colditz GA. Sugar-added beverages and adolescent weight change. Obes Res 2004;12:778 – 88. 28. Raben A, Vasilaras TH, Moller AC, Astrup A. Sucrose compared with artificial sweeteners: different effects on ad libitum food intake and body weight after 10 wk of supplementation in overweight subjects. Am J Clin Nutr 2002;76:721–9. 29. Schulze MB, Manson JE, Ludwig DS, et al. Sugar-sweetened beverages, weight gain, and incidence of type 2 diabetes in young and middle-aged women. JAMA 2004;292:927–34. 30. Almiron Roig E, Chen Y, Drewnowski A. Liquid calories and the failure of satiety: how good is the evidence? Obes Rev 2003;4:201–12. 31. Almiron Roig E, Flores SY, Drewnowski A. No difference in satiety or in subsequent energy intakes between a beverage and a solid food. Physiol Behav 2004;82:671–7. 32. Anderson GH, Woodend D. Consumption of sugars and the regulation of short-term satiety and food intake. Am J Clin Nutr 2003;78(suppl): 843S–9S. 33. Stunkard AJ, Messick S. The three-factor eating questionnaire to measure dietary restraint, disinhibition and hunger. J Psychosom Res 1985;29:71–83. 34. Westerterp Plantenga MS, Rolland V, Wilson SA, Westerterp KR. Satiety related to 24 h diet-induced thermogenesis during high protein/ carbohydrate vs high fat diets measured in a respiration chamber. Eur J Clin Nutr 1999;53:495–502. 35. Senn S. Cross-over trials in statistics in medicine: the first ’25' years. Stat Med 2006;25:3430 – 42. 36. Harper A, James A, Flint A, Astrup A. Increased satiety after intake of an isocaloric, isodense chocolate-milk drink compared with cola, but no difference in ad libitum lunch intake. Obes Rev 2005;6(s1):151 (abstr). 37. Almiron Roig E, Drewnowski A. Hunger, thirst, and energy intakes following consumption of caloric beverages. Physiol Behav 2003;79: 767–73. 38. Anderson GH, Catherine NL, Woodend DM, Wolever TM. Inverse association between the effect of carbohydrates on blood glucose and subsequent short-term food intake in young men. Am J Clin Nutr 2002;76:1023–30. 39. Vaag AA, Holst JJ, Volund A, Beck-Nielsen HB. Gut incretin hormones in identical twins discordant for non-insulin-dependent diabetes mellitus (NIDDM)— evidence for decreased glucagon-like peptide 1 secretion during oral glucose ingestion in NIDDM twins. Eur J Endocrinol 1996; 135:425–32.

40. Adam TC, Westerterp-Plantenga MS. Nutrient-stimulated GLP-1 release in normal-weight men and women. Horm Metab Res 2005;37:111–7. 41. Westerterp KR, Plasqui G, Goris AH. Water loss as a function of energy intake, physical activity and season. Br J Nutr 2005;93:199 –203. 42. Anderson GH. Sugars-containing beverages and post-prandial satiety and food intake. Int J Obes Relat Metab Disord 2006;30(suppl):S52–9. 43. Melanson KJ, Zukley L, Lowndes J, Nguyen V, Angelopoulos TJ, Rippe JM. Effects of high-fructose corn syrup and sucrose consumption on circulating glucose, insulin, leptin, and ghrelin and on appetite in normal-weight women. Nutrition 2007;23:103–12. 44. Anderson GH, Woodend D. Effect of glycemic carbohydrates on shortterm satiety and food intake. Nutr Rev 2003;61(suppl):S17–26. 45. McMillan-Price J, Brand-Miller J. Low-glycaemic index diets and body weight regulation. Int J Obes (Lond) 2006;30(suppl 3):S40 – 6. 46. Foster-Powell K, Holt SH, Brand-Miller JC. International table of glycemic index and glycemic load values: 2002. Am J Clin Nutr 2002;76:5–56. 47. Hung CT. Effects of high-fructose (90%) corn syrup on plasma glucose, insulin, and C-peptide in non-insulin-dependent diabetes mellitus and normal subjects (Abstract). Taiwan Yi Xue Hui Za Zhi 1989;88:883–5. 48. Miller JB, Pang E, Broomhead L. The glycaemic index of foods containing sugars: comparison of foods with naturally-occurring v. added sugars. Br J Nutr 1995;73:613–23. 49. Crapo PA, Kolterman OG, Olefsky JM. Effects of oral fructose in normal, diabetic, and impaired glucose tolerance subjects. Diabetes Care 1980;3:575– 82. 50. Melanson KJ, Westerterp Plantenga MS, Campfield LA, Saris WH. Blood glucose and meal patterns in time-blinded males, after aspartame, carbohydrate, and fat consumption, in relation to sweetness perception. Br J Nutr 1999;82:437– 46. 51. Horowitz M, Cunningham KM, Wishart JM, Jones KL, Read NW. The effect of short-term dietary supplementation with glucose on gastric emptying of glucose and fructose and oral glucose tolerance in normal subjects. Diabetologia 1996;39:481– 6. 52. Mayer J. Glucostatic mechanism of regulation of food intake. N Engl J Med 1953;249:13– 6. 53. Akgun S, Ertel NH. The effects of sucrose, fructose, and high-fructose corn syrup meals on plasma glucose and insulin in non-insulin-dependent diabetic subjects. Diabetes Care 1985;8:279 – 83. 54. Lee BM, Wolever TM. Effect of glucose, sucrose and fructose on plasma glucose and insulin responses in normal humans: comparison with white bread. Eur J Clin Nutr 1998;52:924 – 8. 55. Woodend DM, Anderson GH. Effect of sucrose and safflower oil preloads on short term appetite and food intake of young men. Appetite 2001;37:185–95. 56. Campfield LA, Smith FJ. Blood glucose dynamics and control of meal initiation: a pattern detection and recognition theory. Physiol Rev 2003; 83:25–58. 57. Ravich WJ, Bayless TM, Thomas M. Fructose: incomplete intestinal absorption in humans. Gastroenterology 1983;84:26 –9. 58. McIntyre AS, Thompson DG, Burnham WR, Walker E. The effect of beta-adrenoreceptor agonists and antagonists on fructose absorption in man. Aliment Pharmacol Ther 1993;7:267–74. 59. Buchs AE, Sasson S, Joost HG, Cerasi E. Characterization of GLUT5 domains responsible for fructose transport. Endocrinology 1998;139: 827–31. 60. Havel PJ. Peripheral signals conveying metabolic information to the brain: short-term and long-term regulation of food intake and energy homeostasis. Exp Biol Med (Maywood) 2001;226:963–77. 61. Friedman MI, Granneman J. Food intake and peripheral factors after recovery from insulin-induced hypoglycemia. Am J Physiol 1983;244:R374 – 82. 62. de Kalbermatten N, Ravussin E, Maeder E, Geser C, Jequier E, Felber JP. Comparison of glucose, fructose, sorbitol, and xylitol utilization in humans during insulin suppression. Metabolism 1980;29:62–7. 63. Tappy L, Randin JP, Felber JP, et al. Comparison of thermogenic effect of fructose and glucose in normal humans. Am J Physiol 1986;250:E718 –24. 64. Schwarz JM, Acheson KJ, Tappy L, et al. Thermogenesis and fructose metabolism in humans. Am J Physiol 1992;262:E591– 8. 65. Schwarz JM, Schutz Y, Piolino V, Schneider H, Felber JP, Jequier E. Thermogenesis in obese women: effect of fructose vs. glucose added to a meal. Am J Physiol 1992;262:E394 – 401. 66. Westerterp KR, Donkers JH, Fredrix EW, Boekhoudt P. Energy intake, physical activity and body weight: a simulation model. Br J Nutr 1995; 73:337– 47.

Novel calcium-gelled, alginate-pectin beverage reduced energy intake in nondieting overweight and obese women: interactions with dietary restraint status1–3 Christine L Pelkman, Juan L Navia, Allison E Miller, and Rachael J Pohle ABSTRACT Background: Foods containing strong-gelling fibers may provide a safe and efficacious strategy for reducing food intake by stimulating endogenous satiety signaling. Objective: A novel, 2-part beverage, consisting of alginate-pectin and calcium components, that forms a stable, fibrous gel in the stomach was tested to determine its effects on subjective satiety and food intake in overweight and obese women. Design: The investigation was a within-subjects, double-blind, placebo-controlled study. Subjects (n ҃ 29) ingested a 2-part beverage twice per day (once before breakfast and once midafternoon) for 7 d. Three alginate-pectin formulations were tested: 1.0 g, 2.8 g, and control (no fiber). Subjective satiety and ad libitum food intake were measured on days 1 and 7 of each 1-wk treatment period with a 1-wk washout between testings. Results: A significant reduction in food intake was observed at dinner for both formulations compared with the control formulation. The effects of the gel beverage differed as a function of rigid dietary restraint status. Women in the lower 50th percentile of rigid restraint consumed 12% less energy during the day and 22% less for the evening snack in the 2.8-g condition compared with the control condition. No effect was found for women in the upper 50th percentile of rigid restraint. Conclusions: Consumption of a postingestion, calcium-gelled fiber beverage twice daily reduced energy intake in overweight and obese women with low rigid restraint scores. Use of foods designed to enhance satiety may be an effective adjunctive therapy for weight loss; however, more research is needed to determine how dietary restraint alters this response. Am J Clin Nutr 2007;86: 1595– 602. KEY WORDS Energy intake, satiety, calcium, alginate, pectin, dietary restraint INTRODUCTION

The increasing prevalence of obesity in the United States (1–3) and worldwide (4 – 8) poses a significant threat to public health (9 –13). Obesity results from a positive energy balance, and prevention and treatment efforts have focused on both the intake and expenditure sides of the energy balance equation. Medical approaches focus on reducing food intake by surgical (14) and pharmacologic (15) means. Although effective, these approaches are costly and not without risk of significant side effects (16 –18). Recent advances in understanding the controls of food

intake have made it possible to identify specific targets for reducing food intake that engage endogenous satiety mechanisms and that are potentially safer and less costly than medical approaches. One example is the use of low-energy-dense foods to maximize gastric feedback mechanisms of satiety, thereby reducing food intake (19 –21) and enhancing weight loss (22). Use of low-energy food products that are specifically designed to engage gastrointestinal satiety mechanisms may be efficacious for enhancing weight loss or for preventing weight gain. We report results from a study testing the effects of a novel low-energy, calcium-gelled, alginate-pectin beverage on food intake in overweight and obese women. The product was specifically designed to enhance satiety by forming a thick, stable gel in the stomach. We expect the gel to maintain some integrity as it passes through the upper gastrointestinal tract from anecdotal observation of the gel formed by simply mixing the 2 beverage components in a beaker at ambient temperature, from rheological measurements in an in vitro stomach model, and from examining physical appearance and measurement of flow characteristics of digesta recovered from fistulated Yucatan minipigs that consumed the beverage components (data not shown). In all cases, gel particles were present and contributed to the apparent viscosity of the resultant slurry. Samples that experienced minimal trituration (beaker) contained large gel particles that could be scooped up by the hand-full, whereas samples macerated in the pig digestive tract contained smaller but clearly visible particles. Assuming similar slurry formation in the human digestive tract, the dilution of digesta by this inert material could be expected to slow nutrient absorption and result in prolonged triggering of associated gut satiety signals. We hypothesized that the combination of these signals with gastric signals would enhance satiety and reduce energy intake. Our objective was to assess the effects of the gel beverage on subjective satiety and food intake during several meals. A secondary objective was to assess the effects of sustained consumption of the beverages during a 1-wk period and 1 From the Department of Nutrition and Exercise Sciences, University at Buffalo, Buffalo, NY (CLP, AEM, and RJP), and McNeil Nutritionals, LLC, Fort Washington, PA (JLN). 2 Supported by McNeil Nutritionals, Division of McNeil–PPC, Inc. 3 Reprints not available. Address correspondence to CL Pelkman, Department of Exercise and Nutrition Sciences, University at Buffalo, 15 Farber Hall, Buffalo, NY 14214-8028. E-mail: [email protected]. Received September 8, 2006. Accepted for publication August 2, 2007.

Am J Clin Nutr 2007;86:1595– 602. Printed in USA. © 2007 American Society for Nutrition

1595

1596

PELKMAN ET AL

to evaluate the effect of timing of the beverage; thus, one beverage was ingested with a meal and the other beverage was ingested between meals. Finally, we sought to assess the possible modulating effects of subject characteristics related to eating, such as binge eating and dietary restraint, because previous studies showed that satiety responses may be altered by behavioral characteristics of the subjects (23–27).

separate beverages that, when taken together, constitute a single treatment. The corresponding control beverages were formulated to match the active beverages for flavor and color but without the alginate-pectin blend or calcium. Neither beverage alone was expected to be effective in forming a stable gel; therefore, neither beverage was tested as a separate treatment. Bottles were numbered with arbitrary 4-digit numbers by the manufacturer to maintain the study blind.

SUBJECTS AND METHODS

Test sessions

Subjects

Subjects were asked to report to the laboratory for meals (breakfast, lunch, and dinner) on days 1 and 7 of each test-session week. Test sessions began between 0700 and 0900. Subjects first completed a short questionnaire to ensure they consumed the evening meal and were not ill in the previous week. Immediately before breakfast, they were asked to consume the appropriate 237-mL (1.0 g, 2.8 g, or control) beverage, followed by the 118-mL calcium (or control) beverage. They were asked to consume each beverage over a 3-min interval and were given timers to help pace their consumption. Subjects then consumed breakfast and returned for lunch 4 –5 h later and dinner 9 –10 h later. They were given a cooler (with ice packs) that contained one dose of the test beverage (237-mL and 118-mL components) and a bottle of spring water and were instructed to consume the test beverage 2.5 h after the end of the lunch meal. They were given written instruction reminding them to consume the 237-mL beverage first, followed by the 118-mL beverage and to pace their consumption over 3 min for each beverage. They were asked to refrain from drinking water for 30 min after consumption of the test beverage. Subjects were also instructed not to consume any foods or beverages between breakfast and lunch and between lunch and dinner, except the bottled water and the test beverages provided. Subjects were asked to assess subjective sensations of hunger, fullness, nausea, thirst, and the desire to eat on 100-mm visual analog scales (VASs) before and after each meal and hourly between meals. They were given another cooler after the dinner meal, containing snacks for the evening and enough water and test beverages for a 5-d period. They were told to consume as much or as little of the evening snacks as they desired and to return the leftovers the next morning. They were instructed to consume the test beverages each morning, immediately before breakfast, and in the afternoon (2.5 h after lunch) for the following 5 d and to return the empty containers to the laboratory on day 7. Subjects were also asked to record in a daily diary the time each bottle was consumed and any unusual events or sensations they experienced that day. Returned coolers were immediately checked by study personnel. Empty bottles were counted, and the 4-digit numbers were recorded on a check-in sheet. Any reports of unusual events or symptoms were immediately followed up in a telephone interview conducted by the study coordinator and reviewed by the study physician. The protocol for day 7 was the same as day 1 with the exception that participants were not given test beverages to consume after the test day was completed.

Healthy, premenopausal, nonsmoking, overweight or obese [body mass index (BMI; in kg/m2): 25–35] women between the ages of 20 and 40 y were recruited through media advertisements and flyers. Subjects were screened by self-report questionnaires and anthropometric assessment to ensure that they were in good health, did not have an eating disorder, were weight stable, were not dieting to lose weight, were not depressed, and were willing to consume the foods used in the study. Screening questionnaires included the Beck Depression Inventory (28), the Eating Attitudes Test (29), the Eating Inventory (30), and the Binge Eating Scale (31). Subjects were asked to taste samples of the test beverages to ensure they would be willing to consume the beverages during the study. They were also instructed to consume comparable meals before each test day, to abstain from alcohol for 2 d before each session, to refrain from consumption of any foods after 2200 the evening before each session, and to discontinue use of any vitamin or calcium supplements. This study was approved by the Health Sciences Institutional Review Board of the University at Buffalo, and subjects gave written consent before participation. Study design The study was a within-subjects, placebo-controlled, doubleblind design with 3 formulations tested. Subjects received one formulation for 7 d with a 1-wk washout between test weeks. Subjects consumed the test beverage twice daily: one dose before breakfast and a second dose 앒2.5 h after lunch for a total of 14 doses consumed per 7-d treatment session. Order of treatment was counterbalanced with 5 subjects randomly assigned to each of 6 possible treatment sequences with the use of a random numbers generator. Calcium-alginate beverage Each dose provided 40 kcal and was packaged in 2 parts. The alginate-pectin blend [alginate (앒1:1 Manugel LBA and GHB; ISP, Wayne, NJ) 1.0 or 2.8 g and a 15:85 blend of pectin (USPL220; CP Kelco, Atlanta, GA)] and matched control beverages were formulated in a fruit-flavored aqueous solution (237 mL) sweetened with sucralose. The other part was a 118-mL, fruitflavored beverage also sweetened with sucralose that provided 앒500 mg elemental calcium as the lactate salt (Purac, Lincolnshire, IL) or no calcium. Subjects consumed sequentially one 237-mL and one 118-mL bottle on each occasion the dose was to be consumed. The reaction of the alginate-pectin blend with the calcium occurs in the stomach on ingestion. The cross-linking of polysaccharide chains is fast and yields a thick, stable gel. In this form, it would be unpalatable to subjects and easily distinguishable from the control beverage. The 237-mL alginate-pectin mixtures and the 118-mL calcium salt mixture were formulated as

Meals The study was designed to assess the effects of the calciumalginate beverages on food intake during the course of the day. We varied the timing of the beverages to examine effects on intake at subsequent meals. The first beverage was consumed immediately before breakfast. Food intake at breakfast was standardized. Subjects were given a choice of bagels or cereal and

1597

CALCIUM ALGINATE-PECTIN BEVERAGE REDUCED INTAKE

yogurt for their breakfast and were served this same breakfast, along with other meal-appropriate foods, such as juice, milk, coffee, or tea, on each of the days in the test session and were encouraged to consume all of the meal. Foods presented for lunch, dinner, and evening snack were more varied and designed to allow subjects to choose from a variety of foods. Lunch and dinner meals were served as individual, buffet-style meals that allowed subjects ad libitum selection from a variety of mealappropriate foods, such as sliced meats, bread, cheeses, and fruit at lunch, and hot entrees, including meats (roast beef or chicken breast) and vegetables for dinner. All foods were commercially available products and provided varying amounts of energy and macronutrients, allowing subjects to vary intake of energy, fat, protein, and carbohydrate. Subjects were presented with more food than they were likely to consume and instructed to consume as much or as little as they desired. In addition, subjects were asked to choose from a variety of take-home snacks and beverages, such as potato chips, cookies, chocolate bars, juice, milk, fruit, and fresh vegetables, to be consumed in the evening after dinner. All foods were weighed on an electronic scale to the nearest 0.1 g before and after consumption to determine the amount consumed. Energy and macronutrient composition of the foods were obtained from the manufacturer’s food label or from a standard reference for unlabeled food (32), such as fresh produce. Data analyses Statistical power analyses were conducted with the use of NQUERY (version 4.0; Statistical Solutions, Saugus, MA). We used estimates of mean intake (and SDs) at lunch and dinner and for the daily total reported by Rolls et al (33) in a study that used procedures similar to our research protocol. The analyses were conducted with the use of the paired t test procedure for a onesided t test, and ␣ was adjusted to correct for multiple comparisons because we intended to compare means between 3 conditions (adjusted ␣ ҃ 0.0167). We determined that a sample size of 30 subjects yielded power estimates of 0.81, 0.86, and 0.94 to detect a 15% reduction in food intake at lunch, at dinner, and for total daily intake, respectively. We also conducted post hoc power analyses from the current study to compare with these initial calculations. Outcome data were analyzed with the use of the STATISTICAL ANALYSIS SYSTEM (SAS; version 9.1; SAS Institute, Cary, NC). The mixed model procedure was used to test for treatment differences, with treatment condition (1.0 g, 2.8 g, or control), day (1 or 7), and the interaction of condition and day entered into the statistical models. Subjects were tested on 6 occasions (days 1 and 7 of three, 1-wk treatment periods); thus, we assessed the effects of repeated testing by adding session occasion number (1– 6) as a covariate in all statistical models and kept the variable in the model when found to be significant. The endpoint measurements included the total intake of energy and macronutrients during the day, as well as at each individual meal (breakfast, lunch, dinner, and evening snack). Models were tested to determine the effects of treatment on the VAS ratings of hunger, fullness, prospective consumption, and nausea. In addition, an appetite score was computed to reflect the combined effects on hunger, fullness, and prospective consumption, similar to the strategy used by Anderson et al (34) with the use of the following equation: appetite ҃ [hunger ѿ prospective consumption ѿ (100 Ҁ fullness)]/3. Mixed model analysis of variance

models were tested for each VAS rating and for appetite score with condition, day, time, and all possible interactions entered as factors. Separate models were run to test for effects on ratings taken before meals (immediately before breakfast, before lunch, and before dinner). Ratings taken after breakfast (from immediately after breakfast to the last rating taken before lunch) and after lunch (from immediately after lunch to the last rating taken before dinner) were separated to examine effects on specific postprandial intervals. Ratings during those intervals were also used to calculate incremental values of area under the curve (with the use of the trapezoid rule) for the morning and afternoon periods. Age; BMI; and scores on the Beck Depression Inventory, Eating Attitudes Test, Binge Eating Scale, and the 3 factors of the Eating Inventory, including global, flexible, and rigid restraint (35), were entered as covariates in the mixed models as main effects and in interaction terms to determine whether baseline characteristics of the subjects modulated the effects of beverage ingestion on food intake or subjective satiety. The covariates were tested as continuous and binary variables (with a 50thpercentile cutoff). None of the continuous variables were found to be significant. With the use of a 50th-percentile split, trends for an interaction of global restraint score and condition were found for total daily energy intake (P ҃ 0.059) and intake at evening snack (P ҃ 0.059). The only statistically significant interactions with condition were for rigid restraint, reported in “Results.” The interactions were further examined with the use of the SLICE command in SAS to test for an effect of condition within each rigid restraint group. P 쏝 0.05 was considered to be statistically significant for interaction terms and the effect of condition in the SLICE procedure. Tukey’s post hoc test was used to compare least-squares means when main effects of condition were found. When significant effects of condition were found within rigid restraint groups, adjusted P values were used to compare leastsquares means between the 3 conditions [adjusted P ҃ 1 Ҁ (1 Ҁ P)3]. Data given in the text, figure, and tables are least-squares means (앐SEMs) from the mixed models, unless stated otherwise. RESULTS

Thirty-five women were enrolled in the study, and 29 completed all test sessions (Table 1). Withdrawals were due to job conflicts (n ҃ 2), personal conflicts (n ҃ 1), lack of childcare (n TABLE 1 Baseline subject characteristics1 Variable Age (y) Weight (kg) Height (cm) BMI (kg/m2) Beck Depression Inventory (score) Eating Attitudes Test (score) Binge Eating Scale (score) Subscales of the Eating Inventory (score) Restraint (global score) Rigid restraint Flexible restraint Disinhibition Hunger 1

Value 33.4 앐 6.5 (20–40) 83.0 앐 5.7 (73.0–97.1) 164.6 앐 4.4 (154.1–173.4) 30.6 앐 2.2 (27.5–34.4) 4 앐 3.2 (0–9) 7.7 앐 7.7 (0–30) 11.9 앐 7.3 (2–29) 10.7 앐 4.0 (4–20) 3.4 앐 1.7 (1–7) 3.1 앐 1.6 (1–7) 7.7 앐 3.9 (2–16) 6.0 앐 3.7 (0–14)

All values are x៮ 앐 SD; range in parentheses; n ҃ 29.

1598

PELKMAN ET AL

҃ 1), and competing time commitments (n ҃ 2). Fourteen additional subjects were eligible to participate but withdrew before starting the test meal sessions for similar reasons (n ҃ 9) or for inability to comply with test beverage consumption (n ҃ 5). Some exceptions to the study protocol occurred as a result of subjects’ scheduling conflicts or noncompliance with the study protocol. On 8 occasions subjects had more than 1 wk off between test sessions (2 or 3 wk), one subject required a shorter washout week between test sessions (5 d), and there were 16 instances of subjects having a nonstandard length of test week (from 5 to 8 d). In addition, one subject mistakenly consumed the afternoon test beverage before lunch and another forgot to consume the afternoon test beverage. Two subjects missed a test day (all meals), and one missed an evening snack. All subjects were included in the analyses, and the total sample size was 29, with the exceptions noted above for missing data. Energy and macronutrient intake Consumption of the calcium and alginate-pectin beverages between lunch and dinner significantly reduced energy intake at dinner. Subjects consumed 앒10% less energy in both active-gel conditions compared with the control condition (Table 2). No interaction between condition and day was found. The suppression of food intake at dinner was evident on both days that food intake was measured with an average reduction for the 2 fiber conditions of 9% on day 1 and 12% on day 7 compared with the control condition (data not shown). Approximately two thirds of the reduction in caloric intake observed at the dinner meal (Table 2) was accounted for by the significant reduction in carbohydrate intake at that meal in both conditions. Indeed, significant reductions in carbohydrate intake were observed at breakfast, lunch, and for the day. At breakfast, subjects consumed 앒5% less carbohydrate in the 2 experimental conditions compared with the control condition. Subjects consumed 앒11% less carbohydrate in the 1.0-g condition than in the control condition at lunch and 앒6% less carbohydrate during the day in both gel-beverage conditions than in the control condition (Table 2). The reduction in overall energy intake in each condition for the day, amounting to 앒8.5% of that consumed in the control condition, was not enough to reach statistical significance. Significant interactions with rigid restraint status were found for intake at the evening snack (P ҃ 0.007) and for the total daily intake (P ҃ 0.017) (Figure 1). The rigid restraint subscale of the Eating Inventory comprises seven, 1-point items. Scores for subjects ranged from 1 to 7, and 50% scored 울2. Post hoc testing with the use of the SLICE command showed significant effects of condition for subjects classified in the lower 50th percentile for intake during the day (P ҃ 0.01) and intake at the evening snack (P ҃ 0.046). Subjects in the low rigid restraint category consumed less energy during the day (2541 앐 187 kcal; adjusted P ҃ 0.013) and tended to consume less at the evening snack (536 앐 99 kcal; adjusted P ҃ 0.07) in the 2.8-g condition than in the control condition (2875 앐 186 and 684 앐 98 kcal for daily intake and evening snack, respectively). When the 2 gel-beverage conditions were compared (Figure 1), intake during the day tended to be less with the 2.8-g condition than with the 1.0-g condition (2799 앐 186 kcal; adjusted P ҃ 0.07), but no difference (adjusted P ҃ 0.12) was found between the 2 gel-beverage conditions for the evening snack (668 앐 98 kcal in 1.0-g condition). No significant differences were noted for breakfast (499 앐 34, 476 앐

TABLE 2 Energy and macronutrient intake by test condition1 Control

1.0 g

2.8 g

Breakfast Carbohydrate (kcal) 364 앐 14a,3 345 앐 14 341 앐 14b Protein (kcal) 65 앐 3 63 앐 3 62 앐 3 Fat (kcal) 63 앐 8 65 앐 8 65 앐 8 Total (kcal)4 499 앐 20 479 앐 20 474 앐 20 Lunch Carbohydrate (kcal) 294 앐 22a 262 앐 22b 284 앐 22 Protein (kcal) 114 앐 7 112 앐 7 107 앐 7 Fat (kcal) 366 앐 28 367 앐 28 362 앐 28 Total (kcal)4 807 앐 47 775 앐 47 785 앐 47 Dinner Carbohydrate (kcal) 379 앐 21a 329 앐 21b 324 앐 21b Protein (kcal) 127 앐 6 118 앐 6 116 앐 6 Fat (kcal) 236 앐 14 220 앐 14 216 앐 14 Total (kcal)4 764 앐 37a 689 앐 37b 678 앐 37b Snack Carbohydrate (kcal) 315 앐 36 324 앐 36 314 앐 36 Protein (kcal) 36 앐 4 36 앐 4 40 앐 4 Fat (kcal) 203 앐 27 203 앐 27 210 앐 27 Total (kcal)4 564 앐 58 571 앐 58 575 앐 58 Daily total Carbohydrate (kcal)5 1433 앐 64a 1341 앐 63b 1344 앐 64b Protein (kcal) 342 앐 12 329 앐 12 325 앐 12 Fat (kcal) 868 앐 49 855 앐 49 853 앐 49 Total (kcal)4,5 2716 앐 110 2594 앐 109 2591 앐 109

P2 0.03 0.08 0.73 0.094 0.02 0.15 0.96 0.50 0.002 0.12 0.14 0.01 0.86 0.33 0.92 0.96 0.02 0.12 0.87 0.11

Subjects (n ҃ 29) completed three 1-wk test conditions, in counterbalanced order, with a 1-wk washout period between conditions. In each condition, subjects consumed 1 of 3 beverages (control, 1.0 g, or 2.8 g) before breakfast and in the midafternoon for 7 d. Food intake was measured on days 1 and 7 of each 1-wk test period. Data are collapsed across day (1 and 7) because the effects of day were not significant. Values in the same row with different superscript letters are significantly different, P 쏝 0.05 (Tukey’s post hoc test). 2 For main effect of condition in mixed model. 3 Least-squares x៮ 앐 SEM (all such values). 4 Total energy intake and intake for macronutrients were derived from nutrition facts labels for individual foods consumed; therefore, the total energy consumed at each meal or during the day is not equal to the sum of the energy consumed from each macronutrient. 5 Includes energy from gel or control beverages (80 kcal/d). 1

34, and 454 앐 34 kcal), lunch (855 앐 80, 804 앐 80, and 765 앐 81 kcal), and dinner (756 앐 63, 771 앐 63, and 707 앐 63 kcal) in the control, 1.0- and 2.8-g conditions, respectively. No interactions with day were found. The reduction in energy intake at snack and during the day was due in part to a significant reduction in energy intake from carbohydrates (Table 3). Subjects with low rigid restraint consumed 앒22% less carbohydrate at the evening snack and 12% less carbohydrate during the day in the 2.8-g condition than did subjects in the control condition. In the high rigid restraint group, SLICE analyses showed no significant effects of condition on energy intake (Figure 1). Differences in macronutrient intake between conditions at dinner, at evening snack, and for the daily total were found (Table 3). Subjects with high rigid restraint consumed less protein and fat in the 1.0-g condition than did subjects in the control condition at dinner and during the day. There was no evidence of lowered intake of macronutrients at the evening snack when subjects with high rigid restraint were found to consume more protein in the 2.8-g condition than in the control condition (Table 3). We compared

CALCIUM ALGINATE-PECTIN BEVERAGE REDUCED INTAKE

1599

statistical power analyses. One-tailed tests with an adjusted ␣ of 0.0167 was used in the analyses. We had originally based our sample size (n ҃ 30) on statistical power calculations by using data from a previous study (33) that showed adequate power (쏜0.80) to detect a reduction in food intake of 15% at lunch, dinner, or for the total daily intake. The statistical power to detect a 15% reduction in intake in the current study was high (0.97 for lunch, 0.92 for dinner, and 쏜0.99 for total daily intake), indicating that failure to find an effect on total daily intake in the combined sample of 29 women was not due to a lack of statistical power. The effect was much greater in the group with low rigid restraint. Despite the smaller sample size (n ҃ 10), the power to detect a 15% reduction in total daily intake was 0.97. Power to detect a 15% reduction in intake at lunch, dinner, and evening snack was much lower in the group with low rigid restraint (0.57, 0.42, and 0.24, respectively). Subjective satiety

FIGURE 1. Energy intake by condition and rigid restraint status. Subjects completed three 1-wk test conditions, in counterbalanced order, with a 1-wk washout period between conditions. In each condition, subjects consumed 1 of 3 beverages (control, 1.0 g, or 2.8 g) before breakfast and in the midafternoon for 7 d. Food intake was measured on days 1 and 7 of each 1-wk test period. Values shown are least-squares means from mixed model analyses with condition, day, rigid restraint status, and the interaction of condition and restraint entered as factors and adjusted for the effects of repeated testing (sessions 1– 6). Data are collapsed across day (1 and 7) because the effects of day were not significant. Significant interactions of condition and rigid restraint status were found for total daily energy intake (P ҃ 0.017) and intake at the evening snack (P ҃ 0.007) in mixed model analysis of variance models with condition, day, and rigid restraint status entered as main effects and adjusted for the effects of repeated testing (sessions 1– 6). With the use of the SLICE command, significant main effects of condition were found for total daily energy intake (P ҃ 0.01) and intake at the evening snack (P ҃ 0.046) for subjects in the lower 50th percentile of rigid restraint. The top panel is subjects classified in the lower 50th percentile of rigid dietary restraint, and the bottom panel is subjects classified in the upper 50th percentile of rigid restraint. *Trend for difference between means for total daily intake and intake at the evening snack (0.05 쏜 adjusted P 쏝 0.10). #Significant difference between means for total daily intake (adjusted P ҃ 0.013).

the baseline characteristics of the subjects with high and low rigid restraint with the use of t tests and found no significant differences between groups for binge score, disinhibition, hunger, age, BMI, Beck depression score, or Eating Attitudes Test score (P 쏜 0.20 for all; data not shown). Post hoc power analyses Differences in the least-squares means for intake in the 2.8-g and control conditions (adjusted for the effects of session) and estimates of the SD of the differences between least-squares means from the mixed models were used to perform post hoc

Four-factor interactions were not significant in models testing the effects of condition, day, time, and rigid restraint status on subjective ratings of hunger, fullness, prospective consumption, and appetite score. Significant 2-factor and 3-factor interactions were found; however, the results were inconsistent. For example, subjects with low rigid restraint reported lower ratings of hunger in the morning in the 2.8-g and control conditions than did those in the 1.0-g condition on day 1, but subjects with high rigid restraint reported greater fullness immediately before lunch in the 2.8-g condition than did subjects in the 1.0-g and control conditions on both days. Because the interaction effects involving restraint status and condition were generally small in magnitude and inconsistent, they are not discussed in further detail here. Nausea and adverse events Ratings of nausea were low and ranged on average from 2.6 to 8.3 mm (100-mm scale) during the course of the day (data not shown). No significant effects of condition, restraint status, day, or interactions between variables were found at any time point before meals or during the morning or afternoon. Fourteen incidents of adverse events were reported by 8 subjects. All events were deemed not to be serious by the study physician and included reports of mild stomachache, increased frequency of bowel movements, loose stool, and increased flatulence. Three events were deemed to be temporally unrelated to ingestion of the beverages. Of the remaining 11 reports, 3 were reported after ingesting the control beverage, 4 after ingesting the 1.0-g fiber beverage, and 4 after ingesting the 2.8-g fiber beverage. DISCUSSION

The novel, 2-part beverage tested in the current study is nonviscous before ingestion, but the acidic side-chain groups of the alginate undergo rapid cross-linking reactions with the divalent calcium ions to form a solid gel material on mixing in the stomach. Highly viscous fibers that form strong gels, although more likely to enhance satiety, are likely to be unpalatable. Thus, the delayed-activation beverage system tested in this study represents an alternative approach for the delivery of gels into the gastrointestinal tract that circumvents the aversive effects of orally ingesting a gelled substance.

1600

PELKMAN ET AL

TABLE 3 Macronutrient intake by rigid restraint status and test condition1 Low rigid restraint (n ҃ 10)

Breakfast Carbohydrate (kcal) Protein (kcal) Fat (kcal) Lunch Carbohydrate (kcal) Protein (kcal) Fat (kcal) Dinner Carbohydrate (kcal) Protein (kcal) Fat (kcal) Snack Carbohydrate (kcal) Protein (kcal) Fat (kcal) Daily total Carbohydrate (kcal)6 Protein (kcal) Fat (kcal)

Control

1.0 g

374 앐 244 60 앐 5 61 앐 15

352 앐 24 58 앐 5 62 앐 15

334 앐 36 108 앐 12a 382 앐 48

High rigid restraint (n ҃ 19) 2.8 g

P for interaction2

P for main effect3

Control

1.0 g

2.8 g

335 앐 24 57 앐 5 58 앐 15

359 앐 18 68 앐 4 64 앐 11

342 앐 17 65 앐 4 66 앐 11

344 앐 17 65 앐 4 68 앐 11

0.47 0.89 0.46

0.02 0.11 0.83

306 앐 36 112 앐 12a 354 앐 48

320 앐 37 88 앐 12b 332 앐 48

272 앐 26 118 앐 9 356 앐 35

239 앐 26 112 앐 9 374 앐 35

265 앐 26 116 앐 9 378 앐 35

0.87 0.0035 0.17

0.03 0.02 0.74

378 앐 37 115 앐 11 239 앐 24

352 앐 37 130 앐 11 264 앐 24

329 앐 37 109 앐 11 244 앐 24

379 앐 27 133 앐 8a 235 앐 18a

317 앐 26 112 앐 8b 197 앐 18b

322 앐 27 120 앐 8a,b 201 앐 18b

0.54 0.0035 0.025

0.005 0.21 0.42

388 앐 61a 36 앐 7 243 앐 47

409 앐 61a 34 앐 7 221 앐 46

302 앐 62b 30 앐 7 194 앐 47

275 앐 45 36 앐 5a 181 앐 34

280 앐 45 38 앐 5a,b 194 앐 34

318 앐 45 45 앐 5b 217 앐 34

0.0025 0.045 0.09

0.27 0.83 0.93

1516 앐 107a 318 앐 20a 924 앐 84

1458 앐 1077 334 앐 20a 901 앐 84

1327 앐 107b,7 283 앐 20b 829 앐 85

1327 앐 78a 354 앐 15a 838 앐 61

1218 앐 77b 326 앐 14b 831 앐 61

1289 앐 77a,b 346 앐 14a,b 864 앐 61

0.025 0.00025 0.15

0.006 0.03 0.57

1 Subjects completed three 1-wk test conditions, in counterbalanced order, with a 1-wk washout period between conditions. In each condition, subjects consumed 1 of 3 beverages (control, 1.0 g, or 2.8 g) before breakfast and in the midafternoon for 7 d. Food intake was measured on days 1 and 7 of each 1-wk test period. Values were determined from mixed model analyses with condition, day, rigid restraint status, and the interaction of condition and restraint entered as factors and adjusted for the effects of repeated testing (sessions 1– 6). Data are collapsed across day (1 and 7) because the effects of day were not significant. Within a restraint category, values in the same row with different superscript letters are significantly different, adjusted P 쏝 0.05. Adjusted P ҃ 1 Ҁ (1 Ҁ P)3. 2 Interaction of condition and rigid restraint status in mixed model. 3 Main effect of condition in mixed model. 4 Least-squares x៮ 앐 SEM (all such values). 5 Comparison of least-squares means within restraint categories was conducted when a significant effect of condition was found with the Slice procedure (P 쏝 0.05 for effect of condition within a rigid restraint category). 6 Includes energy from gel or control beverages (80 kcal/d). 7 Trend for a difference between least-squares means within a restraint category (0.05 쏝 adjusted P 쏝 0.10).

The results show that ingesting the gel-forming beverage reduced food intake in weight-stable overweight and obese women without any meaningful change in subjective measures of appetite. In the total sample, dinner intake was suppressed by 앒10% in the 2 fiber conditions, with no evidence of energy compensation later in the evening. The effects, however, were mediated by dietary restraint status, and findings for the subjects with low rigid restraint showed evidence of a threshold response with 앒12% less food intake during the day in the 2.8-g condition than with the control condition, whereas intake in the 1.0-g condition was not less than the control condition and tended to be 앒10% greater than the 2.8-g condition. The effects were not attenuated during a 1-wk period. Intake suppression in the 2.8-g condition in the subjects with low rigid restraint, although not significant at each meal, appeared to be consistent across the day with reductions in food intake at breakfast, lunch, dinner, and evening snack, suggesting both short-term and longer-term effects on satiety. The before-breakfast fiber beverages tended to reduce breakfast intake even though subjects were encouraged to finish the meal. The effect, however, was small in magnitude (앒45 kcal or 9% of the control condition) in subjects with low rigid restraint and 14 kcal (3% of control condition) in subjects with high rigid restraint. The 4-h interval after consumption of the morning

beverage may have been too long to affect intake at lunch. Alternatively, the smaller sample size of the group with low rigid restraint may have reduced our statistical power to detect the 84-kcal (10%) reduction in lunch intake in the 2.8-g condition compared with the control condition. The shorter interval (2.5 h) in the afternoon resulted in reduced intake at dinner in the combined group of women but not in the group with low rigid restraint. In this group, the effects at dinner (앒Ҁ50 kcal; 7% reduction) were not significant, whereas later effects at evening snack were clearly evident (Ҁ141 kcal; 22% reduction). The observed effects are postulated to result from stimulation of gastric and intestinal signals. Recent studies that used magnetic resonance imaging show that high-viscosity, gel-forming fibers form lumps in the stomach, increase gastric volume, and enhance fullness compared with high- or low-viscous meals that do not gel and are more homogenously diluted in the stomach (36, 37). Hoad et al (36) used this approach to compare stronggelling and weak-gelling alginates to guar gum and placebo. The alginate was mixed into a sweetened, milk-based meal replacement beverage that forms a gel on ingestion. Food intake was not assessed; however the results show decreased hunger at 앒2 and 4 h after ingestion of the beverage (38). Our results suggest an immediate effect of the calcium-gelled, alginate-pectin that may be due to increased gastric volume and delayed postconsumptive

CALCIUM ALGINATE-PECTIN BEVERAGE REDUCED INTAKE

effects possibly because of delayed absorption of nutrients by gel lumps. Delayed absorption may cause stimulation of incretin responses that enhance satiety. We conducted exploratory analyses to examine the effects of meal in the statistical models; however, no significant main effects or interactions were found. Further work is needed to assess the time course of the effects of calcium-gelled alginates, with studies designed specifically for that purpose. The amount of fermentable, soluble fiber ingested in the beverage was fairly modest (1.0 or 2.8 g) compared with most highfiber supplements. Higher intake of dietary fiber is associated with lower body weight (39 – 41). Fiber may be associated with reduced body weight because of its effects on satiety and food intake, but the bolus amount of fiber required to manifest a benefit may be larger than most persons are willing to ingest. Soluble, viscous fiber may be more effective for reducing food intake and enhancing satiety than are insoluble forms, an effect postulated to be due to slowing of gastric emptying and prolongation of nutrient absorption in the gastrointestinal tract [see reviews by Pereira and Ludwig (42) and Burton-Freeman (43)]. Recent findings, however, call into question the hypothesis that soluble fiber slows gastric emptying. Large [7.4 g (38)] or small [1.7 g (44)] amounts of psyllium or doses of guar gum ranging from 2.5 to 4.5 g (45) were shown to have no effect on the rate of gastric emptying. French and Read (46) showed that 12 g of guar gum added to a high-fat meal delayed the return of hunger despite a trend for an enhanced rate of gastric emptying. Our findings uncovered an interaction between condition and dietary restraint. Dietary restraint, first introduced by Herman and Mack (47), is defined as the tendency to restrict food intake to control body weight. Herman and Polivy (48) postulated that restrained eaters develop anomalous eating patterns. Persons characterized as having high dietary restraint have been shown to report less hunger (49, 50), be less sensitive to the satiety value of dietary fat (51), more responsive to external cues (24), and less responsive to food palatability (52). Burton-Freeman (25) showed that restrained persons have a blunted cholecystokinin response to a preload, suggesting an aberrant endocrine response to the ingestion of food. Three self-rating questionnaires have been developed to assess dietary restraint (30, 53, 54). We choose the Eating Inventory, developed by Stunkard and Messick (30) and later revised by Westenhoefer et al (35), that delineates 2 types of dietary restraint: rigid restraint, characterized by a maladaptive, all-ornothing approach to dieting and weight loss, or flexible restraint, characterized by a more graduated, adaptive approach. Studies show that higher scores for rigid restraint were correlated with a higher BMI and less successful weight loss, whereas higher scores for flexible restraint were associated with a lower BMI and greater weight-loss success (35). We found differential effects of the gel beverage on food intake as a function of rigid restraint status, whereas no interaction was found for flexible restraint, concordant with Stunkard’s thesis that rigid but not flexible restraint is related to aberrant satiety responses. Subscale scores were highly correlated (r ҃ 0.88 for rigid and global restraint and r ҃ 0.79 for flexible and global restraint). Thus, our finding for a trend for an interaction of global restraint and condition may have been due to our inability to distinguish between subtypes of dietary restraint because of our use of the short version of the Eating Inventory (30). Future studies are needed that use the long version of the Eating Inventory (35), to better classify subjects

1601

into subcategories and to examine relations between restraint and satiety. In conclusion, the results show that ingesting calcium-gelled, alginate-pectin twice per day reduced spontaneous food intake in overweight and obese women. The effect was more pronounced in women with low rigid restraint and evident only for the 2.8-g gel beverage. Further work is needed to determine whether longer-term consumption or increased frequency of consumption or both would reduce energy intake sufficiently to affect body weight. Foods containing strong-gelling calcium and alginate-pectin that is activated in the stomach may be a useful adjunct to current behavioral approaches for weight loss or the maintenance of weight loss. The effects of foods designed to enhance satiety are likely to be modulated by dietary restraint and more successful for persons with low rigid restraint. It is unknown whether therapeutic approaches can be used to change a person’s dietary restraint style from rigid to flexible. Combining such efforts with satiety-enhancing foods or beverages is worthy of further investigation and may lead to substantial improvements to current treatment approaches for controlling body weight. The author’s responsibilities were as follows—CLP (principal investigator): designed the study, performed the data analyses, prepared the manuscript for publication, and supervised subject recruitment, testing, and data collection; AEM: coordinated the study, recruited subjects, conducted subject testing, and performed data input; RJP: assisted in subject recruitment, subject testing, and writing the manuscript: JLN: is an employee of McNeil Nutritionals and contributed to the writing of the manuscript as well as the intellectual development and design of the study. He did not partake in any aspect of subject testing or data analyses. None of the other authors had any personal or financial conflicts of interest with the study sponsor.

REFERENCES 1. Flegal KM, Carroll MD, Ogden CL, Johnson CL. Prevalence and trends in obesity among US adults, 1999 –2000. JAMA 2002;288:1723–7. 2. Flegal KM. Epidemiologic aspects of overweight and obesity in the United States. Physiol Behav 2005;86:599 – 602. 3. Yun S, Zhu BP, Black W, Brownson RC. A comparison of national estimates of obesity prevalence from the behavioral risk factor surveillance system and the national health and nutrition examination survey. Int J Obes (Lond) 2006;30:164 –70. 4. Ma GS, Li YP, Wu YF, et al. [The prevalence of body overweight and obesity and its changes among Chinese people during 1992 to 2002]. Zhonghua Yu Fang Yi Xue Za Zhi 2005;39:311–5 (in Chinese). 5. Mensink GB, Lampert T, Bergmann E. [Overweight and obesity in Germany 1984 –2003. ]. Bundesgesundheitsblatt Gesundheitsforschung Gesundheitsschutz 2005;48:1348 –56 (in German). 6. Vanasse A, Demers M, Hemiari A, Courteau J. Obesity in Canada: where and how many? Int J Obes (Lond) 2006;30:677– 83. 7. Proietto J. Can the obesity epidemic be stopped? Asia Pac J Clin Nutr 2005;14(suppl):S32. 8. Musaiger AO. Overweight and obesity in the Eastern Mediterranean Region: can we control it? East Mediterr Health J 2004;10:789 –93. 9. Seidell JC. Obesity, insulin resistance and diabetes–a worldwide epidemic. Br J Nutr 2000;83(suppl):S5–S8. 10. Flegal KM. Estimating the impact of obesity. Soz Praventivmed 2005; 50:73– 4. 11. Boyle JP, Honeycutt AA, Narayan KM, et al. Projection of diabetes burden through 2050: impact of changing demography and disease prevalence in the U.S. Diabetes Care 2001;24:1936 – 40. 12. Finkelstein E, Fiebelkorn C, Wang G. The costs of obesity among fulltime employees. Am J Health Promot 2005;20:45–51. 13. Li Z, Bowerman S, Heber D. Health ramifications of the obesity epidemic. Surg Clin North Am 2005;85:681–701,v. 14. Khaitan L, Smith CD. Obesity in the United States: is there a quick fix? Pros and cons of bariatric surgery from the adult perspective. Curr Gastroenterol Rep 2005;7:451– 4.

1602

PELKMAN ET AL

15. Bray GA. Drug treatment of obesity. Psychiatr Clin North Am 2005;28: 193–217,ix–x. 16. Vivero LE, Anderson PO, Clark RF. A close look at fenfluramine and dexfenfluramine. J Emerg Med 1998;16:197–205. 17. McCann UD, Seiden LS, Rubin LJ, Ricaurte GA. Brain serotonin neurotoxicity and primary pulmonary hypertension from fenfluramine and dexfenfluramine. A systematic review of the evidence. JAMA 1997; 278:666 –72. 18. Martin JA, Pandolfino JE. Gastrointestinal complications of bariatric surgery. Curr Gastroenterol Rep 2005;7:321– 8. 19. Rolls BJ, Bell EA, Castellanos VH, Chow M, Pelkman CL, Thorwart ML. Energy density but not fat content of foods affected energy intake in lean and obese women. Am J Clin Nutr 1999;69:863–71. 20. Rolls BJ, Castellanos VH, Halford JC, et al. Volume of food consumed affects satiety in men. Am J Clin Nutr 1998;67:1170 –7. 21. Bell EA, Castellanos VH, Pelkman CL, Thorwart ML, Rolls BJ. Energy density of foods affects energy intake in normal-weight women. Am J Clin Nutr 1998;67:412–20. 22. Rolls BJ, Roe LS, Beach AM, Kris-Etherton PM. Provision of foods differing in energy density affects long-term weight loss. Obes Res 2005;13:1052– 60. 23. Brunstrom JM, Yates HM, Witcomb GL. Dietary restraint and heightened reactivity to food. Physiol Behav 2004;81:85–90. 24. Ogden J, Wardle J. Cognitive restraint and sensitivity to cues for hunger and satiety. Physiol Behav 1990;47:477– 81. 25. Burton-Freeman B. Sex and cognitive dietary restraint influence cholecystokinin release and satiety in response to preloads varying in fatty acid composition and content. J Nutr 2005;135:1407–14. 26. Geliebter A, Gluck ME, Hashim SA. Plasma ghrelin concentrations are lower in binge-eating disorder. J Nutr 2005;135:1326 –30. 27. Geliebter A, Yahav EK, Gluck ME, Hashim SA. Gastric capacity, test meal intake, and appetitive hormones in binge eating disorder. Physiol Behav 2004;81:735– 40. 28. Beck AT, Beamesderfer A. Assessment of depression: the depression inventory. Mod Probl Pharmacopsychiatry 1974;7:151– 69. 29. Garner DM, Olmsted MP, Bohr Y, Garfinkel PE. The eating attitudes test: psychometric features and clinical correlates. Psychol Med 1982; 12:871– 8. 30. Stunkard AJ, Messick S. The three-factor eating questionnaire to measure dietary restraint, disinhibition and hunger. J Psychosom Res 1985; 29:71– 83. 31. Gormally J, Black S, Daston S, Rardin D. The assessment of binge eating severity among obese persons. Addict Behav 1982;7:47–55. 32. Pennington JAT. Bowes and Church’s food values of portions commonly used. 17th ed. Philadelphia, PA: Lippincott Williams & Wilkins, 1998. 33. Rolls BJ, Shide DJ, Thorwart ML, Ulbrecht JS. Sibutramine reduces food intake in non-dieting women with obesity. Obes Res 1998;6:1–11. 34. Anderson GH, Catherine NL, Woodend DM, Wolever TM. Inverse association between the effect of carbohydrates on blood glucose and subsequent short-term food intake in young men. Am J Clin Nutr 2002; 76:1023–30. 35. Westenhoefer J, Stunkard AJ, Pudel V. Validation of the flexible and rigid control dimensions of dietary restraint. Int J Eat Disord 1999;26: 53– 64.

36. Hoad CL, Rayment P, Spiller RC, et al. In vivo imaging of intragastric gelation and its effect on satiety in humans. J Nutr 2004;134:2293–300. 37. Marciani L, Gowland PA, Spiller RC, et al. Effect of meal viscosity and nutrients on satiety, intragastric dilution, and emptying assessed by MRI. Am J Physiol Gastrointest Liver Physiol 2001;280:G1227–33. 38. Rigaud D, Paycha F, Meulemans A, Merrouche M, Mignon M. Effect of psyllium on gastric emptying, hunger feeling and food intake in normal volunteers: a double blind study. Eur J Clin Nutr 1998;52:239 – 45. 39. Miller WC, Niederpruem MG, Wallace JP, Lindeman AK. Dietary fat, sugar, and fiber predict body fat content. J Am Diet Assoc 1994;94: 612–5. 40. Alfieri MA, Pomerleau J, Grace DM, Anderson L. Fiber intake of normal weight, moderately obese and severely obese subjects. Obes Res 1995; 3:541–7. 41. Ludwig DS, Pereira MA, Kroenke CH, et al. Dietary fiber, weight gain, and cardiovascular disease risk factors in young adults. JAMA 1999; 282:1539 – 46. 42. Pereira MA, Ludwig DS. Dietary fiber and body-weight regulation. Observations and mechanisms. Pediatr Clin North Am 2001;48:969 – 80. 43. Burton-Freeman B. Dietary fiber and energy regulation. J Nutr 2000; 130(suppl):272S–5S. 44. Frost GS, Brynes AE, Dhillo WS, Bloom SR, McBurney MI. The effects of fiber enrichment of pasta and fat content on gastric emptying, GLP-1, glucose, and insulin responses to a meal. Eur J Clin Nutr 2003;57:293– 8. 45. van Nieuwenhoven MA, Kovacs EM, Brummer RJ, WesterterpPlantenga MS, Brouns F. The effect of different dosages of guar gum on gastric emptying and small intestinal transit of a consumed semisolid meal. J Am Coll Nutr 2001;20:87–91. 46. French SJ, Read NW. Effect of guar gum on hunger and satiety after meals of differing fat content: relationship with gastric emptying. Am J Clin Nutr 1994;59:87–91. 47. Herman CP, Mack D. Restrained and unrestrained eating. J Pers 1975; 43:647– 60. 48. Herman CP, Polivy J. Restrained eating. In: Stunkard AJ, ed. Obesity. Philadelphia, PA: Saunders, 1980. 49. Burley VJ, Leeds AR, Blundell JE. The effect of high and low-fibre breakfasts on hunger, satiety and food intake in a subsequent meal. Int J Obes 1987;11(suppl):87–93. 50. Smith CF, Geiselman PJ, Williamson DA, Champagne CM, Bray GA, Ryan DH. Association of dietary restraint and disinhibition with eating behavior, body mass, and hunger. Eat Weight Disord 1998;3:7–15. 51. Rolls BJ, Kim-Harris S, Fischman MW, Foltin RW, Moran TH, Stoner SA. Satiety after preloads with different amounts of fat and carbohydrate: implications for obesity. Am J Clin Nutr 1994;60:476 – 87. 52. Yeomans MR, Tovey HM, Tinley EM, Haynes CJ. Effects of manipulated palatability on appetite depend on restraint and disinhibition scores from the Three-Factor Eating Questionnaire. Int J Obes Relat Metab Disord 2004;28:144 –51. 53. Herman CP, Polivy J. Anxiety, restraint, and eating behavior. J Abnorm Psychol 1975;84:66 –72. 54. van Strien T, Fritjers JER, Bergers GP, Defares PB. The Dutch Eating Behavior Questionnaire (DEBQ) for assessment of restrained, emotional and external eating behavior. Int J Eat Disord 1986;5:295–315.

Postprandial ghrelin, cholecystokinin, peptide YY, and appetite before and after weight loss in overweight women with and without polycystic ovary syndrome1–3 Lisa J Moran, Manny Noakes, Peter M Clifton, Gary A Wittert, Carel W Le Roux, Mohammed A Ghatei, Stephen R Bloom, and Robert J Norman ABSTRACT Background: Polycystic ovary syndrome (PCOS) is a common condition associated with obesity and with reproductive and metabolic dysfunction. Abnormalities in appetite regulation in PCOS patients may contribute to difficulties in weight management. Objective: We aimed to examine appetite, appetite hormones, and ad libitum food consumption before and after weight loss in overweight women with and without PCOS. Design: Overweight age- and weight-matched women with (n ҃ 14) and without (n ҃ 14) PCOS undertook an 8-wk energy-restricted diet (5185.3 앐 141.6 kJ/d). At baseline and study end, subjects consumed a test meal (936 kJ; 25% of energy from protein, 9% from fat, and 67% from carbohydrate). Subjective appetite and circulating glucose, insulin, ghrelin, cholecystokinin, and peptide YY were assessed at 0, 15, 30, 45, 60, 90, 120, and 180 min. A mixed buffet lunch was then offered to assess ad libitum food intake. Results: Weight loss (4.2 앐 3.9 kg) did not differ significantly between the 2 groups. Women with PCOS had significantly (P ҃ 0.023) lower ghrelin concentrations before and after weight loss than did women without PCOS. The degree of postprandial ghrelin suppression was lower at weeks 0 (P ҃ 0.048) and 8 (P ҃ 0.069) in women with PCOS than in women without PCOS. There were no significant differences between the 2 groups in appetite responses, buffet consumption, or fasting or postprandial peptide YY and cholecystokinin before or after weight loss. Conclusions: PCOS was associated with lower fasting ghrelin and a smaller postprandial ghrelin suppression both before and after weight loss but was not associated with other postprandial gut peptides, subjective satiety, or food intake. It is not clear whether appetite regulation is impaired in PCOS. Am J Clin Nutr 2007;86: 1603–10. KEY WORDS Appetite, obesity, polycystic ovary syndrome, weight loss, ghrelin, cholecystokinin, peptide YY INTRODUCTION

Polycystic ovary syndrome (PCOS) is a common endocrine condition in women of reproductive age; it is associated with menstrual dysfunction, hyperandrogenism, a greater risk of developing type 2 diabetes, and an adverse cardiovascular disease risk profile. Insulin resistance (IR) is implicated in its development through the insulin stimulation of ovarian androgen production and the reduction in hepatic synthesis of sex hormone–

binding globulin (SHBG) (1). Obesity—in particular, central obesity—is present in 10 – 65% of Western women with PCOS (2), and its presence worsens the associated IR and metabolic and endocrine features; weight loss reduces abdominal fat and IR and improves menstrual function (3). We previously showed that overweight women with PCOS have lower postprandial satiety and higher postprandial hunger before and after weight loss than do weight-matched controls (4). There is some evidence that disturbances in appetite regulation in PCOS could account for these reported discrepancies in hunger and satiety. Ghrelin is proposed as an acute meal initiator; ghrelin concentrations increase preprandially and decrease postprandially (5), and ghrelin administration stimulates hunger and food intake (6). The reduction in postprandial ghrelin was less in obese than in lean subjects (7–9) and in overweight PCOS patients than in weight-matched persons without PCOS (4). Cholecystokinin is released from the small intestine postprandially, primarily in response to duodenal protein and fat, and it inhibits gastric emptying and reduces meal size and calorie intake in humans (10). Postprandial cholecystokinin has been reported either to be greater in obese than in lean subjects or to be unaffected by body weight status (11, 12). In overweight women with PCOS, a lower postprandial cholecystokinin response than in weight-matched controls was observed, which further supports impaired appetite regulation in PCOS (13). Peptide YY (PYY), a peptide that is 1 From the Research Centre for Reproductive Health, Discipline of Obstetrics and Gynaecology, University of Adelaide, Adelaide, Australia (LJM and RJN); the Commonwealth Scientific and Industrial Research Organisation (CSIRO) Human Nutrition, Adelaide, Australia (LJM, MN, and PMC); the Discipline of Medicine, University of Adelaide, Adelaide, Australia (WG); and the Department of Metabolic Medicine, Hammersmith Hospital, Imperial College London, London, United Kingdom (WLRC, MAG, and SRB). 2 Supported by a National Health and Medical Research Council Program Grant (to RJN) and by The University of Adelaide Faculty of Health Sciences Small Research Grants Scheme, Colin Matthews Research Grants for Clinically Based Research, and CSIRO Human Nutrition. Unilever donated the meal replacements used in the weight-loss component of the study; McDonalds Australia donated the premeal tolerance test food. 3 Address reprint requests to RJ Norman, Discipline of Obstetrics and Gynaecology, University of Adelaide, 6th Floor, Medical School North, Adelaide, SA5005, Australia. E-mail: [email protected]. Received May 30, 2007. Accepted for publication August 8, 2007.

Am J Clin Nutr 2007;86:1603–10. Printed in USA. © 2007 American Society for Nutrition

1603

1604

MORAN ET AL

synthesized in the gastrointestinal tract, increases postprandially, increases satiety, and reduces food intake. It has additional functions, including inhibition or reduction of gallbladder secretion, gut motility and pancreatic secretion (14, 15). Reductions in total fasting PYY or impaired postprandial PYY (15, 16) or no changes in fasting PYY (17, 18) have been observed in overweight subjects, and it is not known whether PYY is differentially regulated in PCOS. Although fasting ghrelin concentrations rise and the postprandial ghrelin response improves with weight loss, these changes may be impaired in women with PCOS (4). Fasting PYY increases (19) or decreases (17) after diet-induced weight loss, and fasting or postprandial cholecystokinin is unchanged by weight loss (20). The effect of weight loss on cholecystokinin and PYY has not yet been examined in persons with PCOS. The aim of this study was therefore to examine fasting and postprandial subjective appetite, appetite hormones (ie, ghrelin, PYY, and cholecystokinin), and ad libitum buffet meal consumption before and after weight loss in overweight women with and without PCOS. SUBJECTS AND METHODS

Subjects and recruitment Overweight age- and weight-matched women with (n ҃ 14) and without (n ҃ 14) PCOS (age: 32.3 앐 5.9 and 36.2 앐 4.5 y, respectively; weight: 94.5 앐 19.8 and 94.9 앐 15.4 kg, respectively) were recruited through public advertisement. PCOS was diagnosed according to the Rotterdam consensus group as previously described (21, 22). Exclusion criteria were pregnancy, breastfeeding, body mass index (BMI; in kg/m2) 쏝 25, type 2 diabetes mellitus, and the use of oral contraceptives, endocrine hormonal treatment, or insulin-sensitizing agents (subjects were required to cease oral contraceptives 4 wk and hormonal treatment or insulin-sensitizing agents 2 wk before commencement of

the study). The PCOS patients and control subjects were matched for age, BMI, and smoking status. All subjects gave written informed consent. We certify that all applicable institutional and governmental regulations concerning the ethical use of human volunteers were followed during this research. The study was approved by the human ethics committees of the Commonwealth Scientific and Industrial Research Organisation Division of Human Nutrition at The Royal Adelaide Hospital and of the Women’s and Children’s Hospital of South Australia. Study design and dietary treatment The study was conducted on an outpatient basis over a span of 8 wk. Subjects followed an energy-restricted diet in which 2 meals/d were replaced with commercially available meal replacements (Slimfast; Unilever Australasia, Epping, Australia), a supply of which was provided every 2 wk (21). The aim of the dietary intervention was to induce an energy deficit of 앒30% through daily consumption of 2 meal replacements (1800 kJ), a low-fat evening meal, and 욷5 servings of fruit and vegetables (3500 kJ). The dietary composition of the intervention is described in Table 1. Nutrient intakes were calculated with the use of DIET 4/NUTRIENT CALCULATION software (version 4; Xyris Software, Highgate Hill, Australia) by using data from Australian food composition tables. Nutritional intake was assessed from 3-d consecutive dietary food records (1 weekday and 2 weekend days) collected every 2 wk and daily dietary checklists. Dietary compliance was determined by subject adherence to the meal replacement regimen. Subjects attended the clinic every 2 wk and were weighed while wearing light clothes but no shoes (model AMZ14 Mettler scales; A&D Mercury, Kinomoto, Japan). At weeks 0 and 8, waist circumference and total fat mass and total fat-free mass (determined by bioelectrical impedance

TABLE 1 Dietary intake data for subjects at baseline and during the study intervention1 PCOS patients

Energy (MJ) Protein (g) (% of total energy) Carbohydrate (g) (% of total energy) Fat (g) (% of total energy) SFA (% of total energy) MUFA (% of total energy) PUFA (% of total energy) Fiber Cholesterol

Control subjects

Baseline2 (n ҃ 13)

Intervention3 (n ҃ 13)

Baseline2 (n ҃ 13)

Intervention3 (n ҃ 14)

7.3 앐 0.9

5.2 앐 0.2

8.1 앐 0.9

5.2 앐 0.2

93.6 앐 10.8 21.0 앐 0.7

75.8 앐 3.1 23.3 앐 0.5

101.5 앐 11.2 20.1 앐 0.8

69.1 앐 1.8 21.7 앐 0.8

176.0 앐 22.0 40.8 앐 1.2

159.6 앐 6.0 52.2 앐 1.2

200.7 앐 21.1 42.7 앐 1.9

151.7 앐 8 50.0 앐 1.8

73.6 앐 9.4 37.4 앐 1.2 15.3 앐 0.5 13.3 앐 0.5 5.3 앐 0.5 17.3 앐 2.4 319.9 앐 33.0

32.3 앐 42.4 22.8 앐 1.4 12.5 앐 0.9 10.9 앐 1.0 6.4 앐 0.6 20.0 앐 1.1 147.6 앐 9.5

82.4 앐 10.9 36.8 앐 1.5 15.2 앐 0.9 13.0 앐 0.5 5.2 앐 0.4 21.4 앐 2.1 318.6 앐 48.4

37.4 앐 3.5 26.5 앐 1.8 14.2 앐 1.5 13.4 앐 1.3 7.0 앐 0.6 18.8 앐 0.8 131.5 앐 9.8

All values are x៮ 앐 SEM. PCOS, polycystic ovary syndrome; SFA, saturated fatty acids; MUFA, monounsaturated fatty acids; PUFA, polyunsaturated fatty acids. Measurements were made at the week 0 visit (food-frequency questionnaire) or during the study (3-d food records) and were assessed by using a one-factor ANOVA with PCOS status as the fixed factor. 2 Assessed with a food-frequency questionnaire. 3 Assessed with 3-d food records. 1

APPETITE HORMONES IN POLYCYSTIC OVARY SYNDROME

analysis) were measured as previously described (21), and overnight fasting venous blood samples were taken for measurement of testosterone and SHBG. Dietary restraint, disinhibition, and hunger were measured at week 0 by using the Three-Factor Eating Questionnaire (TFEQ) (23). Prestudy dietary intake (the previous 6 mo) was assessed with the use of a food-frequency questionnaire from the Anti-Cancer Foundation. Exercise was assessed at weeks 0 and 8 by using a 7-d physical activity record (24). For the 2 wk before study commencement, subjects weighed themselves daily to ensure weight stability, defined as a weight change of 울2% of initial body weight. At weeks 0 and 8, subjects underwent a meal tolerance test (MTT). All subjects consumed the same meal the evening before the test (3820 kJ; 20% of energy from protein, 17% from fat, and 62% from carbohydrate) and refrained from consuming alcohol for 24 h. A cannula was inserted into a forearm vein, and an overnight fasting venous blood sample was taken between 0800 and 1000 for measurement of plasma glucose and insulin concentrations and serum ghrelin, cholecystokinin, and PYY concentrations. Subjects then completed a validated visual analogue scale (VAS) questionnaire to assess subjective hunger as previously described (25). The change in ratings from baseline was quantified. Subjects consumed a liquid preload of Slimfast [325 mL, 936 kJ, 12 g protein (25% of energy from protein), 2 g fat (9% of energy from fat), and 35 g carbohydrate (67% of energy from carbohydrate)] within 5 min; additional blood samples were taken and VAS questionnaires were completed at 15, 30, 45, 60, 90, 120, and 180 min after meal consumption. At 180 min, subjects were given a mixed buffet-style lunch (12.1 MJ; 15% of energy from protein, 44% from fat, and 41% from carbohydrate); each subject served his or her own meal from designated portions of the foods and ate until satisfied over a 30-min period. Each food was weighed to the nearest gram with the use of digital scales before and after eating. Total (glucose, insulin, cholecystokinin, and PYY), incremental (ghrelin), and net (VAS) areas under the curve (AUCs) during the 3-h MTT were calculated geometrically by using the trapezoidal rule (26). Biochemical measurements Blood for measurement of serum concentrations was collected in tubes with no additives and allowed to clot at room temperature for 30 min. Blood for measurement of plasma concentrations was collected in tubes containing sodium fluoride and EDTA (glucose) or potassium and EDTA and aprotinin (aprotinin concentration: 500 KIU/mL blood; Roche Diagnostics, Indianapolis, IN) (ghrelin, cholecystokinin, and PYY) and stored on ice. Serum and plasma were stored at Ҁ80 °C. Serum SHBG and total testosterone (bound and unbound) (3), total ghrelin (4), insulin and glucose (21), cholecystokinin (11), and total PYY (1–36 and 3–36) (16) were measured as described previously. The homeostatic model assessment (HOMA) was used as a surrogate measure from which to calculate insulin sensitivity according to the following equation (27):

Insulin sensitivity ⫽ [fasting insulin (mU/L) ⫻ fasting glucose (mmol/L)]/22.5 (1) The free androgen index (testosterone/SHBG ҂ 100) and equilibrium-binding equations were used for estimation of free testosterone (28). Biochemical assays were performed in a single

1605

assay at the completion of the study, and all samples for individuals were analyzed in the same assay. Statistical analysis Parametric data were presented as means 앐 SEMs except where specified. Nonparametric data (TFEQ) were presented as median, minimum, and maximum. When data were nonnormally distributed, data were log transformed for analysis. Results are presented for 28 subjects (n ҃ 14 for both the PCOS patients and control subjects); for postprandial glucose, insulin, ghrelin, PYY, and cholecystokinin, the numbers differed slightly (n ҃ 14 and 13 in the PCOS patients and control subjects, respectively) because of incomplete data. Two-tailed statistical analysis was performed with the use of SPSS for WINDOWS software (version 14.0; SPSS Inc, Chicago, IL) with statistical significance set at an ␣ level of P 쏝 0.05. Baseline data were assessed by using a one-factor analysis of variance for parametric data and a Kruskal-Wallis test for nonparametric data (TFEQ). Comparisons between time points were assessed by using a general linear model repeated-measures analysis of variance with PCOS diagnosis as the between-subject factor. In the event of an interaction, Bonferroni post hoc pairwise comparisons were performed. Relations between variables were examined by using bivariate and partial correlations and analysis of covariance after adjustment for weight, fasting and postprandial insulin, and total energy intake. This study had 80% power to detect a preweight-loss difference of 75.9 pmol/L in fasting ghrelin and a change of 31.3 pmol/L in fasting ghrelin during weight loss between subjects with and without PCOS (P 쏝 0.05 for both) (4). RESULTS

Subjects, physical activity, dietary intake, weight loss, body composition, and reproductive hormones Baseline characteristic of the subjects are shown in Table 1. For the TFEQ data, there were no differences in the baseline scores for dietary restraint (median: 9.5; range: 3–16), disinhibition (10; 3–16), or hunger (7; 0 –14) between the PCOS patients and the control subjects. Activity levels did not differ significantly between the 2 groups at week 0 and did not change throughout the study. There were no significant differences in energy or macronutrient intake between the 2 groups at baseline or during the study intervention (Table 2). Decreases in weight (x៮ 앐 SEM: 4.2 앐 3.9 kg or 4.3 앐 3.8%), waist circumference, total fat mass, total fat-free mass, free testosterone, free androgen index, and testosterone and increases in SHBG occurred in all subjects with no differential effect of PCOS status (Table 2). Insulin and glucose homeostasis There was no effect of PCOS status on changes in fasting glucose and no change in fasting glucose over the study duration. There was an interaction between postprandial glucose and PCOS status (P ҃ 0.017) so that the postprandial glucose response decreased significantly (P ҃ 0.043) only in the PCOS patients, and no changes in postprandial glucose occurred with weight loss in the control subjects (Figure 1). Decreases in HOMA (13.6%; P ҃ 0.010), fasting insulin (15.9%; P ҃ 0.004), AUC insulin (20.0 앐 3.9%; P 쏝 0.001), and postprandial insulin (P 쏝 0.001) did not differ significantly between the 2 groups. There was no difference between the 2 groups in the magnitude

1606

MORAN ET AL

TABLE 2 Weight, body composition, lipids, glucose, and reproductive hormones before and after 8 wk of energy restriction1

Weight (kg)2 PCOS patients Control subjects BMI (kg/m2) PCOS patients Control subjects Waist circumference (cm)2 PCOS patients Control subjects Total fat-free mass (kg)2 PCOS patients Control subjects Total fat mass (kg)2 PCOS patients Control subjects Fasting glucose (mmol/L) PCOS patients Control subjects Fasting insulin (mU/L)3 PCOS patients Control subjects HOMA2,3 PCOS patients Control subjects Testosterone (nmol/L)2,3 PCOS patients Control subjects SHBG (nmol/L)3 PCOS patients Control subjects Free androgen index2,3 PCOS patients Control subjects Free testosterone (pmol/L)2,3 PCOS patients Control subjects Glucose AUC (mmol/L/180 min) PCOS patients Control subjects Insulin AUC (mU 䡠 LҀ1 䡠 180 minҀ1)2,3 PCOS patients Control subjects Ghrelin AUC (pmol/L/180 min) PCOS patients Control subjects CCK AUC (pmol 䡠 LҀ1 䡠 180 minҀ1) PCOS patients Control subjects PYY AUC (pmol 䡠 LҀ1 䡠 180 minҀ1) PCOS patients Control subjects

Week 0

Week 8

94.5 앐 5.3 94.9 앐 4.1

90.6 앐 4.8 90.2 앐 3.9

35.3 앐 1.5 35.3 앐 1.3

34.8 앐 1.5 34.4 앐 1.7

113.9 앐 4.3 111.0 앐 3.1

106.9 앐 4.4 103.9 앐 3.1

59.5 앐 3.4 61.1 앐 2.5

58.3 앐 3.2 58.9 앐 2.3

34.5 앐 2.5 34.5 앐 2.0

31.8 앐 2.1 31.3 앐 1.8

5.3 앐 0.2 5.1 앐 0.2

5.3 앐 0.1 5.1 앐 0.2

22.4 앐 3.8 11.8 앐 1.8

18.3 앐 3.8 8.5 앐 1.5

5.5 앐 1.0 2.8 앐 0.5

4.4 앐 0.9 2.0 앐 0.4

3.2 앐 0.3 2.1 앐 0.1

2.7 앐 0.2 1.8 앐 0.1

20.5 앐 2.9 24.9 앐 2.1

21.8 앐 3.0 28.7 앐 3.0

22.0 앐 5.0 9.3 앐 1.2

17.3 앐 4.9 7.3 앐 1.0

81.4 앐 9.9 45.0 앐 3.9

64.6 앐 9.0 36.5 앐 3.3

989 앐 36 935 앐 31

954 앐 26 912 앐 35

11842 앐 2543 6702 앐 890

8818 앐 1466 5039 앐 690

2118 앐 480 3941 앐 919

1943 앐 693 2446 앐 422

369 앐 112 290 앐 69

412 앐 94 304 앐 91

3748 앐 279 3295 앐 194

4036 앐 390 3139 앐 256

All values are x៮ 앐 SEM. PCOS, polycystic ovary syndrome; SHBG, sex hormone– binding globulin; HOMA, homeostasis model assessment; AUC, area under the curve; CCK, cholecystokinin; PYY, peptide YY. n ҃ 14 in each group except for weight; waist circumference; total fat-free mass; total fat mass; and AUC glucose, insulin, ghrelin, CCK and PYY (n ҃ 14 and 13 in the PCOS patients and control subjects, respectively. Data were assessed by using a one-factor ANOVA for week 0 or week 8 data with PCOS status as the fixed factor and repeated-measures ANOVA for changes with time as the within-subject factor and PCOS status as the between-subject factor. For conversion of glucose values from mmol/L to mg/dL, multiply by 18. For conversion of insulin values from mU/L to pmol/L, multiply by 6.95. For conversion of ghrelin values from pmol/L to pg/mL, multiply by 3.38. 2 P 쏝 0.05 for effect of time for all subjects (weeks 0 – 8). 3 P 울 0.05 for difference between subjects with and without PCOS at week 0 and week 8. There was no time ҂ PCOS status interaction for any variable. 1

APPETITE HORMONES IN POLYCYSTIC OVARY SYNDROME 7.0

*

Glucose (mmol/L)

6.0 5.0

3.0 2.0

0.0 0

15

30 45 60 Time (min)

90

120 180

180 160 140 Insulin (mU/L)

of the change, although subjects without PCOS had lower fasting, AUC insulin, and postprandial insulin responses at all time points (Figure 1, Table 1). Fasting ghrelin

4.0

1.0

120 100

† †† **

80 60 40 20 0 0

15

30 45 60 Time (min)

90

120 180

600 500 Ghrelin (pg/mL)

1607

400

Fasting ghrelin was significantly higher in the PCOS patients than in the control subjects at week 0 (486.2 앐 41.1 and 318.1 앐 44.0 pg/mL, respectively; P ҃ 0.01), and there was a trend for it to be higher at week 8 (463.5 앐 36.8 and 347.0 앐 56.0 pg/mL, respectively; P ҃ 0.099). There was no change in fasting ghrelin after weight loss in either group. At weeks 0 and 8, fasting ghrelin in the PCOS patients and control subjects was correlated with fasting insulin (r ҃ Ҁ0.518, P ҃ 0.006 and r ҃ Ҁ0.515, P ҃ 0.006, respectively), HOMA (r ҃ Ҁ0.488, P ҃ 0.008 and r ҃ Ҁ0.508, P ҃ 0.007, respectively), free androgen index (r ҃ Ҁ0.553, P ҃ 0.003 and r ҃ Ҁ0.489, P ҃ 0.010, respectively), and free testosterone (r ҃ Ҁ0.538, P ҃ 0.004 and r ҃ Ҁ0.471, P ҃ 0.013, respectively). After control for weight, all of the above relations remained except fasting ghrelin and HOMA at week 0 (r ҃ Ҁ0.369, P ҃ 0.063). Postprandial ghrelin The PCOS patients had significantly lower ghrelin concentrations at all time points after preload consumption before and after weight loss than did the control subjects (P ҃ 0.023 for betweensubject effect of PCOS) (Figure 1). There was a trend for an effect of weight loss on changes in test meal ghrelin (P ҃ 0.097) and a significant effect of PCOS status on test meal ghrelin (minuteby-PCOS status effect, P ҃ 0.023). In the PCOS patients, the postprandial ghrelin response was more impaired (as indicated by a lesser postprandial decrease) at week 0 and tended to be more impaired at week 8 than in the control subjects (P ҃ 0.048 and 0.069, respectively, for time-by-PCOS status effect). The above differences in ghrelin between subjects with and without PCOS were removed on adjustment for fasting or postprandial insulin at week 0 or 8.

300

Fasting and postprandial PYY and cholecystokinin ‡

200 100 0 0

15

30

45

60

90

120 180

Time (min)

FIGURE 1. Mean (앐SEM) glucose, insulin, and ghrelin concentrations at baseline and 15, 30, 45, 60, 90, 120, and 180 min after the ingestion of a test meal at weeks 0 (—) and 8 (- - -). Week 0 and week 8 data were compared by repeated-measures ANOVA with week and blood sampling time as withinsubject factors and polycystic ovary syndrome (PCOS) status as the betweensubject factor. ⽧, subjects with PCOS, n ҃ 14; 䡺, subjects without PCOS, n ҃ 13. To convert glucose values from mmol/L to mg/dL, multiply by 18. To convert insulin values from mU/L to pmol/L, multiply by 6.95. To convert ghrelin values from pmol/L to pg/mL, multiply by 3.38. *Significant week ҂ min ҂ PCOS status effect, P ҃ 0.017. **Significant effect of time from week 0 to week 8 for fasting (P ҃ 0.004) and postprandial (P 쏝 0.001) insulin. † Significant between-subject effect of PCOS status, P 쏝 0.05. ††Significant difference at 0 min between subjects with and subjects without PCOS at week 0 (P ҃ 0.042) and week 8 (P ҃ 0.009). ‡Significant min ҂ PCOS status effect (P ҃ 0.023).

There was no significant effect of PCOS status on baseline cholecystokinin in either the PCOS patients or the control subjects (1.1 앐 0.3 and 0.7 앐 0.2 pmol/L, respectively; P ҃ 0.276) and PYY (17.2 앐 1.6 and 14.0 앐 0.7 pmol/L, respectively; P ҃ 0.091); on post-weight-loss cholecystokinin (1.0 앐 0.3 and 0.7 앐 0.2 pmol/L, respectively; P ҃ 0.302) and PYY (15.9 앐 2.8 and 13.6 앐 1.0 pmol/L, respectively; P ҃ 0.381); or on the changes in fasting cholecystokinin and PYY with weight loss. PCOS status had no significant effect on postprandial cholecystokinin or PYY or on changes in postprandial cholecystokinin or PYY with weight loss (Table 1). There was no significant effect of weight loss on fasting cholecystokinin (P ҃ 0.919) or PYY (P ҃ 0.404) or postprandial cholecystokinin (P ҃ 0.440) or PYY (P ҃ 0.210) in either group. Cholecystokinin and PYY increased in both groups after test meal consumption at both week 0 and week 8 (P 쏝 0.001). The change in weight was significantly and negatively correlated with the change in AUC PYY (r ҃ Ҁ0.453, P ҃ 0.018) or CCK (r ҃ Ҁ0.443, P ҃ 0.021) with weight loss. Visual analogue scores PCOS patients and control subjects had an increase in their sensation of fasting fullness with weight loss (29.4 앐 3.5 and

1608

MORAN ET AL

38.4 앐 4.8 mm, respectively; P ҃ 0.033) and a decrease in their sensation of postprandial fullness with weight loss (P ҃ 0.005), as reflected in the AUC (3326.5 앐 630.0 and 1130.9 앐 686.4 mm/180 min, respectively; P ҃ 0.009). No significant reduction in fullness after weight loss was observed with adjustment for the change in the total amount of energy consumed at the buffet meal. There were no other significant differences in fasting or postprandial VAS measures before or after weight loss or between the PCOS patients and control subjects (data not shown). None of the postprandial appetite hormones correlated with each other or with VAS. Buffet dietary intake The energy intake at the buffet meal did not differ significantly between the PCOS patients and control subjects before (4625.8 앐 458.6 and 5476.3 앐 354.6 kJ, respectively) or after (3921.4 앐 355.3 and 4314.6 앐 330.1 kJ, respectively) weight loss. Macronutrient intake and fatty acid profile also did not differ significantly before or after weight loss (data not shown). After weight loss, the amount of food eaten at the buffet meal decreased equivalently, by 919.0 앐 260.0 kJ, for all subjects (P ҃ 0.002); there were consequent reductions in total macronutrient intake but no significant changes in proportional macronutrient intake. The amount of food eaten at the buffet correlated significantly with weight (r ҃ 0.489, P ҃ 0.008), BMI (r ҃ 0.447, P ҃ 0.017), total FFM (r ҃ 0.498, P ҃ 0.007), and total FM (r ҃ 0.513, P ҃ 0.005) at week 0 but not at week 8. DISCUSSION

We have confirmed reports of lower fasting ghrelin concentrations (4, 29), and we also reported a smaller postprandial reduction in ghrelin in overweight women with PCOS than in control subjects. In contrast to previous findings (4), we observed no significant change in fasting ghrelin and only a trend for a change in postprandial ghrelin with weight loss. This discrepancy likely is due to the modest weight loss (4.2 kg), because no significant changes in ghrelin were observed after weight losses of 3% to 5% (30). Postmeal ghrelin suppression is related to the energy content of the meal (9), and it correlates with postmeal decreases in hunger and increases in satiety (31). The impairment in postprandial ghrelin secretion observed in obesity (7–9) may be related to the impairment of appetite regulation in overweight humans, which is consistent with reports of delayed satiation (32). Furthermore, positive energy balance may decrease the sensitivity of the central nervous system to ghrelin (33), which suggests that postprandial ghrelin and its regulatory role on appetite may be blunted in obesity. The improvement in postprandial ghrelin that follows weight loss suggests an improvement in appetite regulation, and this possibility is supported by the postweight-loss reduction in food consumption observed here and by other investigators (34). We observed an impaired postprandial ghrelin response in women with PCOS that was partially normalized by weight loss; this same observations was previously made by our group in conjunction with increased postprandial hunger (4). One proposed regulator of ghrelin is insulin, which could acutely suppress ghrelin secretion postprandially (35) and could chronically suppress it in hyperinsulinemic conditions such as obesity or PCOS (4, 36). A recent study found that, in obese children, postprandial ghrelin decreases were positively correlated with

insulin sensitivity and negatively correlated with postprandial insulin (35). A dose-dependent suppression of total ghrelin by hyperinsulinemia was also observed in subjects without type 2 diabetes, whereas supraphysiological insulin concentrations were required to suppress ghrelin concentrations in subjects with type 2 diabetes (37). The ability of insulin to suppress ghrelin may thus be altered by IR. This possibility is consistent with the lower ghrelin concentrations observed in the present study in persons with PCOS and in all subjects before weight loss and with amelioration of the differences in ghrelin between women with and without PCOS after adjustment for insulin concentrations. Furthermore, women with PCOS but without IR have fasting ghrelin concentrations similar to those of control subjects and higher than those of women with PCOS and IR (29). Thus, it is possible that abnormalities in appetite regulation will be more prevalent in women with PCOS and IR than in women with PCOS but not IR. Conversely, our previous study showed lower fasting ghrelin concentrations in women with PCOS than in weight- and insulin-matched control subjects, which suggests a role for PCOS independent of IR in regulating ghrelin (4). A variety of contributory factors to ghrelin regulation in PCOS— including adiponectin (38), fat–free mass (39), and androgens (40)—also warrant investigation. A greater degree of weight loss may be necessary to reduce insulin, androgen, or other factors sufficiently to induce changes in ghrelin. We report for the first time that fasting and postprandial PYY do not differ significantly in age- and weight-matched women with or without PCOS before or after weight loss. Postprandial PYY profiles are proposed to play a role in regulating both acute satiation and longer-term satiety (16, 41), and the similar postprandial PYY may account for the similar food intake between women with and without PCOS. We also report for the first time that postprandial PYY is not altered after diet-induced weight loss. Lower fasting and postprandial PYY concentrations observed previously in obese persons—potentially caused by decreased PYY synthesis or release— could cause reduced postprandial satiety and therefore potentially contribute to obesity (15, 16). Fasting total PYY increases with weight loss, and the increase is positively related to the degree of weight loss (19). The lack of a change in fasting or postprandial PYY with weight loss seen in the present study was unexpected; however, the intervention duration and degree of weight loss (4.2 kg) may have been inadequate to induce changes. Alternatively, this lack of a change may reflect increased central nervous system sensitivity to the effects of PYY or an adaptive response to attempts to restore energy balance. Food intake is similarly reduced after PYY 3–36 infusions in obese and lean subjects (15), which indicates that obese subjects may not suffer PYY resistance. The negative correlation between the degree of weight loss and a change in AUC PYY could also indicate that subjects with the greatest degree of weight loss potentially had a relative compensatory decrease in postprandial PYY; this possibility is supported by recent studies showing decreases in fasting (17) or postprandial (42) PYY after diet- or gastric bypass–induced weight loss. Unlike other investigators (13), we observed no differences in fasting or postprandial cholecystokinin between women with and without PCOS either before or after weight loss. It is not clear whether the similar cholecystokinin profiles of the women with and without PCOS are related to the consequent buffet meal energy intake, because the primary effects of cholecystokinin are an inhibition of gastric emptying, an increase in satiation, and a

APPETITE HORMONES IN POLYCYSTIC OVARY SYNDROME

reduction in meal size (43). However, a recent study suggested a potential role of cholecystokinin in satiety through a prolonged postprandial elevation of cholecystokinin and the possibility that cholecystokinin is a significant predictor of late postprandial appetite (44). Although we observed a significant negative correlation between the degree of weight loss and the change in AUC cholecystokinin in all subjects, fasting and postprandial cholecystokinin did not change with weight loss. This is consistent with previous reports of similar postprandial cholecystokinin profiles between aged-matched lean and obese subjects (11), no changes in postprandial cholecystokinin with diet-induced weight loss (20), and similar satiating effects of cholecystokinin infusions in lean and obese subjects (43). Conversely, postprandial cholecystokinin was higher in overweight than in lean males and females (12). Cholecystokinin is released postprandially from the small intestine, predominantly in response to duodenal protein and fat (10). The relatively low amount of protein or fat in our preload may not have provided a sufficient stimulus (25% of energy from protein, 67% from carbohydrate, and 9% from fat compared with 17% of energy from protein, 54% from carbohydrate, and 28% from fat) (13). The lack of observed differences in PYY may also be affected by the preload macronutrient composition, because postprandial PYY is stimulated more strongly by protein than by fat or carbohydrate (45). Furthermore, PYY is released in proportion to calories ingested (14), and the energy content of the preload (936 kJ) may not have been adequate to allow detection of subtle differences. Whereas altered ghrelin regulation in PCOS potentially indicates impaired appetite regulation and difficulty with weight management, these possibilities are not supported by the findings of similar cholecystokinin and PYY concentrations, subjective appetite, and buffet food intake in women with and without PCOS. In addition to its proposed role in appetite regulation, ghrelin has a variety of functions—including endocrine pancreatic function, glucose metabolism, inflammation, vasodilation, and ovarian function (46 – 48). It is possible that the lower ghrelin concentrations commonly observed in persons with PCOS represent the increased metabolic, diabetic, and reproductive dysfunction associated with the condition rather than abnormalities in appetite regulation. The preprandial increase in ghrelin also may occur as an anticipatory response to feeding rather than as a physiologic meal initiator (49, 50); this possibility is supported by a later onset of the postprandial ghrelin decrease in obese than in lean males, but similar energy intakes in the 2 groups (11). The observed differences in ghrelin may induce subtle alterations in appetite, which would be more detectable in a free-living environment with ad libitum energy intake. Selection of subjects with similar eating behaviors and performance of the MTT at defined stages of the menstrual cycle would further strengthen our observations, because variations in appetite or food intake, appetite hormones (specifically, cholecystokinin), insulin sensitivity, or glucose homeostasis have been reported over the course of the menstrual cycle (51–53) or between restrained and unrestrained eaters (54). Other potential limitations of the current study include the measurement of total ghrelin, which comprises biologically active (Ser3 octanolyated) and inactive ghrelin and which may not be optimal for the assessment of ghrelin regulation in diverse conditions such as obesity, weight loss, and PCOS (55). In conclusion, we observed no differences in postprandial subjective appetite, PYY and cholecystokinin concentrations, and ad libitum food consumption, but we confirmed alterations in

1609

fasting and postprandial ghrelin concentrations in women with PCOS. Ad libitum buffet consumption was decreased after modest weight loss in conjunction with an improved postprandial ghrelin profile, but no changes were seen in subjective appetite or circulating cholecystokinin or PYY concentrations. It is not clear whether women with PCOS have abnormal regulation of energy homeostasis or appetite hormones, and, despite improving some aspects of appetite regulation, the effect of weight loss on postprandial anorexigenic hormones is unclear in overweight women with or without PCOS. We thank Julia Weaver, Vanessa Courage, Rosemary McArthur, Ruth Pinches, Sue Evans, Deborah Roffe, Mark Mano, Candita Sullivan, Paul Orchard, Cathryn Seccafien and Michael Mular for their assistance. The authors’ responsibilities were as follows—LM, MN, PMC, GAW, and RJN: conceived and designed the study and contributed to data analysis and manuscript writing; CWL, MAG, and SRB: contributed to data analysis and manuscript writing; LM: coordinated the study, collected the data, wrote the manuscript, and contributed to the design and implementation of the study dietary protocol; and MN: contributed to the design of the study dietary protocol. None of the authors had a personal or financial conflict of interest.

REFERENCES 1. Dunaif A. Insulin resistance and the polycystic ovary syndrome: mechanism and implications for pathogenesis. Endocr Rev 1997;18:774 – 800. 2. Kiddy DS, Sharp PS, White DM, et al. Differences in clinical and endocrine features between obese and non-obese subjects with polycystic ovary syndrome: an analysis of 263 consecutive cases. Clin Endocrinol (Oxf) 1990;32:213–20. 3. Huber-Buchholz MM, Carey DG, Norman RJ. Restoration of reproductive potential by lifestyle modification in obese polycystic ovary syndrome: role of insulin sensitivity and luteinizing hormone. J Clin Endocrinol Metab 1999;84:1470 – 4. 4. Moran LJ, Noakes M, Clifton PM, et al. Ghrelin and measures of satiety are altered in polycystic ovary syndrome but not differentially affected by diet composition. J Clin Endocrinol Metab 2004;89:3337– 44. 5. Cummings DE, Purnell JQ, Frayo RS, Schmidova K, Wisse BE, Weigle DS. A preprandial rise in plasma ghrelin levels suggests a role in meal initiation in humans. Diabetes 2001;50:1714 –9. 6. Druce MR, Neary NM, Small CJ, et al. Subcutaneous administration of ghrelin stimulates energy intake in healthy lean human volunteers. Int J Obes (Lond) 2006;30:293– 6. 7. English PJ, Ghatei MA, Malik IA, Bloom SR, Wilding JP. Food fails to suppress ghrelin levels in obese humans. J Clin Endocrinol Metab 2002; 87:2984 – 87. 8. Moran LJ, Luscombe-Marsh ND, Noakes M, Wittert GA, Keogh JB, Clifton PM. The satiating effect of dietary protein is unrelated to postprandial ghrelin secretion. J Clin Endocrinol Metab 2005;90:5205–11. 9. le Roux CW, Patterson M, Vincent RP, Hunt C, Ghatei MA, Bloom SR. Postprandial plasma ghrelin is suppressed proportional to meal calorie content in normal-weight but not obese subjects. J Clin Endocrinol Metab 2005;90:1068 –71. 10. Liddle RA, Goldfine ID, Rosen MS, Taplitz RA, Williams JA. Cholecystokinin bioactivity in human plasma. Molecular forms, responses to feeding, and relationship to gallbladder contraction. J Clin Invest 1985; 75:1144 –52. 11. Bowen J, Noakes M, Clifton PM. Appetite regulatory hormone responses to various dietary proteins differ by body mass index status despite similar reductions in ad libitum energy intake. J Clin Endocrinol Metab 2006;91:2913–9. 12. French SJ, Murray B, Rumsey RD, Sepple CP, Read NW. Preliminary studies on the gastrointestinal responses to fatty meals in obese people. Int J Obes Relat Metab Disord 1993;17:295–300. 13. Hirschberg AL, Naessen S, Stridsberg M, Bystrom B, Holtet J. Impaired cholecystokinin secretion and disturbed appetite regulation in women with polycystic ovary syndrome. Gynecol Endocrinol 2004;19:79 – 87. 14. Adrian TE, Ferri GL, Bacarese-Hamilton AJ, Fuessl HS, Polak JM, Bloom SR. Human distribution and release of a putative new gut hormone, peptide YY. Gastroenterology 1985;89:1070 –7.

1610

MORAN ET AL

15. Batterham RL, Cohen MA, Ellis SM, Le Roux CW, Withers DJ, Frost GS, et al. Inhibition of food intake in obese subjects by peptide YY3–36. N Engl J Med 2003;349:941– 8. 16. le Roux CW, Batterham RL, Aylwin SJ, Patterson M, Borg CM, Wynne KJ, et al. Attenuated peptide YY release in obese subjects is associated with reduced satiety. Endocrinology 2006;147:3– 8. 17. Pfluger PT, Kampe J, Castaneda TR, Vahl T, D’Alessio DA, Kruthaupt T, et al. Effect of human body weight changes on circulating levels of peptide YY and peptide YY3–36. J Clin Endocrinol Metab 2007;92: 583– 8. 18. Stock S, Leichner P, Wong AC, Ghatei MA, Kieffer TJ, Bloom SR, et al. Ghrelin, peptide YY, glucose-dependent insulinotropic polypeptide, and hunger responses to a mixed meal in anorexic, obese, and control female adolescents. J Clin Endocrinol Metab 2005;90:2161– 8. 19. Roth CL, Enriori PJ, Harz K, Woelfle J, Cowley MA, Reinehr T. Peptide YY is a regulator of energy homeostasis in obese children before and after weight loss. J Clin Endocrinol Metab 2005;90:6386 –91. 20. Lieverse RJ, van Seters AP, Jansen JB, Lamers CB. Relationship between hunger and plasma cholecystokinin during weight reduction with a very low calorie diet. Int J Obes Relat Metab Disord 1993;17:177–9. 21. Moran LJ, Noakes M, Clifton PM, Wittert GA, Williams G, Norman RJ. Short-term meal replacements followed by dietary macronutrient restriction enhance weight loss in polycystic ovary syndrome. Am J Clin Nutr 2006;84:77– 87. 22. Revised 2003 consensus on diagnostic criteria and long-term health risks related to polycystic ovary syndrome (PCOS). Hum Reprod 2004;19: 41–7. 23. Stunkard AJ, Messick S. The three-factor eating questionnaire to measure dietary restraint, disinhibition and hunger. J Psychosom Res 1985; 29:71– 83. 24. Bouchard C, Tremblay A, Leblanc C, Lortie G, Savard R, Theriault G. A method to assess energy expenditure in children and adults. Am J Clin Nutr 1983;37:461–7. 25. Luscombe-Marsh ND, Noakes M, Wittert GA, Keogh JB, Foster P, Clifton PM. Carbohydrate-restricted diets high in either monounsaturated fat or protein are equally effective at promoting fat loss and improving blood lipids. Am J Clin Nutr 2005;81:762–72. 26. Wolever TM, Jenkins DJ, Jenkins AL, Josse RG. The glycemic index: methodology and clinical implications. Am J Clin Nutr 1991;54:846 – 54. 27. Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF, Turner RC. Homeostasis model assessment: insulin resistance and betacell function from fasting plasma glucose and insulin concentrations in man. Diabetologia 1985;28:412–9. 28. Vermeulen A, Verdonck L, Kaufman JM. A critical evaluation of simple methods for the estimation of free testosterone in serum. J Clin Endocrinol Metab 1999;84:3666 –72. 29. Schofl C, Horn R, Schill T, Schlosser HW, Muller MJ, Brabant G. Circulating ghrelin levels in patients with polycystic ovary syndrome. J Clin Endocrinol Metab 2002;87:4607–10. 30. Weigle DS, Cummings DE, Newby PD, et al. Roles of leptin and ghrelin in the loss of body weight caused by a low fat, high carbohydrate diet. J Clin Endocrinol Metab 2003;88:1577– 86. 31. Lejeune MP, Westerterp KR, Adam TC, Luscombe-Marsh ND, Westerterp-Plantenga MS. Ghrelin and glucagon-like peptide 1 concentrations, 24-h satiety, and energy and substrate metabolism during a high-protein diet and measured in a respiration chamber. Am J Clin Nutr 2006;83:89 –94. 32. Delgado-Aros S, Cremonini F, Castillo JE, et al. Independent influences of body mass and gastric volumes on satiation in humans. Gastroenterology 2004;126:432– 40. 33. Tung YC, Hewson AK, Carter RN, Dickson SL. Central responsiveness to a ghrelin mimetic (GHRP-6) is rapidly altered by acute changes in nutritional status in rats. J Neuroendocrinol 2005;17:387–93. 34. Doucet E, St-Pierre S, Almeras N, Tremblay A. Relation between appetite ratings before and after a standard meal and estimates of daily

35.

36.

37.

38.

39.

40.

41. 42.

43. 44.

45.

46.

47.

48.

49.

50.

51.

52. 53.

54.

55.

energy intake in obese and reduced obese individuals. Appetite 2003; 40:137– 43. Maffeis C, Bonadonna RC, Consolaro A, et al. Ghrelin, insulin sensitivity and postprandial glucose disposal in overweight and obese children. Eur J Endocrinol 2006;154:61– 8. Tschop M, Weyer C, Tataranni PA, Devanarayan V, Ravussin E, Heiman ML. Circulating ghrelin levels are decreased in human obesity. Diabetes 2001;50:707–9. Anderwald C, Brabant G, Bernroider E, et al. Insulin-dependent modulation of plasma ghrelin and leptin concentrations is less pronounced in type 2 diabetic patients. Diabetes 2003;52:1792– 8. Vrbikova J, Dvorakova K, Hill M, et al. Determinants of circulating adiponectin in women with polycystic ovary syndrome. Gynecol Obstet Invest 2005;60:155– 61. Purnell JQ, Cummings D, Weigle DS. Changes in 24-h area-under-thecurve ghrelin values following diet-induced weight loss are associated with loss of fat-free mass, but not with changes in fat mass, insulin levels or insulin sensitivity. Int J Obes (Lond) 2007;31:385–9. Panidis D, Farmakiotis D, Koliakos G, et al. Comparative study of plasma ghrelin levels in women with polycystic ovary syndrome, in hyperandrogenic women and in normal controls. Hum Reprod 2005;20: 2127–32. Batterham RL, Cowley MA, Small CJ, et al. Gut hormone PYY(3–36) physiologically inhibits food intake. Nature 2002;418:650 – 4. Morinigo R, Moize V, Musri M, et al. Glucagon-like peptide-1, peptide YY, hunger, and satiety after gastric bypass surgery in morbidly obese subjects. J Clin Endocrinol Metab 2006;91:1735– 40. Lieverse RJ, Jansen JB, Masclee AM, Lamers CB. Satiety effects of cholecystokinin in humans. Gastroenterology 1994;106:1451– 4. Bowen J, Noakes M, Trenerry C, Clifton PM. Energy intake, ghrelin, and cholecystokinin after different carbohydrate and protein preloads in overweight men. J Clin Endocrinol Metab 2006;91:1477– 83. Batterham RL, Heffron H, Kapoor S, et al. Critical role for peptide YY in protein-mediated satiation and body-weight regulation. Cell Metab 2006;4:223–33. Murata M, Okimura Y, Iida K, et al. Ghrelin modulates the downstream molecules of insulin signaling in hepatoma cells. J Biol Chem 2002;277: 5667–74. Li WG, Gavrila D, Liu X, et al. Ghrelin inhibits proinflammatory responses and nuclear factor-kappaB activation in human endothelial cells. Circulation 2004;109:2221– 6. Skommer J, Katulski K, Poreba E, et al. Gherlin expression in women with polycystic ovary syndrome–a preliminary study. Eur J Gynaecol Oncol 2005;26:553– 6. Drazen DL, Vahl TP, D’Alessio DA, Seeley RJ, Woods SC. Effects of a fixed meal pattern on ghrelin secretion: evidence for a learned response independent of nutrient status. Endocrinology 2006;147:23–30. Natalucci G, Riedl S, Gleiss A, Zidek T, Frisch H. Spontaneous 24-h ghrelin secretion pattern in fasting subjects: maintenance of a mealrelated pattern. Eur J Endocrinol 2005;152:845–50. Escalante Pulido JM, Alpizar Salazar M. Changes in insulin sensitivity, secretion and glucose effectiveness during menstrual cycle. Arch Med Res 1999;30:19 –22. (Published erratum appears in Arch Med Res 1999; 30:265.) Dye L, Blundell JE. Menstrual cycle and appetite control: implications for weight regulation. Hum Reprod 1997;12:1142–51. Frick G, Bremme K, Sjogren C, Linden A, Uvnas-Moberg K. Plasma levels of cholecystokinin and gastrin during the menstrual cycle and pregnancy. Acta Obstet Gynecol Scand 1990;69:317–20. De Castro JM. How can eating behavior be regulated in the complex environments of free-living humans? Neurosci Biobehav Rev 1996;20: 119 –31. Marzullo P, Verti B, Savia G, et al. The relationship between active ghrelin levels and human obesity involves alterations in resting energy expenditure. J Clin Endocrinol Metab 2004;89:936 –9.

Comparison of monounsaturated fat with carbohydrates as a replacement for saturated fat in subjects with a high metabolic risk profile: studies in the fasting and postprandial states1–3 Lars Berglund, Michael Lefevre, Henry N Ginsberg, Penny M Kris-Etherton, Patricia J Elmer, Paul W Stewart, Abby Ershow, Thomas A Pearson, Barbara H Dennis, Paul S Roheim, Rajasekhar Ramakrishnan, Roberta Reed, Kent Stewart, and Katherine M Phillips for the DELTA Investigators ABSTRACT Background: In subjects with a high prevalence of metabolic risk abnormalities, the preferred replacement for saturated fat is unresolved. Objective: The objective was to study whether carbohydrate or monounsaturated fat is a preferred replacement for saturated fat. Design: Fifty-two men and 33 women, selected to have any combination of HDL cholesterol 울 30th percentile, triacylglycerol 욷 70th percentile, or insulin 욷 70th percentile, were enrolled in a 3-period, 7-wk randomized crossover study. The subjects consumed an average American diet (AAD; 36% of energy from fat) and 2 additional diets in which 7% of energy from saturated fat was replaced with either carbohydrate (CHO diet) or monounsaturated fatty acids (MUFA diet). Results: Relative to the AAD, LDL cholesterol was lower with both the CHO (Ҁ7.0%) and MUFA (Ҁ6.3%) diets, whereas the difference in HDL cholesterol was smaller during the MUFA diet (Ҁ4.3%) than during the CHO diet (Ҁ7.2%). Plasma triacylglycerols tended to be lower with the MUFA diet, but were significantly higher with the CHO diet. Although dietary lipid responses varied on the basis of baseline lipid profiles, the response to diet did not differ between subjects with or without the metabolic syndrome or with or without insulin resistance. Postprandial triacylglycerol concentrations did not differ significantly between the diets. Lipoprotein(a) concentrations increased with both the CHO (20%) and MUFA (11%) diets relative to the AAD. Conclusions: In the study population, who were at increased risk of coronary artery disease, MUFA provided a greater reduction in risk as a replacement for saturated fat than did carbohydrate. Am J Clin Nutr 2007;86:1611–20. KEY WORDS

Diet, nutrition, fatty acids, lipids, lipoproteins

INTRODUCTION

Metabolic factors are important predictors of cardiovascular disease (1). Beyond LDL cholesterol, the presence of dyslipidemia with low blood concentrations of HDL cholesterol, high triacylglycerol concentrations, or both; obesity; impaired glucose tolerance; and hypertension have been shown to be associated with cardiovascular disease risk (1–3). Clusters of these factors have been identified as the metabolic syndrome and syndrome X and as indicators of insulin resistance (1, 4 – 6). The constellation of these factors has

been suggested to provide a high-risk metabolic milieu in which the cardiovascular disease risk exceeds that predicted by LDL cholesterol. Furthermore, individuals at increased risk of coronary artery disease (CAD) with low HDL cholesterol, high triacylglycerol, or both together make up a greater fraction of individuals with premature CAD than those with isolated elevated LDL cholesterol (7). In addition, hyperinsulinemia— either directly, or more likely as a surrogate for insulin resistance—may independently contribute to the development of CAD (8). Lifestyle modifications are recommended as first line interventions to improve metabolic risk factors. Current dietary recommendations by the National Cholesterol Education Program (NCEP) (9) and the American Heart Association (10) to reduce 1 From the Department of Medicine, Columbia University College of Physicians and Surgeons, New York, NY (LB, HNG, and RR); the Division of Nutrition and Chronic Disease, Pennington Biomedical Research Center, Baton Rouge, LA (ML); the Nutrition Department, Pennsylvania State University, University Park, PA (PMK-E); the Division of Epidemiology, University of Minnesota School of Public Health, Minneapolis, MN (PJE); the Department of Biostatistics, Collaborative Studies Coordinating Center (PWS) and the Department of Nutrition, School of Public Health and School of Medicine (BHD), The University of North Carolina at Chapel Hill, Chapel Hill, NC; the Division of Cardiovascular Diseases, National Heart, Lung, and Blood Institute, National Institutes of Health, Bethesda, MD (AE); the Research Institute, Mary Imogene Bassett Hospital, Cooperstown, NY (TAP and RR); the Department of Physiology, Louisiana State University Medical School, New Orleans, LA (PSR); and the Department of Biochemistry, Virginia Polytechnic Institute and State University, Blacksburg, VA (KS and KMP). 2 Supported by NIH grants (5-U01-HL 049644, HL 049648, HL 049649, HL 049651, and HL 049659) and M01-NCRR 00645. The following companies made in-kind contributions of products: AARHUS, Bertoli USA, Best Foods, Campbell Soup Company, Del Monte Foods, General Mills, Hershey Foods Corporation, Institute of Edible Oils and Shortenings, Kraft General Foods, Land O’Lakes, McCormick Incorporated, Nabisco Foods Group, Neomonde Baking Company, Palm Oil Research Institute, Park Corporation, Procter and Gamble, Quaker Oats, Ross Laboratories, Swift-Armour and Eckrick, Van Den Bergh Foods, Cholestech, and Lifelines Technology Incorporated. 3 Reprints not available. Address correspondence to L Berglund, Department of Medicine, University of California, Davis, UCD Medical Center, CRISP, 2921 Stockton Boulevard, Suite 1400, Sacramento, CA 95817. Email: [email protected]. Received January 8, 2007. Accepted for publication July 26, 2007.

Am J Clin Nutr 2007;86:1611–20. Printed in USA. © 2007 American Society for Nutrition

1611

1612

BERGLUND ET AL

the intake of total fat, saturated fatty acids (SFAs), and cholesterol target primarily elevated LDL-cholesterol concentrations. Although the efficacy of these diets with regard to LDL cholesterol has been shown (11–17), there is concern over their potential to lower HDL cholesterol, raise triacylglycerol, and cause unfavorable postprandial lipid changes (17–20). Although carbohydrates have been advocated as a replacement for saturated fats, there has been less focus on the specific carbohydrate composition of such replacement diets, which may modulate the extent of diet-induced hypertriacylglycerolemia (21, 22). These concerns, coupled with concerns about increases in the risk of type 2 diabetes, have prompted alternative dietary approaches, some with higher fat intakes (23, 24). Studies in healthy subjects have shown benefits with regard to cardiovascular risk by substituting either a protein-rich diet or a diet rich in unsaturated fat for a carbohydrate-rich diet (25). However, the question remains whether individuals at risk of premature CAD because of metabolic risk factors such as low HDL cholesterol, high triacylglycerol, high fasting insulin, or a combination thereof would achieve a more favorable overall risk factor profile through replacement of saturated fat by carbohydrates or by alternative dietary approaches. The DELTA Program (Dietary Effects on Lipoproteins and Thrombogenic Activity) comprised 2 multicenter controlled diet studies that examined the effect of changes in dietary fat on CAD risk factors. In the first study, we showed that replacement of dietary saturated fat with carbohydrate reduced total and LDLcholesterol concentrations across sex and ethnicity (17). In the present study we sought to determine whether the replacement of dietary saturated fat with monounsaturated fat, as opposed to carbohydrate, would result in a better overall risk factor profile in nondiabetic individuals with one or more of the following: low HDL-cholesterol, high triacylglycerol, or high insulin concentrations. To be able to reflect the postabsorptive physiologic state as well as to assess diet response to a standardized metabolic fat load challenge, we also evaluated 2 separate postprandial conditions in our study. We further explored whether the diet response would differ depending on baseline lipid concentrations or the presence of the metabolic syndrome and insulin resistance.

SUBJECTS AND METHODS

Study population Four research centers (Columbia University, Pennington Biomedical Research Center, Pennsylvania State University, and the University of Minnesota) each enrolled 20 –30 participants between the ages of 21 and 65 y. Recruitment goals focused on enrolling participants who were likely to be at risk of the potential negative effects of low-fat diets. Thus, subjects were eligible if the average of 2 screening measurements met any of the following requirements based on criteria of the third National Health and Nutrition Examination Survey specific for age, sex, and race: 1) HDL cholesterol 울 30th percentile, 2) triacylglycerol 욷 70th percentile, and 3) insulin 욷 70th percentile. Subjects were ineligible if their 1) average screening total cholesterol was 쏝25th percentile or 쏜90th percentile, 2) LDL cholesterol was 쏜4.91 mmol/L, 3) fasting triacylglycerol concentrations were 쏝30th percentile or 쏜5.65 mmol/L, or 4) HDL cholesterol was 쏜70th percentile. Additionally, subjects had to be in good health, free of

chronic disease (including documented heart disease and diabetes) and taking no medications known to affect lipids or thrombotic factors. Subjects were classified as having the metabolic syndrome if any 3 of the 5 defined characteristics of the syndrome were present at baseline (4). Subjects with a baseline homeostasis model assessment index 쏜3 were classified as having insulin resistance. All participants provided written informed consent. The experimental protocol was approved by the Institutional Review Board at each respective site. For subgroup analyses based on eligibility criteria, we recategorized participants using eligibility criteria obtained while participants were consuming an average American diet (AAD) to estimate underlying response differences without the confounding effect of habitual dietary intake. We examined lipid responsiveness to the diets by each eligibility criterion individually, ie, participants with low HDL compared with those with normal HDL, participants with high triacylglycerol compared with those with normal triacylglycerol, and participants with high insulin compared with those with normal insulin. Study protocol Three diets were fed in a randomized, double-blind, 3-way crossover design with each diet period lasting 7 wk. There were “rest periods” of 4 to 6 wk duration between each diet period. Twelve-hour fasting blood samples were obtained from each subject for endpoint determinations once weekly during weeks 5, 6, and 7. With the exception of a self-selected Saturday evening meal, all foods consumed by subjects were provided by the research centers. Subjects were counseled to eat the Saturday evening meal based on the NCEP Adult Treatment Panel Step I guidelines (9). Subjects were weighed twice weekly; if necessary, adjustments were made in energy intake to maintain stable body weight. Compliance with the protocol was assessed each week through a review of daily questionnaires. Diets The AAD was designed to reflect typical consumption patterns of the US population (Table 1). The carbohydratereplacement diet (CHO diet) was designed to meet nutrient specifications of the NCEP Step I diet, whereas the monounsaturated fat–replacement diet (MUFA diet) was designed to not only match the saturated and polyunsaturated fat content of the CHO diet, but to also match the total fat content of the AAD. Thus, 7% of energy from SFAs was replaced with either carbohydrate (primarily as complex carbohydrates) on the CHO diet or with monounsaturated fat on the MUFA diet. In keeping with current guidelines, the CHO diet was designed to contain more fiber than the AAD. All 3 diets were designed to provide the same amount of cholesterol (300 mg/d). The target nutrient intakes for the 3 diets are summarized in Table 1. The methods used for menu development, validation, preparation, and delivery were described previously (26). A diet-monitoring protocol was established to confirm the uniformity of diet composition over time and between the 4 research center kitchens. Daily menus were sampled and shipped frozen to the Food Analysis Laboratory and Control Center at Virginia Polytechnic Institute and State University, where 8-d menu cycle composites were prepared and assayed as described previously for total fat, total carbohydrates, protein, fatty acids, cholesterol, and energy (26). Glucose, fructose, maltose, sucrose, and lactose contents were measured by HPLC (27, 28) and

1613

PREFERRED DIET TO REDUCE CAD RISK TABLE 1 Specified and assayed nutrient values for the 3 different diets1 AAD

MUFA diet

CHO diet

Nutrient

Goal

Assay2

Goal

Assay2

Goal

Assay2

Protein (% of energy) Carbohydrate (% of energy) Sugar (% of energy) Starch (% of energy) Fat (% of energy) SFA (% of energy) MUFA (% of energy) PUFA (% of energy) Cholesterol (mg/d) Fiber (g/1000 kcal)

16 47

15.3 앐 0.2 49.0 앐 0.2 18.5 앐 0.5 26.2 앐 0.3 35.8 앐 0.2 15.6 앐 0.1 14.4 앐 0.1 5.8 앐 0.1 302 앐 8 10.6 앐 1.3

16 47

15.5 앐 0.1 48.8 앐 0.3 18.1 앐 0.4 26.7 앐 0.6 35.7 앐 0.3 8.7 앐 0.2 20.8 앐 0.2 6.2 앐 0.1 293 앐 8 11.0 앐 1.3

16 54

16.1 앐 0.2 54.9 앐 0.23 19.4 앐 0.4 29.2 앐 0.6 29.0 앐 0.2 8.0 앐 0.1 15.5 앐 0.2 5.5 앐 0.1 292 앐 6 17.0 앐 2.0

37 16 14 7 300 7.5

37 8 22 7 300 7.5

30 8 15 7 300 15.0

1 AAD, average American diet; MUFA diet, 7% of energy from saturated fatty acids (SFA) replaced with monounsaturated fatty acids; CHO diet, 7% of energy from SFAs replaced with carbohydrates. 2 x៮ 앐 SEM of 8 menu cycles (8 d/menu cycle) for each diet. 3 More than 50% of the increase in carbohydrates in the CHO diet was in the form of complex carbohydrates.

reported as total sugars, the amount of starch was assayed enzymatically (29), and total dietary fiber was assayed by an enzymatic-gravimetric procedure (30). Measured concentrations of the key components in each experimental diet are summarized in Table 1. As seen in the table, the data demonstrate that the composition of the diets as fed to the participants was close to design targets and very precise over time and feeding sites. Postprandial studies During week 7 of each diet period, 2 d-long studies were performed as part of the DELTA 2 protocol. The studies were designed to test the effect of the consumption of natural food and of a high-fat load. In the first study, blood samples were drawn from 68 participants (43 men and 25 women) during fasting conditions, prelunch, and predinner (fasting and 4 and 8 h) for each of the 3 different diets. The second day-long study was on the last day of each diet period, separated by 욷2 d from the premeal study. After a fasting blood draw, a standardized highfat meal was administered; patients remained in a fasting state during the day, and repeated blood samples were drawn at 4 and 8 h. In both postprandial studies, insulin, glucose, and triacylglycerol concentrations were measured at those time points, and comparisons between studies were based on the area under the curve for each analyte (16). The high-fat meal, prepared within 24 h before administration, included 190.0 g heavy cream (Tuscan heavy cream, grade A; Nestle, Glendale, CA), 90.0 g ice cream (Breyers, Green Bay, WI) 22.0 g safflower oil (Hain Food Group, Melville, NY), 25.0 g of a powered whey protein (Promod; Ross Laboratories, Columbus, OH), 30.0 g syrup (Nestle Quik, Glendale, CA), and 0.2 g Lactaid (McNeil Nutritionals, Ft Washington, PA) as a precaution against lactose intolerance in any of the participants. The nutrient composition of the meal, based on a body surface area of 2 m2, included 105 g fat (75% of total calories; 52 g saturated fat), 48 g carbohydrate (15% of total calories), 32 g protein (10% of total calories), 300 mg cholesterol, and 1237 calories. The body surface area of the subjects was calculated by using the Dubois equation to gauge the appropriate weight of the meal for each subject. The subjects consumed the meal within a 15-min period. The meal containers were rinsed

with chilled distilled water to ensure that the entire content was consumed. Laboratory analyses Standardized blood sampling and processing procedures were validated and used at all 4 research centers. Plasma and serum samples were collected, processed, and enzymatically assayed for total cholesterol, LDL cholesterol, HDL cholesterol, triacylglycerol, glucose, and uric acid at each research center (17). A special lipid standardization protocol administered by the Centers for Disease Control showed that the precision and accuracy of each research center’s laboratory were adequate for the assayed values to be combined. Apolipoprotein (apo) A-I and apo B concentrations were assayed by rate immunonephelometry as previously described (Beckman, Fullerton, CA) (31), lipoprotein(a) [Lp(a)] concentrations were measured by enzyme-linked immunosorbent assay (Terumo, Elkton, MD), and insulin concentrations were evaluated by radioimmunoassay. All these assays were performed centrally. Insulin resistance was estimated by using the homeostasis model assessment of insulin resistance (HOMA-IR) index: fasting plasma glucose concentration (mmol/L) ҂ fasting plasma insulin concentration (␮U/mL)/22.5. A HOMA index 쏜 3.0 was defined as insulin resistance. Statistical analyses The effects of changing diet composition were evaluated in terms of 10 response variables: total cholesterol, LDL cholesterol, HDL cholesterol, triacylglycerol (natural log scale), apo B, apo A-I, Lp(a) (square root scale), glucose, insulin, and uric acid. The linear statistical model, the set of primary hypotheses, the strategy for controlling type I error, and the estimation procedures were all specified a priori. The model assumed 3 components of variance: subject, subject-by-diet, and residual. The means of the conditional distribution of assay values were assumed to be a linear function of 5 categorical factors (number of levels shown in brackets): diet [3], sex-age group [4], race [2], research center [4], feeding period [3], and interaction of diet with research center. Statistical computations for estimation and testing were performed via established methods (32) by using the

1614

BERGLUND ET AL

TABLE 2 Subject characteristics during the reference average American diet

African American Age (y) BMI (kg/m2) Total cholesterol (mmol/L) LDL cholesterol (mmol/L) HDL cholesterol (mmol/L) Triacylglycerols (mmol/L) Apolipoprotein A-I (mg/dL) Apolipoprotein B (mg/dL) Lipoprotein(a) (mg/dL) Glucose (mmol/L) Uric acid (mg/dL) Insulin (␮U/mL)

Men (n ҃ 52)

Women (n ҃ 33)

Total (n ҃ 85)

4 33.3 앐 8.61 27.4 앐 4.2 5.16 앐 0.9 3.33 앐 0.6 1.01 앐 0.2 1.38 (1.06–2.07)3 122.7 앐 11.5 109.9 앐 21.2 5.2 (2.3–14.7) 5.18 앐 0.7 6.44 앐 1.2 11.7 (8.4–14.9)

6 39.0 앐 9.72 27.9 앐 4.6 5.15 앐 0.8 3.29 앐 0.7 1.18 앐 0.32 1.35 (0.90–1.81) 136.9 앐 17.12 111.2 앐 25.1 13.32 (4.7–28) 5.13 앐 0.6 4.53 앐 1.22 9.9 (7.9–14.4)

10 35.5 앐 9.4 27.6 앐 4.4 5.16 앐 0.8 3.32 앐 0.6 1.08 앐 0.2 1.38 (1.02–2.03) 128.2 앐 15.5 110.4 앐 22.7 9.0 (2.7–18.7) 5.16 앐 0.7 5.70 앐 1.5 11.3 (8.2–14.8)

x៮ 앐 SD (all such values). Significantly different from men, P 쏝 0.01. 3 Median; interquartile range in parentheses (all such values). 1 2

mixed-model procedure of the SAS software system, version 9.1.3 (33). Fasting concentrations in weeks 5, 6, and 7 were tested for any time trends. No time trends were seen, which indicated stability in the final 3 wk of each diet period. The 3 values were averaged before further analysis. For each outcome variable, factors whose interactions were nonsignificant (meaning that diet effects were not affected by the level of that factor) were removed from the model one by one; the final model for each outcome variable included, besides the diets, only those factors with a significant interaction. The model also included diet period to allow for seasonal variation in outcome variables. HDL cholesterol was found to be higher in the third period by 0.024 mmol/L (P ҃ 0.002) and triacylglycerol higher in the second period by 앒4% (P ҃ 0.01). For the a priori analysis, a P 쏝 0.01 was chosen as statistically significant. Auxiliary analyses were used to evaluate the sensitivity of the main results to perturbations of the modeling assumptions and to address secondary research questions. Results are presented with the values corrected for the seasonal changes. The results of the statistical analyses were unaffected if no seasonal correction is made.

RESULTS

Of the 110 participants randomly assigned, 85 completed all 3 diet periods. The final study population ranged in age from 21 to 61 y and included 33 women and 10 African Americans (Table 2). In accordance with the recruitment strategy, our study population was characterized by having, at screening, low HDL cholesterol, moderately elevated triacylglycerol and insulin, and near normal LDL cholesterol. When characterized by specific eligibility criteria, 82% of the subjects had HDL cholesterol 울 30th percentile, 58% of the subjects had triacylglycerol 욷 70th percentile, and 22% had insulin values 욷 70th percentile. Thirty-eight percent of the subjects qualified in 2 eligibility categories: low HDL cholesterol with high triacylglycerol was the most common combination (29%). Eighteen percent qualified in all 3 categories. Total cholesterol and LDL cholesterol were, on average, 5.5% and 7.0% lower, respectively, when subjects consumed the CHO or AAD diet, respectively (Table 3). Comparable differences in total cholesterol and LDL cholesterol were observed between the AAD and the MUFA diet (Ҁ6.0% and Ҁ6.3%, respectively), and

TABLE 3 Effect of the 3 diets on the primary study endpoints1

Total cholesterol (mmol/L) LDL cholesterol (mmol/L) HDL cholesterol (mmol/L) Triacylglycerols (mmol/L) Apolipoprotein A-I (mg/dL) Apolipoprotein B (mg/dL) Lipoprotein(a) (mg/dL) Glucose (mmol/L) Insulin (␮U/mL) Uric acid (mg/dL)

AAD

MUFA diet

CHO diet

5.17 앐 0.08 3.31 앐 0.08 1.08 앐 0.03 1.48 앐 0.08 128 앐 2 110 앐 3 9.9 앐 1.4 5.1 앐 0.1 12.2 앐 0.6 5.7 앐 0.2

4.86 앐 0.08 3.10 앐 0.082 1.03 앐 0.032 1.42 앐 0.07 125 앐 22 106 앐 22 11.0 앐 1.52 5.1 앐 0.1 12.3 앐 0.6 5.7 앐 0.2

4.89 앐 0.082 3.08 앐 0.082 1.00 앐 0.022,3 1.59 앐 0.082,3 122 앐 22 107 앐 22 11.9 앐 1.62 5.1 앐 0.1 12.1 앐 0.7 5.6 앐 0.2

2

1 All values are x៮ 앐 SEM, except for triacylglycerols and Lp(a), which are medians 앐 SEM. AAD, average American diet; MUFA diet, 7% of energy from saturated fatty acids replaced with monounsaturated fatty acids; CHO diet, 7% of energy from saturated fatty acids replaced with carbohydrates. 2 Significantly different from AAD on the basis of adjusted values from the a priori linear regression model, P 쏝 0.01. 3 Significantly different from the MUFA diet on the basis of adjusted values from the a priori linear regression model, P 쏝 0.01.

PREFERRED DIET TO REDUCE CAD RISK

LDL-cholesterol concentrations with the MUFA and CHO diets did not differ. In contrast, the CHO and MUFA diets differed significantly with respect to their effects on both HDL cholesterol and triacylglycerol concentrations. Although HDLcholesterol concentrations were lower with both the MUFA and CHO diets than with the AAD (P 쏝 0.01 for each comparison), the difference with the MUFA diet (Ҁ4.3%) was less than the difference observed with the CHO diet (Ҁ7.2%). Furthermore, as seen in Table 3, HDL-cholesterol concentrations with the CHO and the MUFA diets differed significantly (P 쏝 0.01). In addition, whereas plasma triacylglycerol concentrations tended to be lower with the MUFA diet than with the AAD (Ҁ4.9%; P 쏝 0.03), triacylglycerol concentrations were significantly higher with the CHO diet than with either the AAD (6.5%) or the MUFA (11.4%) diet (P 쏝 0.01 for each comparison). Changes in apo A-I and apo B paralleled changes in HDL cholesterol and LDL cholesterol, respectively. As seen in Table 3, average median Lp(a) concentrations were significantly higher during both the CHO and MUFA diets (11% and 20%, respectively) than during the AAD, and Lp(a) concentrations with the 2 latter diets were not significantly different from each other. Finally, although our study population was chosen for a potential predisposition to insulin resistance, the fasting concentrations of glucose, insulin, and uric acid were not affected by changes in diet. We next examined whether the lipid responses differed between subgroups, representing individuals having either low HDL cholesterol, high triacylglycerol, or high insulin on AAD. Relative to the normal-HDL group (n ҃ 40), individuals with low HDL cholesterol (n ҃ 45) had smaller reductions in HDL cholesterol in response to either the MUFA or CHO diet (P ҃ 0.02) (Figure 1). For the CHO diet, the decrease in HDL cholesterol in the low-HDL group was half that observed in the normal-HDL group. When comparing the MUFA and CHO diets, significant differences in HDL-cholesterol concentrations were seen only in the normal-HDL group. In contrast, reductions in LDL cholesterol with either the CHO or MUFA diet tended to be greater in the low-HDL group, whereas changes in triacylglycerol concentrations with either the MUFA or the CHO diets were similar in both HDL groups. The high (n ҃ 39) and normal (n ҃ 46) triacylglycerol groups experienced similar changes in HDL and LDL cholesterol, but differed significantly in their triacylglycerol response to the 2 diets (P ҃ 0.007) (Figure 1). In individuals with high triacylglycerol concentrations, triacylglycerol concentrations were not different when the amount of total dietary fat was decreased from 37% to 30% (2.13 mmol/L with the AAD and 2.18 mmol/L with the CHO diet), whereas triacylglycerol concentrations were significantly lower with the MUFA diet (2.13 mmol/L with the AAD and 1.90 mmol/L with the MUFA diet; P ҃ 0.0005). HDL cholesterol, LDL cholesterol, and triacylglycerol responses to the CHO and MUFA diets did not differ in individuals categorized as having either normal (n ҃ 54) or high (n ҃ 31) fasting insulin concentrations (data not shown). We next examined whether the fasting lipid responses differed by the presence of the metabolic syndrome or insulin resistance. As seen in Table 4, triacylglycerol concentrations with the AAD diet were significantly higher in subjects with the metabolic syndrome (n ҃ 20; 12 men and 8 women) or with insulin resistance (n ҃ 28; 18 men and 10 women) than in subjects who did not fulfill these criteria. Baseline HDL and LDL cholesterol did

1615

not differ across insulin resistance strata, whereas HDLcholesterol concentrations were lower and LDL-cholesterol concentrations were higher in subjects with the metabolic syndrome. As seen in Figure 2, the triacylglycerol and HDL cholesterol responses to the MUFA and the CHO diets, compared with the AAD, did not differ significantly for subjects with or without the metabolic syndrome or insulin resistance. Similar results were seen for total and LDL cholesterol (data not shown). The postprandial responses of triacylglycerol, insulin, and glucose are shown in Figure 3. As outlined above, 2 different postprandial protocols were carried out, a daylong meal study and a standardized metabolic challenge fat-load study. As seen in Figure 3, the triacylglycerol response was not different for the 3 diets during the daylong study, whereas glucose concentrations were lower in the postlunch phase of the daylong study for the CHO diet (Figure 3). Insulin concentrations were highest with the AAD diet during the daylong study (Figure 3). No differences were observed between the 3 diets during the fat-load study (data not shown). We then analyzed the postprandial responses in subjects with and without the metabolic syndrome or insulin resistance. Daylong and post fat-load triacylglycerol or insulin concentrations did not differ between the diets in these subgroups (data not shown). Although no between-diet difference was seen in glucose concentrations during the daylong study, somewhat more pronounced differences were observed during the standardized fat load in subjects with the metabolic syndrome, although the differences were not significant (Figure 4).

DISCUSSION

Diet modification is universally adopted as an important and early intervention approach to modifying risk factors for CAD. It is well documented that dietary SFA is associated with an increased prevalence of CAD and that the intake of polyunsaturated fatty acids (PUFAs) reduces cardiovascular morbidity (34, 35). Meta analyses and reviews of numerous dietary trials have concluded that as a replacement for SFA, either MUFA or carbohydrates lower LDL cholesterol with equal effectiveness (36 – 38). Whether a reduction in SFA intake, particularly via its replacement with MUFAs or carbohydrates, results in a reduction in cardiovascular morbidity or mortality has not been conclusively established. The choice of replacement for SFA might also affect the potential for weight reduction. With an increasing frequency of a clustering of metabolic abnormalities that accentuate cardiovascular risk (1– 6, 39), examination of the preferred replacement for SFA in the diet needs to focus not only on lowering LDL cholesterol but also on the effects of replacement nutrients on other metabolic risk factors. This issue is of particular importance to the large segment of the population who are at risk of CAD because of an unfavorable and complex metabolic milieu that includes low HDL-cholesterol concentrations, high triacylglycerol concentrations, or insulin concentrations. In the present randomized, double-blind, controlled feeding study, we directly compared the ability of 2 approaches for reducing SFA to favorably affect plasma lipids, lipoproteins, and indexes of glucose metabolism in a study population with a high prevalence of one or more metabolic abnormalities. The present study confirms the primary importance of reducing dietary SFAs, independent of changes in total dietary fat, to

1616

BERGLUND ET AL

FIGURE 1. Mean (앐SD) differences in LDL-cholesterol, HDL-cholesterol, and log triacylglycerol (TG) concentrations between diet groups in participants classified by HDL-cholesterol or TG concentrations. Participants were either classified as those with low HDL (HDL cholesterol 울 30th percentile; n ҃ 45) or normal HDL (HDL cholesterol 쏜 30th percentile; n ҃ 40) based on average values obtained while consuming the average American diet (AAD) or as those with high TG (욷70th percentile; n ҃ 39) or normal TG (쏝70th percentile; n ҃ 46) based on average values obtained while consuming the AAD. MUFA diet, 7% of energy from saturated fatty acids replaced with monounsaturated fatty acids; CHO diet, 7% of energy from saturated fatty acids replaced with carbohydrates. Significant diet effects within each HDL or TG group: *P 쏝 0.05, **P 쏝 0.01, ***P 쏝 0.001, ****P 쏝 0.0001.

achieve reductions in LDL cholesterol concentrations. Replacement of 7% of energy from SFAs in the AAD with either carbohydrate or MUFA produced an equivalent 6 –7% reduction in

LDL cholesterol, consistent with previous observations (16, 40). This level of LDL cholesterol lowering would, by itself, be predicted to reduce CAD risk by 앒10% (41). In the recent Women’s

TABLE 4 Classification of subjects as having the metabolic syndrome or insulin resistance on the basis of lipid concentrations during the average American diet1

Total cholesterol (mmol/L) LDL cholesterol (mmol/L) HDL cholesterol (mmol/L) Triacylglycerols (mmol/L)

Metabolic syndrome (n ҃ 20)

No metabolic syndrome (n ҃ 62)

P

HOMA 쏜 3 (n ҃ 28)

HOMA 울 3 (n ҃ 57)

P

5.76 앐 0.78 3.64 앐 0.52 0.98 앐 0.18 2.28 앐 55%

4.96 앐 0.75 3.23 앐 0.65 1.11 앐 0.23 1.29 앐 38%

쏝0.01 쏝0.01 0.02 쏝0.01

5.37 앐 0.75 3.33 앐 0.54 1.03 앐 0.21 2.00 앐 50%

5.04 앐 0.85 3.31 앐 0.67 1.06 앐 0.26 1.27 앐 41%

NS NS NS 쏝0.001

1 All values are x៮ 앐 SD. HOMA, homeostasis model assessment. Subjects were classified as having the metabolic syndrome if any 3 of the 5 defined characteristics of the syndrome were present at baseline. Subjects with baseline HOMA 쏜 3 were classified as having insulin resistance.

1617

0

∆ HDL Cholesterol (mmol/L)

∆ HDL Cholesterol (mmol/L)

PREFERRED DIET TO REDUCE CAD RISK

-0.02 -0.04 -0.06 -0.08 -0.1 MetSyn (n=20) MUFA-AAD

-0.02 -0.04 -0.06 -0.08 -0.1

Others (n=65)

HOMA>3 (n=28)

CHO-AAD

MUFA-AAD

Others (n=57) CHO-AAD

15 ∆TG (%)

15 ∆ TG (%)

0

10 5

10 5

0

0

-5

-5

-10

-10 MetSyn (n=20) MUFA-AAD

Others (n=65) CHO-AAD

HOMA>3 (n=28) MUFA-AAD

Others (n=57) CHO-AAD

FIGURE 2. Mean (앐SD) differences in HDL-cholesterol and triacylglycerol (TG) concentrations between diets in participants classified by the presence or absence of the metabolic syndrome (MetSyn) or insulin resistance, defined as a homeostasis model assessment (HOMA) index 쏜3.0. AAD, average American diet; MUFA diet, 7% of energy from saturated fatty acids replaced with monounsaturated fatty acids; CHO diet, 7% of energy from saturated fatty acids replaced with carbohydrates.

Health Initiative randomized controlled dietary modification trial, where a dietary intervention based on fat reduction and an increase in vegetables, fruit, and grains did not result in any improvement in cardiovascular risk, the reduction in LDL cholesterol achieved was smaller, although still significant, and no changes in triacylglycerol or HDL-cholesterol concentrations were found (42). However, women with the lowest SFA intake (쏝6.1%) had a significantly reduced hazard ratio for cardiovascular disease, which suggested that a more pronounced change in risk factor levels is needed to achieve improvement in cardiovascular risk. With respect to the effect of the diets on HDL cholesterol, 2 findings are of note. First, as previously reported, HDLcholesterol concentrations fell when SFAs were replaced by either carbohydrates or MUFAs. Second, the difference in average HDL-cholesterol concentrations between the CHO and the MUFA diets was only 0.03 mmol/L. Earlier studies that showed beneficial effects of MUFA diets on HDL-cholesterol concentrations examined greater contrasts in total fat levels (38 –50% compared with 20 –25% of energy as fat) (43– 47). However, when diets high in MUFAs have been compared against diets with total fat levels consistent with AHA Step 1 recommendations, differences in HDL-cholesterol concentrations have been typically 쏝0.05 mmol/L, similar to our findings (16, 48, 49). Furthermore, our results are very similar to the findings in the OmniHeart study (25). We found a negative correlation between the AAD HDLcholesterol concentration and the difference in HDL-cholesterol concentrations between the AAD diet and the CHO diet. Furthermore, compared with participants with normal HDLcholesterol concentrations, participants with AAD HDL cholesterol 쏝 30th percentile had smaller reductions in HDL

cholesterol with both the CHO and MUFA diet. Thus, although there was a clear advantage of an MUFA diet in maintaining HDL-cholesterol concentrations in participants with normal HDL-cholesterol concentrations, this advantage was substantially diminished in participants with low HDL cholesterol. In addition, there are potentially unfavorable effects of dietary changes on plasma triacylglycerol concentrations. We observed moderate differences between the replacement diets in the present study: relative to the AAD, triacylglycerol concentrations were higher with the CHO diet and lower with the MUFA diet. This finding is in agreement with the results for healthy subjects by Appel et al (25). Notably, essentially all of the reductions in triacylglycerol associated with the MUFA diet could be attributed to the response of those participants with AAD triacylglycerol concentrations 욷 70th percentile, where triacylglycerol concentrations were 10% lower. In this group, no difference in triacylglycerol concentrations was seen between the AAD and CHO diets. This suggests that, in some individuals, SFAs are both hypercholesterolemic and hypertriacylglycerolemic (50, 51). We also investigated the responses to the 3 diets between subgroups of subjects with or without the metabolic syndrome and with or without insulin resistance. The differences in the lipid responses to replacement of SFAs with CHO or MUFA were similar to those observed for the overall study group. These results reinforce the conclusion that diet modulation for subjects with a more broadly identified metabolic risk milieu is of importance in the prevention of CAD. In addition to fasting lipid concentrations, postprandial lipemia is an emerging indicator of cardiovascular risk (20, 52). We did not observe any significant difference between the diets in response to either daylong meals or a high-fat load.

1618

BERGLUND ET AL

risk factors, the overall emphasis on LDL-cholesterol concentrations as the target of therapy suggests that reductions in LDL cholesterol should be a primary focus. Our study indicates that reductions in HDL cholesterol associated with a reduced saturated fat intake were variable and significantly less during the MUFA diet than during the CHO diet. We recognize that our study had some limitations. We maintained the weights of our subjects during the study and could not, therefore, address the issue of dietary effects on lipid concentrations under “free-living conditions.” Reductions in body weight and body fat are associated with favorable changes in coronary heart disease risk factors, and, in outpatient studies, weight loss with low-fat diets is often observed (11, 57, 58). However the extent of weight loss that can be achieved with a CHO diet in a free-living population may be modest (42, 59) and insufficient to reverse the potential adverse changes in HDL cholesterol and triacylglycerol. Overall, our results suggest that, in individuals considered at increased risk of CAD because of an unfavorable metabolic setting, the replacement of dietary SFA with MUFA rather than CHO is preferred because of associated smaller reductions in HDL cholesterol and a trend toward a reduction in fasting triacylglycerol concentrations. Diets lower in SFAs and higher in MUFAs may be particularly beneficial in individuals with normal HDL-cholesterol concentrations or with higher triacylglycerol concentrations. Our study further suggests that rather than relying on a single dietary recommendation for everyone at risk of CAD, individualized dietary recommendations

FIGURE 3. Mean (앐SD) changes from baseline in concentrations of triacylglycerol (TG), glucose, and insulin in 68 subjects (43 men and 25 women) in response to daylong meals or a standardized fat load. Blood samples were drawn before the fat load (fasting) and 4 and 8 h after the fat load. AAD, average American diet; MUFA diet, 7% of energy from saturated fatty acids replaced with monounsaturated fatty acids; CHO diet, 7% of energy from saturated fatty acids replaced with carbohydrates.

A secondary aim of this study was to investigate dietary effects on plasma Lp(a) concentrations. It has been shown that reductions in saturated fat intake have increased Lp(a) concentrations (53–56). In agreement with these findings, we previously reported that reductions in dietary fat and SFAs led to significant increases in Lp(a) (17). Because Lp(a) concentrations increased relative to the AAD with both the CHO and the MUFA diets, our results suggest a specific effect of SFA reductions on Lp(a) concentrations. Thus, a reduction in saturated fat intake has broad net effects on lipid and lipoprotein concentrations: reduced LDLcholesterol concentrations, increased Lp(a) concentrations, and reduced HDL-cholesterol concentrations. Although these changes indicate a somewhat complex pattern with regard to lipid

FIGURE 4. Mean (앐SD) changes from baseline in glucose in response to daylong meals or to a standardized fat load in subjects with the metabolic syndrome. Blood samples were drawn before the fat load (fasting) and 4 and 8 h after the fat load. AAD, average American diet (n ҃ 17); MUFA diet, 7% of energy from saturated fatty acids replaced with monounsaturated fatty acids (n ҃ 18 for daylong and n ҃ 16 for fat load); CHO diet, 7% of energy from saturated fatty acids replaced with carbohydrates (n ҃ 17).

PREFERRED DIET TO REDUCE CAD RISK

based on underlying risk factors may be a more effective approach for CAD prevention. DELTA Research Investigators: Columbia University (Henry N Ginsberg, Principal Investigator; Rajasekhar Ramakrishnan; Wahida Karmally; Lars Berglund; Maliha Siddiqui; Niem-Tzu Chen; Steve Holleran; Colleen Johnson; Roberta Holeman; Karen Chirgwin; Kellye Stennett; Lencey Ganga; Tajudeen T Towalawi; Minnie Myers; Colleen Ngai; Nelson Fontanez; Jeff Jones; Carmen Rodriguez; and Norma Useche), Pennington Biomedical Research Center (Michael Lefevre and Paul S Roheim, Co-Principal Investigators; Donna H Ryan; Marlene M Most; Catherine M Champagne; Donald Williamson; Richard Tulley; Ricky Brock; Deonne Bodin; Betty Kennedy; Michelle Barkate; and Elizabeth Foust), Pennsylvania State University (Penny Kris-Etherton, Principal Investigator; Satya S Jonnalagadda; Janice Derr; Abir Farhat-Wood; Vikkie A Mustad; Kate Meaker; Edward Mills; Mary-Ann Tilley; Helen Smiciklas-Wright; Madeleine SigmanGrant; Jean-Xavier Guinard; Pamela Sechevich; C Channa Reddy; Andrea M Mastro; and Allen D Cooper), University of Minnesota (Patricia Elmer, Principal Investigator; Aaron Folsom; Nancy Van Heel; Christine Wold; Kay Fritz; Joanne Slavin; and David Jacobs), University of North Carolina at Chapel Hill, Coordinating Center (Barbara H Dennis, Principal Investigator; Paul Stewart; CE Davis; James Hosking; Nancy Anderson; Susan Blackwell; Lynn Martin; Hope Bryan; W Brian Stewart; Jeffrey Abolafia; Malachy Foley; Conroy Zien; Szu-Yun Leu; Marston Youngblood; Thomas Goodwin; Monica Miles; and Jennifer Wehbie), Mary Imogene Bassett Research Institute (Thomas A Pearson and Roberta Reed), University of Vermont (Russell Tracy and Elaine Cornell), Virginia Polytechnic and State University (Kent K Stewart and Katherine M Phillips), Southern University (Bernestine B McGee and Brenda Williams), Beltsville Agricultural Research Center (Gary R Beecher, Joanne M Holden, and Carol S Davis), and the National Heart, Lung, and Blood Institute (Abby G Ershow, David J Gordon, Michael Proschan, and Basil M Rifkind). All authors contributed to the planning of the studies, the recruitment and study visits, the laboratory and statistical analyses, and the drafting of the manuscript. None of the authors had a conflict of interest.

12. 13. 14. 15.

16.

17. 18. 19. 20. 21. 22. 23. 24. 25.

REFERENCES 1. Reaven GM. Banting lecture 1988. Role of insulin resistance in human disease. Diabetes 1988;37:1595– 607. 2. Grundy SM. Does a diagnosis of metabolic syndrome have value in clinical practice? Am J Clin Nutr 2006;83:1248 –51. 3. Reaven GM. The metabolic syndrome: is this diagnosis necessary? Am J Clin Nutr 2006;83:1237– 47. 4. Grundy SM, Cleeman JI, Daniels SR, et al. Diagnosis and management of the metabolic syndrome: a statement for health care professionals. An American Heart Association/National Heart, Lung and Blood Institute scientific statement. Circulation 2005;112:2735–52. 5. Reaven GM. Role of insulin resistance in human disease (syndrome X): an expanded definition. Annu Rev Med 1993;44:121–31. 6. Kahn R, Buse J, Ferrannini E, Stern M. The metabolic syndrome: time for a critical appraisal: joint statement from the American Diabetes Association and the European Association for the Study of Diabetes. Diabetes Care 2005;28:2289 –304. 7. Genest J Jr, McNamara JR, Ordovas JM, et al. Lipoprotein cholesterol, apolipoprotein A-I and B and lipoprotein (a) abnormalities in men with premature coronary artery disease. J Am Coll Cardiol 1992;19:792– 802. 8. Despres JP, Lamarche B, Mauriege P, et al. Hyperinsulinemia as an independent risk factor for ischemic heart disease. N Engl J Med 1996; 334:952–7. 9. Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults. Executive Summary of the Third Report of the National Cholesterol Education Program (NCEP) Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (Adult Treatment Panel III). JAMA 2000;285:2486 –97. 10. American Heart Association Nutrition Committee, Lichtenstein AH, Appel LJ, Brands M, et al. Diet and lifestyle recommendations revision 2006: a scientific statement from the American Heart Association Nutrition Committee. Circulation 2006;114:82–96. 11. Bae CY, Keenan JM, Wenz J, McCaffrey DJ. A clinical trial of the

26. 27. 28. 29. 30. 31.

32. 33. 34. 35.

36.

1619

American Heart Association step one diet for treatment of hypercholesterolemia. J Fam Pract 1991;33:249 –54. Grundy SM, Nix D, Whelan MF, Franklin L. Comparison of three cholesterol-lowering diets in normolipidemic men. JAMA 1986;256: 2351–5. Denke MA. Individual responsiveness to a cholesterol-lowering diet in postmenopausal women with moderate hypercholesterolemia. Arch Intern Med 1994;154:1977– 82. Denke MA, Grundy SM. Individual responses to a cholesterol-lowering diet in 50 men with moderate hypercholesterolemia. Arch Intern Med 1994;154:317–25. Barr SL, Ramakrishnan R, Johnson C, Holleran S, Dell RB, Ginsberg HN. Reducing total dietary fat without reducing saturated fatty acids does not significantly lower total plasma cholesterol concentrations in normal males. Am J Clin Nutr 1992;55:675– 81. Ginsberg HN, Barr SL, Gilbert A, et al. Reduction of plasma cholesterol levels in normal men on an American Heart Association CHO diet or a CHO diet with added monounsaturated fat. N Engl J Med 1990;322: 574 –9. Ginsberg HN, Kris-Etherton PM, Dennis B, et al. Effects of reducing dietary saturated fatty acids on plasma lipids and lipoproteins in healthy subjects. Arterioscler Thromb Vasc Biol 1998;18:441–9. Mensink RP, Katan MB. Effect of dietary fatty acids on serum lipids and lipoproteins. A meta-analysis of 27 trials. Arterioscler Thromb 1992;12: 911–9. Grundy SM, Denke MA. Dietary influences on serum lipids and lipoproteins. J Lipid Res 1990;31:1149 –72. Hyson D, Rutledge JC, Berglund L. Postprandial lipemia and cardiovascular disease. Curr Atheroscler Rep 2003;5:437– 44. Parks E, Hellerstein MK. Carbohydrate-induced hypertriacylglycerolemia: an historical perspective and review of biological mechanisms. Am J Clin Nutr 2000;71:412– 43. Howard BV, Wylie-Rosett J. Sugar and cardiovascular disease. Circulation 2002;106:523–7. Lara-Castro C, Garvey WT. Diet, insulin resistance, and obesity: zoning in on data for Atkins dieters living in South Beach. J Clin Endocrinol Metab 2004;89:4197–205. Schwenke DC. Insulin resistance, low-fat diets, and low-carbohydrate diets: time to test new menus. Curr Opin Lipidol 2005;16:55– 60. Appel LJ, Sacks FM, Carey VJ, et al for the OmniHeart Collaborative Research Group. Effects of protein, monounsaturated fat, and carbohydrate intake on blood pressure and serum lipids. Results of the OmniHeart Randomized Trial. JAMA 2005;294:2455– 64. Dennis BH, Stewart P, Wang CH, et al. Diet design for a multicenter controlled feeding trial: The DELTA Program. J Am Diet Assoc 1998; 98:766 –76. Li BW, Schuhmann PJ, Wolf WR. Chromatographic determination of sugars and starch in a diet composite reference material. J Agric Food Chem 1985;33:531– 6. Li BW, Marshall MW, Andrews KW, Adams TT. Analysis of individual foods for the validation of sugars and starch contents of composited diets. J Food Comp Anal 1988;1:152– 8. Thivend P, Mercier C, Guilbot A. Determination of starch with glycoamylase. In: Whistler RC, BeMiller JN, eds. Methods in carbohydrate chemistry. Vol 6. New York, NY: Academic Press, 1972:100 –5. Cranker KJ, Phillips KM, Stewart KK. Fine tuning a bile-enzymaticgravimetric total dietary fiber method. J Assoc Off Anal Chem Int 1997; 80:89 –94. Tuck CH, Holleran S, Berglund L. Hormonal regulation of lipoprotein (a) levels: effect of estrogen replacement therapy on lipoprotein (a) and acute phase reactants in postmenopausal women. Arterioscler Thromb Vasc Biol 1997;17:1822–9. Muller KE, Stewart PW. Linear model theory: univariate, multivariate, and mixed models. New York, NY: Wiley, 2006. SAS Institute Inc. SAS/STAT user’s guide, release 6.07. Cary, NC: SAS Institute Inc, 1992. (Technical report P-229.) Kris-Etherton PM, Hecker KD, Binkoski AE. Polyunsaturated fatty acids and cardiovascular health. Nutr Rev 2004;62:414 –26. Wang C, Harris WS, Chung M, et al. nҀ3 Fatty acids from fish or fish-oil supplements, but not ␣–linolenic acid, benefit cardiovascular disease outcomes in primary- and secondary-prevention studies: a systematic review. Am J Clin Nutr 2006;84:5–17. Derr J, Kris-Etherton PM, Pearson TA, Seligson FH. The role of fatty acid saturation on plasma lipids, lipoproteins, and apolipoproteins: II.

1620

37. 38. 39. 40.

41. 42. 43.

44. 45.

46. 47.

BERGLUND ET AL

The plasma total and low-density lipoprotein cholesterol response of individual fatty acids. Metabolism 1993;42:130 – 4. Hegsted DM, Ausman LM, Johnson JA, Dallal GE. Dietary fat and serum lipids: an evaluation of the experimental data. Am J Clin Nutr 1993;57:875– 83. Kris-Etherton PM, Yu S. Individual fatty acid effects on plasma lipids and lipoproteins: human studies. Am J Clin Nutr 1997;65(suppl):1628S– 44S. Ford ES. Risks for all-cause mortality, cardiovascular disease, and diabetes associated with the metabolic syndrome. Diabetes Care 2005;28: 1769 –78. Yu-Poth S, Zhao G, Etherton T, Naglak M, Jonnalagadda S, KrisEtherton PM. Effects of the National Cholesterol Education Program’s Step I and Step II dietary intervention programs on cardiovascular disease risk factors: a meta-analysis. Am J Clin Nutr 1999;69:632– 46. Anderson KM, Wilson PFW, Odell PM, Kannel WB. An updated coronary risk profile. A statement for health professionals. Circulation 1991;83:356 – 62. Howard BV, Van Horn L, Hsia J, et al. Low-fat dietary pattern and risk of cardiovascular disease. The Women’s Health Initiative randomized controlled dietary modification trial. JAMA 2006;295:655– 66. Garg A, Bonanome A, Grundy SM, Zhang ZJ, Unger RH. Comparison of a high-carbohydrate diet with a high-monounsaturated-fat diet in patients with non-insulin-dependent diabetes mellitus. N Engl J Med 1988;319:829 –34. Mensink RP, Katan MB. Effect of monounsaturated fatty acids versus complex carbohydrates on high-density lipoproteins in healthy men and women. Lancet 1987;1:122–5. Mensink RP, de Groot MJ, van den Broeke LT, Severijnen-Nobels AP, Demacker PN, Katan MB. Effects of monounsaturated fatty acids v complex carbohydrates on serum lipoproteins and apoproteins in healthy men and women. Metabolism 1989;38:172– 8. Grundy SM. Comparison of monounsaturated fatty acids and carbohydrates for lowering plasma cholesterol. N Engl J Med 1986;314:745– 8. Grundy SM, Florentin L, Nix D, Whelan MF. Comparison of monounsaturated fatty acids and carbohydrates for reducing raised levels of plasma cholesterol in man. Am J Clin Nutr 1988;47:965–9.

48. Garg A, Bantle JP, Henry RR, et al. Effects of varying carbohydrate content of diet in patients with non-inusulin-dependent diabetes mellitus. JAMA 1994;271:1421– 8. 49. Baggio G, Pagnan A, Muraca M, et al. Olive-oil-enriched diet: effect on serum lipoprotein levels and biliary cholesterol saturation. Am J Clin Nutr 1988;47:960 – 4. 50. Vega GL, Groszek E, Wolf R, Grundy SM. Influence of polyunsaturated fats on composition of plasma lipoproteins and apolipoproteins. J Lipid Res 1982;23:811–22. 51. Cortese C, Levy Y, Janus ED, et al. Modes of action of lipid-lowering diets in man: studies of apolipoprotein B kinetics in relation to fat consumption and dietary fatty acid composition. Eur J Clin Invest 1983;13: 79 – 85. 52. Ginsberg HN, Illingworth DR. Postprandial dyslipidemia: an atherogenic disorder common in patients with diabetes mellitus. Am J Cardiol 2001;88:9H–15H. 53. Tholstrup T, Marckmann P, Vessby B, Sandstrom B. Effects of fats high in individual saturated fatty acids on plasma lipoprotein[a] levels in young healthy men. J Lipid Res 1995;36:1447–52. 54. Clevidence BA, Judd JT, Schaefer EJ, et al. Plasma lipoprotein (a) levels in men and women consuming diets enriched in saturated, cis-, or transmonounsaturated fatty acids. Arterioscler Thromb Vasc Biol 1997;17: 1657– 61. 55. Silaste ML, Rantala M, Alfthan G, et al. Changes in dietary fat intake alter plasma levels of oxidized low-density lipoprotein and lipoprotein (a). Arterioscler Thromb Vasc Biol 2004;24:498 –503. 56. Berglund L. Diet and drug therapy for lipoprotein (a). Curr Opin Lipidol 1995;6:48 –56. 57. Dattilo AM, Kris-Etherton PM. Effects of weight reduction on blood lipids and lipoproteins: a meta-analysis. Am J Clin Nutr 1992;56:320 – 8. 58. Hunninghake DB, Stein EA, Dujovne CA, et al. The efficacy of intensive dietary therapy alone or combined with lovastatin in outpatients with hypercholesterolemia. N Engl J Med 1993;328:1213–9. 59. Katan MB, Grundy SM, Willett WC. Beyond low-fat diets. N Engl J Med 1997;337:563– 6.

Comparison of the effects of fish and fish-oil capsules on the n–3 fatty acid content of blood cells and plasma phospholipids1–3 William S Harris, James V Pottala, Scott A Sands, and Philip G Jones ABSTRACT Background: n–3 Fatty acids (FAs) have been shown to be beneficial for cardiovascular health. Whether n–3 FAs from oily fish consumed weekly or from fish-oil capsules taken daily are equally bioavailable is not clear. Objective: The purpose of this study was to compare the rate and extent of enrichment of blood cell membranes [ie, red blood cells (RBCs)] and plasma phospholipids with n–3 FAs from these 2 sources. Design: Healthy premenopausal female volunteers were randomly assigned to consume a daily average of 485 mg eicosapentaenoic (EPA) and docosahexaenoic (DHA) acids either from 2 servings of oily fish (ie, salmon and albacore tuna) per week or from 1–2 capsules/d. Results: After 16 wk, EPAѿDHA in RBCs in the fish group (n ҃ 11) increased from 4.0 앐 0.6% of total FAs to 6.2 앐 1.4%, whereas it rose from 4.3 앐 1.0% to 6.2 앐 1.4% in the capsule group (P 쏝 0.0001 for both; NS for group effect). Similar results were observed in plasma phospholipids. EPAѿDHA stabilized in the latter after 4 wk but continued to rise through week 16 in RBCs. EPA in RBCs increased significantly (P ҃ 0.01) more rapidly in the fish group than in the capsule group during the first 4 wk, but rates did not differ significantly between groups thereafter. Total FA variances were less in RBCs than in plasma phospholipids (P ҃ 0.04). Conclusion: These findings suggest that the consumption of equal amounts of EPA and DHA from oily fish on a weekly basis or from fish-oil capsules on a daily basis is equally effective at enriching blood lipids with n–3 FAs. Am J Clin Nutr 2007;86:1621–5. KEY WORDS n–3 Fatty acids, eicosapentaenoic acid, docosahexaenoic acid, fish, fish oil, erythrocytes, phospholipids INTRODUCTION

A greater intake of the long-chain n–3 fatty acids (FAs) eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA) has been recommended by a variety of scientific and regulatory bodies to reduce population risk for coronary artery disease (CAD) (1). The American Heart Association has divided its recommendations into 3 categories: persons without known CAD, those with CAD, and those with elevated serum triacylglycerol concentrations (2). The intakes suggested for persons in those categories are, respectively, 욷2 fish meals/wk (preferably oily fish), 앒1 g EPAѿDHA/d (from oily fish or from capsules), and 2– 4 g EPAѿDHA/d from capsules. A presumption of an equivalence between oily fish and fish-oil capsules is inherent in these recommendations, and yet few studies have examined whether n–3

FAs are equally bioavailable from these 2 very different matrixes. In a pilot study including 4 – 8 volunteers who had consumed either salmon or fish-oil supplements, Visioli et al (3) found similar increases in serum phospholipid EPA (per mg EPA consumed) from either source, but a 3 times greater increase in serum DHA concentrations when the DHA came from salmon than when it came from fish-oil capsules. This study was small and retrospective; moreover, the EPA and DHA intakes from these 2 sources were not matched, and the study was not randomized. A larger study compared the effects of eating salmon and taking cod liver oil on serum n–3 FAs; it found that the former was a more efficient vehicle (4). However, in neither of these studies were EPA and DHA intakes prospectively matched. In the present study, we sought to compare the effects of equivalent, nutritionally relevant intakes of marine n–3 FAs from oily fish given 2 times/wk and of daily fish-oil capsule supplementation on the EPA and DHA content of blood cell membranes [(ie, red blood cells, RBCs, or the omega-3 index, a proposed risk marker for cardiovascular mortality (5)] and of the plasma phospholipid (PPL) fraction. We also compared the tolerability of each approach. SUBJECTS AND METHODS

Subjects Women between 21 and 49 y old who were premenopausal and not pregnant or nursing and who had a body mass index (in kg/m2) 쏝30 were recruited. Exclusions were gastrointestinal disorders that could interfere with fat absorption, an intention to lose weight, consumption of 쏜2 alcoholic drinks/d, and regular (쏜2 times/mo) consumption of tuna or salmon supplemented with fish-oil or flaxseed oil capsules. Written informed consent was obtained from all participants. The protocol was approved by the St Luke’s Hospital Institutional Review Board. 1 From the Lipid and Diabetes Research Center, Saint Luke’s Mid America Heart Institute, Kansas City, MO. 2 Supported by grants from the Saint Luke’s Hospital Foundation, Kansas City, MO, and a Fisheries Scholarship grant from the National Fisheries Institute (to WSH). 3 Reprints not available. Address correspondence to WS Harris, Nutrition and Metabolic Diseases Research Institute, Sanford Research/USD and the Sanford School of Medicine, University of South Dakota, 1400 West 22nd Street, Sioux Falls, SD 57105. E-mail: [email protected]. Received March 18, 2007. Accepted for publication August 7, 2007.

Am J Clin Nutr 2007;86:1621–5. Printed in USA. © 2007 American Society for Nutrition

1621

1622

HARRIS ET AL

Study design After qualifying for the trial, women were randomly assigned to either the fish (n ҃ 11) or capsule (n ҃ 12) group. Those in the former group were asked to consume every 2 wk three 171-g (6-ounce) cans of albacore tuna (StarKist; Del Monte Foods, San Francisco, CA) and one 171-g fillet of Norwegian Atlantic– farmed salmon (American Seafood International, New Bedford, MA). The cans of tuna and frozen salmon fillets were provided (along with recipes) for consumption at home. The fish could be eaten for more than one meal, but all servings had to be consumed within 24 h. The tuna and salmon could not be consumed on the same day. The importance of complete consumption was stressed but not directly monitored. Those women randomly assigned to the capsule group were asked to take 17 n–3 FA capsules (Omega-3; CardioTabs, Kansas City, MO) at a rate of 1–2/d according to a provided schedule. This particular n–3 FA supplement was chosen because its FA composition is more reflective of that of salmon and tuna (ie, it is DHA rich) than is that of most supplements, which typically are EPA rich. To ascertain the number of capsules to prescribe, we analyzed 3 capsules, 3 fillets of salmon, and 3 cans of albacore tuna— each in triplicate. For the capsules, an oil aliquot was removed and weighed. The internal standard (17:0) was added, the sample was methylated and analyzed by gas chromatography (GC) as described below. For the fish, the complete cooked salmon fillet and the contents of the entire can of albacore tuna were weighed and homogenized, and 3 aliquots were extracted according to a method of Bligh and Dyer (6). The albacore tuna averaged 185 앐 29 mg EPA and 1010 앐 150 mg DHA (1195 mg EPAѿDHA) per can. The salmon averaged 777 앐 222 mg EPA and 2429 앐 797 mg DHA per fillet. Accordingly, 3 cans of albacore tuna and one salmon fillet provided 1333 mg EPA and 5460 mg DHA over 2 wk, for a total daily average of 485 mg EPAѿDHA (95 mg ѿ 390 mg, respectively). Each capsule contained 86 앐 2 mg EPA and 311 앐 12 mg DHA. Seventeen capsules, taken over a 2-wk span, provided an average total of 482 mg EPAѿDHA/d (104 mg ѿ 378 mg, respectively). The average amount of linoleic acid and arachidonic acid (AA) provided by the fish and the capsules was 22 and 44 mg/d and 8 and 8 mg/d, respectively. The duration of the study was 16 wk, and clinic visits were scheduled every 2 wk. Tolerability of the fish and the fish-oil capsules was compared by using a questionnaire. Subjects were asked 3 questions at the end of the study: 1) Did you experience any fishy aftertaste (ie, “burps”) during the study? 2) If so, how frequent were the “burps”? 3) If so, how unpleasant were they?

쏝5% of the blood samples), no value for LDL cholesterol was reported. CVs for all of these assays are 쏝3%. Analysis of n–3 fatty acids Blood cell membranes Frozen whole blood was thawed, hemolyzed in water (1:14), and spun for 5 min at 4 °C at 2800 ҂ g in an ultracentrifuge (TL100 equipped with a TLA-100.3 rotor; Beckman Instruments, Fullerton, CA). The supernatant fluid (containing hemoglobin and serum lipids) was discarded, and the pellet (almost entirely composed of RBC membranes) was suspended in 1 mL boron-trifluoride methanol (BF3; Sigma, St Louis, MO) and transferred to a screw-cap test tube. The tubes were heated for 10 min at 100 oC to hydrolyze and methylate the membrane glycerophospholipid FAs (9). After the tubes were cooled, water and hexane (1:1) were added, and the tube was briefly shaken and then centrifuged for 3 min at 1500 ҂ g and at room temperature to separate the layers. The upper (hexane) layer was removed, the solvent was evaporated under nitrogen, and the blood sample was resuspended in decane and transferred to a vial for analysis by flame ionization GC. Plasma phospholipids Plasma lipids were extracted according to the method of Carlson (10) with the use of methanol, methylene chloride, and saline, and the phospholipid fraction was isolated by using thinlayer chromatography on silica gel G (Analtech Inc, Newark, DE) with heptane:diethyl ether:formic acid (80:20:2). The phospholipid band was collected and heated for 10 min at 100 oC in BF3 to produce FA methyl esters, which were recovered and prepared for GC analysis as described below. Gas chromatography Fused silica capillary columns (100-m length, 0.25-mm internal diameter, 0.25-␮m film thickness; SP-2560; Supelco, Bellefonte, PA) were used to determine FA composition. The methyl esters were analyzed in gas chromatographs (GC14Al; Shimadzu Scientific Instruments, Columbia, MD for those derived from PPL; GC9A; Shimadzu Scientific Instruments for those derived from RBC membranes). The inclusion of a weighed external standard FA mixture (GLC673b; NuCheck Prep, Elysian, MN) allowed for control of the differences in response factors between the instruments (the response factor for palmitic acid was assumed to be 1.0). Statistical analysis

Analysis of lipids and lipoproteins For measurement of fasting plasma lipids and lipoproteins, blood was drawn after a fast of 욷10 h into tubes containing 1 mg EDTA/mL. Whole plasma triacylglycerol and cholesterol concentrations were measured enzymatically (Cholesterol/HP; Roche Diagnostics, Indianapolis, IN) and by using a triacylglycerol reagent (GPO-Trinder; Bayer Diagnostics, Tarrytown, NY) on a Cobas Fara analyzer (Roche Analytic Instruments Inc, Nutley, NJ) according to the manufacturer’s instructions. Plasma HDL cholesterol was measured after precipitation of the apolipoprotein B– containing lipoproteins (7). LDL-cholesterol concentrations were calculated by using the Friedewald equation (8). If the triacylglycerol concentration was 쏜400 mg/dL (as was seen in

Data not normally distributed (ie, EPA, AA, total cholesterol, triacylglycerol) were log transformed for analysis. The mean response profile was examined across time by treatment group with the use of lowess smoothing, which suggested (see Results) that a piece-wise general linear model would fit well for both the RBC and PPL sample types. Akaike’s Information Criterion was used to compare models with different timepoints, and a knot (ie, inflection point) at 4 wk produced the best fit. After we examined all residual pair-wise correlations from an ordinary least-squares model, we implemented a Toeplitz correlation structure for the repeated measures. Estimated total variances between RBC and PPL FAs were compared by using the F(21, 21) distribution. A post hoc power analysis based on the observed variances found

1623

FISH VERSUS CAPSULES AND BLOOD n–3 FATTY ACIDS

that we would not have been able to detect a between-group difference in RBC EPAѿDHA of 울1.5. Data are presented as means 앐 SDs, and a P value 쏝 0.05 was considered statistically significant. Analyses were performed by using SAS software (version 9.1; SAS Institute Inc, Cary, NC).

TABLE 1 Effects of fish feeding versus capsule supplementation on the proportion of total fatty acids present as eicosapentaenoic acid (EPA), docosahexaenoic acid (DHA), and arachidonic acid (AA) in blood cell membranes and plasma phospholipids1 RBC membranes

RESULTS

Subjects All women were premenopausal. The 2 groups (fish group, n ҃ 11; capsules group, n ҃ 12) did not differ with respect to body mass index (26.1 앐 2.1 and 25.5 앐 2.1, respectively) or racial backgrounds (n ҃ 9, 2, and 0 and 9, 2 and 1 for whites, blacks, and Asians, respectively, in the fish and capsules groups, respectively). They did, however, differ significantly (P ҃ 0.01) in age (35 앐 8.7 and 43 앐 3.8 y, respectively), but age was not a significant response predictor in any of the longitudinal analyses. There was no significant weight change during the study in either group (a gain of 0.1 kg and 0.7 kg, respectively). Compliance with the fish-oil capsule protocol was 97 앐 8%, and the participants uniformly reported complete consumption of the provided fish. Baseline red blood cells and plasma phospholipid fatty acid content The baseline EPA content of RBC membranes was 20% lower in the fish group than in the capsule group (P ҃ 0.03), but the EPA content of the PPL did not differ significantly between the groups (Table 1). There were no baseline differences between groups with either the RBC or PPL sample type. Rate of fatty acid increase in plasma phospholipid and red blood cells: capsules compared with fish Rates (ie, slopes) were compared between the PPL and RBC sample types and between groups in 2 phases: the first 4 wk and the last 12 wk (Table 2). The only significant group ҂ time interaction for any FA of interest in either sample type in either period was that for EPA in RBCs during the first 4 wk: the rate of increase of EPA in RBCs was 앒7% faster in the fish group than in the capsule group (P ҃ 0.01). Because there were no other significant group interactions, data from all 23 women (both the fish and capsule groups) were combined, and the responses over time between the PPL and RBC sample types were evaluated. PPL concentrations of EPA rose by 19%/wk over the first month, which was significantly greater than the slope in RBC EPA in the capsule group (P ҃ 0.0003) but not significantly different from that in the fish group (Table 2). EPA did not rise further over the next 3 mo in either sample type. For DHA, during the first month, concentrations rose in PPL more than twice as quickly as in RBCs (Table 2), but RBC DHA continued to rise over the next 3 mo, whereas PPL DHA did not. The summed metric EPAѿDHA had the same response pattern as did DHA alone. AA concentrations decreased significantly (P ҃ 0.004) in PPL (but not in RBCs) during the first phase, and neither slope was different from zero in the second phase. Fatty acid variances in plasma phospholipids and red blood cells Estimated total variances for all 4 FA variables (EPA, DHA, EPAѿDHA, and AA) in PPL and RBCs were analyzed to determine whether one marker was more biologically stable than

EPA Week 0 2 4 6 8 10 12 14 16 DHA Week 0 2 4 6 8 10 12 14 16 AA Week 0 2 4 6 8 10 12 14 16

Plasma phospholipids

Fish

Capsules

Fish

Capsules

0.80 앐 0.12 1.04 앐 0.27 1.43 앐 0.35 1.21 앐 0.17 1.39 앐 0.34 1.26 앐 0.44 1.15 앐 0.36 1.39 앐 0.41 1.34 앐 0.35

0.99 앐 0.172 1.22 앐 0.14 1.27 앐 0.21 1.26 앐 0.30 1.20 앐 0.25 1.32 앐 0.31 1.21 앐 0.18 1.29 앐 0.26 1.30 앐 0.40

0.53 앐 0.25 0.98 앐 0.49 1.83 앐 1.09 1.25 앐 1.10 1.30 앐 0.81 1.10 앐 0.83 1.09 앐 0.67 1.93 앐 1.05 1.52 앐 0.88

0.57 앐 0.11 1.00 앐 0.25 1.00 앐 0.17 1.09 앐 0.32 1.10 앐 0.38 1.12 앐 0.38 1.04 앐 0.39 0.90 앐 0.29 1.02 앐 0.35

3.22 앐 0.58 3.72 앐 0.52 4.13 앐 0.84 4.13 앐 0.67 4.52 앐 0.61 4.50 앐 0.83 4.61 앐 1.06 4.81 앐 1.17 4.83 앐 1.16

3.34 앐 0.79 3.86 앐 0.73 4.02 앐 0.78 4.20 앐 0.86 4.33 앐 0.94 4.54 앐 0.95 4.60 앐 0.99 4.66 앐 1.00 4.86 앐 1.10

3.22 앐 0.55 4.54 앐 0.91 5.18 앐 1.36 5.33 앐 1.23 5.79 앐 1.88 4.65 앐 1.12 5.12 앐 1.06 5.17 앐 0.94 5.28 앐 1.04

2.84 앐 0.80 4.03 앐 0.78 4.31 앐 0.86 5.31 앐 1.76 4.92 앐 1.74 4.73 앐 1.20 4.52 앐 1.15 4.70 앐 1.81 4.53 앐 1.35

15.5 앐 1.9 15.3 앐 1.6 15.1 앐 2.1 15.0 앐 1.9 15.7 앐 1.1 15.5 앐 1.4 15.4 앐 1.7 14.7 앐 1.9 14.8 앐 1.5

16.0 앐 1.4 16.2 앐 1.5 16.1 앐 2.1 15.8 앐 1.9 16.0 앐 2.5 15.4 앐 2.5 15.8 앐 3.0 15.6 앐 1.7 15.7 앐 1.3

13.0 앐 2.5 11.9 앐 1.4 12.2 앐 2.3 11.3 앐 2.2 11.6 앐 1.7 11.1 앐 1.9 11.6 앐 1.9 11.8 앐 1.6 11.7 앐 1.6

12.2 앐 2.6 11.5 앐 2.0 11.3 앐 2.4 11.3 앐 2.6 11.0 앐 2.3 11.4 앐 1.9 11.5 앐 2.2 11.1 앐 2.0 11.1 앐 2.1

1 All values are x៮ 앐 SD. Baseline differences between groups were evaluated by t test. The group ҂ time interaction was significant (P ҃ 0.01) only for red blood cell EPA: the fish group increased more during the first 4 wk than did the capsule group. For time trends by sample type, see Table 2. 2 P 쏝 0.01.

the other. Variances (% of total FAs) were as follows: 0.24 versus 0.6 (P ҃ 0.001) for EPA, 1.66 versus 0.9 (P ҃ 0.09) for DHA; 2.62 versus 1.31 (P ҃ 0.06) for EPAѿDHA; and 0.03 versus 0.01 (P ҃ 0.04) for AA. Hence, variances for RBC FAs were 25–50% of those for PPL FAs. Serum lipid and lipoproteins Effects of capsules and fish on serum lipids were first evaluated for group ҂ time interactions (Figure 1). There were no interactions for total, LDL, or HDL cholesterol; consequently, the groups were pooled and the time effects on each lipid were evaluated. There were time effects for total and LDL cholesterol (Table 2). Mean LDL cholesterol increased from 106 to 115 mg/dL (P ҃ 0.01). There were no effects on HDL cholesterol. There was a significant group ҂ time interaction for triacylglycerol (P ҃ 0.01). Mean plasma triacylglycerols increased from 68

1624

HARRIS ET AL

TABLE 2 Predicted change in fatty acids of interest over time1 EPA2 0 – 4 wk

4 –16 wk

0 – 4 wk

% Red blood cells Plasma phospholipids

3

Interaction 194

EPA ѿ DHA

DHA 4 –16 wk

0 – 4 wk

% 4,5

NS NS

6.3 15.34

AA2

4 –16 wk

0 – 4 wk

% 4,5

2 NS

4,5

7.6 15.14

4 –16 wk %

4,5

1.6 NS

5

NS Ҁ2.24

NS NS

1

EPA, eicosapentaenoic acid; DHA, docosahexaenoic acid; AA, arachidonic acid. Change was measured as the percentage change per week from baseline. Lognormal distribution; otherwise normal. 3 The fish group increased 13.6%/wk, and the capsule group increased 6.3%/wk, for a 7%/wk difference between groups, P ҃ 0.01. 4 Slope significantly different from 0, P 쏝 0.0004. 5 Slope significantly different from plasma phospholipid, P 쏝 0.02. 2

to 85 mg/dL (P ҃ 0.03) in the capsule group but decreased from 104 to 94 mg/dL (P ҃ 0.22) in the fish group over 16 wk. Side effects The frequency of fishy aftertaste was significantly (P 쏝 0.001) higher in the capsule group than in the fish group (10/12 and 1/11, respectively). For the 7 subjects in the capsule group, it occurred 욷1 time/wk, and it was most often considered “mildly unpleasant.” DISCUSSION

This study showed that, regardless of whether n–3 FAs are consumed from oil-rich fish or fish-oil capsules, there was, after 16 wk of treatment, no difference in the effect on the major long-chain n– 6 or n–3 FAs, whether measured in RBCs or in PPL. Over that period, the consumption of 앒485 mg EPAѿDHA/d from fish or capsules produced a 40 –50% rise in RBC EPAѿDHA and a 60 – 80% rise in PPL EPAѿDHA. We found that EPAѿDHA concentrations rose more quickly in PPL than in RBCs; the former stabilized by 4 wk and the latter continued to rise over the span of the 4-mo test period. This finding confirms previous studies indicating that the turnover of EPA and DHA in plasma is faster than that in RBCs (11). The EPA content

Lipid concentration (mg /dL)

250 200 150

*

100 50 0 Fish Total cholesterol

Fish Triacylglycerol

Fish LDL cholesterol

Fish HDL cholesterol

FIGURE 1. Mean (앐SD) effects of twice-weekly fish consumption (n ҃ 11) and daily fish-oil capsule supplementation (n ҃ 12) on serum lipids and lipoproteins at baseline (f) and week 16 (䡺). Significant (P ҃ 0.01) week ҂ group interactions were seen only for triacylglycerols. Week effects were significant for total and LDL cholesterol. For analysis of effects on total, LDL, and HDL cholesterol, groups were pooled. In the aggregate, total and LDL cholesterol increased significantly over time (P ҃ 0.01). *P ҃ 0.025 versus baseline.

of RBCs over weeks 0 – 4 was the only FA compartment (and the only time frame) in which a difference between fish and capsules was detectable; EPA concentrations rose 7% faster in the former than the latter, but only during the first month, after which time the concentrations did not differ significantly between groups. This suggests that, at least in the short-term, EPA may be more bioavailable from fish than from capsules. AA concentrations decreased by 앒2%/wk in PPL during the first month, but they did not change significantly thereafter or in RBCs at any time. Finally, the biological variability of FAs in RBCs was 앒50% of that observed in PPL. Such a difference in variances is not unexpected, because the latter (transported as it is in plasma lipoproteins) is more likely to be subject to day-to-day variation in composition than is the RBC membrane. Essential FAs are known to transfer directly from plasma to RBC membranes (12), and, for linoleic acid at least, to reach a new steady state within 앒2 wk (13). DHA, at 1 g/d, was reported to reach a steady state in PPL within 4 wk and in RBCs in 4 – 6 mo (14). Studies with labeled DHA showed that albumin-bound, nonesterified DHA was poorly transferred to RBCs, whereas DHA esterified in lysophosphatidylcholine was incorporated into RBC (and platelet) membranes within hours of ingestion (15). The factors that influence the rate of incorporation of EPA and DHA into both plasma and tissues deserve further investigation. The AHA has recommended the intake of 앒2 fish meals (preferably oily fish)/wk for primary prevention of CAD. In the present study, that intake provided 앒485 mg EPAѿDHA/d, which is approximately the intake currently recommended by government health agencies in the United Kingdom Britain (16) and Australia and New Zealand (17). The AHA also recommends 앒1 g EPAѿDHA/d for patients with known cardiovascular disease (2). That amount is about twice the dose provided here and thus would be expected to produce twice the increase in RBC EPAѿDHA (ie, an increase to 8% instead of to 6%). Such an effect was previously observed (5). The former concentration has been proposed as a “cardioprotective target value” (5). Minimal effects of this intake of EPAѿDHA on plasma lipids and lipoproteins were anticipated. We found, however, that the capsules produced a small increase in triacylglycerols (which remained within the normal range), but the fish did not. Total and LDL cholesterol rose slightly in the combined groups. Nevertheless, LDL-cholesterol and triacylglyerol concentrations remained in the normal range in both groups. Past studies of fish feeding have rarely fed this small amount of n–3 FAs, but, when

FISH VERSUS CAPSULES AND BLOOD n–3 FATTY ACIDS

examined, neither triacylglycerol nor LDL cholesterol was affected (18). Higher doses of n–3 FAs do lower triacylglycerol and, in some subjects, raise LDL cholesterol (19). Fish consumption was associated with fewer episodes of a fishy aftertaste than was capsule consumption, which suggests increased tolerability of the former. Fish also provides high-quality protein and trace minerals (especially selenium and iodine) that are not provided by supplements. In contrast, capsules are more convenient to consume, and they provide none of the mercury or chemical contaminants (20) that can be found in fish (21). However, a recent risk-benefit analysis indicated that the cardiovascular benefits of consuming fish (ie, salmon) far outweigh (앒400:1) any risks due to the potential presence of these contaminants (21). The present study differed in several ways from previous studies exploring the question of bioavailability (3, 22–24). It examined nutritionally achievable intakes of n–3 FAs (ie, 앒500 mg/d) that are currently recommended by several health organizations and agencies. It compared daily capsules with twice-weekly fish consumption, and 2 forms of n–3 FA were used (ie, ethyl esters in capsules and FAs carried in triacylglycerols and phospholipids in fish). The study was randomized and prospectively designed to compare the effects of these 2 sources of n–3 FAs on 2 commonly used measures of n–3 status: RBCs and PPLs. It is important that both EPA and DHA intakes (not just total n–3 FAs) were matched in the 2 groups. Finally, blood samples were taken at a frequency that would allow tracking of the rate of rise, and (given that the RBC lifespan is 앒16 wk) the study was planned to be long enough to achieve a new steady state in both n–3 markers. This steady state, however, was not achieved: the EPAѿDHA content was still increasing at 16 wk. This was a small study, and we would not have been able to detect relatively small response differences between the groups. Fish and capsule consumption was not directly supervised, so we cannot be certain of daily intakes. Future studies should include a larger sample size, both men and women, a wider variety of ages, and various doses of n–3 FAs; they should last for 쏜4 mo; and they could also include a washout phase to track the rate of clearance of these FAs from plasma and RBC membranes. In conclusion, the EPAѿDHA content of RBCs or PPLs did not differ significantly when equivalent doses of n–3 FAs were provided twice a week from fish or daily from capsules for 4 mo. Accordingly, either source could be used to bolster tissue n–3 concentrations, and evidence from past randomized trials suggests that both would be expected to result in a lower risk of CAD events. We appreciate the critical contributions to the project made by Carrie Robinson (study coordinator), Sheryl Windsor (research unit manager), and Alan Forker (study physician). The authors’ responsibilities were as follows: WSH: conceived of the project, wrote the protocol, obtained the funding and institutional review board approval, and contributed significantly to the manuscript; SAS: conducted laboratory analyses and wrote the first draft of the manuscript; and JVP and PJG: performed the statistical analyses. WSH is a scientific advisor to Monsanto and Reliant Pharmaceuticals, and SAS is employed by OmegaMetrix, LLC (a company that offered n–3 FA blood testing and that is now defunct). The other authors had no personal or financial conflict of interest.

REFERENCES 1. Harris WS. International recommendations for long-chain omega-3 fatty acids. J Cardiovasc Med (in press).

1625

2. Kris-Etherton PM, Harris WS, Appel LJ. Fish consumption, fish oil, omega-3 fatty acids, and cardiovascular disease. Circulation 2002;106: 2747–57. 3. Visioli F, Rise P, Barassi MC, Marangoni F, Galli C. Dietary intake of fish vs. formulations leads to higher plasma concentrations of n–3 fatty acids. Lipids 2003;38:415– 8. 4. Elvevoll EO, Barstad H, Breimo ES, et al. Enhanced incorporation of n–3 fatty acids from fish compared with fish oils. Lipids 2006;41:1109 – 14. 5. Harris WS, von Schacky C. The omega-3 index: a new risk factor for death from coronary heart disease? Prev Med 2004;39:212–20. 6. Bligh EG, Dyer WJ. A rapid method of total lipid extraction and purification. Can J Biochem Physiol 1959;37:911–7. 7. Warnick GR, Benderson J, Albers JJ. Dextran sulfate-Mgѿ2 precipitation procedure for quantitation of high density lipoprotein cholesterol. Clin Chem 1982;28:1379 – 88. 8. Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concentration of low-density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge. Clin Chem 1972;19:499 –502. 9. Morrison WR, Smith LM. Preparation of fatty acid methyl esters and dimethylacetals from lipids with boron fluoride-methanol. J Lipid Res 1964;5:600 – 8. 10. Carlson LA. Extraction of lipids from human whole serum and lipoproteins and from rat liver tissue with methylene chloride-methanol: a comparison with extraction with chloroform-methanol. Clin Chim Acta 1985;149:89 –93. 11. Katan MB, Deslypere JP, van Birgelen AP, Penders M, Zegwaard M. Kinetics of the incorporation of dietary fatty acids into serum cholesteryl esters, erythrocyte membranes, and adipose tissue: an 18-month controlled study. J Lipid Res 1997;38:2012–22. 12. Reed CF. Phospholipid exchange between plasma and erythrocytes in man and the dog. J Clin Invest 1968;47:749 – 60. 13. Skeaff CM, Hodson L, McKenzie JE. Dietary-induced changes in fatty acid composition of human plasma, platelet, and erythrocyte lipids follow a similar time course. J Nutr 2006;136:565–9. 14. Arterburn LM, Hall EB, Oken H. Distribution, interconversion, and dose response of n–3 fatty acids in humans. Am J Clin Nutr 2006;83(suppl): 1467S–76S. 15. Brossard N, Croset M, Normand S, et al. Human plasma albumin transports [13C]docosahexaenoic acid in two lipid forms to blood cells. J Lipid Res 1997;38:1571– 82. 16. Scientific Advisory Committee on Nutrition. Advice on fish consumption: risks and benefits. Internet: http://www.food.gov.uk/news/ newsarchive/2004/jun/fishreport2004 (accessed 2 August 2007). 17. Nutrient reference values for Australia and New Zealand. Internet: http:// www.nhmrc.gov.au/publications/synopses/n35syn.htm (accessed 2 August 2007). 18. Jacques H, Noreau L, Moorjani S. Effect on plasma lipoproteins and endogenous sex hormones of substituting lean white fish for other animal-protein sources in diets of postmenopausal women. Am J Clin Nutr 1992;55:896 –901. 19. Harris WS. N–3 Fatty acids and serum lipoproteins: human studies. Am J Clin Nutr 1997;65(suppl):1645S–54S. 20. Foran SE, Flood JG, Lewandrowski KB. Measurement of mercury levels in concentrated over-the-counter fish oil preparations: is fish oil healthier than fish? Arch Pathol Lab Med 2003;127:1603–5. 21. Mozaffarian D, Rimm EB. Fish intake, contaminants, and human health: evaluating the risks and the benefits. JAMA 2006;296:1885–99. 22. Vidgren HM, Agren JJ, Schwab U, Rissanen T, Hanninen O, Uusitupa MIJ. Incorporation of n–3 fatty acids into plasma lipid fractions, and erythrocyte membranes and platelets during dietary supplementation with fish, fish oil, and docosahexaenoic acid-rich oil among healthy young men. Lipids 1997;32:697–705. 23. Fahrer H, Hoeflin F, Lauterburg BH, Peheim E, Levy A, Vischer TL. Diet and fatty acids: can fish substitute for fish oil? Clin Exper Rheum 1991;9:403– 6. 24. Agren JJ, Hanninen O, Julkunen A, et al. Fish diet, fish oil and docosahexaenoic acid rich oil lower fasting and postprandial plasma lipid levels. Eur J Clin Nutr 1996;50:765–71.

Dietary fiber intake and retinal vascular caliber in the Atherosclerosis Risk in Communities Study1–3 Haidong Kan, June Stevens, Gerardo Heiss, Ronald Klein, Kathryn M Rose, and Stephanie J London ABSTRACT Background: Dietary fiber appears to decrease the risk of cardiovascular morbidity and mortality. Microvascular abnormalities can be observed by retinal examination and contribute to the pathogenesis of various cardiovascular diseases. The impact of dietary fiber on the retinal microvasculature is not known. Objective: We aimed to examine the association between dietary fiber intake and retinal vascular caliber. Design: At the third visit (1993–1995) of the Atherosclerosis Risk in Communities (ARIC) Study, a population-based cohort of adults in 4 US communities, the retinal vascular caliber of 10 659 participants was measured and summarized from digital retinal photographs. Usual dietary intake during the same period was assessed with a 66-item food-frequency questionnaire. Results: After control for potential confounders including hypertension, diabetes, lipids, demographic factors, cigarette smoking, total energy intake, micronutrients intake, and other cardiovascular disease risk factors, higher intake of fiber from all sources and from cereal were significantly associated with wider retinal arteriolar caliber and narrower venular caliber. Participants in the highest quintile of fiber intake from all sources had a 1.05-␮m larger arteriolar caliber (P for trend ҃ 0.012) and a 1.11-␮m smaller venular caliber (P for trend ҃ 0.029). Conclusions: Dietary fiber was related to wider retinal arteriolar caliber and narrower venular caliber, which are associated with a lower risk of cardiovascular disease. These data add to the growing evidence of the benefits of fiber intake on various aspects of cardiovascular pathogenesis. Am J Clin Nutr 2007;86:1626 –32.

convey important information regarding the state of the microcirculation in the eyes and in other vascular beds (23). Several recent prospective studies have shown that a narrower retinal arteriolar diameter independently predicts incident severe hypertension (24), coronary heart disease (25, 26), and diabetes mellitus (27, 28). A wider retinal venular diameter has been associated with an increased risk of stroke, cerebral infarction (29), and cerebral small vessel disease (30). We hypothesized that a higher intake of dietary fiber is associated with a wider retinal arteriolar and a narrower venular caliber. We examined this hypothesis in a population-based cohort of middle-aged men and women. We also examined potential modifying effects of cardiovascular disease risk factors, including sex, smoking status, diabetes status, hypertension, and physical activity. SUBJECTS AND METHODS

Subjects The design and objectives of the Atherosclerosis Risk in Communities (ARIC) Study have been reported in detail (31). Briefly, the ARIC Study is a prospective epidemiologic study of new and established risk factors for atherosclerosis and community trends in coronary heart disease. The study population was selected as a probability sample of 15 792 men and women aged 45– 64 y in Forsyth County, NC; Jackson, MS; selected suburbs of Minneapolis, MN; and Washington County, MD. Eligible participants 1

KEY WORDS Dietary fiber, cardiovascular diseases, microcirculation, retinal abnormalities, cereal INTRODUCTION

Dietary fiber intake is associated with a reduced risk of cardiovascular diseases, including ischemic heart disease (1–7), stroke (2, 7–9), peripheral arterial disease (10), hypertension (11), and atherosclerosis (12–14). The underlying mechanisms of the effect of dietary fiber on the cardiovascular system remain poorly understood, although previous studies have shown that fiber intake can affect blood pressure, systemic inflammation, serum lipid concentrations, postprandial absorption of carbohydrates, insulin sensitivity, fibrinolysis, coagulation, and endothelial cell function (15–21). Microvascular dysfunction has long been implicated as a possible pathogenic factor in the development of various cardiovascular disorders (22). Observation of retinal vascular caliber may

1626

From the Epidemiology Branch, National Institute of Environmental Health Sciences, National Institutes of Health, Department of Health and Human Services, Research Triangle Park, NC (HK and SJL); the Departments of Nutrition and Epidemiology (JS) and the Department of Epidemiology (GH and KMR), School of Public Health, the University of North Carolina at Chapel Hill, NC); and the Department of Ophthalmology and Visual Sciences, University of Wisconsin School of Medicine and Public Health, Madison, WI (RK). 2 Supported by grant Z01 ES043012 from the Intramural Research Program, National Institute of Environmental Health Sciences. The Atherosclerosis Risk in Communities Study is carried out as a collaborative study supported by National Heart, Lung, and Blood Institute contracts N01-HC55015, N01-HC-55016, N01-HC-55018, N01-HC-55019, N01-HC-55020, N01-HC-55021, and N01-HC-55022. 3 Reprints not available. Address correspondence to SJ London, Epidemiology Branch, National Institute of Environmental Health Sciences, PO Box 12233, Mail Drop A3-05, Research Triangle Park, NC 27709. E-mail: [email protected]. Received March 29, 2007. Accepted for publication July 27, 2007.

Am J Clin Nutr 2007;86:1626 –32. Printed in USA. © 2007 American Society for Nutrition

DIETARY FIBER AND RETINAL VASCULAR CALIBER

were interviewed at home and then invited to a baseline clinical examination in 1987–1989. Three further examinations were carried out at approximately 3-y intervals, and participants were contacted annually by telephone between visits to the clinic. Participants for the current analysis are limited to 12 887 who attended the third visit of the ARIC study cohort (1993–1995), at which time the retinal examinations occurred. This represented 86% of cohort survivors. We excluded persons who were of an ethnicity other than African American or white (n ҃ 38) and who were missing data on retinal vascular caliber (n ҃ 1849) or dietary fiber (n ҃ 341). The final study sample consisted of 10 659 adults. The Institutional Review Board of the 4 participating centers approved the study. Measurement of retinal vascular caliber The procedures for retinal photography and the assessment of photographs were described in detail previously (32). Briefly, photographs of the retina were taken from a randomly selected eye after 5 min of dark adaptation. Trained graders, masked to all participant characteristics, used a standardized protocol to evaluate the photographs for microvascular signs. Retinal arteriolar and venular calibers were measured with a computer-assisted technique, whereby photographs were digitized with a high-resolution scanner, and the diameters of all arterioles and venules in an area half to one disc diameter from the optic disc were measured. These diameters were summarized as the central retinal artery equivalent (CRAE) and the central retinal venular equivalent (CRVE), which represented average calibers of retinal arterioles and venules, respectively. A smaller CRAE value represents narrower retinal arterioles, and a higher CRVE value represents wider venular diameters. Quality control procedures were previously reported (32). For the retinal vascular caliber, reliability coefficients were 0.84 for within-grader and 0.79 for between-grader agreement. Dietary assessment The usual dietary intake of the participants at the third visit over the preceding year was assessed by using a 66-item semiquantitative food-frequency questionnaire. The questionnaire was a modified version of the 61-item instrument designed and validated by Willett et al for self-administration. The correlation coefficient of energy-adjusted crude fiber between the questionnaire and four 1-wk dietary records was 0.58 (33). To improve data quality and completeness, the questionnaire was administered by trained interviewers. Participants were asked to report the frequency of consumption of each food on the basis of 9 categories, which ranged from never or 쏝1 time/mo to 욷6 times/d. Interviewers also obtained additional information, including the brand name of the breakfast cereal usually consumed. All dietary factors in our analysis were adjusted for total energy by using the residual method (34).

1627

mellitus was defined as a fasting glucose concentration of 욷126 mg/dL (7.0 mmol/L), a nonfasting glucose concentration of 욷200 mg/dL (11.1 mmol/L), or a self-reported history of or treatment for diabetes. Anthropometric measures (weight and height) were determined by trained, certified technicians who followed a detailed, standardized protocol (35). BMI was calculated as weight (kg)/[height squared (m)]. Blood collection and processing for concentrations of HDL cholesterol, LDL cholesterol, and triacylglycerol are described elsewhere (35). Trained and certified interviewers also collected information on age, ethnicity, sex, smoking, alcohol consumption status, medical history, occupation, education, and physical activity. We used the sports index, derived from the survey of Baecke et al (36), as a measure of physical activity. The index ranged from 1 (low) to 5 (high) for physical activity from sports during leisure time. Statistical analysis For this analysis, CRAE and CRVE were used in combination with fiber intake data from the same period in a cross-sectional analysis. SAS (version 9.1.2; SAS, Cary, NC) software was used for all statistical analyses. The distributions of CRAE and CRVE were continuous and relatively normally distributed in this population; therefore, we used linear regression models to examine the association of retinal vascular caliber with fiber intake. We analyzed energy-adjusted intake of fiber according to quintiles. To assess for confounding, multivariate linear regression models were used. Our base model adjusted for age, sex, race and center. Several known and potential confounding factors were included in the multivariate models, either as indicator variables [sex, race, center, smoking status (never, former, and current smokers), occupation, education, alcohol intake, diabetes status, and hypertension] or continuous variables [age, smoking years, age at which smoking started, cigarettes smoked per day, BMI, physical activity, long-term systolic and diastolic blood pressure, serum lipids (HDL, LDL, and triacylglycerol), dietary factors from both food and supplements (total energy intake, glycemic index, carotenoids, folate, nҀ3 fatty acids, and vitamins B-6, B-12, C, and E), and other sources of fiber (total fiber intake not adjusted for the specific fiber types)]. Because CRAE and CRVE are correlated and might be confounders for each other (37), we included CRAE and CRVE in the models simultaneously (23, 38). Taking the lowest quintile of fiber intake as the reference, we estimated the difference of CRAE and CRVE with fiber intake after adjustment for the abovementioned covariates. In addition, we also conducted the stratified analysis by sex, smoking status, diabetes status, hypertension, and physical activity. Given that the measurement error of dietary assessment may bias our findings, we repeated our analysis with the dietary data at the first visit (1987–1989). We also examined the association of fiber intake with frank retinal microvascular abnormalities such as arteriovenous nicking and retinopathy.

Other covariates

RESULTS

Blood pressure was measured with a random-zero sphygmomanometer according to a standardized protocol (35). We used the average values over the first 3 examinations (9-y mean blood pressure) to approximate the long-term blood pressure level. Hypertension was defined as a systolic blood pressure of 욷140 mm Hg, a diastolic blood pressure of 욷90 mm Hg, or the use of antihypertensive medication during the previous 2 wk. Diabetes

The descriptive characteristics of the ARIC participants at visit 3 stratified by quintiles of energy-adjusted total dietary fiber are shown in Table 1. Participants with a higher fiber intake generally had a higher intake of carotenoids (P 쏝 0.001), folate (P 쏝 0.001), nҀ3 fatty acids (P 쏝 0.001), and vitamins B-6 (P 쏝 0.001), B-12 (P ҃ 0.009), C (P 쏝 0.001), and E (P 쏝 0.001). Subjects in the highest quintile of fiber intake were generally

1628

KAN ET AL

TABLE 1 Characteristics of participants of the Atherosclerosis Risk in Communities (ARIC) Study at the third visit (1993–1995) by quintiles of energy-adjusted total dietary fiber1 Quintiles

Dietary intake3 Total fiber intake (g) Cereal fiber (g) Fruit fiber (g) Vegetable fiber (g) Carotenoids (IU) Folate (␮g) Vitamin B-6 (mg) Vitamin B-12 (␮g) Vitamin C (mg) Vitamin E (mg) nҀ3 fatty acids (g) Subject characteristics Age (y) Female sex (%) BMI (kg/m2) Black race (%) Drinker status Current drinker (%) Former drinker (%) Never drinker (%) Smoking status Current smoker (%) Former smoker (%) Never regular (%) Sports index Systolic blood pressure (mm Hg) Diastolic blood pressure (mm Hg) Diabetes (%)

1 (lowest)

3

5 (highest)

P for trend2

10.00 앐 0.064 2.52 앐 0.03 1.54 앐 0.03 2.32 앐 0.04 4350 앐 95 255.1 앐 4.0 3.1 앐 0.2 9.2 앐 0.2 182.2 앐 5.2 60.8 앐 3.2 0.201 앐 0.006

16.84 앐 0.02 3.43 앐 0.03 3.16 앐 0.04 4.68 앐 0.04 8365 앐 115 338.4 앐 4.2 4.8 앐 0.3 10.0 앐 0.3 229.3 앐 5.9 75.8 앐 3.8 0.261 앐 0.005

26.84 앐 0.14 4.28 앐 0.06 5.14 앐 0.08 9.09 앐 0.11 15 549 앐 281 440.3 앐 5.0 5.5 앐 0.4 10.1 앐 0.3 317.6 앐 7.2 101.2 앐 4.4 0.335 앐 0.007

쏝0.001 쏝0.001 쏝0.001 쏝0.001 쏝0.001 쏝0.001 쏝0.001 0.009 쏝0.001 쏝0.001 쏝0.001

52.8 앐 0.1 40.3 28.7 앐 0.1 21.5

53.7 앐 0.1 60.0 28.7 앐 0.1 19.3

54.6 앐 0.1 62.2 28.0 앐 0.1 20.9

쏝0.001 쏝0.001 쏝0.001 0.253

64.2 18.8 17.0

51.7 22.4 25.9

44.7 24.6 30.7

쏝0.001 쏝0.001 쏝0.001

27.5 40.9 31.6 2.38 앐 0.02 121.8 앐 0.3 73.3 앐 0.2 11.9

15.3 41.6 43.0 2.55 앐 0.02 121.0 앐 0.3 72.3 앐 0.2 15.3

12.8 40.8 46.4 2.68 앐 0.02 121.2 앐 0.3 71.7 앐 0.2 15.3

쏝0.001 0.570 쏝0.001 쏝0.001 0.442 쏝0.001 쏝0.001

n ҃ 10 659. Logistic regression and general linear models were used for categorical and continuous variables, respectively. For categorical variables with 쏜2 levels (drinker status and smoking status), P values are for comparisons of the indicated level with all other levels combined. 3 All dietary factors were energy-adjusted and were from both food and supplements. 4 x៮ 앐 SEM (all such values). 1 2

slightly older (P 쏝 0.001), had lower BMI values (P 쏝 0.001) and diastolic blood pressure (P 쏝 0.001), were more likely to be female (P 쏝 0.001), had more physical activity (P 쏝 0.001), and were less likely to be current drinkers (P 쏝 0.001), smokers (P 쏝 0.001), or diabetes patients (P 쏝 0.001). The mean (앐SEM) retinal arteriolar caliber was 162.3 앐 0.2 ␮m, and the venular caliber was 193.1 앐 16.7 ␮m. Consistent with previous literature (36), we found that sex, age, BMI, alcohol drinking, smoking status, physical activity, blood pressure, and serum lipids (HDL and triacylglycerol) independently predicted retinal vascular caliber in our analysis (data not shown). We found a statistically significant dose-response relation between CRAE and dietary fiber from all sources and from cereal (Table 2). After adjustment for CRVE, age, sex, race, center, BMI, smoking, alcohol drinking, occupation, education, physical activity, diabetes status, and other dietary factors (multivariate model 1), total fiber consumption was positively associated with arteriolar caliber (P for trend ҃ 0.002); CRAE was 1.42 ␮m higher (95% CI: 0.42, 2.42 ␮m) in the highest quintile of intake than in the lowest quintile. Sports activity accounted for most of the difference between base model and multivariate model 1

(change in slope: Ҁ10%). After further adjustment for current hypertension, long-term systolic and diastolic blood pressure and lipids (HDL, LDL, and triacylglycerol) (multivariate model 2), the dose-response relation for total fiber attenuated (change in slope compared with base model: Ҁ30%) but remained significant (P for trend ҃ 0.012); CRAE was 1.05 ␮m higher (95% CI: 0.09, 2.01 ␮m) in the highest quintile of intake than in the lowest quintile. A similar pattern of relation with CRAE was found for cereal fiber. The association of fruit fiber with CRAE was not significant in multivariate model 1 (P for trend ҃ 0.114), although it became significant after further adjustment for current hypertension, long-term systolic and diastolic blood pressure, and lipids (multivariate model 2) (P for trend ҃ 0.028). Vegetable fiber was not significantly associated with CRAE in either base model or after multivariate analyses (data not shown). Similarly, we found significantly inverse dose-response associations between CRVE and fiber intake from all sources and from cereal, both before and after adjustment for covariates (Table 3). After adjustment for CRAE, age, sex, race, center, BMI, smoking, alcohol drinking, occupation, education, physical activity, diabetes, and dietary factors (multivariate model 1), the

1629

DIETARY FIBER AND RETINAL VASCULAR CALIBER TABLE 2 Differences in retinal arteriolar caliber (␮m) across increasing quintiles of energy-adjusted fiber intake compared with the lowest quintile1 Quintiles of energy-adjusted fiber intake Model2

1 (lowest)

2

3

4

P for trend3

5

Change in slope compared with base model %

Total fiber Median intake (g) Base model Multivariate model 1 Multivariate model 2 Fiber from cereal Median intake (g) Base model Multivariate model 1 Multivariate model 2 Fiber from fruit Median intake (g) Base model Multivariate model 1 Multivariate model 2

10.77 0 0 0

14.41 0.48 (Ҁ0.37, 1.34) 0.52 (Ҁ0.36, 1.40) 0.19 (Ҁ0.65, 1.03)

16.80 0.55 (Ҁ0.30, 1.41) 0.53 (Ҁ0.36, 1.43) 0.36 (Ҁ0.50, 1.22)

19.58 1.37 (0.50, 2.23) 1.29 (0.37, 2.21) 0.92 (0.03, 1.8)

24.80 1.56 (0.69, 2.44) 1.42 (0.42, 2.42) 1.05 (0.09, 2.01)

쏝0.001 0.002 0.012

— Ҁ10 Ҁ30

1.52 0 0 0

2.37 0.06 (Ҁ0.79, 0.91) 0.13 (Ҁ0.73, 1.00) Ҁ0.18 (Ҁ1.01, 0.66)

3.07 1.05 (0.20, 1.90) 1.05 (0.17, 1.93) 0.62 (Ҁ0.23, 1.47)

3.94 1.41 (0.56, 2.26) 1.52 (0.63, 2.40) 1.06 (0.21, 1.92)

5.73 2.57 (1.72, 3.43) 2.45 (1.54, 3.36) 1.56 (0.69, 2.44)

쏝0.001 쏝0.001 쏝0.001

— Ҁ5 Ҁ34

0.85 0 0 0

1.91 0.92 (0.06, 1.77) 0.83 (Ҁ0.05, 1.70) 0.58 (Ҁ0.26, 1.42)

2.85 0.73 (Ҁ0.13, 1.58) 0.68 (Ҁ0.20, 1.56) 0.55 (Ҁ0.30, 1.39)

3.93 1.50 (0.64, 2.37) 1.31 (0.41, 2.21) 1.16 (0.30, 2.02)

5.93 1.07 (0.20, 1.94) 0.79 (Ҁ0.14, 1.71) 0.94 (0.05, 1.83)

0.015 0.114 0.028

— Ҁ31 Ҁ8

n ҃ 10 659. 95% CIs in parentheses. Base model adjusted for central retinal venular equivalent, age, sex, race, and center; multivariate model 1 adjusted for central retinal venular equivalent, age, sex, race, center, BMI, smoking (smoking status, smoking years, age at which smoking started, and cigarettes smoked per day), alcohol intake, occupation, education, physical activity, diabetes status, dietary factors from both food and supplements (total energy intake, glycemic index, carotenoids, folate, nҀ3 fatty acids, and vitamins B-6, B-12, C, and E), and other sources of fiber (total fiber intake not adjusted for the specific fiber types); multivariate model 2 adjusted as for model 1 and for hypertension, long-term systolic and diastolic blood pressure, and lipids (HDL, LDL, and triacylglycerol). 3 Based on quintiles scaled by the quintile medians. 1 2

difference of CRVE in the highest quintile was Ҁ1.31 ␮m (95% CI: Ҁ2.27, Ҁ0.35 ␮m) relative to the lowest quintile of total fiber intake (P for trend ҃ 0.011). Smoking and physical activity accounted for most of the difference between base model and

multivariate model 1 (change in slope: Ҁ56%). The inverse association of total fiber with CRVE remained significant (change in slope compared with base model: Ҁ62%; P for trend ҃ 0.029) after further adjustment for current hypertension, long-term sys-

TABLE 3 Differences in retinal venular caliber (␮m) across increasing quintiles of energy-adjusted fiber intake compared with the lowest quintile1 Quintiles of energy-adjusted fiber intake Model2

1 (lowest)

2

3

4

5

P for trend3

Change in slope compared with base model %

Total fiber Median intake (g) Base model Multivariate model 1 Multivariate model 2 Fiber from cereal Median intake (g) Base model Multivariate model 1 Multivariate model 2 Fiber from fruit Median intake (g) Base model Multivariate model 1 Multivariate model 2

10.77 0 0 0

14.41 16.80 19.58 24.80 Ҁ1.50 (Ҁ2.33, Ҁ0.66) Ҁ1.52 (Ҁ2.36, Ҁ0.68) Ҁ2.27 (Ҁ3.12, Ҁ1.43) Ҁ2.83 (Ҁ3.68, Ҁ1.98) 쏝0.001 Ҁ0.68 (Ҁ1.52, 0.17) Ҁ0.43 (Ҁ1.29, 0.43) Ҁ0.79 (Ҁ1.67, 0.10) Ҁ1.31 (Ҁ2.27, Ҁ0.35) 0.011 Ҁ0.51 (Ҁ1.35, 0.34) Ҁ0.29 (Ҁ1.15, 0.58) Ҁ0.60 (Ҁ1.49, 0.28) Ҁ1.11 (Ҁ2.08, Ҁ0.15) 0.029

— Ҁ56 Ҁ62

1.52 0 0 0

2.37 3.07 3.94 5.73 Ҁ1.00 (Ҁ1.83, Ҁ0.17) Ҁ1.95 (Ҁ2.79, Ҁ1.12) Ҁ1.97 (Ҁ2.80, Ҁ1.14) Ҁ3.21 (Ҁ4.05, Ҁ2.38) 쏝0.001 Ҁ0.38 (Ҁ1.22, 0.45) Ҁ0.82 (Ҁ1.67, 0.02) Ҁ0.68 (Ҁ1.53, 0.17) Ҁ1.39 (Ҁ2.27, Ҁ0.52) 0.002 Ҁ0.23 (Ҁ1.07, 0.61) Ҁ0.75 (Ҁ1.60, 0.11) Ҁ0.57 (Ҁ1.42, 0.29) Ҁ1.16 (Ҁ2.04, Ҁ0.27) 0.009

— Ҁ58 Ҁ64

0.85 0 0 0

1.91 2.85 3.93 5.93 Ҁ1.07 (Ҁ1.91, Ҁ0.24) Ҁ1.56 (Ҁ2.40, Ҁ0.73) Ҁ1.85 (Ҁ2.70, Ҁ1.01) Ҁ2.39 (Ҁ3.25, Ҁ1.54) 쏝0.001 Ҁ0.03 (Ҁ0.87, 0.81) Ҁ0.16 (Ҁ1.01, 0.69) Ҁ0.24 (Ҁ1.10, 0.63) Ҁ0.76 (Ҁ1.65, 0.13) 0.066 0.01 (Ҁ0.83, 0.85) Ҁ0.12 (Ҁ0.97, 0.73) Ҁ0.23 (Ҁ1.10, 0.63) Ҁ0.88 (Ҁ1.77, 0.02) 0.032

— Ҁ65 Ҁ59

n ҃ 10 659. 95% CIs in parentheses. Base model adjusted for central retinal artery equivalent, age, sex, race, and center; multivariate model 1 adjusted for central retinal artery equivalent, age, sex, race, center, BMI, smoking (smoking status, smoking years, age at which smoking started, and cigarettes smoked per day), alcohol intake, occupation, education, physical activity, diabetes status, dietary factors from both food and supplements (total energy intake, glycemic index, carotenoids, folate, nҀ3 fatty acids, and vitamins B-6, B-12, C, and E), and other sources of fiber (total fiber intake not adjusted for the specific fiber types); multivariate model 2 adjusted as for model 1 and for hypertension, long-term systolic and diastolic blood pressure, and lipids (HDL, LDL, and triacylglycerol). 3 Based on quintiles scaled by the quintile medians. 1 2

1630

KAN ET AL

tolic and diastolic blood pressure, and lipids (multivariate model 2); the CRVE was 1.11 ␮m lower (95% CI: 0.15, 2.08 ␮m) in the highest quintile of intake than in the lowest quintile. A similar pattern of relation with CRVE was found for cereal fiber. The association with CRVE was marginally significant for fruit fiber (P for trend ҃ 0.066 in multivariate model 1) and was not significant for vegetable fiber. We examined whether sex, smoking status, diabetes, hypertension, and physical activity modified the associations of total fiber with arteriolar caliber (Table 4). We found no significant interaction terms. Similar patterns were found for venular caliber. Using the dietary data at the first visit (1987–1989), we found similar associations of dietary fiber with wider retinal arteriolar caliber and narrower venular caliber as we did with diet at visit 3 (1993–1995) when the retinal exams were done. We found no significant association of fiber intake with arteriovenous nicking or retinopathy.

DISCUSSION

In this cross-sectional analysis of a population-based cohort of middle-aged adults, we found significant associations of higher fiber intake from all sources and from cereal with wider retinal arteriolar caliber and narrower venular caliber. These associations were not explained by other dietary factors, including antioxidants, B vitamins, nҀ3 fatty acids, glycemic index, and fruit and vegetable fiber or by a large array of risk factors for this

TABLE 4 Adjusted differences in retinal arteriolar caliber (␮m) between the highest and lowest quintiles of energy-adjusted intakes of total fiber, by sex, smoking status, diabetes status, hypertension, and physical activity1

Sex Female (n ҃ 5947) Male (n ҃ 4712) Smoking status Never or past (n ҃ 8790) Current (n ҃ 1864) Diabetes status No (n ҃ 9096) Yes (n ҃ 1515) Hypertension No (n ҃ 6409) Yes (n ҃ 4197) Sports index 쏝2.5 (n ҃ 4893) 욷2.5 (n ҃ 5720)

Adjusted differences2

P for interaction3

0.68 (Ҁ0.68, 2.03) 1.37 (Ҁ0.03, 2.78)

0.744

1.22 (0.17, 2.28) 0.80 (Ҁ1.63, 3.22)

0.242

1.08 (0.05, 2.11) 1.01 (Ҁ1.71, 3.72)

0.678

1.59 (0.36, 2.83) 0.48 (Ҁ1.04, 2.00)

0.656

1.69 (0.28, 3.10) 0.26 (Ҁ1.07, 1.59)

0.305

n ҃ 10 659. 95% CIs in parentheses. Comparison of the highest with the lowest quintiles of energy-adjusted intakes after adjustment for central retinal venular equivalent, age, sex, race, center, BMI, smoking (smoking status, smoking years, age at which smoking started, and cigarettes smoked per day), alcohol intake, occupation, education, physical activity, diabetes status, hypertension, long-term systolic and diastolic blood pressure, lipids (HDL, LDL, and triacylglycerol), and dietary factors from both food and supplements (total energy intake, glycemic index, carotenoids, folate, nҀ3 fatty acids, and vitamins B-6, B-12, C, and E). 3 Likelihood ratio test for interaction (effect modification) by sex, smoking, diabetes, hypertension, or physical activity. 1 2

condition, including smoking, physical activity, hypertension, diabetes, and serum lipids. Several mechanisms could underlie the associations we observed. Fiber intake may reduce known risk factors for smaller retinal arteriolar caliber and wider venular caliber. The primary risk factor for retinal arteriolar narrowing is hypertension (39). Several clinical trials and prospective studies suggest that fiber may protect against hypertension (40 – 42). In our analysis, the effect of fiber intake on CRAE or CRVE attenuated after adjustment for current hypertension and long-term blood pressure, which supports the hypothesis that the protective effect of fiber on arteriolar narrowing or venular widening may be mediated, in part, through its direct or indirect effects on blood pressure. Fiber intake may also reduce dyslipidemia, which is a risk factor for retinal microvascular abnormalities (39); attenuation of the association between dietary fiber and retinal vascular caliber when serum lipids were included in the regression model supports this potential mechanism. In addition, higher cereal fiber intake has been associated with reduced incident diabetes in the ARIC cohort (43), which is related with retinal venular widening (39). However, it should be noted that the significant associations between dietary fiber and retinal vascular caliber remained after we carefully controlled for hypertension, long-term blood pressure, lipids, and diabetes, which suggests that other mechanism may also play a role in the protective effect of dietary fiber. For example, fiber intake appears to reduce systemic inflammation, an important contributor to arteriolar narrowing and venular widening (19 –21, 39); however, the markers of systemic inflammation, such as C-reactive protein, fibrinogen, and white blood cell count, were not available for most subjects at the third visit of the ARIC Study. Moreover, fiber intake was found to benefit endothelial cell function (18); several small clinical studies have suggested that endothelial dysfunction may influence retinal vascular caliber (44, 45). Fiber consumption may also replace intake of other foods with potentially detrimental effects on the microcirculation. Another possibility is that some constituents of dietary fiber, such as trace elements, may reduce cardiovascular disease risk (46). As in most observational studies, residual confounding is possible. However, we found significant associations of dietary fiber with retinal vascular caliber after detailed adjustment for known and potential cardiovascular disease risk and protective factors (eg, hypertension, long-term blood pressure, lipids, diabetes, smoking, physical activity, alcohol intake, total energy intake, glycemic index, nҀ3 fatty acids, antioxidant vitamins, and specific sources of fiber), which suggests an independent role of dietary fiber in the etiology of retinal microvascular abnormalities. Although the concern may be raised that a diet high in fiber might be a marker of a healthy lifestyle, including less frequent smoking (Table 1), we carefully adjusted for smoking [smoking status (current, past, and never smokers), smoking years, age at which smoking started, and cigarettes smoked per day]. Although residual confounding by smoking could occur despite our careful control, we also found a protective effect of fiber in never smokers, which suggests that the benefits of fiber intake are not due to the correlation with smoking behavior (47). In adjusted analyses, we found significant associations for total fiber and fiber from cereal, but not for vegetable fiber. The lack of an association of retinal vascular caliber and vegetable fiber is consistent with several prior reports on other cardiovascular outcomes (3– 6, 9), which suggests that the effect of dietary

DIETARY FIBER AND RETINAL VASCULAR CALIBER

fiber may vary depending on the food sources. However, the biological mechanisms for these differences are unclear. We found no significant association of fiber with arteriovenous nicking and retinopathy, which suggests that fiber might be protective in the earlier stages of pathogenesis. The heterogeneity of these associations may reflect different pathophysiologic processes related with specific retinal microvascular signs (48). On stratification by hypertension, subjects with hypertension were the smaller group. Although we did not observe a significant effect of fiber among subjects with hypertension (Table 4), there was no suggestion of interaction. This finding suggests limited power for this stratified analysis. However, it is possible that the effect of hypertension on retinal vascular caliber may dominate to such an extent that the additional exposure to fiber does not enhance effects in the same pathways. The limitations of our analysis should be noted. We used a food-frequency questionnaire to characterize dietary fiber intake. Although fiber intake assessed by the food-frequency questionnaire was reasonably well correlated with intake measured by diet records, measurement error likely limited our ability to detect associations. In addition, caution must be made when interpreting the findings described herein that the current analyses were cross-sectional; thus, a temporal relation between fiber intake and retinal vascular caliber cannot be established. However, it should be noted that ARIC subjects would not have been aware of their retinal vascular caliber in advance and thus could not have changed their diet based on this result. A major strength of our analysis was that it was based on carefully collected data on retinal abnormalities in a large cohort of the general population from 4 US communities. ARIC is also one of the largest studies of risk factors for these retinal microvascular signs. Confounding by hypertension, lipids, and diabetes was addressed by direct measurements made during the ARIC visit, and detailed data were available on other potential confounders. In summary, in this cross-sectional analysis, a higher intake of fiber from all sources and from cereal was related to wider retinal arteriolar caliber and narrower venular caliber, both of which have been found to be associated with a lower risk of cardiovascular disease. These associations were independent of smoking, hypertension, diabetes, serum lipids, and other risk factors for cardiovascular disease. These data add to the evidence of a protective role for fiber in various aspects of the pathogenesis of cardiovascular disease. The authors’ responsibilities were as follows—HK: contributed to the data analysis and manuscript preparation; JS, GH, RK, and KMR: contributed to the study design and manuscript preparation; and SJL: contributed to the study design, data analysis, and manuscript preparation. None of the authors had any financial or personal interest, except for JS, whose institution received unrestricted gifts from Sanofi-Aventis and the Gatorade Corporation.

REFERENCES 1. Bazzano LA, He J, Ogden LG, Loria CM, Whelton PK. Dietary fiber intake and reduced risk of coronary heart disease in US men and women: the National Health and Nutrition Examination Survey I Epidemiologic Follow-up Study. Arch Intern Med 2003;163:1897–904. 2. Mozaffarian D, Kumanyika SK, Lemaitre RN, Olson JL, Burke GL, Siscovick DS. Cereal, fruit, and vegetable fiber intake and the risk of cardiovascular disease in elderly individuals. JAMA 2003;289:1659 – 66.

1631

3. Wolk A, Manson JE, Stampfer MJ, et al. Long-term intake of dietary fiber and decreased risk of coronary heart disease among women. JAMA 1999;281:1998 –2004. 4. Pereira MA, O’Reilly E, Augustsson K, et al. Dietary fiber and risk of coronary heart disease: a pooled analysis of cohort studies. Arch Intern Med 2004;164:370 – 6. 5. Rimm EB, Ascherio A, Giovannucci E, Spiegelman D, Stampfer MJ, Willett WC. Vegetable, fruit, and cereal fiber intake and risk of coronary heart disease among men. JAMA 1996;275:447–51. 6. Pietinen P, Rimm EB, Korhonen P, et al. Intake of dietary fiber and risk of coronary heart disease in a cohort of Finnish men. The AlphaTocopherol, Beta-Carotene Cancer Prevention Study. Circulation 1996; 94:2720 –7. 7. Steffen LM, Jacobs DR Jr, Stevens J, Shahar E, Carithers T, Folsom AR. Associations of whole-grain, refined-grain, and fruit and vegetable consumption with risks of all-cause mortality and incident coronary artery disease and ischemic stroke: the Atherosclerosis Risk in Communities (ARIC) Study. Am J Clin Nutr 2003;78:383–90. 8. Oh K, Hu FB, Cho E, et al. Carbohydrate intake, glycemic index, glycemic load, and dietary fiber in relation to risk of stroke in women. Am J Epidemiol 2005;161:161–9. 9. Ascherio A, Rimm EB, Hernan MA, et al. Intake of potassium, magnesium, calcium, and fiber and risk of stroke among US men. Circulation 1998;98:1198 –204. 10. Merchant AT, Hu FB, Spiegelman D, Willett WC, Rimm EB, Ascherio A. Dietary fiber reduces peripheral arterial disease risk in men. J Nutr 2003;133:3658 – 63. 11. Whelton SP, Hyre AD, Pedersen B, Yi Y, Whelton PK, He J. Effect of dietary fiber intake on blood pressure: a meta-analysis of randomized, controlled clinical trials. J Hypertens 2005;23:475– 81. 12. Wu H, Dwyer KM, Fan Z, Shircore A, Fan J, Dwyer JH. Dietary fiber and progression of atherosclerosis: the Los Angeles Atherosclerosis Study. Am J Clin Nutr 2003;78:1085–91. 13. Liese AD, Schulz M, Fang F, et al. Dietary glycemic index and glycemic load, carbohydrate and fiber intake, and measures of insulin sensitivity, secretion, and adiposity in the Insulin Resistance Atherosclerosis Study. Diabetes Care 2005;28:2832– 8. 14. Erkkila AT, Herrington DM, Mozaffarian D, Lichtenstein AH. Cereal fiber and whole-grain intake are associated with reduced progression of coronary-artery atherosclerosis in postmenopausal women with coronary artery disease. Am Heart J 2005;150:94 –101. 15. Pereira MA, Pins JJ. Dietary fiber and cardiovascular disease: experimental and epidemiologic advances. Curr Atheroscler Rep 2000;2:494 – 502. 16. Anderson JW. Dietary fiber prevents carbohydrate-induced hypertriglyceridemia. Curr Atheroscler Rep 2000;2:536 – 41. 17. Burke V, Hodgson JM, Beilin LJ, Giangiulioi N, Rogers P, Puddey IB. Dietary protein and soluble fiber reduce ambulatory blood pressure in treated hypertensives. Hypertension 2001;38:821– 6. 18. Katz DL, Nawaz H, Boukhalil J, et al. Effects of oat and wheat cereals on endothelial responses. Prev Med 2001;33:476 – 84. 19. Bo S, Durazzo M, Guidi S, et al. Dietary magnesium and fiber intakes and inflammatory and metabolic indicators in middle-aged subjects from a population-based cohort. Am J Clin Nutr 2006;84:1062–9. 20. King DE, Egan BM, Woolson RF, Mainous AG III, Al-Solaiman Y, Jesri A. Effect of a high-fiber diet vs a fiber-supplemented diet on C-reactive protein level. Arch Intern Med 2007;167:502– 6. 21. Ma Y, Griffith JA, Chasan-Taber L, et al. Association between dietary fiber and serum C-reactive protein. Am J Clin Nutr 2006;83:760 – 6. 22. Marcus ML, Chilian WM, Kanatsuka H, Dellsperger KC, Eastham CL, Lamping KG. Understanding the coronary circulation through studies at the microvascular level. Circulation 1990;82:1–7. 23. Liew G, Sharrett AR, Kronmal R, et al. Measurement of retinal vascular caliber: issues and alternatives to using the arteriole to venule ratio. Invest Ophthalmol Vis Sci 2007;48:52–7. 24. Smith W, Wang JJ, Wong TY, et al. Retinal arteriolar narrowing is associated with 5-year incident severe hypertension: the Blue Mountains Eye Study. Hypertension 2004;44:442–7. 25. Wong TY, Klein R, Sharrett AR, et al. Retinal arteriolar narrowing and risk of coronary heart disease in men and women. The Atherosclerosis Risk in Communities Study. JAMA 2002;287:1153–9. 26. Wang JJ, Liew G, Wong TY, et al. Retinal vascular caliber and the risk of coronary heart disease-related mortality. Heart 2006;92:1583–7. 27. Wong TY, Klein R, Sharrett AR, et al. Retinal arteriolar narrowing and

1632

28. 29. 30. 31. 32.

33. 34. 35.

36. 37.

KAN ET AL

risk of diabetes mellitus in middle-aged persons. JAMA 2002;287: 2528 –33. Wong TY, Shankar A, Klein R, Klein BE, Hubbard LD. Retinal arteriolar narrowing, hypertension, and subsequent risk of diabetes mellitus. Arch Intern Med 2005;165:1060 –5. Ikram MK, de Jong FJ, Bos MJ, et al. Retinal vessel diameters and risk of stroke: the Rotterdam Study. Neurology 2006;66:1339 – 43. Ikram MK, De Jong FJ, Van Dijk EJ, et al. Retinal vessel diameters and cerebral small vessel disease: the Rotterdam Scan Study. Brain 2006; 129:182– 8. The ARIC investigators. The Atherosclerosis Risk in Communities (ARIC) Study: design and objectives. Am J Epidemiol 1989;129:687– 702. Hubbard LD, Brothers RJ, King WN, et al. Methods for evaluation of retinal microvascular abnormalities associated with hypertension/sclerosis in the Atherosclerosis Risk in Communities Study. Ophthalmology 1999;106:2269 – 80. Willett WC, Sampson L, Stampfer MJ, et al. Reproducibility and validity of a semiquantitative food frequency questionnaire. Am J Epidemiol 1985;122:51– 65. Willett WC, Howe GR, Kushi LH. Adjustment for total energy intake in epidemiologic studies. Am J Clin Nutr 1997;65(suppl):1220S– 8S; discussion 1229S–31S. National Heart, Lung, and Blood Institute, University of North Carolina School of Public Health ARIC Coordinating Center. Atherosclerosis Risk in Communities Study operations manual no. 2: cohort component procedures. Version 2.0. Chapel Hill, NC: National Heart, Lung, and Blood Institute, 1988. Baecke JA, Burema J, Frijters JE. A short questionnaire for the measurement of habitual physical activity in epidemiological studies. Am J Clin Nutr 1982;36:936 – 42. Liew G, Wong TY, Mitchell P, Wang JJ. Are narrower or wider retinal venules associated with incident hypertension? Hypertension 2006;48: e10; author reply e11.

38. Wong TY, Kamineni A, Klein R, et al. Quantitative retinal venular caliber and risk of cardiovascular disease in older persons: the cardiovascular health study. Arch Intern Med 2006;166:2388 –94. 39. Wong TY, Islam FM, Klein R, et al. Retinal vascular caliber, cardiovascular risk factors, and inflammation: the multi-ethnic study of atherosclerosis (MESA). Invest Ophthalmol Vis Sci 2006;47:2341–50. 40. Streppel MT, Arends LR, van ’t Veer P, Grobbee DE, Geleijnse JM. Dietary fiber and blood pressure: a meta-analysis of randomized placebo-controlled trials. Arch Intern Med 2005;165:150 – 6. 41. Ascherio A, Rimm EB, Giovannucci EL, et al. A prospective study of nutritional factors and hypertension among US men. Circulation 1992; 86:1475– 84. 42. Ascherio A, Hennekens C, Willett WC, et al. Prospective study of nutritional factors, blood pressure, and hypertension among US women. Hypertension 1996;27:1065–72. 43. Stevens J, Ahn K, Juhaeri, Houston D, Steffan L, Couper D. Dietary fiber intake and glycemic index and incidence of diabetes in AfricanAmerican and white adults: the ARIC study. Diabetes Care 2002;25: 1715–21. 44. Delles C, Michelson G, Harazny J, Oehmer S, Hilgers KF, Schmieder RE. Impaired endothelial function of the retinal vasculature in hypertensive patients. Stroke 2004;35:1289 –93. 45. Kawagishi T, Matsuyoshi M, Emoto M, et al. Impaired endotheliumdependent vascular responses of retinal and intrarenal arteries in patients with type 2 diabetes. Arterioscler Thromb Vasc Biol 1999;19:2509 –16. 46. Slavin JL, Martini MC, Jacobs DR Jr, Marquart L. Plausible mechanisms for the protectiveness of whole grains. Am J Clin Nutr 1999;70(suppl): 459S– 63S. 47. Henley SJ, Flanders WD, Manatunga A, Thun MJ. Leanness and lung cancer risk: fact or artifact? Epidemiology 2002;13:268 –76. 48. Wong TY, Klein R, Klein BE, Tielsch JM, Hubbard L, Nieto FJ. Retinal microvascular abnormalities and their relationship with hypertension, cardiovascular disease, and mortality. Surv Ophthalmol 2001;46:59 – 80.

Sex-specific association of fatty acid binding protein 2 and microsomal triacylglycerol transfer protein variants with response to dietary lipid changes in the 3-mo Medi-RIVAGE primary intervention study1–3 Marguerite Gastaldi, Sophie Dizie`re, Catherine Defoort, Henri Portugal, Denis Lairon, Michel Darmon, and Richard Planells ABSTRACT Background: The dietary guidelines targeted at reducing cardiovascular risk lead to largely heterogeneous responses in which genetic determinants are largely involved. Objectives: We evaluated the effect of fatty acid binding protein 2 (FABP2) Ala54Thr and microsomal triacylglycerol transfer protein (MTTP) Ҁ493G/T allelic variations on plasma lipid markers, at baseline and on the response to the 3-mo Medi-RIVAGE primary prevention study. Design: Subjects with moderate cardiovascular disease risk (n ҃ 169) were advised to reduce total and saturated dietary fats and to increase intake of monounsaturated and polyunsaturated fats. They were genotyped for FABP2 Ala54Thr and MTTP Ҁ493G/T allelic variations, and plasma was processed for cardiovascular risk marker analyses. Results: At baseline, men and women homozygous for Thr54 presented a significant opposite profile for plasma oleic acid (18:1), triacylglycerol-rich lipoprotein (TRL) cholesterol, and TRL phospholipids. In addition, all Thr/Thr men presented higher 18:1 values than did women. For the MTTP Ҁ493G/T polymorphism, although all TT subjects presented high apolipoprotein B-48, a genotype ҂ sex interaction was present for palmitic acid, linolenic acid, eicosatrienoic acid, and insulin. The prudent diet clearly improved plasma lipid markers. FABP2 genotype did not interact much with the amplitude of the response. However, for MTTP polymorphism, men homozygous for the T allele displayed a significantly more pronounced response than did men carrying the G allele, which is particularly evident by their larger decrease in the Framingham score. Conclusions: These 2 polymorphic loci are thus differently associated with the baseline lipid markers as well as with the response to nutritional recommendations, but both presented a marked sexspecific profile, with the response to diet being particularly efficient in men homozygous for the MTTP Ҁ493T allele. Am J Clin Nutr 2007;86:1633– 41. KEY WORDS Serum lipids, dietary fats, fatty acid binding protein 2, FABP2 polymorphism, microsomal triacylglycerol transfer protein, MTTP polymorphism, Mediterranean diet, risk assessment

INTRODUCTION

Cardiovascular diseases (CVDs) represent one of the main causes of death in developed countries and are thus one of the greatest concerns in public health. The various dietary guidelines that are provided by the nutrition research community are aimed at decreasing the incidence of such diseases (1). However, although nutritional recommendations are targeted to the whole population, responses to diet turned out to be largely heterogeneous, and a role for genetic determinants in the interindividual variation is now clearly admitted (2). One of the most widespread recommendations is to replace saturated fat with monounsaturated and polyunsaturated fats, and the variation of the response to such a dietary challenge may be due to the presence of allelic variants in the genes involved in fatty acid (FA) absorption and metabolism. In the epithelial cells of the small intestine, the absorption of dietary long-chain FAs (LCFAs) represents a multistep process that involves 1) the uptake of LCFAs at the apical side of the cell, 2) the intracellular transport of LCFAs and their esterification, and 3) the formation and secretion of triacylglycerol-rich lipoproteins (TRLs) at the basal side of the cell. Among other processes, the assembly and secretion of TRLs involve the intracellular LCFA trafficking by FA binding proteins (FABPs) and the lipid transfer on the nascent apolipoprotein (apo) B polypeptide chain by the microsomal triacylglycerol transfer protein (MTP). The human FA binding protein 2 (FABP2) gene encodes the intestinal FABP isoform (I-FABP) that is specifically expressed in enterocytes. In 1995, Baier et al (3) reported that a single 1 From the Institut National de la Sante´ et de la Recherche Medicale (INSERM), U476 “Nutrition Humaine et lipides,” Marseille, F-13385 France; Institut national de la Recherche Agronomique (INRA), UMR 1260, Marseille, F-13385 France; and Univ Méditerranée Aix-Marseille 2, Faculté de Médecine, IPHM-IFR 125, Marseille, F-13385 France. 2 Supported by the French Research Ministry (AQS grant), INSERM (IDS grant), the Provence-Alpes-Côte d’Azur Régional council, the Bouches du Rhône Général council, and the CRITT-PACA. 3 Reprints not available. Address correspondence to M Gastaldi, Univ Méditerranée Aix-Marseille 2, Faculté de Médecine, IPHM-IFR 125, Marseille, F-13385 France. E-mail: [email protected]. Received May 3, 2007. Accepted for publication July 31, 2007.

Am J Clin Nutr 2007;86:1633– 41. Printed in USA. © 2007 American Society for Nutrition

1633

1634

GASTALDI ET AL

nucleotide polymorphism (G to A) within the second exon at codon 54 predicted an amino acid substitution (Ala54Thr) that increases the in vitro binding affinity of the protein to LCFAs. Later, it was reported that, when challenged with an oral fat load, Thr54 homozygotes displayed a significantly greater increase in plasma chylomicron triacylglycerol and VLDL triacylglycerol (4). Another study found an association between Thr54 carriers and higher cholesterol concentration in chylomicron after ingesting olive oil (4, 5). All these modifications could be ascribed to an increased trafficking of FAs toward chylomicron assembly sites. The role played by the MTP protein in the chylomicron or VLDL assembly was highlighted by studies showing that MTP activity was absent in abetalipoproteinemic subjects (6) and that a liver-specific deletion of the gene coding for MTP (MTTP) produced a dramatic reduction in plasma VLDL triacylglycerol and apo B-100 secretion. The most studied promoter polymorphism at the MTTP locus is located 493 bp (base pair) upstream from the transcriptional start site (Ҁ493G/T), but the association between this polymorphism and biological markers for risk of CVD is still controversial (7–12) and has never been addressed by intervention studies. The present study was undertaken to examine whether the FABP2 and MTTP gene variants were associated with different responses to a 3-mo dietary intervention targeted to reduce total fat intake concomitantly with a replacement of saturated FAs by monounsaturated and polyunsaturated FAs. SUBJECTS AND METHODS

Subjects The design and methods of the Medi-RIVAGE study have been previously reported (13), and a marked reduction of cardiovascular risk was shown (14). Briefly, 169 volunteers with moderate cardiovascular risks were provided with nutritional recommendations. They consumed a low-fat diet for 3 mo: either a Mediterranean-type diet adapted from the traditional model or an adaptation of the commonly prescribed low-fat American Heart Association–type diet. The compliance with dietary recommendations was followed by dietitians. Three-day food records (at entry and after 3 mo) and 24-h unscheduled dietary recalls (once a month) were realized. No differences were observed in the dietary compliance according to sex. The GENI program nutritional database (version 6.1; Micro6, Nancy, France) was used and is based on the French REGAL food database (14). At entry and at the end of the 3-mo period, biochemical analyses were performed as previously described (14). Informed consent was obtained for each subject, and the study was approved by the institution’s ethics committee (ethics committee no. 98/25). Polymorphism detection Genomic DNA was prepared from white blood cells by a standard proteinase K-phenol method. The FABP2 Ala54Thr polymorphism and the MTTP Ҁ493G/T polymorphism were genotyped by a polymerase chain reaction restriction fragment length polymorphism assay (3, 7). For FABP2 genotyping the forward primer was 5'-CAGGTGTTAATATAGTGAAAAGG, and the reverse primer was 5'-TTACCCTGAGTTCAGTTCCG; the restriction cleavage was performed by Hha1 enzyme. For

MTTP genotyping the forward primer was 5'-AGTTTCACACATAAGGACAATCATCTA and the reverse primer was 5'-GGATTTAAATTTAAACTGTTAATTCATATCAC. The restriction cleavage was performed by Hph1 enzyme. Statistical analyses Statistical analyses were performed with the SAS statistical software (SAS Institute, Cary, NC). A chi-square test was used to determine whether genotypes at each gene locus were in HardyWeinberg equilibrium. Logarithmic transformation was performed on individual values of apo B-48, linolenic acid (18:3), stearidonic acid (18:4), eicosapentaenoic acid (20:5), triacylglycerols, insulin, TRL cholesterol, TRL triacylglycerols, TRL phospholipids, and Framingham score to improve the normality of their distribution. Because the patients were assigned for 3 mo to 2 different diets, it was first checked whether those 2 populations could be merged for the present analysis. A repeatedmeasures (baseline and 3 mo) general linear model showed that the type of diet had no effect on the variables studied. In addition, it was checked by chi-square test that the distribution of the subjects as a function of their genotypes was not affected by the type of diet. Before testing the effect of genotypes on the dependent variables, interfering covariables (adjustment factors) were identified. First, each dependent variable measured at baseline was tested in univariate general linear models with independent qualitative variables. Age was always included in the adjustment. Second, linear correlations between the dependent variables and the quantitative covariables were performed, and correlations significant at the 0.05 level were retained. Identified covariables were included as adjustment factors for testing the effect of genotypes at baseline and 3 mo. Professional activity (activity, retirement, inactivity), body mass index (in kg/m2), alcohol consumption, antihypertensive treatment (yes or no), smoking status (smoker, former smoker, never smoker), and menopausal status in women (yes, no or treated) were entered as covariables into the models. The effects of the genotypes at baseline were tested with general linear models. The effect of the genotypes on the response to the diet was tested in repeated-measures general linear models. Interactions of genotype by time and by sex were tested. Results were given for men and women separately when the interaction was significant. RESULTS

The distribution of FABP2 genotypes was 0.51 for Ala54/ Ala54, 0.40 for Ala54/Thr54, and 0.09 for Thr54/Thr54. The frequency of MTTP genotypes were 0.41 for Ҁ493GG, 0.49 for Ҁ493GT, and 0.10 for Ҁ493TT. The FABP2 Thr54-encoding allele had a frequency of 0.29, and the MTTP Ҁ493T variant had a frequency of 0.35. At both FABP2 and MTTP loci, the distribution of genotypes was not significantly different from that expected under the Hardy-Weinberg equilibrium (P ҃ 0.920 and P ҃ 0.593, respectively). At entry, subjects homozygous for the Thr54-encoding allele had a significantly higher percentage of plasma stearic acid (16:0) (Table 1). Interestingly, a significant interaction between genotype and sex was observed for the percentage of plasma oleic acid (18:1). This interaction points out that men and women exhibited a significant opposite profile with highest values for Thr/Thr men and lowest values for Thr/Thr women. In addition,

1635

FABP2, MTTP VARIANTS: SEX, DIET, PLASMA LIPIDS TABLE 1 Plasma fatty acid composition according to the Ala54Thr polymorphism of the fatty acid binding 2 gene

P1

Thr/Thr

Ala/Thr

Ala/Ala

% of total fatty acids

% of total fatty acids

% of total fatty acids

26.71 앐 3.924 25.12 앐 3.27

24.35 앐 3.72 23.54 앐 2.70

23.78 앐 2.88 22.94 앐 2.71

0.012 0.637

2.75 앐 0.87 2.21 앐 0.84

2.55 앐 1.18 2.28 앐 1.05

2.45 앐 0.87 2.32 앐 0.98

0.455 0.171

6.69 앐 1.88 6.52 앐 1.62

6.42 앐 1.67 6.60 앐 1.40

6.88 앐 1.50 6.63 앐 1.40

0.178 0.055

21.83 앐 5.53 21.90 앐 3.65

20.63 앐 2.84 21.09 앐 3.31

20.13 앐 2.82 20.69 앐 3.39

0.378 0.764

25.79 앐 5.26 26.12 앐 3.79

28.58 앐 4.09 27.91 앐 4.33

28.24 앐 3.99 27.63 앐 4.41

0.130 0.740

0.35 앐 0.20 0.31 앐 0.20

0.31 앐 0.24 0.29 앐 0.20

0.37 앐 0.24 0.40 앐 0.28

0.110 0.254

0.04 앐 0.09 0.04 앐 0.05

0.07 앐 0.12 0.06 앐 0.13

0.07 앐 0.14 0.06 앐 0.13

0.463 0.944

1.43 앐 0.52 1.51 앐 0.58

1.44 앐 0.72 1.42 앐 0.66

1.70 앐 0.69 1.59 앐 0.60

0.086 0.369

6.61 앐 2.05 6.62 앐 1.55

6.91 앐 1.77 6.88 앐 1.92

7.44 앐 1.68 7.33 앐 1.71

0.126 0.867

0.67 앐 0.44 1.21 앐 0.96

0.91 앐 0.69 1.13 앐 1.03

0.94 앐 1.06 1.20 앐 0.72

0.885 0.312

2.18 앐 0.69 3.12 앐 0.92

2.54 앐 1.11 2.91 앐 1.18

2.64 앐 1.07 3.14 앐 1.08

0.289 0.258

25.26 앐 5.31 24.17 앐 3.50

21.06 앐 3.01 22.09 앐 3.24

20.40 앐 2.42 22.33 앐 3.41

0.002 0.076

17.91 앐 2.18 19.31 앐 1.48

20.32 앐 2.72 20.36 앐 3.20

19.96 앐 3.04 19.65 앐 2.96

0.105 0.419

2,3

16:0 (palmitic acid) Baseline 3 mo 16:1 (palmitoleic acid)2,5 Baseline 3 mo 18:0 (stearic acid)2 Baseline 3 mo 18:1 (oleic acid)2,6 Baseline 3 mo 18:2 (linoleic acid)2,7 Baseline 3 mo 18:3 (linolenic acid)2 Baseline 3 mo 18:4 (stearidonic acid)2 Baseline 3 mo 20:3 (eicosatrienoic acid)2,6 Baseline 3 mo 20:4 (arachidonic acid)2 Baseline 3 mo 20:5 (eicosapentaenoic acid)2,8 Baseline 3 mo 22:6 (docosahexaenoic acid)2,8 Baseline 3 mo 18:16 Men9,10 Baseline 3 mo Women10,11 Baseline 3 mo 1

Comparison between polymorphisms at baseline (tested with general linear models) and in their response to diet (tested with repeated-measures general linear models). 2 n ҃ 15 Thr/Thr, 67 Ala/Thr, 87 Ala/Ala. 3 Adjusted for professional activity. 4 x៮ 앐 SD (all such values). 5 Adjusted for BMI, menopausal status in women, and smoking status. 6 Adjusted for alcohol consumption. 7 Adjusted for menopausal status in women and smoking status. 8 Adjusted for BMI. 9 n ҃ 8 Thr/Thr, 28 Ala/Thr, 33 Ala/Ala. 10 Significant interaction of genotype ҂ sex at baseline (P ҃ 0.0001) and significant interaction of genotype ҂ sex ҂ time after 3 mo (P ҃ 0.028). 11 n ҃ 7 Thr/Thr, 39 Ala/Thr, 54 Ala/Ala.

men, but not women, homozygous for the Thr54-encoding allele had a significantly higher plasma percentage of oleic acid (18:1) than did men carrying the Ala-encoding allele. No significant differences could be found in any of the biochemical markers (Table 2), but, again, according to genotypes, men and women showed significantly opposite profiles for TRL cholesterol and TRL phospholipids. The lowest values for TRL cholesterol and

TRL phospholipids were found in women homozygous for the Thr54-encoding allele, whereas highest values were found in men homozygous for the Thr54-encoding allele. For the MTTP Ҁ493G/T polymorphism, Table 3 shows that all subjects homozygous for the T allele exhibited significant lower percentages for stearic acid (18:0) and higher percentages for 18:1. In addition, a significant genotype ҂ sex interaction was

1636

GASTALDI ET AL

TABLE 2 Fasting biochemical markers according to the Ala54Thr polymorphism of the fatty acid binding protein 2 gene1

Apo A-1 (g/L)3 Baseline 3 mo Apo B (g/L)3 Baseline 3 mo Apo B-48 (g/L)3 Baseline 3 mo Apo E (mg/L)3,5 Baseline 3 mo Total cholesterol (mmol/L)3,5 Baseline 3 mo HDL cholesterol (mmol/L)3,6 Baseline 3 mo LDL cholesterol (mmol/L)3,7 Baseline 3 mo Triacylglycerols (mmol/L)3,8 Baseline 3 mo TRL cholesterol (mmol/L)3 Baseline 3 mo TRL triacylglycerols (mmol/L)3,6 Baseline 3 mo TRL phospholipids (mmol/L)3,8 Baseline 3 mo Glucose (mmol/L)3,9 Baseline 3 mo Insulin (mU/L)3,6 Baseline 3 mo Framingham score3 Baseline 3 mo TRL cholesterol at baseline (mmol/L)10 Men11 Women12 TRL phospholipids at baseline (mmol/L)8,13 Men11 Women12 1

Thr/Thr

Ala/Thr

Ala/Ala

P2

1.43 앐 0.374 1.45 앐 0.38

1.54 앐 0.27 1.46 앐 0.26

1.46 앐 0.29 1.39 앐 0.25

0.141 0.337

1.30 앐 0.27 1.19 앐 0.28

1.26 앐 0.26 1.17 앐 0.26

1.21 앐 0.21 1.20 앐 0.23

0.337 0.033

0.21 앐 0.15 0.29 앐 0.21

0.28 앐 0.24 0.26 앐 0.26

0.22 앐 0.18 0.28 앐 0.26

0.258 0.065

43.62 앐 15.61 46.13 앐 30.08

42.32 앐 11.96 40.07 앐 12.33

41.45 앐 12.62 39.52 앐 11.90

0.757 0.722

6.54 앐 1.10 6.18 앐 0.88

6.60 앐 1.03 6.09 앐 1.03

6.45 앐 0.88 6.15 앐 0.91

0.599 0.246

1.45 앐 0.52 1.45 앐 0.50

1.57 앐 0.40 1.58 앐 0.52

1.52 앐 0.47 1.51 앐 0.48

0.758 0.827

4.13 앐 1.12 3.70 앐 0.82

4.26 앐 1.03 3.84 앐 0.89

4.21 앐 0.83 3.92 앐 0.76

0.845 0.596

1.74 앐 1.20 1.72 앐 1.33

1.52 앐 0.71 1.29 앐 0.54

1.51 앐 0.98 1.38 앐 0.86

0.821 0.672

2.02 앐 2.70 1.99 앐 2.46

1.10 앐 0.60 0.97 앐 0.50

1.11 앐 0.82 1.06 앐 0.82

0.910 0.678

1.83 앐 2.31 1.55 앐 1.58

0.99 앐 0.59 0.91 앐 0.52

1.02 앐 0.73 0.92 앐 0.69

0.980 0.130

0.52 앐 0.56 0.49 앐 0.48

0.34 앐 0.19 0.32 앐 0.18

0.34 앐 0.22 0.32 앐 0.22

0.958 0.248

5.46 앐 0.60 5.27 앐 0.49

5.27 앐 0.62 5.01 앐 0.58

5.15 앐 0.66 5.06 앐 0.62

0.158 0.226

11.78 앐 6.35 8.78 앐 4.17

10.00 앐 7.51 7.81 앐 4.37

10.80 앐 6.21 9.20 앐 5.87

0.169 0.716

5.93 앐 3.54 5.53 앐 3.72

5.72 앐 3.19 4.47 앐 3.49

6.09 앐 3.13 5.36 앐 2.96

0.195 0.326

3.29 앐 3.24 0.56 앐 0.39

1.28 앐 0.60 0.98 앐 0.58

1.37 앐 0.94 0.98 앐 0.73

0.072 0.312

0.83 앐 0.62 0.17 앐 0.10

0.38 앐 0.18 0.32 앐 0.20

0.41 앐 0.25 0.30 앐 0.19

0.072 0.122

Apo, apolipoprotein; TRL, triacylglycerol-rich lipoprotein. Comparison between polymorphisms at baseline (tested with linear models) and in their response to diet (tested with repeated-measures general linear models). 3 n ҃ 15 Thr/Thr, 67 Ala/Thr, 87 Ala/Ala. 4 x៮ 앐 SD (all such values). 5 Adjusted for menopausal status in women. 6 Adjusted for BMI. 7 Adjusted for smoking status. 8 Adjusted for menopausal status in women and BMI. 9 Adjusted for BMI, professional activity, and alcohol consumption. 10,13 Significant interaction of genotype ҂ sex: 10 P ҃ 0.035, 13 P ҃ 0.013. 11 n ҃ 8 Thr/Thr, 28 Ala/Thr, 33 Ala/Ala. 12 n ҃ 7 Thr/Thr, 39 Ala/Thr, 54 Ala/Ala. 2

1637

FABP2, MTTP VARIANTS: SEX, DIET, PLASMA LIPIDS TABLE 3 Plasma fatty acid composition according to the Ҁ493 G/T polymorphism of the microsomal triglyceride transfer protein

P1

T/T

G/T

G/G

% of total fatty acids

% of total fatty acids

% of total fatty acids

24.95 앐 3.794 23.04 앐 2.17

24.69 앐 3.49 23.92 앐 2.97

23.61 앐 3.15 22.80 앐 2.66

0.024 0.043

2.66 앐 1.22 2.25 앐 1.14

2.58 앐 1.12 2.37 앐 1.07

2.41 앐 0.79 2.21 앐 0.85

0.416 0.576

6.23 앐 1.56 6.31 앐 1.22

6.48 앐 1.64 6.53 앐 1.54

7.03 앐 1.54 6.77 앐 1.29

0.026 0.236

22.05 앐 1.95 20.29 앐 2.78

20.57 앐 3.03 21.13 앐 3.65

19.98 앐 3.45 20.92 앐 3.18

0.029 0.033

26.71 앐 3.78 28.93 앐 5.54

27.89 앐 4.06 26.83 앐 4.13

28.82 앐 4.39 28.22 앐 4.12

0.134 0.027

0.35 앐 0.22 0.30 앐 0.18

0.34 앐 0.24 0.35 앐 0.25

0.35 앐 0.24 0.37 앐 0.26

0.941 0.964

0.10 앐 0.14 0.06 앐 0.09

0.08 앐 0.15 0.07 앐 0.16

0.05 앐 0.08 0.05 앐 0.07

0.604 0.396

1.42 앐 0.77 1.12 앐 0.70

1.58 앐 0.71 1.53 앐 0.66

1.60 앐 0.67 1.59 앐 0.53

0.424 0.132

6.68 앐 1.62 7.03 앐 1.69

7.02 앐 1.76 7.00 앐 1.69

7.44 앐 1.79 7.21 앐 1.94

0.092 0.270

0.94 앐 0.71 1.46 앐 1.59

0.87 앐 0.58 1.14 앐 0.70

0.94 앐 1.18 1.13 앐 0.82

0.631 0.971

2.56 앐 1.22 3.12 앐 0.99

2.57 앐 1.16 3.04 앐 1.12

2.55 앐 0.90 3.04 앐 1.13

0.961 0.931

29.44 앐 4.55

25.27 앐 3.48

23.80 앐 2.89

0.006

4.87 앐 0.67 6.48 앐 1.49 0.09 앐 0.07 0.51 앐 0.68 0.52 앐 0.80

6.36 앐 1.66 6.29 앐 1.48 0.39 앐 0.28 1.45 앐 0.61 0.87 앐 0.53

6.71 앐 1.60 6.34 앐 1.21 0.33 앐 0.25 1.44 앐 0.62 0.78 앐 0.48

0.187 0.023 0.046 0.025 0.123

23.57 앐 2.28

24.27 앐 3.46

23.45 앐 3.38

0.482

6.65 앐 1.53 6.26 앐 1.19 0.43 앐 0.19 1.70 앐 0.55 1.08 앐 0.66

6.56 앐 1.63 6.71 앐 1.58 0.30 앐 0.21 1.67 앐 0.77 0.86 앐 0.62

7.29 앐 1.47 7.14 앐 1.26 0.36 앐 0.23 1.73 앐 0.70 1.07 앐 1.53

0.056 0.241 0.268 0.931 0.051

2,3

16:0 (palmitic acid) Baseline 3 mo 16:1 (palmitoleic acid)2,5 Baseline 3 mo 18:0 (stearic acid)2 Baseline 3 mo 18:1 (oleic acid)2,6 Baseline 3 mo 18:2 (linoleic acid)2,7 Baseline 3 mo 18:3 (linolenic acid)2 Baseline 3 mo 18:4 (stearidonic acid)2 Baseline 3 mo 20:3 (eicosatrienoic acid)2,6 Baseline 3 mo 20:4 (arachidonic acid)2 Baseline 3 mo 20:5 (eicosapentaenoic acid)2,8 Baseline 3 mo 22:6 (docosahexaenoic acid)2,8 Baseline 3 mo Men9 16:0 at baseline3,10 18:0 Baseline 3 mo11 18:3 at baseline12 20:3 at baseline6,10 20:5 at baseline8,13 Women14 16:0 at baseline3,10 18:0 Baseline 3 mo11 18:3 at baseline12 20:3 at baseline6,10 20:5 at baseline8,13 1

Comparison between polymorphisms at baseline (tested with general linear models) and in their response to diet (tested with repeated-measures general linear models). 2 n ҃ 17 T/T, 83 G/T, 69 G/G. 3 Adjusted for professional activity. 4 x៮ 앐 SD (all such values). 5 Adjusted for BMI, menopausal status in women, and smoking status. 6 Adjusted for alcohol consumption. 7 Adjusted for menopausal status in women and smoking status. 8 Adjusted for BMI. 9 n ҃ 4 T/T, 34 G/T, 31 G/G. 10, 12, 13 Significant interaction of genotype ҂ sex: 10 P ҃ 0.049, 12 P ҃ 0.015, 13 P ҃ 0.008. 11 Significant interaction of genotype ҂ sex ҂ time, P ҃ 0.010. 14 n ҃ 13 T/T, 48 G/T, 38 G/G.

1638

GASTALDI ET AL

observed when considering 16:0, 18:3, and eicosatrionioic acid (20:3). Interestingly, only TT men displayed higher 16:0 and lower 18:3 and 20:3 values. For the biochemical markers (Table 4), all TT subjects presented significant higher concentrations of apo B-48. Insulin values displayed a genotype ҂ sex interaction with a different pattern in men and women. As expected, at the end of the 3-mo diet, all subjects reduced their fat intake with a sharp decrease in saturated FAs and a slight increase in monounsaturated and polyunsaturated FA consumption (14). Interestingly, when considering the FABP2 Ala54Thr polymorphism, the only difference observed in plasma FA composition concerned the 18:1 percentage: men and women exhibited a different profile of response linked to genotype. However, in all subjects homozygous for the Thr allele, the diet induced a significant higher decrease in fasting apo B than did the other 2 genotypes (Table 2). As for the MTP Ҁ493G/T polymorphism, this work shows that there were genotype-specific differences in the response to diet. Indeed, TT subjects significantly reduced 16:0 and 18:1 and increased 18:2 (Table 3). In addition, a genotype ҂ sex interaction was also observed: TT men significantly increased 18:0. For the biochemical markers (Table 4), subjects homozygous for MTP Ҁ493T also displayed a genotype-specific response in fasting apo B-48 values, total cholesterol, fasting triacylglycerols, and fasting TRL phospholipids. Moreover, this greater genotype-specific susceptibility appears also to be sex specific. Indeed, when we consider the Framingham score, which represents a “global” evaluation of risk, we observed in men a marked decrease of this score that is significantly more important than in women. DISCUSSION

Our results show the interaction of sex with 2 gene polymorphisms and how the response to diet can be modulated by these polymorphisms. In this study, we did not find any relation between the Ala54Thr polymorphism at the FAB2 locus and fasting glucose or insulin concentrations, which is consistent with some previous data (15–19). An association between this polymorphism and insulin resistance was previously described in the Pima population, a group known to have a high prevalence of type 2 diabetes mellitus (3). Nevertheless, it was shown later that this association is likely to be linked to promoter variations, which are in complete genotypic concordance with the Ala54Thr substitution in Pima but not in white populations (20). Moreover, in young French-Canadians, among the components of the metabolic syndrome, only plasma triacylglycerol concentrations were shown to display an interaction with the FABP2 polymorphism (11). In our study, only men homozygous for the Thr allele showed a tendency to higher triacylglycerol concentrations (data not shown). Previous experiments showed that in vitro, compared with the Ala isoform, the human I-FABP Thr protein presented a 2-fold higher affinity for 18:1 FAs (21). In addition, expression studies showed that Caco-2 cells transfected with the Thr-encoding allele exhibited a 2-fold increase in FA uptake and a 5-fold increase in triacylglycerol secretion than did Caco-2 cells transfected with the Ala-encoding allele (22). Finally, it was shown that human intestinal explants from carriers of the Thr allele displayed a marked increase in triacylglycerol secretion (23). In agreement with those studies, we showed here a significant association between the Thr54 variant and an altered fasting plasma FA

profile (increase in 16:0 associated in men with an increase in 18:1). Our study also showed an interaction between the Ala/Thr polymorphism and sex because the Thr/Thr men showed high TRL-cholesterol and TRL-phospholipid concentrations. These latter alterations together with the increases in 16:0 and 18:1 might be associated with an increased risk of CVD. Such an interaction between this polymorphism and sex has already been described. Although not strictly comparable, data reported from the Framingham cohort showed that women but not men homozygous for the Ala54-encoding allele displayed lower total and LDL-cholesterol concentrations (24). Such a sex-specific difference, as observed in our study, can also be linked to recent data obtained from I-FABP knockout mice (25) in which females did not present any altered liver histology, whereas males showed centrolobular vacuolated hepatocytes. Our study also highlighted a sex-specific link between allelic variation at the MTTP Ҁ493 locus and lipid metabolism at baseline. MTP is known to play a critical role in the assembly of triacylglycerol-rich particles in liver and intestine. However, some clues seem to indicate that the expression of the MTTP gene could be different for the tissue and could lead to compensatory effects. For example, an intestine-specific MTP deficiency in mice led to an increase in hepatic VLDL secretion (26). MTP inhibitors also may present different effects on intestine or liver (27). The MTTP T allele was shown to be associated with an increase in gene expression (7) and with an increased production of small apo B-48 – containing lipoproteins in the postprandial state (28). It is thus not surprising to observe in our study an increase in plasma apo B-48 concentration, the specific chylomicron apolipoprotein, in the subjects homozygous for the T allele. A longitudinal study in adult men has shown that T allele carriers presented a significantly increased risk of coronary heart disease (8). Paradoxically, however, this study also put forward that this increased risk was linked to a decrease in plasma total cholesterol in this subpopulation. Other works however did not show any modification in circulating cholesterol (10 –12). In our study we can only observe a tendency to an increase in the T homozygous population. The increased risk observed by Ledmyr et al (8) might be associated to the modification of other lipid compounds that have long been linked to CVD risk. As a matter of fact, we observed in men homozygous for the T allele, an increase in the saturated FA 16:0 together with a decrease in polyunsaturated FAs (18:3 and 20:3). This pattern is known to be associated with CVD risk. Interestingly, this FA pattern appears to be sex specific because it is restricted to the male population. Because several studies have cast light on the possible regulation of the MTTP gene expression by insulin (29, 30), we have searched for a possible link between fasting plasma insulin concentrations and the MTTP Ҁ493 polymorphism. Although no significant modification of this marker could be found according to genotype in the whole population, a significant genotype ҂ sex interaction was observed. The effect of the 3-mo diet resulted in a clear improvement of most biological markers (14). We must note that no differences in the dietary intakes of men and women were observed at baseline and at the end of the diet period. However, the response to diet differed according to the 2 loci. Indeed, the response did not seem to be greatly modulated by the Ala/Thr FABP2 polymorphism, except for the decrease in apo B in the Thr/Thr population and the different pattern of response in Thr/Thr men and women

1639

FABP2, MTTP VARIANTS: SEX, DIET, PLASMA LIPIDS TABLE 4 Fasting biochemical markers according to the Ҁ493 G/T polymorphism of the microsomal triglyceride transfer protein1

Apo A-1 (g/L)3 Baseline 3 mo Apo B (g/L)3 Baseline 3 mo Apo B-48 (g/L)3 Baseline 3 mo Apo E (mg/L)3 Baseline 3 mo Total cholesterol (mmol/L)3,5 Baseline 3 mo HDL cholesterol (mmol/L)3,6 Baseline 3 mo LDL cholesterol (mmol/L)3,7 Baseline 3 mo Triacylglycerols (mmol/L)3,8 Baseline 3 mo TRL cholesterol (mmol/L)3 Baseline 3 mo TRL triacylglycerols (mmol/L)3,6 Baseline 3 mo TRL phospholipids (mmol/L)3,8 Baseline 3 mo Glucose (mmol/L)3,9 Baseline 3 mo Insulin (mU/L)3,6 Baseline 3 mo Framingham score3,10 Baseline 3 mo Insulin at baseline (mU/L)6,11 Men12 Women13 Framingham score10,14 Men12 Baseline 3 mo13 Women13 Baseline 3 mo13 1

T/T

G/T

G/G

P2

1.48 앐 0.214 1.40 앐 0.30

1.47 앐 0.28 1.41 앐 0.28

1.51 앐 0.32 1.44 앐 0.26

0.475 0.883

1.29 앐 0.24 1.17 앐 0.28

1.24 앐 0.25 1.22 앐 0.26

1.22 앐 0.22 1.15 앐 0.21

0.349 0.179

0.30 앐 0.18 0.24 앐 0.27

0.27 앐 0.23 0.28 앐 0.22

0.20 앐 0.17 0.27 앐 0.28

0.011 0.015

45.4 앐 13.5 36.9 앐 8.5

42.7 앐 12.2 41.2 앐 12.9

40.3 앐 12.7 40.2 앐 17.4

0.294 0.057

6.98 앐 0.89 6.11 앐 1.09

6.49 앐 1.05 6.16 앐 1.00

6.43 앐 0.84 6.09 앐 0.87

0.088 0.030

1.52 앐 0.48 1.56 앐 0.56

1.51 앐 0.42 1.48 앐 0.40

1.56 앐 0.47 1.59 앐 0.58

0.337 0.293

4.64 앐 0.75 4.00 앐 0.80

4.20 앐 1.03 3.93 앐 0.83

4.15 앐 0.83 3.76 앐 0.80

0.099 0.122

1.62 앐 0.87 1.11 앐 0.46

1.58 앐 0.88 1.44 앐 0.80

1.46 앐 0.94 1.36 앐 0.86

0.256 0.036

1.36 앐 0.76 0.90 앐 0.48

1.18 앐 0.86 1.16 앐 0.85

1.18 앐 1.39 1.11 앐 1.30

0.112 0.087

1.10 앐 0.65 0.77 앐 0.45

1.12 앐 0.87 1.02 앐 0.76

1.05 앐 1.15 0.97 앐 0.86

0.258 0.079

0.38 앐 0.18 0.26 앐 0.14

0.38 앐 0.26 0.36 앐 0.24

0.33 앐 0.29 0.32 앐 0.27

0.135 0.042

5.24 앐 0.65 5.00 앐 0.71

5.19 앐 0.56 5.04 앐 0.53

5.28 앐 0.73 5.09 앐 0.65

0.340 0.483

9.69 앐 5.08 7.39 앐 4.44

10.13 앐 5.98 8.38 앐 4.41

11.31 앐 7.90 9.13 앐 6.10

0.825 0.370

6.76 앐 3.09 4.47 앐 3.70

6.09 앐 3.21 5.33 앐 3.37

5.52 앐 3.14 4.78 앐 3.03

0.191 0.002

12.63 앐 6.47 8.79 앐 4.49

11.63 앐 5.94 9.09 앐 5.85

10.96 앐 8.61 11.59 앐 7.38

0.221 0.136

7.00 앐 4.97 2.75 앐 5.85

6.42 앐 3.06 5.91 앐 2.97

5.72 앐 2.67 4.65 앐 3.27

0.247 0.008

6.69 앐 2.56 5.09 앐 2.70

5.86 앐 3.33 4.94 앐 3.60

5.37 앐 3.48 4.89 앐 2.86

0.448 0.089

Apo, apolipoprotein; TRL, triacylglycerol-rich lipoprotein. Comparison between polymorphisms at baseline (tested with general linear models) and in their response to diet (tested with repeated-measures general linear models). 3 n ҃ 17 T/T, 83 G/T, 69 G/G. 4 x៮ 앐 SD (all such values). 5 Adjusted for menopausal status in women. 6 Adjusted for BMI. 7 Adjusted for smoking status. 8 Adjusted for menopausal status in women and BMI. 9 Adjusted for menopausal status in women, BMI, professional activity, alcohol consumption, and antihypertensive treatment. 10 Adjusted for menopausal status in women, BMI, smoking status, and professional activity. 11 Significant interaction of genotype ҂ sex, P ҃ 0.045. 12 n ҃ 4 T/T, 34 G/T, 31 G/G. 13 n ҃ 13 T/T, 48 G/T, 38 G/G. 14 Significant interaction of genotype ҂ sex ҂ time, P ҃ 0.036. 2

1640

GASTALDI ET AL

for 18:1. This lack of reactivity is not surprising because, to date, few studies reported the influence of the FABP2 Ala54Thr polymorphism on the efficiency of a diet in terms of modulation of the magnitude of the response. One study reported on a greatest high-fiber diet-induced response in plasma total and LDL cholesterol among dyslipidemic subjects homozygous for the Threncoding allele (31). A second study was performed in normolipidemic subjects who were submitted to a change from saturated fat to monounsaturated FAs. That study showed that carriers of the Thr-encoding allele presented an enhanced decrease in insulin sensitivity compared with Ala/Ala homozygous subjects. Although in that latter study and in ours, the type of diets was comparable, we did not detect any diet-induced difference in insulin sensitivity: all subjects had decreased fasting plasma insulin concentrations regardless of genotypes. Finally, our study showed a clear interaction between the MTTP Ҁ493G/T variation and the magnitude in the response to the diet. Indeed, although biological markers for the risk of CVD clearly decreased after the 3-mo intervention, this improvement is particularly marked in men homozygous for the T allele than in the other subjects. The TT subjects, as a whole, showed a significantly higher response in reducing fasting cholesterol, apo B-48, TRL phospholipids, 16:0, and 18:1 and in increasing 18:2 values, but only TT men showed a major improvement of the Framingham score. This sex-related sensitivity to diet can be linked to the recent finding that, in men with familial hypercholesterolemia but not in women, this polymorphism modulated after-treatment plasma triacylglycerol values (32). In conclusion, important new findings in this study show that FABP2 Ala54Thr and MTTP Ҁ493G/T polymorphisms are differently involved in the contribution to cardiovascular risk in men and women. These 2 polymorphisms interact with sex 1) at baseline on fasting biological markers showing an increased risk of CVD for FABP2 Thr/Thr and MTTP TT men and 2) after the 3-mo diet leading to a better response in MTTP TT men.

7.

8.

9.

10.

11.

12.

13.

14.

15.

16.

We thank Chantal Bideau, Danielle Iniesta, and Nicole Peyrol for their technical help in SNP determination. The author’s responsibilities were as follows—MG: was involved in the interpretation of the data and wrote the manuscript; SD and MD: performed statistical analysis of data; HP and RP: were involved in blood analyses, CD: was involved in the collection of data; DL: coordinated the study, RP: designed the genetic study and contributed to the writing of the manuscript. None of the authors had a personal or financial conflict of interest.

17.

REFERENCES

20.

1. Elmadfa I, Freisling H. Fat intake, diet variety and health promotion. Forum Nutr 2005;1–10. 2. Corella D, Ordovas JM. Single nucleotide polymorphisms that influence lipid metabolism: interaction with dietary factors. Annu Rev Nutr 2005; 25:341–90. 3. Baier LJ, Sacchettini JC, Knowler WC, et al. An amino acid substitution in the human intestinal fatty acid binding protein is associated with increased fatty acid binding, increased fat oxidation, and insulin resistance. J Clin Invest 1995;95:1281–7. 4. Agren JJ, Valve R, Vidgren H, Laakso M, Uusitupa M. Postprandial lipemic response is modified by the polymorphism at codon 54 of the fatty acid-binding protein 2 gene. Arterioscler Thromb Vasc Biol 1998; 18:1606 –10. 5. Dworatzek PD, Hegele RA, Wolever TM. Postprandial lipemia in subjects with the threonine 54 variant of the fatty acid-binding protein 2 gene is dependent on the type of fat ingested. Am J Clin Nutr 2004;79:1110 –7. 6. Wetterau JR, Aggerbeck LP, Bouma ME, et al. Absence of microsomal

18.

19.

21.

22.

23.

24.

25.

triglyceride transfer protein in individuals with abetalipoproteinemia. Science 1992;258:999 –1001. Karpe F, Lundahl B, Ehrenborg E, Eriksson P, Hamsten A. A common functional polymorphism in the promoter region of the microsomal triglyceride transfer protein gene influences plasma LDL levels. Arterioscler Thromb Vasc Biol 1998;18:756 – 61. Ledmyr H, McMahon AD, Ehrenborg E, et al. The microsomal triglyceride transfer protein gene-493T variant lowers cholesterol but increases the risk of coronary heart disease. Circulation 2004;109:2279 – 84. Juo SH, Han Z, Smith JD, Colangelo L, Liu K. Common polymorphism in promoter of microsomal triglyceride transfer protein gene influences cholesterol, ApoB, and triglyceride levels in young African American men: results from the coronary artery risk development in young adults (CARDIA) study. Arterioscler Thromb Vasc Biol 2000;20:1316 –22. Couture P, Otvos JD, Cupples LA, Wilson PW, Schaefer EJ, Ordovas JM. Absence of association between genetic variation in the promoter of the microsomal triglyceride transfer protein gene and plasma lipoproteins in the Framingham Offspring Study. Atherosclerosis 2000;148: 337– 43. Stan S, Lambert M, Delvin E, et al. Intestinal fatty acid binding protein and microsomal triglyceride transfer protein polymorphisms in FrenchCanadian youth. J Lipid Res 2005;46:320 –7. Yamada Y, Ando F, Shimokata H. Association of a microsomal triglyceride transfer protein gene polymorphism with blood pressure in Japanese women. Int J Mol Med 2006;17:83– 8. Vincent S, Gerber M, Bernard MC, et al. The Medi-RIVAGE study (Mediterranean Diet, Cardiovascular Risks and Gene Polymorphisms): rationale, recruitment, design, dietary intervention and baseline characteristics of participants. Public Health Nutr 2004;7:531– 42. Vincent-Baudry S, Defoort C, Gerber M, et al. The Medi-RIVAGE study: reduction of cardiovascular disease risk factors after a 3-mo intervention with a Mediterranean-type diet or a low-fat diet. Am J Clin Nutr 2005;82:964 –71. Pihlajamaki J, Rissanen J, Heikkinen S, Karjalainen L, Laakso M. Codon 54 polymorphism of the human intestinal fatty acid binding protein 2 gene is associated with dyslipidemias but not with insulin resistance in patients with familial combined hyperlipidemia. Arterioscler Thromb Vasc Biol 1997;17:1039 – 44. Lei HH, Coresh J, Shuldiner AR, Boerwinkle E, Brancati FL. Variants of the insulin receptor substrate-1 and fatty acid binding protein 2 genes and the risk of type 2 diabetes, obesity, and hyperinsulinemia in AfricanAmericans: the Atherosclerosis Risk in Communities Study. Diabetes 1999;48:1868 –72. Galluzzi JR, Cupples LA, Meigs JB, Wilson PW, Schaefer EJ, Ordovas JM. Association of the Ala54-Thr polymorphism in the intestinal fatty acid-binding protein with 2-h postchallenge insulin levels in the Framingham Offspring Study. Diabetes Care 2001;24:1161– 6. Tahvanainen E, Molin M, Vainio S, et al. Intestinal fatty acid binding protein polymorphism at codon 54 is not associated with postprandial responses to fat and glucose tolerance tests in healthy young Europeans. Results from EARS II participants. Atherosclerosis 2000;152:317–25. Okada T, Sato NF, Kuromori Y, et al. Thr-encoding allele homozygosity at codon 54 of FABP 2 gene may be associated with impaired delta 6 desaturase activity and reduced plasma arachidonic acid in obese children. J Atheroscler Thromb 2006;13:192– 6. Formanack ML, Baier LJ. Variation in the FABP2 promoter affects gene expression: implications for prior association studies. Diabetologia 2004;47:349 –51. Storch J, Veerkamp JH, Hsu KT. Similar mechanisms of fatty acid transfer from human anal rodent fatty acid-binding proteins to membranes: liver, intestine, heart muscle, and adipose tissue FABPs. Mol Cell Biochem 2002;239:25–33. Baier LJ, Bogardus C, Sacchettini JC. A polymorphism in the human intestinal fatty acid binding protein alters fatty acid transport across Caco-2 cells. J Biol Chem 1996;271:10892– 6. Levy E, Menard D, Delvin E, et al. The polymorphism at codon 54 of the FABP2 gene increases fat absorption in human intestinal explants. J Biol Chem 2001;276:39679 – 84. Galluzzi JR, Cupples LA, Otvos JD, Wilson PW, Schaefer EJ, Ordovas JM. Association of the A/T54 polymorphism in the intestinal fatty acid binding protein with variations in plasma lipids in the Framingham Offspring Study. Atherosclerosis 2001;159:417–24. Agellon LB, Li L, Luong L, Uwiera RR. Adaptations to the loss of

FABP2, MTTP VARIANTS: SEX, DIET, PLASMA LIPIDS

26. 27.

28.

29.

intestinal fatty acid binding protein in mice. Mol Cell Biochem 2006; 284:159 – 66. Xie Y, Newberry EP, Young SG, et al. Compensatory increase in hepatic lipogenesis in mice with conditional intestine-specific Mttp deficiency. J Biol Chem 2006;281:4075– 86. Aggarwal D, West KL, Zern TL, Shrestha S, Vergara-Jimenez M, Fernandez ML. JTT-130, a microsomal triglyceride transfer protein (MTP) inhibitor lowers plasma triglycerides and LDL cholesterol concentrations without increasing hepatic triglycerides in guinea pigs. BMC Cardiovasc Disord 2005;5:30. Lundahl B, Hamsten A, Karpe F. Postprandial plasma ApoB-48 levels are influenced by a polymorphism in the promoter of the microsomal triglyceride transfer protein gene. Arterioscler Thromb Vasc Biol 2002; 22:289 –93. Hagan DL, Kienzle B, Jamil H, Hariharan N. Transcriptional regulation

1641

of human and hamster microsomal triglyceride transfer protein genes. Cell type-specific expression and response to metabolic regulators. J Biol Chem 1994;269:28737– 44. 30. Qiu W, Taghibiglou C, Avramoglu RK, et al. Oleate-mediated stimulation of microsomal triglyceride transfer protein (MTP) gene promoter: implications for hepatic MTP overexpression in insulin resistance. Biochemistry 2005;44:3041–9. 31. Hegele RA, Wolever TM, Story JA, Connelly PW, Jenkins DJ. Intestinal fatty acid-binding protein variation associated with variation in the response of plasma lipoproteins to dietary fibre. Eur J Clin Invest 1997;27:857– 62. 32. Garcia-Garcia AB, Gonzalez C, Real JT, et al. Influence of microsomal triglyceride transfer protein promoter polymorphism -493 GT on fasting plasma triglyceride values and interaction with treatment response to atorvastatin in subjects with heterozygous familial hypercholesterolaemia. Pharmacogenet Genomics 2005;15:211– 8.

Body mass index and fat mass are the primary correlates of insulin resistance in nondiabetic stage 3– 4 chronic kidney disease patients1–3 M Luisa Trirogoff, Ayumi Shintani, Jonathan Himmelfarb, and T Alp Ikizler ABSTRACT Background: Insulin resistance has been noted in patients with chronic kidney disease (CKD). The determinants of insulin resistance have not been well-studied in CKD patients. Objective: The objective of this study was to examine the degree and determinants of insulin resistance in persons without diabetes but with stage 3– 4 CKD. Design: Demographic characteristics, metabolic hormones, and inflammatory markers were measured in 95 nonobese stage 3– 4 CKD patients without prior diagnosis of diabetes mellitus and 36 control subjects without CKD. The estimated glomerular filtration rate (eGFR) was measured by using the Modification of Diet in Renal Disease study equation. Insulin resistance was measured with the use of the homeostasis model assessment of insulin resistance (HOMA-IR). Results: After age and sex adjustments, HOMA-IR scores were significantly and positively correlated with body mass index (BMI) and percentage body fat. After control for age, race, adiponectin concentrations, sex, and eGFR in a multivariate regression model, BMI remained as the only significant predictor of insulin resistance (standardized regression coefficient ҃ 0.55; P 쏝 0.001). When substituted for BMI, percentage body fat also was an independent predictor of insulin resistance. The prevalence of abnormal HOMA did not differ significantly between CKD patients (98%) and BMImatched control subjects (94%). Conclusion: Whereas insulin resistance is highly prevalent in stage 3– 4 CKD, the primary determinant of insulin resistance in this population is BMI, specifically, fat mass. Am J Clin Nutr 2007; 86:1642– 8. KEY WORDS Insulin resistance, chronic kidney disease, homeostasis model assessment, body mass index, adiposity

INTRODUCTION

Both the incidence and prevalence of chronic kidney disease (CKD) and end-stage renal disease (ESRD) continue to increase at an alarming rate in the United States. Much investigation has been focused on ESRD patients, but an increasing recognition of the high prevalence of moderate-to-severe CKD has redirected the attention to this patient population to identify risk factors associated with hospitalization, death, and progression to ESRD. Indeed, studies have shown that there is a greater risk of atherosclerotic events and a higher risk of death in patients with mildto-moderate CKD than in those without kidney disease (1). Furthermore, CKD is accompanied by numerous metabolic derangements

1642

such as oxidative stress, chronic inflammation, and endothelial dysfunction (2). Insulin resistance (IR) in advanced kidney disease has been well recognized since the seminal work by DeFronzo et al (3) using hyperinsulinemic euglycemic clamp techniques. IR was reported to be an independent risk factor for cardiovascular morbidity and mortality in patients with ESRD (4). IR associated with mild-to-moderate CKD has also been described, albeit in reports mainly from European and Japanese populations. To our knowledge, few studies have investigated IR in CKD patients in the United States, where 11% of the adult population is estimated to have CKD (5, 6), and potential determinants of IR in the US population have not been studied in detail. Greater attention is being focused on the role of inflammation, adiposity, and its associated adipokines such as adiponectin in the general and CKD population in the United States; however, their potential relation to IR and cardiovascular disease risk has yet to be clearly defined. The growing prevalence of obesity and metabolic syndrome in the United States, the complex relation of both conditions with CKD, and their association with cardiovascular disease risk underlie the importance of recognizing and defining the risk factors for IR in this patient population. In the present study, we aimed to evaluate potential determinants of IR in a population of patients without diabetes but with stage 3– 4 CKD. We hypothesized that the estimated glomerular filtration rate (eGFR) and body mass index (BMI; in kg/m2) would each be closely associated with levels of IR in persons with CKD. To test this hypothesis, we examined the relation among eGFR, BMI, and insulin resistance, as determined by using the homeostasis model assessment of IR (HOMA-IR) in 95 nondiabetic persons with moderate-to-severe (stage 3– 4) CKD. We 1 From the Division of Nephrology (MLT and TAI) and the Department of Biostatistics (AS), Vanderbilt University Medical Center, Nashville, TN, and the Division of Nephrology, Maine Medical Center, Portland, ME (JH). 2 Supported by National Institutes of Health grants no. R01 DK45604 and K24 DK62849 and Diabetes Research and Training Center grant no. DK20593 from the National Institute of Diabetes, Digestive and Kidney Diseases; grant no. R01 HL070938 from the National Heart, Lung, and Blood Institute; and grant no. M01 RR-00095 from the National Center for Research Resources (to the Vanderbilt General Clinical Research Center). 3 Reprints not available. Address correspondence to TA Ikizler, Division of Nephrology, Vanderbilt University School of Medicine, Medical Center North, S-3223, 1161 21st Avenue, Nashville, TN 37232-2372. E-mail: [email protected]. Received March 22, 2007. Accepted for publication August 7, 2007.

Am J Clin Nutr 2007;86:1642– 8. Printed in USA. © 2007 American Society for Nutrition

INSULIN RESISTANCE AND ADIPOSITY IN MODERATE CKD

compared results in this group with those in a group of 36 subjects with normal kidney function who were frequency matched for race, sex, and BMI.

This value was calculated from fasting concentrations of insulin and glucose by using the following equation (9 –11):

HOMA-IR ⫽ fasting serum insulin 共␮U/mL兲 ⫻ fasting serum glucose 共mg/dL兲/405

SUBJECTS AND METHODS

1643

(2)

Patients

Inflammatory biomarkers

Subjects were recruited from among the patients attending the outpatient nephrology clinics at the Maine Medical Center (Portland, ME) and the Vanderbilt University Medical Center (Nashville, TN). Criteria for study participation included age 쏜 18 y and CKD due to any cause, being followed in one of the above nephrology clinics, and stage 3– 4 CKD as defined by an eGFR between 15 and 59 mL/min. The eGFR was calculated by using the abbreviated equation described in the Modification of Diet in Renal Disease (MDRD) study (7):

Cytokine concentrations were measured in duplicate by using an enzyme-linked immunosorbent assay kit (ELISA; BioSource International, Camarillo, CA). Interleukin (IL)1␤, IL-6, and tumor necrosis factor-␣ (TNF-␣) were measured in plasma, and IL-8 and IL-10 were measured in serum. The assay analytic sensitivity was 2.0 pg/mL for Il-1␤ and IL-6, 3.0 pg/mL for TNF-␣, 0.7 pg/mL for IL-8, and 1.0 pg/mL for IL-10. Interassay and intraassay variability for the cytokine measurements was as follows: 5% and 4% for Il-1␤, 6% and 8% for IL-6, 10% and 5% for TNF-␣, 5% and 5% for IL-8, and 3% and 4% for IL-10, respectively. Serum C-reactive protein (CRP) concentrations were measured by using the highsensitivity particle-enhanced immunoturbidimetric assay (Roche Modular System, Indianapolis, IN). Analytic sensitivity of the CRP assay was 0.003 mg/dL. Adiponectin and resistin analysis was evaluated by using the Human Serum Adipokine (Panel A) Lincoplex Kit [Linco Research (now Millipore), Billerica, MA]. The assay’s analytic sensitivity was 6.7 pg/mL for resistin and 145.4 pg/mL for adiponectin. Intraassay and interassay variability was 1.4 –7.9% and 쏝21%, respectively.

Estimated GFR ⫽ 186 ⫻ SCr共 ⫺ 1.154兲 ⫻ age共 ⫺ 0.203兲 ⫻ 0.742 (if female) ⫻ 1.21 (if black)

(1)

where SCr is serum creatinine. Patients with a prior diagnosis of diabetes mellitus, current use of oral hypoglycemics or insulin, or a fasting glucose concentration 쏜 126 mg/dL (8) were excluded. Patients with acute inflammatory illnesses (eg, AIDS, active hepatitis B or C, malignancy, or systemic lupus erythematosus), hospitalization for cardiac or infection-related morbidity within the past 6 wk, severe comorbid complications, previous kidney transplantation, and current participation in experimental drug protocols; pregnant women; and prison inmates were also excluded from the study. Control subjects between the ages of 45 and 80 y were frequency matched to the CKD group for BMI, race, and sex. Control subjects were recruited from the Vanderbilt University Medical Center vie E-mail communication; they had a normal eGFR and no prior diagnosis of diabetes mellitus. Demographic data, anthropometric measurements, nutritional and hormonal values, and total percentage body fat (%BF) were obtained. All patients provided written informed consent before study enrollment. The study was approved by the institutional review board of each center. Analytic procedures Blood samples All blood draws were performed either at the General Clinical Research Center (Vanderbilt) or the Research Core Laboratory (Maine Medical Center). Venous blood was drawn into Vacutainer tubes (Becton-Dickinson, Franklin Lakes, NJ) containing EDTA supplemented with 1000 U catalase/mL and into serum separator tubes containing clot activator for plasma and serum separation, respectively. Samples for plasma collection were transported on ice and immediately centrifuged at 4 °C and 1700 ҂ g for 15 min; the samples for serum collection were allowed to clot at room temperature before centrifugation. Plasma and serum samples were thereafter stored at Ҁ70 °C until analysis. Plasma glucose concentrations were measured by using the glucose oxidase method (Model II Glucose Analyzer; Beckman Instruments, Fullerton, CA). Plasma insulin was measured by using a double-antibody radioimmunoassay (Linco Research Inc, St Charles, MO). HOMA-IR was used as a measure of IR.

Bioelectrical impedance analysis Lean body mass and total %BF was determined by using a bioelectrical impedance analyzer (RJL Systems, Clinton Township, MI). The subjects were placed in a supine position with their arms at, but not touching, their sides and with their legs apart. Disposable impedance plethysmography source electrodes were positioned on the dorsal surface of the wrist on the right side and the anterior surface of the ipsilateral ankle. The proximal detector electrodes were placed between the distal prominences of the radius and ulna and between the malleoli of the ankle. A current of 800 ␮A at 50 kHz was applied to the subject at the distal electrodes. A voltage drop detected through the proximal electrodes records impedance. Resistance and reactance are measured as current flows through the compartments of the body and are used to determine overall %BF (12). Statistical analysis Statistical analyses were performed in 2 phases. In the first phase, baseline characteristics, the prevalence of IR and the presence of the potential determinants of IR were compared in CKD patients and control subjects by using the Mann-Whitney U test for continuous variables and the chi-square or Fisher’s exact test for categorical variables. Associations between HOMA-IR and the inflammatory cytokines, adipokines, hormonal markers, BMI, and %BF were assessed in study participants by using Spearman’s correlation coefficients (rs). To adjust for age and sex, multiple linear regression models were used with HOMA-IR as the outcome variable. BMI was further categorized into tertiles (BMI: 쏝24.9, 25–30, or 쏜30), and mean HOMA-IR values within each BMI category were graphically presented separately for

1644

TRIROGOFF ET AL

TABLE 1 Baseline characteristics1 Stage 3 CKD patients (n ҃ 65) Male [n (%)] African American [n (%)] Age (y) BMI (kg/m2) Body fat (%) HOMA Serum CRP (mg/L) Glomerular filtration rate (mL/min) Adiponectin (␮g/mL) Resistin (ng/mL)

35 (53.8) 4 (6.2) 66.8 앐 12.42 28 (25–31)3 31.7 앐 12.8 2.9 (1.7–3.8) 2.6 (1.2–6.8) 40 앐 7.9 18.1 (9.69–28.7) 16.0 (12.8–19.2)

Stage 4 CKD patients (n ҃ 30)

P

Stage 3– 4 CKD patients (n ҃ 95)

Matched control subjects (n ҃ 36)

15 (50) 4 (13.3%)

0.73 0.28

66.5 앐 16 27.5 (25–31) 30.2 앐 13.9 2.8 (1.9–4.5) 2.1 (1.1–3.2) 24.1 (14.3)

0.93 0.95 0.46 0.54 0.19 쏝0.001

66.7 앐 13.6 28 (25–32) 31.2 앐 13.1 2.88 (1.92–3.86) 2.3 (1.3–5.8) 35.1 앐 10.2

58.3 앐 10.5 28 (26–30) 31.2 앐 10.8 2.54 (1.95–4.07) 1.2 (0.6–3.5) 82.0 앐 18.3

쏝0.001 0.92 0.95 0.74 0.018 쏝0.001

0.15 0.11

19.66 (10.57–33.24) 16.49 (12.89–22.98)

15.3 (9.01–26.8) 10.65 (7.74–13.13)

0.13 쏝0.001

32.9 (14.4–46.7) 20.4 (14.2–25.1)

50 (52.6) 8 (8.4%)

16 (47.1) 5 (13.8%)

P 0.58 0.59

1

CKD, chronic kidney disease; HOMA-IR, homeostasis model assessment of insulin resistance; CRP, C-reactive protein. P values were calculated by using the Mann-Whitney U test for continuous variables and the chi-square test for categorical variables. 2 x៮ 앐 SD (all such values). 3 Median; interquartile range in parentheses (all such values).

CKD patients and control subjects. We conducted a test for linear trend to assess the equivalence in HOMA-IR scores among BMI categories. In the second phase, CKD patients and control subjects were analyzed separately. Multivariate linear regression was conducted to assess the independent effect of BMI on HOMA-IR after adjustment for covariates including race, sex, age, adiponectin, and eGFR. Covariates were chosen a priori within an allowable number to prevent the model from overfitting, because they were considered to be associated with the outcome variable, HOMA-IR. To assess whether the observed effect of BMI and %BF is caused by a factor that was correlated to both variables, we further combined the 2 variables into a factor by using principal components analysis. This combined variable was then assessed in a similar multivariable regression. Regression residuals were verified for normality. If residual analysis did not fulfill the assumptions of normality, sensitivity analysis was performed with natural log transformation of the dependent variable, which did not affect the overall significance. We used SPSS software (version 13.0; SPSS Inc, Chicago, IL) for analyses and a 2-sided 5% significance level for all statistical inferences. RESULTS

Baseline characteristics The baseline characteristics of the CKD patients and the matched control subjects are shown in Table 1; the CKD patients were subdivided by disease stage. Race, sex, and BMI did not differ significantly between the groups. CKD patients were significantly older than the control subjects. Both the CKD and the control groups were composed of overweight and obese persons (BMI range: 25–32 and 26 –30, respectively). Mean %BF for the CKD and control groups also did not differ significantly. In this nondiabetic population, the mean HOMA-IR score was 3.7 앐 3.2 in CKD patients and 3.1 앐 1.8 in control subjects (P ҃ 0.73). As expected, baseline eGFR was significantly (P 쏝 0.001) lower in the CKD patients than in the control subjects. The only significant (P 쏝 0.001) difference

between patients with stage 3 CKD and those with stage 4 CDK was in baseline eGFR. Significant differences were noted by sex in baseline weight, %BF, and adiponectin concentrations; therefore, subsequent analyses were adjusted for age and sex. Correlation between homeostasis model assessment of insulin resistance and markers of inflammation, body mass index, adiposity, and kidney function The potential relation between IR and inflammation among study participants was examined by comparing HOMA-IR scores with concentrations of inflammatory cytokines and with BMI, %BF, and eGFR (Table 2). In the unadjusted analysis, there was a significant (P 쏝 0.05) negative correlation between HOMA-IR and plasma concentrations of IL-1␤, IL-8, and TNF-␣. No significant correlations were found between TABLE 2 Association between homeostasis model assessment of insulin resistance and inflammatory cytokines and adipokines, body mass index, body fat percentage, and estimated glomerular filtration rate (eGFR) in study participants with and without chronic kidney disease1

IL-1␤ (pg/mL) IL-6 (pg/mL) IL-10 (pg/mL) IL-12 (pg/mL) IL-8 (pg/mL) Adiponectin (␮g/mL) Resistin (ng/mL) TNF-␣ (mg/L) CRP (mg/L) BMI (kg/m2) Body fat (%) eGFR

Value

rs

16.9 앐 91.63 3.9 앐 3.3 2.4 앐 7 4.25 앐 16.4 9.2 앐 10.9 23.9 앐 17.8 16.6 앐 8.5 1.4 앐 5.0 4.22 앐 6.4 29.1 앐 6.2 31.6 앐 13.0 45.9 앐 24

Ҁ0.241 Ҁ0.057 Ҁ0.113 Ҁ0.133 Ҁ0.221 Ҁ0.240 0.035 Ҁ0.213 0.147 0.496 0.286 Ҁ0.01

P 0.012 0.561 0.247 0.172 0.022 0.007 0.701 0.028 0.096 쏝0.001 0.001 0.910

Padj2 0.214 0.195 쏝0.001 0.074 0.058 0.095 0.492 0.465 0.639 쏝0.001 쏝0.001 0.647

1 n ҃ 107. IL, interleukin; TNF-␣, tumor necrosis factor-␣; CRP, C-reactive protein. Associations were calculated by using Spearman’s correlations. eGFR was calculated by modified MDRD equation: 186 ҂ sCrҀ1.154 ҂ ageҀ0.208 ҂ 0.742 (if female) ҂ 1.212 (if African American). 2 Adjusted for age and sex. 3 x៮ 앐 SD (all such values).

1645

INSULIN RESISTANCE AND ADIPOSITY IN MODERATE CKD

Control subjects CKD patients

HOMA-IR

8.00

6.00

4.00

2.00

*

Predictors of insulin resistance in chronic kidney disease patients by multivariate analysis In a multivariable regression model after control for age, African American race, adiponectin, sex, and eGFR, BMI was a significant predictor of IR in CKD patients but not in control subjects. In a separate analysis, %BF measured by bioelectrical impedance analysis was substituted for BMI and was an independent predictor of IR. Because of the high correlation between BMI and %BF, data reduction methods were performed to combine these 2 variables into a single variable. This combination variable was also found to be a significant predictor of IR in multivariable analysis after control for age, eGFR, race, sex, and adiponectin concentrations (Table 3). The individual cytokines found to be significant in the age-adjusted analysis were also combined by data reduction methods and were placed in the model. Even with the addition of this variable, BMI and %BF remained significant predictor of HOMA-IR (data not shown).

< 24.9

DISCUSSION

BMI range FIGURE 1. Homeostasis model assessment of insulin resistance (HOMA-IR) versus BMI ranges in patients with chronic kidney disease (CKD) and control subjects. Mean HOMA-IR scores in CKD patients and control subjects by BMI ranges of normal (쏝24.9), overweight (25–30), and obese (쏜30) differed significantly between the 3 ranges (P 쏝 0.001, linear trend test). Error bars represent 95% CIs. BMI 쏝24.9 data from control subjects were omitted because n ҃ 3.

HOMA-IR and other proinflammatory cytokines. The inverse correlation between HOMA-IR and adiponectin concentrations was significant (P ҃ 0.007), but no significant correlation was found between HOMA-IR and resistin concentrations. However, IL-10, BMI, and %BF remained significantly associated after adjustment for age and sex. In both the CKD patients and the control subjects, HOMA-IR showed a highly significant (P 쏝 0.001) correlation with BMI. When BMI was divided into tertiles of 쏝24.9, 25–30, and 쏜30, HOMA-IR scores differed significantly (P 쏝 0.001, test for linear trend) in the CKD group (Figure 1). BMI and %BF also were significantly correlated with CRP in CKD patients (rs ҃ 0.308, P ҃ 0.002 and rs ҃ 0.235, P ҃ 0.025, respectively). CRP concentrations also differed significantly according to BMI tertile in the CKD and control groups (P ҃ 0.008 and 0.007, respectively; Kruskall-Wallis test).

It has long been recognized that a complex relation exists among uremia, glucose dispersion, and insulin function. Alterations in insulin function associated with CKD were reported as early as 1951 (13, 14), and the effects of kidney disease on renal uptake and excretion of insulin were reported as early as 1970 (15). In a seminal series of studies, DeFronzo et al (3, 16) and DeFronzo (17) used euglycemic insulin clamp techniques to characterize uremic IR in patients with ESRD who required dialysis. Thus, the pathophysiology of uremic IR in patients undergoing dialysis has been relatively well recognized for many years. Dialysis-dependent patients are under severe physiological stress, and it is likely that additional metabolic abnormalities contribute to uremic IR in these patients. In contrast, few investigations have focused on understanding IR in the larger population with less severe CKD. In the present study, we examined the determinants of IR in nondiabetic patients with stage 3– 4 CKD on the basis of the hypothesis that worsening kidney function would be associated with increasing IR. To our surprise, our results show that IR in this CKD patient cohort is primarily determined by BMI and not by eGFR. Furthermore, %BF, when substituted for BMI, was also predictive of IR in CKD patients. Further analysis indicated that the %BF of BMI is the relevant component predicting IR (ie, HOMA-IR); this relation was maintained in multivariate regression analysis. Adjustment for IL-6 and TNF-␣ did not change

TABLE 3 Results of 3 separate multivariable linear regression models and significant predictors of insulin resistance with BMI, percentage body fat, and combined BMI and percentage body fat among patients with chronic kidney disease (CKD) and control subjects1 Control subjects2

CKD patients

Variable BMI Percentage body fat Combined BMI and percentage body fat3

Standardized regression coefficient

P

Standardized regression coefficient

P

0.55 0.930 0.670

쏝0.001 쏝0.001 쏝0.001

Ҁ0.09 Ҁ0.159 Ҁ0.145

0.13 0.103 0.119

1 Insulin resistance was calculated by log HOMA (homeostasis model assessment). All 3 models were controlled for age, sex, estimated glomerular filtration rate, African American race, and adiponectin concentrations. 2 Covariates were combined into a single factor by using principal components analysis. 3 Variables were combined into a factor by using principal components analysis.

227 (white)

51.3 앐 8.410 50.9 앐 8.211 63 (38–96)12

112.8 앐 89

HOMA-IR

HOMA-IR

No

23.5 앐 2.410 22.0 앐 2.911 25.2 앐 3.8

Yes

No controls

No

Not specified

Yes (multivariate analysis)

Yes (univariate analysis) No (multivariate analysis)

Not noted

Yes

No

Not studied

GFR associated with HOMA-IR

Yes

Yes

26.3 앐 8

26.8 앐 3.6

25.7 앐 1.3 25.9 앐 0.7 25.4 앐 1.1 21.4 앐 3.0

22.8 (21.9–28.5)

BMI

Controls matched for BMI?

2

GFR, glomerular filtration rate; HOMA-IR, homeostasis model assessment of insulin resistance. x៮ ; range in parentheses (all such values). 3 GFR was assessed by using the Cockcroft-Gault equation. 4 The GFR and BMI values are by tertiles of serum creatinine concentrations: 쏝1.3, 1.3–3.0, and 쏜3.0 mg/dL, respectively. 5 x៮ 앐 SD (all such values). 6 GFR was assessed by using inulin clearance. 7 The method of GFR assessment was not specified. 8 GFR was assessed by using 24-h creatinine clearance; the value is for all subjects. 9 GFR was assessed by using the equation from the Modification of Diet in Renal Disease study. Approximately 123 of the subjects had estimated GFR 쏝60 mL 䡠 minҀ1 䡠 1.73 mҀ2. 10 In subjects with hypertensive kidney disease. 11 In normotensive subjects with chronic kidney disease. 12 GFR was assessed by using iothalamate clearance.

1

Kanauchi et al, 2004 (11) Becker et al, 2005 (20)

6453 (0.6% nonHispanic black; 0.5% Mexican American) 120 (Japanese)

Chen et al, 2003 (5)

Oral-glucose-tolerance test and hyperinsulinemic euglycemic clamp HOMA-IR

쏝908

321

Sechi et al, 2002 (23)

119 앐 55,6 67 앐 46 25 앐 26 25.8 앐 267

50 (white)

29 (Japanese)

Short insulin tolerance test Frequently sampled intravenous glucose tolerance test Hyperinsulinemic euglycemic clamp

䡠 1.73m

⫺2

Insulin sensitivy assessment

102 (81–125)2,3

ml 䡠 min

⫺1

Mean GFR

15 (white)

No. of cases (race-ethnicity)

Kobayashi et al, 2005 (22)

Vareesangthip et al, 1997 (21) Fliser et al, 1997 (19)4

Authors

TABLE 4 Summary of literature evaluating insulin resistance in chronic kidney disease patients without diabetes1

Yes

Not specified

Not studied

Not specified

Yes

Yes

Not studied

BMI associated with HOMA-IR

1646 TRIROGOFF ET AL

INSULIN RESISTANCE AND ADIPOSITY IN MODERATE CKD

these results. Whereas a significant association between BMI and IR is well recognized in the general population, the relation between body composition (in particular, %BF and IR) in stage 3– 4 CKD patients has been less well studied, and it constitutes a novel aspect of the present study. To our knowledge, this study is one of the first to describe BMI as the primary determinant of IR in CKD patients, and it is the first to evaluate the relative contribution of fat mass in this relation. Recent studies have suggested a complex relation between IR and CKD. A cross-sectional study utilizing participants from the third National Health and Nutrition Examination Survey (NHANES III) examined associations between metabolic syndrome and CKD and found that a person’s odds of having kidney disease increased as the number of metabolic syndrome components possessed by him or her increased. This association remained significant after adjustment for the presence of hypertension and diabetes, 2 well-known causes of CKD (5). Kurella et al (18) conducted a prospective study using the Atheroslcerosis Risk in Communities study cohort to establish the metabolic syndrome as an independent risk factor for CKD in nondiabetic adults. Their data indicated that obesity and other components of the metabolic syndrome may contribute to the development or progression of CKD, but the data did not indicate whether the development of CKD also contributes to IR. We observed that, in our stage 3– 4 CKD patient group, eGFR did not correlate with the degree of IR. This has also been noted by other investigators who evaluated the presence of IR in CKD (19, 20), regardless of the method by which IR or GFR was measured. Previous studies that examined IR in kidney disease patients are summarized in Table 4. Kobayashi et al (22) described a relation between eGFR and IR that was calculated with the use of the hyperinsulinemic euglycemic clamp technique, but that relation was not maintained in multivariate analysis. Thus, whereas IR is present in these patients, the severity of underlying CKD does not seem to be the principal cause of the metabolic derangement—at least in our study population. The prevalence of obesity continues to increase in the United States, and thus much attention is being focused on the role of adipose tissue—in particular, visceral adipose tissue—as an active secretory organ modulating endocrine systems. The adipokine adiponectin is known for its role in regulating insulin sensitivity (24). Although adiponectin is secreted by adipocytes, its concentrations are lower in obese subjects than in lean subjects (25). This counterintuitive relation is not completely understood, but feedback inhibition of adiponectin’s production by inflammatory cytokines such as TNF-␣ (26), which are higher with greater visceral obesity, may contribute to it (27). Low adiponectin concentrations have been associated with the development of IR in mouse models of obesity (28). The correlation between IR and adiponectin in the present study approached significance only in the adjusted analysis. Visceral fat contains greater amounts of inflammatory mediators—including CRP, IL-6, and TNF-␣—than does subcutaneous fat, and these mediators are thought to contribute to the development of IR (29). In the ESRD population, Axelsson et al (30) found an association between the inflammatory biomarkers and regional fat distribution, in which greater truncal fat mass correlated with higher concentrations of IL-6 and CRP. It is interesting that the data in the present study showed a negative correlation between HOMA-IR and the concentrations of individual cytokines. The cause of this counterintuitive relation is not

1647

clear, and that lack of clarity calls for further studies examining the mechanisms underlying these observations. There are several limitations to our study, in particular the relatively small size of the study population. In addition, the cross-sectional nature of this study, although showing an association between BMI and IR, does not provide information regarding causal relations. Moreover, the control subjects were significantly younger than the CKD patients, which may have accounted for the differences noted. However, subsequent analysis was adjusted for age and sex. Rather than direct measurement, the GFR in both groups was estimated by the use of the abbreviated equation described in the MDRD study (7). This equation has been found to be an adequate predictor of GFR when 24-h creatinine clearance or inulin clearance is not available (31). Ideally, the use of the hyperinsulinemic euglycemic clamp would have provided the best measure of IR in this study, but its use is laborious and time-consuming for a study of this size. Shoji et al (9) showed that HOMA-IR scores correlate well with the hyperinsulinemic euglycemic clamp as a measure of IR in individuals with a wide range of GFRs. Whereas we provide intriguing data regarding body composition and IR, the %BF measured in the CKD patients and in the control subjects in the current study did not differentiate between truncal and nontruncal fat, which may have resulted in an underestimation of the relation of %BF, adiponectin concentrations, and IR. Finally, limiting our study population to persons with stage 3 or 4 CKD limited the extrapolation of our findings to other stages of kidney disease and hindered the detection of a correlation between GFR and IR over a wider range of kidney functions. In summary, our data show that BMI measures, particularly %BF, are the major determinant of IR in nondiabetic stage 3– 4 CKD patients. Whereas the IR of uremia may be seen in the population of ESRD patients undergoing dialysis, who experience greater metabolic stress (3), body composition likely plays a more significant role in the development of IR in patients with less severe renal disease. Prospective studies are needed to more clearly define this relation and to determine whether interventions targeting IR in this patient population can decrease cardiovascular morbidity and mortality, as well as progression to ESRD. The authors thank Karen Majchrzak, Cindy Booker, Andrew Vincz, and the Vanderbilt General Clinical Research Center nursing staff and Jane Kane at Maine Medical Center for their excellent technical assistance. The authors’ responsibilities were as follows—TAI and JH: contributed equally to designing the experiment, collecting and analyzing the data, and writing the manuscript; MLT: analyzed the data and wrote the manuscript; and AS: performed data analyses. None of the authors had a personal or financial conflict of interest.

REFERENCES 1. Go AS, Chertow GM, Fan D, McCulloch CE, Hsu CY. Chronic kidney disease and the risks of death, cardiovascular events, and hospitalization. N Engl J Med 2004;351:1296 –305. 2. Himmelfarb J, Stenvinkel P, Ikizler TA, Hakim RM. The elephant in uremia: oxidant stress as a unifying concept of cardiovascular disease in uremia. Kidney Int 2002;62:1524 –38. 3. DeFronzo RA, Alvestrand A, Smith D, Hendler R, Hendler E, Wahren J. Insulin resistance in uremia. J Clin Invest 1981;67:563– 8. 4. Shinohara K, Shoji T, Emoto M, et al. Insulin resistance as an independent predictor of cardiovascular mortality in patients with end-stage renal disease. J Am Soc Nephrol 2002;13:1894 –900. 5. Chen J, Muntner P, Hamm LL, et al. Insulin resistance and risk of chronic

1648

6.

7.

8. 9. 10.

11.

12.

13.

14.

15.

16. 17. 18.

TRIROGOFF ET AL

kidney disease in nondiabetic US adults. J Am Soc Nephrol 2003;14: 469 –77. Coresh J, Astor BC, Greene T, Eknoyan G, Levey AS. Prevalence of chronic kidney disease and decreased kidney function in the adult US population: third National Health and Nutrition Examination Survey. Am J Kidney Dis 2003;41:1–12. Levey AS, Bosch JP, Lewis JB, Greene T, Rogers N, Roth D. A more accurate method to estimate glomerular filtration rate from serum creatinine: a new prediction equation. Modification of Diet in Renal Disease Study Group. Ann Intern Med 1999;130:461–70. Genuth S, Alberti KG, Bennett P, et al. Follow-up report on the diagnosis of diabetes mellitus. Diabetes Care 2003;26:3160 –7. Shoji T, Emoto M, Nishizawa Y. HOMA index to assess insulin resistance in renal failure patients. Nephron 2001;89:348 –9. Matthews DR, Hosker JP, Rudenski AS, Naylor BA, Treacher DF, Turner RC. Homeostasis model assessment: insulin resistance and betacell function from fasting plasma glucose and insulin concentrations in man. Diabetologia 1985;28:412–9. Kanauchi M, Akai Y, Hashimoto T. Validation of simple indices to assess insulin sensitivity and pancreatic Beta-cell function in patients with renal dysfunction. Nephron 2002;92:713–5. Lukaski HC JP, Bolonchuk WW, Lykken GI. Assessment of fat-free mass using bioelectrical impedance measurements of the human body. Am J Clin Nutr 1985;41:810 –7. Runyan JW Jr, Hurwitz D, Robbins SL. Effect of Kimmelstiel-Wilson syndrome on insulin requirements in diabetes. N Engl J Med 1955;252: 388 –91. Zubrod CG, Eversole SL, Dana GW. Amelioration of diabetes and striking rarity of acidosis in patients with Kimmelstiel-Wilson lesions. N Engl J Med 1951;245:518 –28. Rabkin R, Simon NM, Steiner S, Colwell JA. Effect of renal disease on renal uptake and excretion of insulin in man. N Engl J Med 1970;282: 182–7. DeFronzo RA, Smith D, Alvestrand A. Insulin action in uremia. Kidney Int Suppl 1983;16:S102–14. DeFronzo RA. Pathogenesis of glucose intolerance in uremia. Metabolism 1978;27:1866 – 80. Kurella M, Lo JC, Chertow GM. Metabolic syndrome and the risk for chronic kidney disease among nondiabetic adults. J Am Soc Nephrol 2005;16:2134 – 40.

19. Fliser D, Pacini G, Engelleiter R, et al. Insulin resistance and hyperinsulinemia are already present in patients with incipient renal disease. Kidney Int 1998;53:1343–7. 20. Becker B, Kronenberg F, Kielstein JT, et al. Renal insulin resistance syndrome, adiponectin and cardiovascular events in patients with kidney disease: the Mild and Moderate Kidney Disease Study. J Am Soc Nephrol 2005;16:1091– 8. 21. Vareesangthip K, Tong P, Wilkinson R, Thomas TH. Insulin resistance in adult polycystic kidney disease. Kidney Int 1997;52:503– 8. 22. Kobayashi S, Maesato K, Moriya H, Ohtake T, Ikeda T. Insulin resistance in patients with chronic kidney disease. Am J Kidney Dis 2005; 45:275– 80. 23. Sechi LA, Catena C, Zingaro L, Melis A, De Marchi S. Abnormalities of glucose metabolism in patients with early renal failure. Diabetes 2002; 51:1226 –32. 24. Chandran M, Phillips SA, Ciaraldi T, Henry RR. Adiponectin: more than just another fat cell hormone? Diabetes Care 2003;26:2442–50. 25. Arita Y, Kihara S, Ouchi N, et al. Paradoxical decrease of an adiposespecific protein, adiponectin, in obesity. Biochem Biophys Res Commun 1999;257:79 – 83. 26. Wang B, Jenkins JR, Trayhurn P. Expression and secretion of inflammation-related adipokines by human adipocytes differentiated in culture: integrated response to TNF-alpha. Am J Physiol Endocrinol Metab 2005;288:E731– 40. 27. Tsigos C, Kyrou I, Chala E, et al. Circulating tumor necrosis factor alpha concentrations are higher in abdominal versus peripheral obesity. Metabolism 1999;48:1332–5. 28. Yamauchi T, Kamon J, Waki H, et al. The fat-derived hormone adiponectin reverses insulin resistance associated with both lipoatrophy and obesity. Nat Med 2001;7:941– 6. 29. Fain JN, Madan AK, Hiler ML, Cheema P, Bahouth SW. Comparison of the release of adipokines by adipose tissue, adipose tissue matrix, and adipocytes from visceral and subcutaneous abdominal adipose tissues of obese humans. Endocrinology 2004;145:2273– 82. 30. Axelsson J, Rashid Qureshi A, Suliman ME, et al. Truncal fat mass as a contributor to inflammation in end-stage renal disease. Am J Clin Nutr 2004;80:1222–9. 31. Lewis J, Agodoa L, Cheek D, et al. Comparison of cross-sectional renal function measurements in African Americans with hypertensive nephrosclerosis and of primary formulas to estimate glomerular filtration rate. Am J Kidney Dis 2001;38:744 –53.

Accuracy of the Atwater factors and related food energy conversion factors with low-fat, high-fiber diets when energy intake is reduced spontaneously1–3 Maggie L Zou, Paul J Moughan, Ajay Awati, and Geoffrey Livesey ABSTRACT Background: Systems to calculate metabolizable energy (ME) in foods and diets are often based on Atwater factors. The accuracy of these factors with low-fat diets high in fiber is unknown when food intake is reduced spontaneously. Objective: The objective was to evaluate the accuracy of Atwater factors and other systems for calculating ME available from low-fat, high-fiber diets when food intake was reduced spontaneously. Design: The ME contents of a high-fat, low-fiber diet and 2 low-fat diets, one high in fruit and vegetable fiber and the other high in cereal fiber, were determined in a randomized parallel study in humans (n ҃ 27) and compared with various factorial and empirical models for calculating ME. Results: Food intakes decreased with both the high fruit and vegetable fiber and cereal fiber diets. The difference between ME calculated by using Atwater and similar factors and determined ME values was up to 4% for the refined diet and up to 11% for the low-fat, high-fiber diets. Various factorial and empirical systems for calculating food energy failed to reflect the results of the direct determinations. Conclusion: Atwater factors were inaccurate with low-fat, highfiber diets. Although modified Atwater factors may be accurate under standardized conditions of zero-nitrogen and zero-energy balance, they overestimate energy availability from high-fiber fruit and vegetable and cereal diets when food intake is reduced spontaneously in addition to when intake is reduced voluntarily. Am J Clin Nutr 2007;86:1649 –56. KEY WORDS Atwater factors, energy density, dietary fiber, digestibility, metabolizable energy

INTRODUCTION

The incidence of obesity and related health disorders in developed nations is a matter of major concern (1). One approach to help combat obesity is the inclusion of low-energy foods in the diet and careful monitoring of dietary energy intakes versus energy expenditures to achieve (zero) energy balance at maintenance or negative energy balance if a person is overweight. To develop such foods and dietary regimens, however, implies a need to determine the available energy content of foods with appropriate accuracy, such that foods or components of food with different energy contents can be differentiated. Moreover, zero energy balance means unchanged energy content in the body

rather than zero difference between metabolizable energy (ME) intake and equivalent energy expenditure. This is because some fuels are used inefficiently in metabolism, thereby having physiologic fuel values different from their ME (2). The Atwater general factors (3), although not originally intended to be used generically, are commonly applied to estimate the ME content of foods (4) and may be used in the United States and other regions for food labeling purposes (5). Atwater described “available” food energy in terms of Physiologic Fuel Value (PFVs). He and Rubner researched food and thermogenesis essentially with a view toward a net ME system. However, PFVs have subsequently been discussed and implemented ignoring thermogenesis (3). Nevertheless, the conversion of ME factors to net ME factors is now well defined with good agreement with theory and with animal and human studies (2, 6). Greater uncertainty arises because of the difficulty in predicting energy losses, which affect metabolizability and ME and therefore is the focus of this study. The Atwater factors have important shortcomings. The gross energy contents of dietary proteins, fats, and carbohydrates are not constants. Other chemical components in foods other than protein, fat, and available carbohydrates contribute energy and may influence the ME of the principal chemical components. The digestibilities of macronutrients are variable, as is the energy per unit nitrogen excretion in urine, and some aspects of analytic methods also raise questions. There is little published information on low-fat, high-fiber diets when food intakes are not strictly controlled. High-fiber diets or diets of low energy density are generally recommended to lower food intake, and when volunteers deliberately reduce their food intake by consuming a high-fiber diet, current food-energyassessment systems have poor accuracy and overestimate energy availability (7). The present study, in which ME was determined in subjects consuming a low-fiber, high-fat diet or 2 low-fat, high-fiber diets (cereal or fruit and vegetable based), allowed an assessment to 1

From the Riddet Centre, Massey University, Palmerston North, New Zealand (MLZ, PJM, and AA), and Independent Nutrition Logic Ltd, Pealerswell House, Wymondham, Norfolk, United Kingdom (GL). 2 Supported by the Riddet Centre, Massey University, Palmerston North, New Zealand. 3 Address reprint requests and correspondence to ML Zou, Riddet Centre, Massey University, Private Bag 11 222, Palmerston North, New Zealand. E-mail: [email protected]. Received April 18, 2007. Accepted for publication July 5, 2007.

Am J Clin Nutr 2007;86:1649 –56. Printed in USA. © 2007 American Society for Nutrition

1649

1650

ZOU ET AL

TABLE 1 Physical characteristics of each group of subjects receiving 1 of 3 experimental diets1

No. of subjects Total Women Men Age (y) x៮ SE Range Height (cm) x៮ SE Range Body weight (kg) x៮ SE Range

Refined diet

Fruit and vegetable diet

Cereal diet

9 5 4

9 5 4

9 5 4

35.9 2.1 26–46

35.2 3.7 24–59

38.8 3.9 23–52

and were not receiving any medication. None of the subjects had received antibiotics for at least 3 mo before the study. Written informed consent was obtained from all participants, and confidentiality was maintained throughout the study. Approval to conduct the study was given by Massey University Human Ethics Committee (HEC 98/123). Subjects were well educated and highly motivated. Diets

168.8 3.3 155.2–188.1

168.1 2.5 155.4–179

169.6 3.5 154–180

73.4 6.5 54.3–120.4

75.8 7.2 52.2–123

71.2 4.9 45.8–91.6

1 There were no significant differences in physical characteristics between groups.

be made of differences in nutrient digestibility and urinary energy excretion among such diets under conditions in which energy intake was reduced spontaneously (at will or in response to physiologic cues). A comparison of determined ME values with ME values predicted by using Atwater and related factors and several empirical models is described. SUBJECTS AND METHODS

Subjects Twenty-seven adult subjects completed the study (Table 1). Subjects were visibly healthy, had no gastrointestinal problems,

Three diets were formulated with ingredients of stable composition that were also easy to handle, store, and cook. Diet 1, designated the “refined diet,” contained mainly high-fat, lowfiber foods (eg, white bread, butter, peanuts, homogenized milk, ham, cookies, chocolate bars, cheese, and mince pie). Diet 2, designated the “fruit and vegetable diet,” contained large amounts of fruit and vegetables (eg, dried apricots, sultanas, prunes, fruit bar, and fruit bread). Diet 3, designated the “cereal diet,” contained large amounts of cereal-based foods (eg, wholemeal bread, bran cereal, crisp bread, bran and apple muffin). The 27 free-living volunteers were randomly allocated to 3 arms of a study of parallel design, each with 1 of the 3 diets (9 subjects per diet). Each subject was allocated 1 of 5 preset amounts of food according to prior recall of 24-h habitual intake and an estimate of energy requirement based on body weight and activity. After allocation, most of the subjects consuming the 2 high-fiber diets indicated that they were uncomfortable trying to consume the volume of food and were thus reallocated to the next lower preset intake level. Once allocated to a food intake level, the subjects consumed the given amount of food for the duration of the study. Although dietary intake was individualized for each subject, the proportions of the nutrients within any one diet, and thus the proportion of energy derived from the respective energyyielding nutrients, remained the same within dietary treatments regardless of differences in absolute intakes. Diet compositions as eaten were the same across participants within diets, whereas they differed between diets (Table 2). Subjects were provided

TABLE 2 Determined chemical composition (g/100 g dry food) of the 3 experimental diets1 Component

Refined diet

Fruit and vegetable diet

58.31 17.62 27.59 3.21 51.57 45.87 49.36 46.66 20.14 25.72 4.90 4.54 0.36 3.51 0.45

55.66 13.13 13.13 3.26 70.48 58.21 60.93 62.59 34.74 23.47 7.89 6.83 1.06 5.64 0.81

Cereal diet

% Moisture Crude protein Fat Ash Total CHO by difference2 Available CHO as carbohydrate weight3 Available CHO as monosaccharide equivalents Available CHO by difference4 Sugars Starch Total dietary fiber Insoluble dietary fiber Soluble dietary fiber Nonstarch polysaccharide Resistant starch 1

CHO, carbohydrate. Calculated as 100 Ҁ fat (%) Ҁ crude protein (%) Ҁ ash (%) Ҁ moisture (%). 3 Represents the sum of the individual mono- and disaccharides and starch expressed as the weight of the carbohydrate. 4 Represents available carbohydrate calculated as total carbohydrate by difference Ҁ total dietary fiber (%). 2

65.96 15.83 10.91 4.18 69.08 55.15 58.78 59.06 26.56 28.59 10.02 9.94 0.07 7.82 0.64

FOOD ENERGY FROM LOW-ENERGY-DENSITY DIETS

clear instructions, which were reinforced throughout the study, to not eat foods other than those provided and to not imbibe alcoholic or other energy-containing beverages. Tea, coffee, pepper, salt, and sugar were supplied to each subject, and each subject accurately recorded the actual quantities of these items consumed daily. Each subject’s actual dietary energy intake was corrected by taking into account the amount of tea, coffee, and sugar ingested and any food not consumed. Experimental procedures Diets were eaten for 10 d, including a 4-d preliminary period followed by 6 d of the balance period, when feces and urine were collected. Urine samples were acidified at collection with small quantities of 6 mol/L HCl. Excreta were kept on ice after collection, transferred to the laboratory twice daily, weighed, and stored frozen (Ҁ20 °C). Subsequently, feces were thawed, bulked over days, freeze-dried, finely ground (1-mm mesh), and mixed. Urine was also bulked over days. For each subject, representative samples of excreta were freeze-dried before chemical analysis. On 5 occasions during the experimental period, an entire day’s duplicate meals were collected (as eaten), weighed, homogenized, sampled, and frozen (Ҁ20 °C). The composite samples were freeze-dried and finely ground (1-mm mesh) before nutrient analysis. Chemical analysis Diet, uneaten foods, feces, and urine were analyzed per subject in duplicate for dry matter (DM), ash, total nitrogen, total fat, gross energy, and other constituents as described below. The DM and ash contents of the feces and food samples were determined after drying the samples in an oven at 105 °C for 16 h, which was followed by ashing in a Muffle furnace (FR-550; Gallenkamp, London, United Kingdom) at 500 °C for 16 h (8). The DM contents of the total diet samples, which contained large amounts of sugar and fat, were determined by drying in a 70 °C vacuum oven until a constant weight was achieved (앒24 h) (8). Heats of combustion (gross energy) of the samples were determined by using an adiabatic bomb calorimeter with benzoic acid thermochemical standard (Gallenkamp Co Ltd, London, United Kingdom) (9). Total nitrogen was determined on both food and feces samples by using the Dumas method on a LECO FP3000 CNS auto analyzer (8, 10). The nitrogen content of urine samples was determined on a Kjeltech 1030 auto analyzer (Tecator, Hoganas, Sweden) by following a standard Kjeldahl procedure (8). Crude protein was calculated as total nitrogen ҂ 6.25, with the exception of coffee, for which the conversion factor 5.3 was used (US Department of Agriculture, SR13). Ammonia, urea, uric acid, and creatinine in urine samples were determined on a Cobas Fara II autoanalyzer (Hoffman-La Roche, Basel, Switzerland), following the procedures outlined by Tiffany et al (11), Fossati et al (12), and Larsen (13), respectively. The Soxhlet method (8) was used to measure the fat content of dietary fecal samples. All fecal samples were dried overnight in a 60 °C oven, which was followed by extraction with petroleum spirit (40 – 60 °C) for 7 h (8). For the dietary samples, the material was acid hydrolyzed (3 mol/L HCl) before fat extraction (8). The fat content of tomato sauce was determined according to the Mojonnier method (8). A one-step extraction-transesterification procedure was used to determine total fatty acids in the dietary and fecal samples (14).

1651

Samples (50 –300 mg) of freeze-dried material containing 10 –50 mg fatty acids were treated with a solvent mixture consisting of methanol-toluene-acetyl chloride (27:20:3) at 70 °C for 2 h. The organic layer was transferred into a screw-capped tube (Kimax; Kimble Glass Inc, Vineland, NJ) and dried, and the pigments were removed by adding anhydrous sodium sulfate and florisil. The fatty acid composition was subsequently determined by gas chromatography (Shimadzu GC-8A, packed column with 15% Eggs-X on chromosorb W, 100 –120 mesh; Shimadzu Corporation, Kyoto, Japan) with nitrogen as the carrier gas, FID as the detector, and pentadecanoic acid (C15:0) as internal standard. The amount of total carbohydrate (%) in the diets and feces was defined as the difference between 100 and the sum of the percentage of water, protein, total fat, and ash (15). The amounts of total, soluble, and insoluble dietary fiber in the dietary samples were analyzed by using an enzymatic-gravimetric method (8). Total nonstarch polysaccharide (NSP) in the dietary samples was determined as described by Englyst and Cummings (16, 17). Total NSP was equal to the sum of neutral sugars and uronic acids expressed as polysaccharide. The amount of available carbohydrate in the dietary samples was determined as the sum of the individual mono- and disaccharides, and starch was expressed as the weight of the carbohydrate (18). Sugars were extracted with aqueous alcohol followed by derivatization with Tri-Sil Z (TMS-Imidazole in pyridine; Pierce Ltd, Bonn, Germany), then measured by gas chromatography (Simadzu GC, OV17 column, temperature program 170-240 °C at 5 °C/min, N2 as carrier gas). Starch was determined by using a commercial kit (Total Starch Kit AA/AMG; Megazyme Australia, Sydney, Australia) following a standard procedure (8). Samples were completely dissolved in dimethylsulfoxide, and hydrolyzed with thermostable ␣-amalyse and amyloglucosidase (AMG). The amount of released glucose was determined by spectrophotometry with the use of Chromogen reagent glucose oxidase/peroxidase reagent (GOPOD) (Megazyme International Ireland Ltd, Wicklow, Ireland). The level of available carbohydrate was also calculated as the sum of the individual mono- and disaccharides and starch but was expressed as monosaccharide equivalents. Resistant starch in the dietary samples was measured by using a commercial kit (Resistant Starch Kit; Megazyme Australia) following a standard procedure (AOAC 2002.02). Samples (defatted if the fat content was 쏜10%) were incubated at 37 °C in a shaking water bath with pancreatic ␣-amylase and AMG for 16 h to remove nonresistant starch. The resistant starch was then recovered as a pellet by centrifugation and was dissolved in 2 mol/L KOH. This solution was then neutralized and hydrolyzed with AMG. The resistant starch was quantified by measuring the amount of released glucose with GOPOD (19). Data analysis Dietary ME values were calculated by using Equation 1 (3), whereas apparent digestibility of gross energy, fatty acids analyzed, and macronutrients were calculated by using Equation 2:

ME (MJ/kg) ⫽ [GE food (MJ/d) ⫺ GE feces (MJ/d) ⫺ GE urine (MJ/d)]/food intake (kg/d)

(1)

Apparent digestibility of analyte (%) ⫽ 100 ⫻ [analyte food (units/d) ⫺ analyte feces (units/d)]/analyte food (units/d)

(2)

1652

ZOU ET AL

TABLE 3 Nonstarch polysaccharide (NSP) components in each diet Component

Refined diet

Arabinose Xylose Mannose Galactose Glucose Uronic acid

23.1 23.6 5.1 10.3 23.6 14.2

Fruit and vegetable diet

Cereal diet

% of total NSP 19.7 16.1 11.5 15.4 16.3 20.7

27.5 35.8 2.0 6.8 19.2 8.7

Data were analyzed by using analysis of variance. If a significant effect of diet was found, differences between the 3 diet groups were compared by using Tukey’s test. Student’s t test was used to compare determined with calculated ME values for each of the 3 diets, respectively. Results were considered statistically significant at P 쏝 0.05. All statistical procedures were performed with the use of SAS (version 9; SAS Institute, Inc, Cary, NC). RESULTS

The fruit and vegetable diet and the cereal diet were lower in crude protein and fat than was the refined diet but the former 2 diets had considerably higher fiber contents (Table 2). As expected, the fruit and vegetable diet also had the highest content of uronic acid, and the cereal diet the highest content of neutral saccharides typical of mainly cellulose and hemicellulose (Table 3). The full amino acid and fatty acid compositions of the diets were reported elsewhere (20). When a comparison was made in the percentage difference between estimated and actual GE intakes between the high-fiber diets and the refined diet, subjects consuming the high-fiber diets ingested 17.8% less energy than that predicted to meet their normal energy intake (P 쏝 0.05; Table 4). With the high-fiber diets of lower energy density, subjects elected to choose a lower food intake than that initially allocated. Food intake was spontaneously reduced. Three of the subjects consuming the highfiber diets did not elect to choose a lower food intake, and, when the data for these 3 subjects were excluded from the statistical analysis, daily GE intake was also found to be significantly (P 쏝

0.05) lower with the high-fiber diets than with the refined diet. The overall results of the study were unchanged when the data for the 3 subjects were excluded. Fecal and urinary energy excretions were highly variable between subjects on a diet (Table 4). Fecal bulking (g fecal dry matter/kg dry food intake) was significantly higher (P 쏝 0.005) with the cereal diet than with either the refined or the fruit and vegetable diet (Table 4). Urinary energy excretion did not differ significantly between diets, but urinary nitrogen was lower so that the energy to nitrogen ratio of urine increased with fiber intake, more so with the fruit and vegetable fiber diet than with the cereal fiber diet (Table 5). However, the daily food energy intakes were lower than the estimated energy requirements with the high-fiber diets, and subjects who consumed the lower energy intakes may have had a negative nitrogen balance. Urea was the major urinary nitrogen constituent. Urinary creatinine nitrogen excretion was higher with the fruit and vegetable diet than with the other 2 diets (P 쏝 0.01 for both diets), whereas the uric acid nitrogen content was significantly higher with the cereal diet than with the refined and fruit and vegetable diets (P 쏝 0.005 for both diets). As expected, the digestibilities of energy and crude protein were lower with the higher-fiber diets (P 쏝 0.001) (Table 6). The digestibility of fat was lower with the fruit and vegetable diet (P 쏝 0.001) than with the refined diet, but this was not so for the cereal diet (Table 6). The digestibility of monounsaturated fatty acids decreased with increasing chain length with all diets (Table 6). Unsaturation improved digestibility of the 18-carbon series of fatty acids from the lower digestible fatty acids (18:O; stearic acid) to the higher digestible fatty acids (18.1,18.2, or C:18.3) (Table 6). The ME of the high-fiber diets, when calculated by a variety of methods (refer Table 7), differed from the experimentally determined values (Table 8). However, whether calculated or determined, ME was always lower (P 쏝 0.001) with the high-fiber diets than with the low-fiber diet (Table 8), with metabolizabilities (ME/GE) of 0.90, 0.86, and 0.85 for the low-fiber and high fruit, vegetable fiber, and high-cereal-fiber diets, respectively. It is clear (Table 8) that application of the Atwater factors led to statistically significant and practically relevant differences between predicted and determined ME values for the 3 diets tested. The use of Food and Drug Administration (FDA) and British

TABLE 4 Daily food and food energy intakes and energy excretion by the subjects in the 3 experimental diet groups1

Fresh food intake (g/d) Food dry matter intake (g/d) Daily GE intake (MJ/d) Fecal excretion (MJ/d) Urinary excretion (MJ/d) Digestible energy intake (MJ/d) Difference between estimated and actual GE intake (%)2,3 Dry fecal bulking (g fecal dry matter/kg dry food intake) Wet fecal bulking (g wet fecal weight/kg dry food intake)

Refined diet

Fruit and vegetable diet

Cereal diet

1244 앐 87 524 앐 40 12.250 앐 0.94 0.801 앐 0.06 0.388 앐 0.02 11.449 앐 0.88 1.4 앐 1.9 66.2 앐 2.3a 257.6 앐 25.9a

1245 앐 106 555 앐 47 10.961 앐 0.93 1.119 앐 0.16 0.395 앐 0.03 9.842 앐 0.81 Ҁ12.3 앐 6.4 82.3 앐 7.3a 440.4 앐 82.6a,b

1442 앐 156 492 앐 54 9.753 앐 1.06 1.067 앐 0.12 0.417 앐 0.04 8.686 앐 0.95 Ҁ23.3 앐 7.4 108.9 앐 2.3b 559.8 앐 32.0b

All values are x៮ 앐 SEM; n ҃ 9. GE, gross energy. Means in the same row with different superscript letters are significantly different, P 쏝 0.05 (ANOVA followed by Tukey’s test). 2 Percentage difference ҃ (actual daily GE intake Ҁ estimated daily GE intake)/actual daily GE intake ҂ 100. 3 None of the differences between diets for intakes or fecal and urinary excretions were significant (P 쏜 0.05), except for the percentage difference between estimated and actual GE intakes, for which there was a significant (P 쏝 0.05) difference between the high-fiber diets (combined) and the refined diet. 1

1653

FOOD ENERGY FROM LOW-ENERGY-DENSITY DIETS

TABLE 5 Daily food total nitrogen (N) intakes and excretions, the composition of urinary N, the ratio of urinary energy to urinary N, and the ratio of urinary energy to food N for subjects in the 3 experimental diets groups1

Intake and excretion of N Daily intake (g/d) Fecal excretion (g/d) Urinary excretion (g/d) Urinary nitrogenous constituent (g/100 g total urinary N) Creatinine N Ammonia N Uric acid N Urea N Urinary energy:urinary N (kJ/g N) Urinary energy:food N (kJ/g N)

Refined diet

Fruit and vegetable diet

Cereal diet

14.74 앐 1.15a (10.85–20.22) 1.45 앐 0.31b (1.12–1.87) 11.48 앐 2.06a (8.26–14.33)

11.63 앐 1.00a (6.97–13.99) 2.18 앐 0.23a (1.11–3.03) 8.94 앐 0.55b (6.43–11.25)

12.45 앐 1.35a (9.29–18.67) 2.13 앐 0.22a (1.41–3.41) 11.11 앐 1.14a,b (6.96–17.63)

3.5 앐 0.14a (3.0–4.2) 3.8 앐 0.10a (3.2–4.3) 0.8 앐 0.09a (0.5–1.5) 87.4 앐 0.33a (86.0–89.4) 33.9 앐 0.33a (32.5–35.4) 26.9 앐 1.22a (19.8–30.8)

4.4 앐 0.16b (3.8–5.5) 3.9 앐 0.27a (2.9–5.2) 0.8 앐 0.07a (0.5–1.2) 85.4 앐 1.95a (80.3–96.1) 44.1 앐 0.90b (39.1–47.7) 34.9b 앐 1.81b (23.8–42.9)

3.7 앐 0.15a (2.9–4.4) 3.9 앐 0.26a (2.9–5.4) 1.3 앐 0.08b (0.9–1.6) 85.0 앐 0.75a (82.2–89.9) 37.6 앐 0.39c (35.2–39.1) 33.9 앐 1.20b (28.3–39.6)

All values are x៮ 앐 SE; range in parentheses. n ҃ 9. Means in the same row with different superscript letters are significantly different, P 쏝 0.05 (ANOVA followed by Tukey’s test). 1

modified Atwater factors, for which dietary carbohydrate is corrected for insoluble dietary fiber (FDA) or is determined as available carbohydrate (British), gave better agreement but practically important differences (4%) remained. In general, the empirically derived prediction equations that were tested did not lead to a higher level of accuracy of prediction compared with the factorial models. DISCUSSION

We sought to understand variation in energy availability and the predictive accuracy of Atwater factors (3) for a range of complex mixed diets in healthy persons. We focused on 3 aspects: analysis of dietary carbohydrate (available and unavailable), digestibility and metabolizability, and the importance of

the amount of food consumed. Our results confirm that the Atwater calculation system can overestimate the ME content of diets by up to 11%, a finding consistent with others (3, 7, 29, 30). A new observation from this study was that energy availability is particularly low in association with spontaneously reduced energy intakes from high-fiber diets. The spontaneous reduction in food energy may have been secondary to keeping a constant intake of dry mass per kg body weight (Table 4) or to physiologic cues other than energy. The considerable degree of variability across diets for macronutrient digestibility (Table 6) and urinary energy excretion per unit nitrogen (Table 5) highlights a shortcoming of calculation factors, such as the Atwater factors, and also underscores the potential to explore such variation in the development of weight-loss foods.

TABLE 6 Apparent digestibility of gross energy, crude protein, fat, total carbohydrate, and individual fatty acids for the 3 experimental diets1 Refined diet Digestibility of energy (%) Digestibility of protein (%) Digestibility of fat (%) Digestibility of total CHO (%) Digestibility of individual fatty acids (%) Saturated 10:0 12:0 14:0 16:0 18:02 Monounsaturated 16:12 18:12 Polyunsaturated 18:22 18:32

Fruit and vegetable diet

Cereal diet

93.5 앐 0.19 (92.4–94.1) 90.0 앐 0.44a (88.4–92.6) 95.7 앐 0.29a (94.3–97.3) 94.4 앐 0.39a (92.9–96.1)

90.0 앐 0.79 (85.4–93.4) 81.4 앐 0.92b (78.7–85.6) 87.0 앐 1.00b (81.7–91.6) 95.5 앐 0.66a (91.1–97.4)

89.1 앐 0.23b (87.9–90.0) 82.7 앐 0.73b (80.3–86.7) 95.1 앐 0.33a (93.4–96.9) 91.1 앐 0.26b (89.9–92.1)

99.9 앐 0.08a (99.3–100.0) 98.7 앐 0.75a (93.1–100.0) 99.0 앐 0.20a (98.3–100.0) 94.7 앐 0.58a (92.7–97.6) 86.3 앐 2.24a,b (75.2–96.8)

87.4 앐 1.46b (81.6–93.0) 80.7 앐 1.29b (76.4–86.7) 79.2 앐 1.16b (75.4–85.4) 87.8 앐 0.58b (85.6–91.2) 84.0 앐 2.16a (74.4–92.1)

98.2 앐 0.89a (94.1–100.0) 99.7 앐 0.18a (98.3–100.0) 99.4 앐 0.23a (98.2–100.0) 96.4 앐 0.11c (95.8–96.9) 91.1 앐 1.03b (85.5–96.2)

99.6 앐 0.42a (96.2–100.0) 94.3 앐 0.55a,b (92.5–96.9)

100.0 앐 0a (100.0–100.0) 93.1 앐 1.04a (87.1–96.4)

100.0 앐 0a (100.0–100.0) 95.9 앐 0.24b (94.7–97.1)

94.2 앐 1.29a (88.3–98.6) 93.2 앐 0.78a (88.7–96.5)

92.2 앐 2.22a (77.6–97.7) 96.8 앐 0.65b (93.2–98.8)

96.7 앐 0.50a (94.5–98.7) 97.7 앐 0.38b (95.8–99.3)

a

b

1 All values are x៮ 앐 SE; range in parentheses. n ҃ 9. CHO, carbohydrate. Means in the same row with different superscript letters are significantly different, P 쏝 0.05 (ANOVA followed by Tukey’s test). 2 Digestibilies for the 18-carbon series of fatty acids were compared between groups by using a 2-factor ANOVA (including factors for diet and degree of unsaturation). The digestibility of 18:0 was significantly lower (P 쏝 0.005) than the digestibilities of 18:1, 18:2, and 18:3 for all 3 diets. The digestibilities of monounsaturated fatty acids were compared between groups by using a 2-factor ANOVA (including factors for diet and chain length). The digestibility of 18:1 was significantly lower (P 쏝 0.001) than that of 16:1 with all 3 diets.

1654

ZOU ET AL

TABLE 7 Models used to predict the metabolizable energy values (kJ/g) of the diets1 Model and reference

Model

Factorial model Atwater, 1910 (21) British food tables, 1991 (22)2 FDA, 1993 (23) Atwater, modified, 1998 (7, 24) Empirical model Levy et al, 1958 (25) Miller and Payne, 1959 (9) Southgate, 1975 (26) Miller and Judd, 1984 (27) Livesey, 1991 (28)

MEAtwater ҃ 16.7P ѿ 37.6F ѿ 16.7C MEBritish ҃ 16.7P ѿ 37.6F ѿ 15.7Cm MEFDA ҃ 16.7P ѿ 37.6F ѿ 16.7 (C Ҁ isDF) MEAtwater modified ҃ 16.7P ѿ 37.6F ѿ 16.7 AC ѿ 8.4UC MELevy ҃ 0.976E Ҁ 33.3N Ҁ 250 MEM&P ҃ 0.95E Ҁ 31.4N MESouthgate ҃ 0.977E Ҁ 16.7UC Ҁ 27.6N MEM&J ҃ (0.95E Ҁ DF%) Ҁ 31.4N MELivesey ҃ 0.96E Ҁ 8.4U Ҁ 30N

1

ME (kJ/g), metabolizable energy; P (g), dietary protein; F (g), dietary fat; C (g), total carbohydrate by difference; Cm (g), determined available carbohydrate expressed as equivalent weight of monosaccharide; isDF (g), insoluble dietary fiber; AC, available carbohydrate (determined by difference); UC (g), unavailable carbohydrate determined as Southgate dietary fiber or total dietary fiber; E (kJ/g), gross energy of diet; N (g), dietary nitrogen; DF%, total dietary fiber as a percentage of the dry weight of the diet; U (g), unavailable complex carbohydrate; FDA, Food and Drug Administration. 2 For legislative purposes, the Atwater approach is used in the European Union; fiber has zero energy by default (C ҃ total carbohydrate minus total dietary fiber).

The calculation of food energy assumes accurate analysis of food components. Currently, the approaches used to calculate such differ between regions in relation to total, available, and unavailable carbohydrates. Total carbohydrate in foods and diets can be derived in at least 3 ways (Table 9). With both of the present high-fiber diets, direct measurements gave lower values than did indirect measures determined from differences. Should the direct measures be used, a discrepancy of 앒5% of the dry weight of food arises and would lead toward a lower food energy value of 앒5%. For this reason, care needs to be taken over the approach used to analyze carbohydrate.

Also important is the method for estimating unavailable carbohydrate. NSP, resistant starch, and the sum of these 2 components each underestimates total dietary fiber according to the methods used (Table 10). Total dietary fiber most closely represents the measure used in the derivation of food energy factors for unavailable carbohydrate (2, 28). The ratio of total dietary fiber to NSP in the refined, the fruit and vegetable, and the cereal diets was similar at about an expected value of 1.3 (Table 10). This is of parenthetical interest because the lack of variability suggests that neither NSP nor total dietary fiber would have demonstrable superiority over the other in indicating a healthful diet, which is of particular interest at

TABLE 8 Determined metabolizable energy (ME) values, calculated ME values based on the application of different models, and differences between determined ME and calculated ME for the 3 experimental diets1 Refined diet Value

Fruit and vegetable diet

Difference2

Value

%

Difference2

Cereal diet Value

%

Difference2 %

3

Determined (kJ/g DM diet) GE ME Calculated with factorial models (kJ/g DM diet)4 MEAtwater MEFDA MEBritish MEAtwater modified Calculated with empirical models (kJ/g DM diet) MELevy MEM&P MEM&J MESouthgate MELivesey

23.397 앐 0.03a 21.111 앐 0.06a

— —

19.741 앐 0.02b 17.043 앐 0.14b

— —

19.840 앐 0.01c 16.814 앐 0.05b

— —

21.884 21.134 21.057 21.520

3.74 0.1 Ҁ0.3 1.9

18.873 17.738 16.809 18.245

10.85 4.15 Ҁ1.3 7.15

18.274 16.616 16.122 17.450

8.35 Ҁ1.24 Ҁ4.15 3.85

21.398 21.344 21.295 21.486 21.211

1.44 1.14 0.96 1.85 0.5

18.084 18.096 18.018 17.761 17.661

6.25 6.25 5.85 4.35 3.75

17.966 18.052 18.030 17.378 17.445

6.95 7.45 7.25 3.45 3.85

Means in the same row with different superscript letters are significantly different, P 쏝 0.001 (ANOVA followed by Tukey’s test). Calculated as 100 ҂ (MEcalculated Ҁ MEdetermined)/MEdetermined. 3 Values are x៮ 앐 SEM; n ҃ 9. 4 P 쏝 0.01 (Student’s t test). 5 P 쏝 0.001 (Student’s t test). 6 P 쏝 0.05 (Student’s t test). 1

2

1655

FOOD ENERGY FROM LOW-ENERGY-DENSITY DIETS TABLE 9 Three different approaches to calculating total carbohydrate (g/100 g dry food)1 Method of analysis A) Total CHO by difference B) Available CHO2 ѿ total dietary fiber C) Available CHO ѿ NSP ѿ RS Difference A minus B Difference A minus C 2

Fruit and vegetable diet

Cereal diet

51.6 50.8 49.8 1.6 1.7

70.5 66.1 64.7 6.2 5.8

69.1 65.2 63.6 5.7 5.5

Refer to Table 2 for individual components. CHO, carbohydrate; NSP, nonstarch polysaccharide; RS, resistant starch. Determined directly and expressed as the weight of the carbohydrate (18).

the present time to WHO/FAO/Codex deliberations on the definition of dietary fiber. The calculation of food energy values by factorial approaches, such as the Atwater system, assumes also that the gross energy in fat, carbohydrate, and protein is reasonably well represented by those gross energies used in the derivation of Atwater factors (5.65 kcal/g protein, 9.3 kcal/g fat, and 4.1 kcal/g total carbohydrate; 1 kcal ҃ 4.184 kJ). With the use of these factors, the calculated gross energy contents of the 3 diets were 23.747, 20.303, and 19.838 kJ/g for the refined, fruit and vegetable, and cereal diets, respectively, which differ from the determined values for gross energy in the diets by 1.5%, 2.8%, and 0.0% respectively. These differences appear to be within experimental error, although the overestimation for the fruit and vegetable diet may have resulted because of the lower gross energy of the protein in the fruit and vegetables (with a higher proportion of free amino acids of low energy density), the higher proportion of sugars with a lower gross energy value than starch (Table 2), the higher proportion of organic acids (of lower energy density than carbohydrate), and the higher uronic acid content of fiber (Table 3) and thus the lower gross energy content of the fiber (앒16.5 kJ/g on average for the fruit and vegetable fiber diet compared with 앒17.5 kJ/g on average for the cereal fiber diet) (31). TheMEvaluesforthehigh-fiberdietswerenotcalculatedaccurately by any of the various systems of food energy assessment (Table 8). At firstthisseemssurprising;however,acombinationofthepresentresults with those of Brown et al (7) provides useful insight (Figure 1). Brown et al used a modified Atwater approach in which total carbohydrate is separated into its component parts of available carbohydrate and unavailable carbohydrate, with assignment of separate energy values. The application of this system to food items provides results that are little differentfromthoseobtainedwiththeuseofthefood-specificsystemof

TABLE 10 Contrasting approaches to the assessment of unavailable carbohydrate (g/100 g dry food)1

Component A) TDF B) NSP C) RS D) Sum NSP ѿ RS (B plus C) A minus D A:B 1

Refined diet

Fruit and vegetable diet

Cereal diet

4.9 3.51 0.45 3.96 0.94 1.396

7.89 5.64 0.81 6.45 1.44 1.399

10.02 7.82 0.64 8.46 1.56 1.281

TDF, total dietary fiber; NSP, nonstarch polysaccharide; RS, resistant starch.

food energy assessment (3, 25). The results of the present study combined with those of Brown et al (7) (Figure 1) indicate that the modified Atwater approach may be suitable for application with high-fiber diets, but not when food intake is reduced. When food intake decreases, as it may with consumption of high-fiber diets, fecal excretion exceeds that predicted by the Atwater modified system. This also indicates a need to maintaintheuseofstandardizedmethodologyduringfoodenergyevaluations, for which zero nitrogen balance and zero energy balance have been suggested (2, 6). It also could explain some of the variability in results with unavailable carbohydrates found in the literature (reviewed by Livesey; 30). In the present study, gross energy was expressed as kJ/kg body weight (Figure 1). It remains to be seen, however, whether body weight, the departure from maintenance energy intakes, or some other factor was the key determinant of the variability; whether energy intakes differ between men and women; and the extent to which the slope (Figure 1) might vary with the amount and type of fiber. Importantly, matching energy requirements and energy intake should take account of this response. It will also be important to ascertain whether the lower than expected energy availability from the high-fiber diets takendeliberatelyorspontaneouslyatsubmaintenancelevelspersistsin the long term.

Gross energy intake (kJ/kg body weight) 0 Reduced energy metabolizability compared with prediction (%)

1

Refined diet

50

100

150

200

2 Brown et al 1998

0 -2 Fruit and Vegetable diet (Present study)

-4

Cereal diet (Present study)

-6 -8 -10

Brown et al 1998

FIGURE 1. Mean (앐SEM) reduction in available metabolizable energy associated with low food intakes from high-fiber diets calculated by using a modified Atwater system (see Table 7) with correction for errors in predicted gross energy so as to view the change in metabolizable energy due to biological processes alone:

100/MEdetermined ⫻ [(MEcalculated ⫺ MEdetermined) ⫺ (GEcalculated ⫺ GEdetermined)]

(3)

Values shown are for high-fiber maintenance and submaintenance diets from Brown et al (7) combined with the present results with the high-fiber fruit and vegetable diet and the cereal fiber diet.

1656

ZOU ET AL

A further issue for matching food energy with energy requirements concerns differences in thermogenesis between different food components (2, 32). In particular, protein and fiber are more thermogenic than is available carbohydrate and this contributes an energy loss equivalent to 20% of the ME in protein and 25% of the ME in fiber (above that for available carbohydrate) (2). The sum of absolute errors in the ME system of food energy on such account, for the present study, amounts to 2.4% of ME corrected for thermal losses (net ME). However, in the presentstudytherewasadecreaseinproteinintakesimultaneouslywith an increase in fiber intake, so that we would not expect important differences between the refined and high-fiber diets in the present cases. Indeed, we estimate a difference of only 0.2% of net ME. This differs from many specialized foods intended for obesity control, which use a high proportion of both fiber and protein (2). In developing and evaluating specialized low-calorie foods, for which very small differences in ME may be of practical significance, the use of Atwater or modified Atwater factors is likely to be too inaccurate. Fiber preparations of very high fermentability, very low fermentability, or particularly viscous fiber and gel-forming fibers are expected to differ from the general trend for foodstuffs, which have a narrower range of fermentability, viscosity, and gelling characteristics. For an accurate evaluation of the “available” energy content, especially of novel formulated weight-loss foods, it would be more accurate to determine digested and fermented nutrients (at both the ileal and fecal levels), to predict potential ATP production using stoichiometric relations, to predict urinary energy losses, and thus derive a net ME (6) value. In conclusion, we confirm lower energy availability from high fiber diets. We also show modest (spontaneous) lower food intake is sufficient to make food energy assessment systems inadequate as predictors of food energy value in high fiber diets, thus extending similar prior findings made with sub maintenance high fiber diets. In this study we found that the high-fiber fruit and vegetable diets and the cereal diets were not distinguishable. However, we did not invalidate the modified Atwater factor system for comparison of ME values of foods in general when consumed at zero nitrogen balance and zero energy balance (meaning energy intake minus energy expenditure). The system does not account for differences in thermal energy losses associated with the metabolism of the macronutrients. We thank Mainland Products Ltd, New Zealand, for providing foods and materials and the subjects who participated in the study for their loyal and devoted cooperation. The authors’ responsibilities were as follows—PJM: designed the study, oversaw the conduct of the study, and led the drafting of the manuscript; MLZ: oversaw the recruitment of subjects and led the conduct of the study, data analysis, and manuscript writing; AA: assisted with the data analysis and contributed to the writing of the manuscript; GL: assisted with data analysis and contributed to the writing of the manuscript. None of the authors had a personal or financial conflict of interest.

REFERENCES 1. FAO/WHO/UNU. Human energy requirements. Geneva, Switzerland: WHO, 2004. 2. FAO. Food energy—methods of analysis and conversion factors.. Rome, Italy: Food and Agriculture Organization, 2003. (FAO Food and Nutrition Paper no. 77.) 3. Merrill AL, Watt BK. Energy value of foods: basis and derivation. Agriculture handbook 74. Washington, DC: US Department of Agriculture, Agricultural Research Service, 1973. 4. CAC. Guidelines on nutrition labeling. Rome, Italy: FAO/WHO, 1993. 5. USDA. Nutrient database for standard reference, release 12. Washington, DC: US Department of Agriculture, Agricultural Research Service, 1998.

6. Livesey G. A perspective on food energy standards for nutrition labelling. Br J Nutr 2001a;85:271– 87. 7. Brown J, Livesey G, Roe M, et al. Metabolizable energy of high nonstarch polysaccharide-maintenance and weight-reducing diets in men: experimental appraisal of assessment systems. J Nutr 1998;128:986 –95. 8. AOAC. Official methods of analysis. Washington, DC: Association of Official Analytical Chemists, 2000. 9. Miller DS, Payne PR. A ballistic bomb calorimeter. Br J Nutr 1959;13:501–8. 10. Bellomonte G, Costantini A, Giammarioli S. Comparison of modified automatic Dumas method and the traditional Kjeldahl method for nitrogen determination in infant food. J Assoc Off Anal Chem 1987;70:227–9. 11. Tiffany TO, Jansen JM, Burtis CA, Overton JB, Scott CD. Enzymatic kinetic rate and endpoint analyses of substrate, by use of a GeMSAEC Fast Analyzer. Clin Chem 1972;18:829 – 40. 12. Fossati P, Prencipe L, Berti G. Use of 3,5-dichloro-2-hydroxybenzenesulfonic acid-4-aminophenazone chromogenic system in direct enzymic assay of uricacid in serum and urine. Clin Chem 1980;26:227–31. 13. Larsen K. Creatinine assay by a reaction-kinetic principle. Clin Chim Acta 1972;41:209 –17. 14. Sukhija PS, Palmquist DL. Rapid method for determination of total fatty-acid content and composition of feedstuffs and feces. J Agric Food Chem 1988;36:1202– 6. 15. Henneberg W, Stohmann F. Beitrage zur Begrundung einer rationellen Futterung der Wiederkauer I & II. (Contributions to establish a rational feeding for ruminant I & II.) Braunschweig, 1860 (in German). 16. Englyst HN, Cummings JH. Improved method for measurement of dietary fiber as non-starch polysaccharides in plant foods. J Assoc Off Anal Chem 1988;71:808 –14. 17. Englyst HN, Cummings JH. Simplified method for the measurement of total non-starch polysaccharides by gas-liquid-chromatography of constituent sugars as alditol acetates. Analyst 1984;109:937– 42. 18. Athar N, Spriggs TW, Liu P. The concise New Zealand food composition 4th tables. 4th ed. Wellington, New Zealand: New Zealand Institute for Crop & Food Research Ltd, Palmerston North, New Zealand and Ministry of Health, 1999. 19. McCleary BV, McNally M, Rossiter P, et al. Measurement of resistant starch by enzymatic digestion in starch and selected plant materials: collaborative study. J Assoc Off Anal Chem 2002;85:1103–11. 20. Zou ML. Variation in the fecal digestibility and urinary excretion of energy in three diets for humans. MSc thesis. Palmerston North, New Zealand, Massey University, 2007. 21. Atwater WO. Principles of nutrition and nutritive values of food. United States Farmers’ Bulletin 1910;142. 22. Holland B, Welch AA, Unwin ID, Buss DH, Paul AA, Southgate DAT. McCance and Widdowson’s the compostion of foods. 5th ed. Cambridge, United Kingdom: Royal Society of Chemistry, Ministry of Agriculture, Fisheries and Food, 1991. 23. FDA. Nutritional labeling of food regulation. Title 21 of the Code of Federal Regulations (21F). Fed Regist 1993;53:2175. 24. Livesey G. Attributes of foods not diets will enable consumer choice. In: Palou A, Bornet M, Serra F, eds. Study on ‘Obesity and functional foods in Europe.’ Brussels, Belgium: European Commission, 2001b:366 –75. 25. Levy LM, Bernstein LM, Grossman MI. The calorie content of urine of human beings and the estimation of the metabolizable energy of human foodstuffs. Denver, CO: US Army Medical Research and Nutrition Laboratory, 1958. (Report 26.) 26. Southgate DAT. Fiber and other unavailable carbohydrates and energy effects in the diet. Acton, MA: Publishing Science Group Inc, 1975. 27. Miller DS, Judd PA. The metabolizable energy value of foods. J Sci Food Agric 1984;35:111– 6. 28. Livesey G. Calculating the energy values of foods—towards new empirical formulas based on diets with varied intakes of unavailable complex carbohydrates. Eur J Clin Nutr 1991;45:1–12. 29. Southgate DAT, Durnin JVGA. Calorie conversion factors—an experimental reassessment of factors used in calculation of energy value of human diets. Br J Nutr 1970;24:517–35. 30. Livesey G. Energy values of unavailable carbohydrate and diets—an inquiry and analysis. Am J Clin Nutr 1990;51:617–37. 31. Livesey G. Fiber as energy in man. In: Kritchevsky D, Bonefield C, eds. Dietary fiber in health and disease. St Paul, MN: Eagan Press, 1995:46 –57. 32. Livesey G. Thermogenesis associated with fermentable carbohydrate in humans, validity of indirect calorimetry, and implications of dietary thermogenesis for energy requirements, food energy and body weight. Int J Obes Relat Metab Disord 2002;26:1553– 69.

Dose response to vitamin D supplementation among postmenopausal African American women1–3 Sonia A Talwar, John F Aloia, Simcha Pollack, and James K Yeh ABSTRACT Background: Reports on the dose response to vitamin D are conflicting, and most data were derived from white men and women. Objective: The objective was to determine the response of serum 25-hydroxyvitamin D [25(OH)D] to oral vitamin D3 supplementation in an African American population. Design: Healthy black postmenopausal women (n ҃ 208) participated in a vitamin D3 supplementation trial for a period of 3 y. Analyses were done in the vitamin D supplementation arm (n ҃ 104) to quantify the response in serum 25-hydroxyvitamin D concentrations at a steady state vitamin D input. The participants received 20 ␮g/d (800 IU) oral vitamin D3 for the initial 2 y and 50 ␮g/d (2000 IU) for the third year. Results: Supplementation with 20 ␮g/d (800 IU/d) vitamin D3 raised the mean serum 25(OH)D concentration from a baseline of 46.9 앐 20.6 nmol/L to 71.4 앐 21.5 nmol/L at 3 mo. The mean (앐SD) concentration of serum 25(OH)D was 87.3 앐 27.0 nmol/L 3 mo after supplementation increased to 50 ␮g/d (2000 IU/d). All participants achieved a serum 25(OH)D concentration 쏜35 nmol/L, 95% achieved a concentration 쏜50 nmol/L, but only 60% achieved a concentration 쏜75 nmol/L. All patients had concentrations 쏝153 nmol/L. On the basis of our findings, an algorithm for prescribing vitamin D so that patients reach optimal serum concentrations was developed. The algorithm suggests a dose of 70 ␮g (2800 IU/d) for those with a concentration 쏜45 nmol/L and a dose of 100 ␮g (4000 IU/d) for those with a concentration 쏝45 nmol/L. Conclusions: Supplementation with 50 ␮g/d (2000 IU/d) oral vitamin D3 is sufficient to raise serum 25-hydroxyvitamin D concentrations to 쏜50 nmol/L in almost all postmenopausal African American women. However, higher doses were needed to achieve concentrations 쏜75 nmol/L in many women in this population. Am J Clin Nutr 2007;86:1657– 62.

may be just as important for other nonskeletal effects, such as the improvement of the immune system and the prevention of certain cancers (5). Up to 42% of African American women and 4.2% of white women of childbearing age have serum 25(OH)D concentrations 쏝62.5 nmol/L during the summer (6). The prevalence is expected to be much higher during the winter. Experts in the field are now disputing the vitamin D requirements set forth as adequate intakes in 1997 by the Panel on Calcium and Related Nutrients. There is an emerging consensus that 25(OH)D concentrations 쏜75 nmol/L may be optimal for bone health and extraskeletal effects (7–12). Heaney (13) recently described that an oral intake of 욷55 ␮g/d (2200 IU/d) may be required in addition to the prevailing intake of vitamin D to raise 25(OH)D concentrations to near 80 nmol/L or higher. Blacks produce less vitamin D3 than do whites in response to usual levels of sun exposure and have lower serum 25(OH)D concentrations in winter and summer (14, 15). Weaver and Fleet (16) estimated that blacks need 46 – 62 ␮g/d of vitamin D3 supplements. However, this assumption is based on a solitary study performed in white adults (17). Thus far, there is a lack of sufficient evidence to make ethnically specific recommendations. We analyzed the dose response of vitamin D supplementation in a cohort of postmenopausal African American women receiving daily vitamin D3 supplementation for a period of 3 y in a double-blind, placebo-controlled longitudinal trial conducted at our center from 1998 to 2004 (18). The aim of this report was to quantify the response of serum 25-hydroxyvitamin D concentrations to a steady state vitamin D input.

KEY WORDS Ethnicity, vitamin D, 25-hydroxyvitamin D, osteoporosis, vitamin D deficiency, African Americans

Participants

INTRODUCTION

Clinicians often measure serum 25-hydroxyvitamin D [25(OH)D] to determine vitamin D status. Although concentrations 쏝20 nmol/L are well known to cause clinical osteomalacia and rickets, concentrations between 20 and 75 nmol/L (vitamin D insufficiency) have more recently been suggested to have an adverse influence on the skeleton (1, 2). Vitamin D insufficiency in the elderly is associated with low bone mass due to secondary hyperparathyroidism and, as a result, a higher incidence of fractures (2– 4). It has also been appreciated that sufficient vitamin D

SUBJECTS AND METHODS

Two hundred eight healthy postmenopausal African American women not receiving hormone replacement therapy were recruited from the Long Island community. One hundred four participants were randomly assigned to receive daily vitamin D supplements and 104 received placebo. All of the participants 1 From the Bone Mineral Research Center, Winthrop University Hospital, Mineola, NY. 2 Supported by the National Institute of Aging (RO1 AG15325), NIH. 3 Reprints not available. Address correspondence to JF Aloia, 222 Station Plaza North, Suite 510, Mineola, NY 11501. E-mail: [email protected]. Received October 26, 2006. Accepted for publication July 20, 2007.

Am J Clin Nutr 2007;86:1657– 62. Printed in USA. © 2007 American Society for Nutrition

1657

1658

TALWAR ET AL

provided written informed consent, and the trial was approved by the Institutional Review Board of Winthrop University Hospital. All of the procedures followed were in accordance with the ethical standards of our Institutional Review Board on human experimentation in accordance with the Helsinki Declaration of 1975 as revised in 1983. African American ancestry of the participants was assessed by self-declaration that both parents and 욷3 out of 4 grandparents were African American. Exclusion criteria included previous treatment with bone active agents and any medication or illness that affects skeletal metabolism. Postmenopausal status was confirmed on the basis of serum follicle-stimulating hormone concentrations 쏜23 mIU/L. A medical history and physical examination were conducted by a physician on site in all subjects. Exclusion criteria included previous treatment with bisphosphonates or fluoride; use of estrogen, calcitonin, glucocorticoids, androgens, phosphate, anabolic steroids, or 쏜400 IU/d vitamin D 6 mo before entry; history of previous hip fracture; uncontrolled diabetes, anemia, or thyroid disease; history of current liver, renal, neurologic, or malignant disease; malabsorption or alcoholism; history of hypercalciuria, nephrolithiasis, or active sarcoidosis; smoking 쏜10 cigarettes a day; unexplained weight loss; use of medications known to interfere with calcium or vitamin D absorption or metabolism, such as anticonvulsants; severe osteoarthritis or scoliosis that would interfere with bone density assessment of the spine or hip; and participation in weight training or elite athletic training. Study design The participants were randomly assigned with the use of a computer-generated sequence to receive either 20 ␮g/d (800 IU/d) oral vitamin D3 or a matched placebo. At the completion of 24 mo of supplementation, the dose of vitamin D3 was raised to 50 ␮g/d (2000 IU/d) in the vitamin D group, and the study continued for an additional year. Calcium intake was assessed with a food-frequency questionnaire at each visit, and supplements were given to both active and placebo groups to ensure a total daily intake of 1200 –1500 mg Ca. Vitamin D3 (20- and 50-␮g capsules) and matched placebo capsules were custommanufactured for the study (Tishcon Corp, Westbury, NY) and were acquired in 3 separate shipments to avoid a spontaneous decline in potency. The content of vitamin D was also assessed and confirmed by an independent laboratory (Vitamin D, Skin, and Bone Research Laboratory, Department of Medicine, Boston University School of Medicine, Boston, MA). The calcium supplements were provided as calcium carbonate. Outcome variables A fasting blood sample was collected for analysis of serum 25(OH)D at baseline and at 3, 6, 12, 18, 24, 27, 30, and 36 mo. These samples were collected throughout the year but during the same month at the annual visits to avoid seasonal effects. Serum 25(OH)D was measured by radioimmunoassay (RIA) with the use of a kit manufactured by DiaSorin Inc (Stillwater, MN) (19). The intraassay CV was 4.1%, and the interassay CV was 7.0%. Our laboratory participates in DEQAS, an international quality assurance program to ensure accuracy in the measurement of serum 25(OH)D (20). Other laboratory measurements made but not analyzed in this study included serum chemistries, calcium, serum parathyroid hormone, 1,25-dihydroxyvitamin D, osteocalcin, and cross-laps (18). Body fat was measured every 6 mo by using dual-energy X-ray absorptiometry.

Statistical analysis Multiple linear regression was used to model vitamin D response as a function of predictor variables, including dose, baseline values, season, body mass index, and percentage body fat. Slope at a specific time point was defined as the change in serum 25(OH)D from baseline divided by the dose assigned during the preceding 3 mo. Within-subject change in slope was analyzed with the paired t test. Pearson correlation was used to quantify the linear association between variables. Continuous change across time between the active and placebo groups were analyzed by using a mixed-model analysis of variance framework implemented in PROC MIXED (version 9.1; SAS Institute, Cary, NC). The analysis of correlated data arising from repeated measurements used Generalized Estimating Equations (GEEs) implemented in the SAS Procedure GENMOD. A 2-tailed P value 쏝0.05 was deemed statistically significant. Results are expressed as means 앐 SD. Analyses were done both with all available data and with the use of only subjects with complete data. Because no differences in the results were found, only the results with the use of all available data are reported (ie, an intent-to-treat analysis). Similarly, nonparametric analyses and data transformation were applied, but, because they did not lead to different results, only the parametric analyses of the raw data are reported. Optimal dosing algorithm development The results from multiple regressions and other analyses were used to identify variables to be included in an empiric algorithm for prescribing vitamin D. Variables that were considered for inclusion were dose, age, amount of body fat, basal serum 25(OH)D concentration, and time of year that a patient’s vitamin D was measured. Because these data included only black women, race and sex were not included. Only dose, basal concentration, and season were found to predict vitamin D response. Data found by Vieth (21) suggest that the 25(OH)D response to each 1 ␮g vitamin D/d is approximately constant for doses 쏜35 ␮g/d. We therefore approached the problem of finding the optimal vitamin D dose by multiplying a patient’s observed slope on 50 ␮g/d by a wide range of possible doses. This projected response is then added to the basal value to obtain a projected endpoint on that dose. (We did not adjust for season because the projection was made for the same time period during which the slope was measured; thus, the slope already includes the effects of season.) We had the computer search through a wide range of possible doses from 35 to 150 ␮g/d and calculated the projected vitamin D levels for each patient if everyone would be taking the same dose. We defined an optimal dose as one at which the concentrations of all patients are projected to remain 쏝250 nmol/L, whereas 욷90% exceeded 75 nmol/L. A dose of 95 ␮g/d satisfied those criteria, but some patients were projected to be too close to 250 nmol/L. For safety purposes we tried another class of dosing rule that was based on the reasoning that not all patients need take the same dose. Patients whose basal value was below a to-be-determined threshold would be given a higher dose, whereas those with a basal value above that threshold would be given a lower dose, ie, after a single basal 25(OH)D vitamin D concentration measurement is made (Dobs), a high dose (DoseH)is prescribed if the Dobs is below a to-be-determined threshold (T) and a lower dose (DoseL)if the Dobs is above it. Potentially, these 3 variables (T,

1659

DOSE RESPONSE TO VITAMIN D TABLE 1 Demographic characteristics, bone mineral density, and laboratory values at baseline1 Characteristic Age (y) Height (cm) Weight (kg) BMI (kg/m2) Smoking (%) Current user Ever Dietary vitamin D intake (␮g/d) Calcium intake (mg/d) 25(OH)D (nmol/L) 1,25(OH)2D (pmol/L) PTH (pg/mL)

Placebo group (n ҃ 104)

Vitamin D group (n ҃ 104)

61.2 앐 6.32 161.4 앐 6.1 79.2 앐 12.6 30 앐 4

59.9 앐 6.2 162.7 앐 6.6 78.0 앐 13.6 29 앐 4

7 40 4.6 앐 4.2 756 앐 541 43.2 앐 16.8 118.8 앐 39.2 42.4 앐 18.4

7 35 4.6 앐 4.8 762 앐 623 46.9 앐 20.6 121.8 앐 39.6 44.2 앐 19.3

1 The data are from reference 18. 25(OH)D, 25-hydroxyvitamin D; 1,25(OH)2D, 1,25-dihydroxyvitamin D; PTH, parathyroid hormone. There were no significant differences between groups by independent t test. 2 x៮ 앐 SD (all such values).

DoseL, and DoseH) could vary as a function of a patient’s covariates (characteristics), such as age or weight, but we did not find this to be the case. The combinations of T, DoseL, and DoseH that satisfied the criteria that all patients are projected to have concentrations 쏝250 nmol/L (for safety) and that 욷90% of the population would achieve 욷75 nmol/L (for efficacy) was found by an SAS computer program by searching all possible combinations of T, DoseL, and DoseH. The nature of this algorithm does not allow for the estimation of parameters and their CIs. RESULTS

Baseline characteristics of the vitamin D group are shown in Table 1. Thirteen percent of the women were taking calcium and vitamin supplements before entry into the study.

FIGURE 1. Mean (앐SD) changes from baseline in 25-hydroxyvitamin D [25(OH)D] concentrations in the vitamin D (solid line) group and the placebo group (dashed line) throughout the 36-mo study period. A significant groupby-time interaction was observed (P 쏝 0.0001).

In the vitamin D group, the concentration of serum 25(OH)D increased significantly (P 쏝 0.0001) over the first 3 mo at the dose of 20 ␮g vitamin D3/d (Table 2 and Figure 1). The placebo group did not change its mean vitamin D concentration from baseline (time-by-group interaction: P 쏝 0.0001). Over this same period, parathyroid hormone decreased significantly (P 쏝 0.0001; Table 2). The response when assessed as the change in serum 25(OH)D concentrations per 1 ␮g vitamin D3 supplemented is equivalent to a slope of 1.1 앐 0.9 nmol 䡠 LҀ1 䡠 ␮gҀ1. When the dose of vitamin D3 was raised from 20 to 50 ␮g/d during the final year of the study, the mean serum 25(OH)D concentration achieved after 3 mo at the higher dose was 87.3 앐 27.0 nmol/L. The mean 25(OH)D concentration, despite a 250% increase in dose, increased by 17.2 앐 22.2 nmol/L (22% increase). The slope was 0.76 앐 0.53 nmol 䡠 LҀ1 䡠 ␮gҀ1 for the higher dose. The maximum increase in serum 25(OH)D was seen at 27 mo (3 mo after higher dose began); 60% of the participants achieved concentrations 쏜75 nmol/L. At 36 mo, the mean serum

TABLE 2 Laboratory values at baseline and at 3, 24, and 27 mo in the vitamin D and placebo groups by intake of vitamin D3 (n ҃ 104)1 Measure and group

Baseline (0 ␮g/d)

3 mo (20 ␮g/d)

24 mo (50 ␮g/d)

27 mo (50 ␮g/d)

92.0 앐 66.03 86.3 앐 49.7

108.9 앐 73.3 118.1 앐 78.8

107.4 앐 66.6 118.8 앐 69.3

110.8 앐 64.0 108.2 앐 64.3

9.0 앐 0.5 9.0 앐 0.6

9.1 앐 0.5 9.2 앐 0.6

9.3 앐 0.3 9.3 앐 0.6

9.3 앐 0.3 9.3 앐 .35

43.2 앐 16.8 46.9 앐 20.6

39.1 앐 18.2 71.4 앐 21.5

41.6 앐 18.1 65.9 앐 22.4

45.2 앐 21.4 87.2 앐 27.0

119.2 앐 39.0 121.3 앐 39.2

97.2 앐 34.8 128.0 앐 50.7

87.4 앐 28.8 107.6 앐 33.6

104.3 앐 32.4 128.2 앐 43.1

40.7 앐 19.0 42.3 앐 20.1

34.4 앐 15.5 33.0 앐 14.4

38.2 앐 15.3 39.3 앐 17.7

35.5 앐 15.9 36.3 앐 15.0

2

24-h urine excretion Placebo Vitamin D Serum calcium (mg/dL)2 Placebo Vitamin D 25(OH)D (nmol/L)4 Placebo Vitamin D 1,25(OH)2D (pmol/L)2,4 Placebo Vitamin D PTH (pg/mL)2 Placebo Vitamin D 1

Differences across time within and between groups were tested with a mixed-model ANOVA. No significant differences were observed at baseline. The time trend was significant, P 쏝 0.0001. 3 x៮ 앐 SD (all such values). 4 The group-by-time interaction was significant, P 쏝 0.0001. 2

1660

TALWAR ET AL

25(OH)D concentration waned to 73.9 앐 26.8 nmol/L, and only 50% of the participants achieved serum 25(OH)D concentrations 쏜75 nmol/L at the end of the study. In a multiple regression analysis, neither age, weight, body mass index, percentage body fat nor grams of body fat significantly influenced the response to vitamin D3 supplementation. The 3-mo change per 1 ␮g vitamin D, ie, the response slope, was inversely dependent on the basal 25(OH)D concentration. The slope was also inversely dependent on the dose used for supplementation. The slope was not constant across the 2 doses: the higher dose of 50 ␮g produced a smaller change per 1 ␮g vitamin D in serum 25(OH)D (slope ҃ 0.76 앐 0.5) than did the lower dose of 20 ␮g (slope ҃ 1.1 앐 0.9) (paired t (53) ҃ 2.8, P ҃ 0.007). The lower the baseline measure, the greater the vitamin D response. This was indicated by the negative correlation between baseline vitamin D and the change in vitamin D from baseline to specific time points. For example, among the active patients, the correlation between baseline and change was Ҁ0.38 (P ҃ 0.0005) at 3 mo and was Ҁ0.35 (P ҃ 0.007) at 27 mo. A very similar pattern of equally large negative correlation coefficients was observed among the placebo patients as well: the correlation between baseline and change at 3 mo was Ҁ0.26 (P ҃ 0.017) and at 27 mo was Ҁ0.42 (P ҃ 0.0009). This correlation was evident even when both measurements were taken in the same season 1 y apart, ie, at 12, 24, or 36 mo, which ruled out seasonal variation as an explanation for this association. Effects of season We noted empirically a pronounced peak in our serum 25(OH)D data during the months of June to September. At baseline, the pre- and postsummer periods were not statistically different with respect to the mean vitamin D concentration (37.3 앐 18.0 and 42.9 앐 16.8 nmol/L, respectively; P ҃ 0.12), but were both statistically lower than during the summer period (50.5 앐 18.6 nmol/L). Pooling the 2 nonsummer periods for comparison with summer resulted in a mean difference of 10.4 nmol/L (the SE for the linear contrast was 2.5, t ҃ 4.1, P ҃ 0.0001). Age- and several weight-related variables were not seen to be significant factors associated with the change in vitamin D. Only basal vitamin D concentration and season are included as factors in the formula for prescribing vitamin D.

Daily calcium intake including supplements was 1349 앐 204 mg/d. Adverse events There were 8 serious adverse events in this subset of subjects, none of which was considered to be related to the study medication. Specific study-related adverse events included 6 isolated incidents of mild hypercalcemia in this group. The hypercalcemia resolved on repeat fasting sampling. Similarly, isolated episodes of elevated 24-h urinary calcium excretion (defined as 쏜5 mg 䡠 kgҀ1 䡠 dҀ1) were observed among 3 participants. Calcium supplements had to be discontinued in one participant because of persistent hypercalciuria, which resolved the abnormality. In the other 2 participants, the condition resolved spontaneously on repeat analysis of 24-h urine samples with no alteration in study supplements. Overall, there was a slight increase in serum calcium and urinary calcium excretion over 3 y with vitamin D supplementation. However, this increase was similar to the increase seen with calcium supplementation alone (placebo arm). In addition, the concentrations remained within the reference range for healthy adults in all participants. There were no episodes of nephrolithiasis. There was a slight increase in serum creatinine in both groups over 3 y that also remained within the reference range for healthy adults in all participants. Twenty-four– hour urinary calcium adjusted for body weight (mg 䡠 dҀ1 䡠 kgҀ1) was not statistically different between the vitamin D and placebo groups during the course of the study (Figure 2). Very few patients ever exceeded 5 mg 䡠 dҀ1 䡠 kgҀ1 and, when retested, were found to be below the threshold of 5, except in one instance. Although vitamin D did not seem to adversely affect the calcium economy, we did note a statistically positive correlation between serum calcium (measured across all active patients after baseline, when calcium supplementation began) and 25(OH)D (r ҃ 0.22, n ҃ 626 observations). Because the observations were not independent, a GEE analysis controlling for multiple observations per individual resulted in a P value 쏝0.0001 for the association between serum calcium and 25(OH)D. Correlations with a similar magnitude were noted at individual time points.

Results of computer search of optimal T, DoseL, and DoseH Computer analysis of our data arrived at only one solution that satisfied our stated criteria. This solution suggests a dose of 70 ␮g for those with a concentration 쏜45 nmol/L and a dose of 100 ␮g for those with a concentration 쏝45 nmol/L. In an improvement over the dosing rule of “all patients take 95 ␮g/d” we found that the rule incorporating a threshold resulted in all patients projected to remain 쏝220 nmol/L after 3 mo on the prescribed dose and 90% projected to achieve 욷75 nmol/L. Everyone was projected to reach 59 nmol/L. To guarantee that 97% of the subjects would have a concentration 쏜75 nmol/L, a larger dose must be prescribed to result in 4 participants projected to have values of 25(OH)D 쏜250 nmol/L. Adherence Vitamin D pill compliance after the randomization visit was 87%; 앒96% of the subsequent visits were kept by our patients.

FIGURE 2. Mean (앐SD) urinary calcium excretion by weight in the vitamin D group (solid line) and the placebo group (dashed line) throughout the 36-mo study period. Calcium excretion remained stable throughout the study. No significant group-by-time interaction was observed.

DOSE RESPONSE TO VITAMIN D

The mean serum calcium concentration among those in the highest quartile of serum vitamin D was 0.25 mg/dL higher than that among those in the lowest quartile (P 쏝 0.0001). This single change takes on more significance when it is contrasted with the small amount of overall variability in serum calcium. Because the SD across the study was 0.48 mg/dL, vitamin D shifts the population more than half an SD. DISCUSSION

This study is the first report of dose responses to oral vitamin D3 supplementation among African Americans. Our data show that the current recommended daily allowance of vitamin D3 of 400 – 600 IU/d will not optimize vitamin D nutrition in this population. Furthermore, higher amounts than the recommended upper daily allowance of 50 ␮g/d (2000 IU/d) may be required to achieve concentrations of 25(OH)D 쏜 75 nmol/L in most of the African American population. Our study showed that a dose of 50 ␮g/d can raise the population serum 25(OH)D concentration to an average of 앒75 nmol/L. However, to raise the 25(OH)D concentration to 쏜75 nmol/L in all individuals, a dose of 70 ␮g (2800 IU/d) for those with a concentration 쏜45 nmol/L and a dose of 100 ␮g (4000 IU/d) for those with a concentration 쏝45 nmol/L are required in an African American population. The response to vitamin D3 supplementation in the literature yielded somewhat variable results. Barger-Lux et al (12) showed that in a relatively replete group of white subjects, 25 ␮g vitamin D3/d resulted in an increase of 13 nmol/L from a mean of 67 to 80 nmol/L. This amount of supplementation left a significant proportion of the study group at suboptimal concentrations. The basal 25(OH)D concentration was negatively correlated with response, and the dose was inversely correlated with the response per 1 ␮g. Likewise, Heaney et al (17) treated a group of healthy volunteers with a basal 25(OH)D concentration of 72 nmol/L with either 25 or 250 ␮g/d vitamin D3. They reported a dose response of 0.7 nmol/L per 1 ␮g oral vitamin D3 supplemented. The mean 25(OH)D concentration was 84 nmol/L after 5 mo of supplementation with the 25-␮g/d dose, which left many subjects with serum 25(OH)D concentrations 쏝75 nmol/L. In this relatively vitamin D–replete group of subjects, Heaney et al (17) suggest that 앒114 ␮g/d would be required to achieve optimal 25(OH)D in most of the subjects. In another study, Vieth et al (23) showed that 25 ␮g/d raised the mean concentration of 25(OH)D from 47 nmol/L to only 68.7 nmol/L, whereas 100 ␮g/d raised it to 96 nmol/L. Eighty-eight percent of the participants receiving 100 ␮g/d achieved 25(OH)D concentrations 쏜80 nmol/L compared with only 35% of those receiving 25 ␮g/d (23). Although 100 ␮g/d resulted in optimum serum 25(OH)D concentrations in most of the subjects in Vieth et al’s study, it is likely that many of the subjects would not require this much vitamin D3 to achieve optimal concentrations. Furthermore, the above data are applicable only to the white population. Other than our previous work, very little data regarding vitamin D dosing in African Americans is available in the literature (24). Another finding of our study was that the basal serum 25(OH)D concentration is a predictor of the response to vitamin D3 supplementation: higher increases are seen at lower basal 25(OH)D concentrations, a finding consistent with previous studies (12). However, we believe that this finding is a statistical artifact due to regression to the mean (25); therefore, we did not include this factor in our dose finding algorithm. On the other

1661

hand, a basal value is still useful because it determines the degree of insufficiency in an individual and the change required to attain optimal concentrations. A comparison of our results on vitamin D responses with those in the literature among whites suggests that the response to oral vitamin D supplementation is not blunted in African Americans. If one presumes that the slope of the response to oral vitamin D3 is 0.7 nmol/L per 1 ␮g oral vitamin D3 supplemented, as reported by Heaney et al (26), then by inference, a dose of 50 ␮g vitamin D3/d is expected to raise serum concentrations of 25(OH)D from 앒47 nmol/L (our baseline) to 82 nmol/L in whites. In fact, the mean concentration achieved in our participants was 87 nmol/L at 27 mo. Application of the Barger-Lux formula for predicting the response to 1 ␮g vitamin D (12) to our initial dose of 20 ␮g yields a dose-response slope of 1.57. For the dose of 50 ␮g, the predicted slope is 1.16. The ratio of these 2 numbers is almost exactly the ratio of our observed slopes, which suggests structural similarities between blacks and whites in terms of their response. Heaney’s (13) recent recommendation of an oral intake of 55 ␮g/d (2200 IU/d) in addition to the prevailing recommended intake of vitamin D to raise 25(OH)D concentrations near 80 nmol/L is based on the population achieving a mean concentration of 80 nmol/L instead of each individual achieving an optimal concentration. Our simplified one-measurement, one-dose adjustment algorithm gave satisfactory results. But given the individual variability in the responses to vitamin D, a better, more precise result would be expected if the concentration of 25(OH)D and the dose of vitamin D was adjusted a second time. Application of this or similar algorithms to other populations may not be as effective. The effect of sunlight is lower in the African-American population, and our study was performed in a northern latitude. Thus, in light-skinned populations, or where sun exposure is greater, seasonal adjustments would be a greater consideration. It is also possible that age, percentage body fat, or some other measurable variable may influence the response among whites. Still, we are encouraged that a simplified dosing scheme can ultimately be developed for wide clinical application, and these data may be used to make recommendations for populations in whom the baseline 25(OH)D is known. In our study, we found no influence of increasing the vitamin D intake on bone loss. African Americans differed from whites in that they have a more efficient calcium economy. Blacks conserve urinary calcium more efficiently and yet have relative resistance to the effects of parathyroid hormone on the skeleton (27, 28). Their bone mass is superior to whites, and their risk of fracture is lower. Thus, blacks have a lower requirement for calcium than do whites for a skeletal endpoint. The optimal calcium and vitamin D status in African Americans will be determined in the future by their extracalcemic effects in protecting against hypertension, obesity, diabetes, autoimmune diseases, and certain cancers (29 –31). We thank Sharon Sprintz for her expertise as a dual-energy X-ray absorptiometry technician, Jane Greensher for her expertise as the Nurse Coordinator, and Marty Feuerman for her contribution to the data and statistical analyses and to the literature review. We also thank Lynn Maier for preparing the typescript. The authors’ responsibilities were as follows—SAT: helped design and write the manuscript and responsible for the medical supervision of the study participants; JFA: helped design and write the manuscript and designed and supervised the study; SP (study statistician): responsible for the data and

1662

TALWAR ET AL

statistical analyses and helped write the manuscript; and JKY (laboratory director): responsible for the biochemical assays. None of the authors had a personal or financial conflict of interest.

REFERENCES 1. Heaney R. Functional indices of vitamin D status and ramifications of vitamin D deficiency. Am J Clin Nutr 2004;80(suppl):1706S–9S. 2. Guillemant J, Taupin P, Le HT, et al. Vitamin D status during puberty in French healthy male adolescents. Osteoporos Int 1999;10:222–5. 3. Chapuy MC, Arlot ME, Duboeuf F, et al. Vitamin D3 and calcium to prevent hip fractures in the elderly women. N Engl J Med 1992;327(23): 1637– 42. 4. Ooms ME, Roos JC, Bezemer PD, van der Vijgh WJ, Bouter LM, Lips P. Prevention of bone loss by vitamin D supplementation in elderly women: a randomized double-blind trial. J Clin Endocrinol Metab 1995; 80(4):1052– 8. 5. Bischoff-Ferrari HA, Giovannucci E, Willett WC, Dietrich T, DawsonHughes B. Estimation of optimal serum concentrations of 25hydroxyvitamin D for multiple health outcomes. Am J Clin Nutr 2006; 84(1):18 –28. 6. Looker AC, Dawson-Hughes B, Calvo MS, Gunter EW, Sahyoun NR. Serum 25-hydroxyvitamin D status of adolescents and adults in two seasonal subpopulations from NHANES III. Bone 2002;30(5):771–7. 7. NIH. Osteoporosis prevention, diagnosis, and therapy. NIH Consensus Statement 2000;17(1):1– 45. 8. Chapuy MC, Preziosi P, Maamer M, et al. Prevalence of vitamin D insufficiency in an adult normal population. Osteoporos Int 1997;7(5): 439 – 43. 9. Holick MF. Vitamin D importance in the prevention of cancers, type 1 diabetes, heart disease, and osteoporosis. Am J Clin Nutr 2004;79(3): 362–71. 10. Bischoff HA, Stahelin HB, Dick W, et al. Effects of vitamin D and calcium supplementation on falls: a randomized controlled trial. J Bone Miner Res 2003;18(2):343–51. 11. Guillemant J, Taupin P, Le HT, et al. Vitamin D status during puberty in French healthy male adolescents. Osteoporos Int 1999;10(3):222–5. 12. Barger-Lux MJ, Heaney RP, Dowell S, Chen TC, Holick MF. Vitamin D and its major metabolites: serum levels after graded oral dosing in healthy men. Osteoporos Int 1998;8(3):222–30. 13. Heaney RP. The Vitamin D requirement in health and disease. J Steroid Biochem Mol Biol 2005;97(1–2):13–9. 14. Harris SS, Soteriades E, Coolidge JA, Mudgal S, Dawson-Hughes B. Vitamin D insufficiency and hyperparathyroidism in a low income, multiracial, elderly population. J Clin Endocrinol Metab 2000;85(11): 4125–30.

15. Harris SS, Dawson-Hughes B. Seasonal changes in plasma 25hydroxyvitamin D concentrations of young American black and white women. Am J Clin Nutr 1998;67(6):1232– 6. 16. Weaver C, Fleet J. Vitamin D requirements: current and future. Am J Clin Nutr 2004;80(suppl):1735S–9S. 17. Heaney RP, Davies KM, Chen TC, Holick MF, Barger-Lux MJ. Human serum 25-hydroxycholecalciferol response to extended oral dosing with cholecalciferol. Am J Clin Nutr 2003;77:204 –10. 18. Aloia JF, Talwar SA, Pollack S, Yeh J. A randomized controlled trial of vitamin D3 supplementation in African American women. Arch Intern Med 2005;165:1618 –23. 19. Hollis B. Relative concentrations of 25-hydroxyvitamin D2/D3 and 1,25dihydroxyvitamin D2/D3 in maternal plasma at delivery. Nutr Res 1984; 4:27. 20. Carter G, Carter R, Gunter E, et al. Measurement of vitamin D metabolites: an international perspective on methodology and clinical interpretation. J Steroid Biochem Mol Biol 2004;89 –90:467–71. 21. Vieth R. Vitamin D supplementation, 25-hydroxyvitamin D concentrations, and safety. Am J Clin Nutr 1999;69(5):842–56. 22. Dawson-Hughes B, Harris SS, Dallal GE. Plasma calcidiol, season, and serum parathyroid hormone concentrations in healthy elderly men and women. Am J Clin Nutr 1997;65(1):67–71. 23. Vieth R, Chan PC, MacFarlane GD. Efficacy and safety of vitamin D3 intake exceeding the lowest observed adverse effect level. Am J Clin Nutr 2001;73(2):288 –94. 24. Bischoff-Ferrari HA, Giovannucci E, Willett W, Dietrich T, DawsonHughes B. Estimation of optimal serum concentrations of 25hydroxyvitamin D for multiple health outcomes. Am J Clin Nutr 2006; 84:18 –28. 25. Bonate P. Analysis of pretest-posttest designs. Boca Raton, FL: Chapman & Hall, 2000:71. 26. Heaney RP, Davies KM, Chen TC, Holick MF, Barger-Lux MJ. Human serum 25-hydroxycholecalciferol response to extended oral dosing with cholecalciferol. Am J Clin Nutr 2003;77(1):204 –10. 27. Aloia JF, Mikhail M, Pagan CD, Arunachalam A, Yeh JK, Flaster E. Biochemical and hormonal variables in black and white women matched for age and weight. J Lab Clin Med 1998;132(5):383–9. 28. Cosman F, Morgan DC, Nieves JW, et al. Resistance to bone resorbing effects of PTH in black women. J Bone Miner Res 1997;12(6):958 – 66. 29. Rejnmark L, Jorgensen ME, Pedersen MB, et al. Vitamin D insufficiency in Greenlanders on a westernized fare: ethnic differences in calcitropic hormones between Greenlanders and Danes. Calcif Tissue Int 2004; 74(3):255– 63. 30. Heaney RP. Ethnicity, bone status, and the calcium requirement. Nutr Res 2002;22:153–78. 31. Heaney RP. Low calcium intake among African Americans: effects on bones and body weight. J Nutr 2006;136(4):1095– 8.

Same genetic components underlie different measures of sweet taste preference1–3 Kaisu Keskitalo, Hely Tuorila, Tim D Spector, Lynn F Cherkas, Antti Knaapila, Karri Silventoinen, and Markus Perola ABSTRACT Background: Sweet taste preferences are measured by several often correlated measures. Objective: We examined the relative proportions of genetic and environmental effects on sweet taste preference indicators and their mutual correlations. Design: A total of 663 female twins (324 complete pairs, 149 monozygous and 175 dizygous pairs) aged 17– 80 y rated the liking and intensity of a 20% (wt/vol) sucrose solution, reported the liking and the use-frequency of 6 sweet foods (sweet desserts, sweets, sweet pastry, ice cream, hard candy, and chocolate), and completed a questionnaire on cravings of sweet foods. The estimated contributions of genetic factors, environmental factors shared by a twin pair, and environmental factors unique to each twin individual to the variance and covariance of the traits were obtained with the use of linear structural equation modeling. Results: Approximately half of the variation in liking for sweet solution and liking and use-frequency of sweet foods (49 –53%) was explained by genetic factors, whereas the rest of the variation was due to environmental factors unique to each twin individual. Sweet taste preference–related traits were correlated. Tetravariate modeling showed that the correlation between liking for the sweet solution and liking for sweet foods was due to genetic factors (genetic r ҃ 0.27). Correlations between liking, use-frequency, and craving for sweet foods were due to both genetic and unshared environmental factors. Conclusion: Detailed information on the associations between preference measures is an important intermediate goal in the determination of the genetic components affecting sweet taste preferences. Am J Clin Nutr 2007;86:1663–9. KEY WORDS Twin study, sweet taste, genetic effects, heritability, taste preferences

creased risk of dental caries. Dietary guidelines worldwide discourage the consumption of added sugar (4). Several methods for measuring sweet taste preferences have been developed. In chemosensory tests, aqueous solutions of sucrose have often been used as the taste stimulus. However, the preference for sugar in water may poorly represent the liking for sweet foods, not to mention their actual use. In a cross-cultural study of 122 students, Holt et al (5) found that liking for sweetness in an aqueous solution did not predict the degree of liking for sweetness in orange juice, custard, or shortbread biscuits. An easier and less expensive way to collect data on sweetness preferences is to use postal or electronic questionnaires. Usually, a list of foods is presented to a subject and he or she is requested to evaluate liking or use-frequency of the foods. In the case of sweet foods, the liking and the use-frequency of a food item are often correlated (6, 7). In addition, behavioral questionnaires measuring the tendency to crave sweet foods (8, 9) or attitudes toward them (6, 10) have been developed. The outcomes of different sweet taste preference–related measures are often correlated (11), but it is not known whether this correlation is due to an underlying genetically determined preference for sweet taste or environmental factors. Our earlier family study showed that sweet taste preference–related traits were inherited, but we were unable to separate the effects of shared genes and family environment (3). In this study, our aims were 1) to test whether the variation shared by family members is due to genetic or shared environmental factors and 2) to examine whether the correlations between different sweet taste preference measures are due to genetic or environmental factors in a genetically informative sample of monozygotic and dizygotic female twins. 1

INTRODUCTION

Humans have innate preference for a sweet taste (1), but the degree of liking for sweetness varies greatly among individuals (2). This variation is likely to have environmental roots, but it may also have a genetic component (3). Although individual differences exist, most people find sweet foods palatable, which has led to an extensive supply and consumption of sugarcontaining products. Nutritionally beneficial foods, such as fruit, often naturally contain sugars, but foods with added sugars are disadvantageous to health because of extra calories and an in-

From the Departments of Food Technology (KK, HT, and AK), Public Health (KS), and Medical Genetics (MP),University of Helsinki, Helsinki, Finland; the Department of Molecular Medicine, National Public Health Institute, Helsinki, Finland (KK, AK, and MP); and the Twin Research and Genetic Epidemiology Unit, St Thomas’ Hospital, Kings College London, London, United Kingdom (TDS and LFC). 2 Supported by the Academy of Finland (grants 206327 and 108297), the GenomeEUtwin Project (QLG2-CT-2002-01254), and the Finnish Heart Association. 3 Address reprint requests and correspondence to K Keskitalo, Department of Food Technology, PO Box 66, University of Helsinki, FI-00014 Helsinki, Finland. Email: [email protected]. Received April 26, 2007. Accepted for publication July 30, 2007.

Am J Clin Nutr 2007;86:1663–9. Printed in USA. © 2007 American Society for Nutrition

1663

1664

KESKITALO ET AL

SUBJECTS AND METHODS

Subjects The subjects were recruited from the UK Adult Twin Registry (Twins UK) in 2005 (12). The registry consists predominantly of same-sex female twin pairs; thus, no males were included in this study. Only subjects who had both participated in the taste test and completed the questionnaires were included in the analyses, which yielded a sample of 663 females with a mean (앐SD) age of 55.6 앐 12.4 (range: 17.3– 80.7 y). This sample comprised 303 monozygotic (149 complete pairs) and 360 dizygotic (175 complete pairs) twins. Zygosity was determined by using the “Peas in the Pod” questionnaire (13, 14); if zygosity was still uncertain, it was checked by genotyping. The study was approved by Guy’s and St. Thomas’ Hospital Ethics Committee, and all participants gave informed consent. Subjects were recruited without selecting for a particular trait or disease. The average clinic visit lasted between 3 and 6 h, during which the subjects participated in several clinical tests. Chemosensory test In our previous family study (3), we determined the heritability of 3 suprathreshold sucrose concentrations: 3%, 7.5%, and 18.75% (wt:vol) sucrose in water. We found that the highest heritability (41% of variation) was obtained for the liking for the 18.75% solution. In the present study, we wanted to have a sweet taste preference test that would be easy and fast to prepare and to administer; therefore, only one very sweet solution was used. The sample was a 20% (wt:vol) sucrose solution that was prepared by pouring a 4-g prepackaged sugar sachet (Finnsugar, Kantvik, Finland) into a standard size plastic cup marked for 4 cL and 2 cL (Polarcup, Ha¨meenlinna, Finland). The cup was filled with water until the mark of 2 cl was reached, and the solution was stirred gently until the sugar had dissolved completely. The test administrator prepared the samples, which were stored overnight in the refrigerator (7 °C) and brought to room temperature before serving. Pilot testing indicated that the presentation of this intensely sweet stimulus in a single stimulus condition resulted in ratings similar to those given to the stimulus as part of the sample series used in the earlier study. The ratings for the 20% sucrose solution were not different from those of the 18.75% sucrose solution—a fact predicted by the Weber ratio (ie, just noticeable difference from a reference concentration) of sucrose in water that varies between 0.08 and 0.20 (15). Subjects visited the clinic after fasting overnight. The instructions for the taste test were given both orally and in written form, and the test administrator was present throughout the testing procedure. Before tasting, the subjects were not told that the solution was sweet. If there were more than one subject present at the time, they were told to refrain from communicating during the test. Subjects were requested to first rinse their mouths with water and to then put the entire 20-mL solution into their mouths, to swirl it around for a short while (5–10 s), and to expectorate. They then rated the degree of liking or disliking and the intensity of the taste per a 120-mm vertical Labeled Affective Magnitude Scale (LAM; 16) and Labeled Magnitude Scale (LMS; 17), respectively. LAM and LMS are relatively new instruments, but they have been validated against more conventional scales (16 – 19). With an extreme stimulus, ie, a very high sweetness, we wanted to use scales that allow ratings without the risk of the ceiling effect. Our pilot testing with 31 subjects indicated that the

LAM and LMS resulted in better discrimination than did the conventional 9-point category scales. The verbal labels and their positions on the line from the bottom of the scale were as follows for the LAM: “the greatest imaginable dislike” (0 mm), “dislike extremely” (13 mm), “dislike very much” (26 mm), “dislike moderately” (39 mm), “dislike slightly” (54 mm), “neither like nor dislike” (60 mm), “like slightly” (66 mm), “like moderately” (81 mm), “like very much” (94 mm), “like extremely” (107 mm), and “the greatest imaginable like” (120 mm). In the LMS, the verbal labels were “barely detectable” (2 mm), “weak” (7 mm), “moderate” (20 mm), “strong” (42 mm), “very strong” (59 mm), and “the strongest imaginable sensation” (120 mm). In addition, the intensity rating of a 6-n-propylthiouracil (PROP) filter paper (20) was included in the study as a positive control for heritability: the contribution of genetic factors on the variance of the intensity rating of PROP is known to be 쏜50% (21). The preparation of PROP filter papers was made as described earlier (3). Subjects first tasted pure filter paper to be later able to distinguish the taste of PROP from that of paper. After rinsing their mouths, they set the PROP filter paper on their tongues for 앒10 s. After waiting a short while (the strongest sensation of PROP intensity often comes with a delay), they rated the intensity using a similar 120-mm vertical LMS as for the intensity rating of the sucrose solution. Questionnaire data Before the clinic visit, the twins were sent postal questionnaires, which were completed at home and brought with them to the visit. The questionnaire included the ratings of like and dislike and use-frequency for 34 foods. The response alternatives for liking and disliking were 1 ҃ dislike very much, 2 ҃ dislike moderately, 3 ҃ dislike slightly, 4 ҃ neither like nor dislike, 5 ҃ like slightly, 6 ҃ like moderately, and 7 ҃ like very much. For use-frequency, the ratings were 1 ҃ never, 2 ҃ a couple times of a year or more rarely, 3 ҃ a couple times of a month, 4 ҃ a couple times a week, 5 ҃ once a day, and 6 ҃ several times a day. The foods were categorized by using factor analysis with maximum likelihood extraction and orthogonal Varimax rotation. A group of 6 sweet foods (sweet desserts, sweets, sweet pastry, ice cream, hard candy, and chocolate) was identified, and composite measures for liking and use-frequency of sweet foods were calculated as the mean of ratings given to 6 sweet foods items. Thus, the theoretical range was 1–7 for liking and 1– 6 for use-frequency of sweet foods. The reliability of the scales was further studied by Cronbach’s ␣ values; the values for liking and use-frequency of sweet foods were 0.84 and 0.71, respectively. The questionnaire also included a Craving for Sweet Foods scale, which is a subscale of the Health and Taste Attitude Scales (9, 22). This validated scale measures the tendency to crave sweet foods with 6 statements, each evaluated according to a 7-point Likert scale (1 ҃ strongly disagree, 2 ҃ moderately disagree, 3 ҃ slightly disagree, 4 ҃ neither agree nor disagree, 5 ҃ slightly agree, 6 ҃ moderately agree, and 7 ҃ strongly agree). The score for craving for sweet foods is calculated as the mean of ratings given to the 6 items; thus, the theoretical range was 1–7. The Cronbach’s ␣ value for the scale was 0.70. Quantitative genetic analysis Classic twin modeling relies on the assumption that monozygotic twins are genetically identical, whereas dizygotic twins

1665

ASSOCIATIONS BETWEEN SWEETNESS PREFERENCE MEASURES

share, on average, half of their segregating genes (23). Genetic variation can be divided into additive genetic variation, which consists of the sum of the allelic effects on the phenotype over all relevant loci, and nonadditive genetic variation, which includes the interaction of alleles in the same locus (dominance). The epistatic effect, ie, interaction between alleles in different loci, is assumed to be absent. The correlations of both additive and nonadditive genetic effects are 1 within monozygotic pairs. Within dizygotic pairs, the correlations are 0.5 for additive and 0.25 for nonadditive genetic effects. Environmental variation can be divided into environmental factors shared and unshared by cotwins. The shared environment, having a similar effect on monozygotic and dizygotic pairs, includes all environmental factors that make the twin pair similar for the trait, such as maternal nonheritable factors, shared childhood experiences, parental socioeconomic status, and the same friends. The unshared environment includes all environmental factors and experiences that make siblings in the family dissimilar, including measurement error. Thus, the correlations of shared and unshared environmental effects are defined as 1 and 0, respectively, within both monozygotic and dizygotic twin pairs. Random mating with respect to the traits in question and the absence of gene-environment interactions is also assumed in the model. Assortative mating of parents may increase dizygotic correlations and thus inflate the estimates of shared environmental variance and reduce the genetic variance. The possible effect of gene-environment interaction is estimated as part of additive genetic component to the extent that the environmental factors interacting with the genes are shared within the pair, which thus may also reflect genetically based differences in susceptibility to environmental factors. To the extent that such environmental factors are not shared between twin pairs, the unique environmental component will absorb the effect (24). On the basis of these assumptions, the phenotypic variance of a trait can be decomposed to additive genetic effects (A), dominant (nonadditive) genetic effects (D), shared (common) environmental effects (C), and unshared environmental effects (E). In genetic modeling, these variance components are treated as latent (unmeasured) and standardized independent variables, which are used to explain the variation of the trait, treated as the dependent variable in the model. The variance components explaining the total observed phenotypic variance can be calculated by squaring the path coefficients (regression coefficients) in the model. Because we had only twins reared together, but not

adopted twins or other relatives in these data, we were unable to estimate shared environment (C) and dominant genetic (D) variance components simultaneously (24). Genetic modeling was carried out with the Mx statistical package, version 1.7 (23). We first built univariate models estimating relative proportions of additive genetic (A), shared environmental (C), and unshared environmental effects (E) on the variation of each trait separately. The assumptions of the twin model were tested by comparing the chi-square change (⌬␹2) between the twin model and the saturated model, which did not make any of the assumptions of the twin model. Relying on the final models and on the correlations between the phenotypes, we hypothesized which traits may have common underlying genetic or environmental effects and included these in a multivariate model. Cholesky decomposition was chosen as the general overall multivariate model. Cholesky decomposition assumes that specific genetic and environmental factors affect each phenotype, but these factors can also affect other phenotypes. Thus, we could study whether a correlation between phenotypes was due to shared genetic or environmental factors. Starting with the full model, we first tested whether all 3 variance components—A, C, and E—were necessary to explain the variance of and covariance between the traits. Second, we tested whether the specific variance components affecting a trait were unique to that trait or also affected the other traits. The fit of the model was estimated by using chi-square goodness-of-fit statistics. If the change in chi-square values compared with the change in the df measured by a P value was 쏜0.05 between 2 nested models, the more parsimonious model was assumed to provide a better fit to the data. RESULTS

The mean ratings, SDs, and within-pair intraclass correlations of the traits for monozygotic and dizygotic twins are presented in Table 1. The differences in means and variances between monozygotic and dizygotic twins were negligible and were not statistically significant. Age did not significantly correlate with any of the phenotypes (Pearson’s r values between Ҁ0.16 and 0.09). For most of the traits, the within-pair correlations of the monozygotic twins were higher than those of the dizygotic twins, which implies that genetic effects probably underlie the traits. Because the within-pair correlations of monozygotic twins were

TABLE 1 Ratings and within-pair intraclass correlations of monozygotic and dizygotic pairs (n ҃ 663) Monozygotic twins (n ҃ 303)

Dizygotic twins (n ҃ 360)

149 complete pairs Variable Liking for sweet solution Intensity of sweet solution Intensity of PROP3 Liking for sweet foods Use-frequency of sweet foods Craving for sweet foods 1

175 complete pairs

x¯ 앐 SD

Within-pair r

x¯ 앐 SD

Within-pair r1

61.1 앐 24.2 34.1 앐 23.0 46.7 앐 31.9 5.5 앐 1.0 2.9 앐 0.5 4.2 앐 1.1

0.48 0.32 0.54 0.59 0.54 0.34

63.2 앐 23.7 32.7 앐 21.2 48.5 앐 34.5 5.5 앐 1.0 2.9 앐 0.6 4.1 앐 1.2

0.262 0.23 0.444 0.362 0.252 0.33

1

Pearson correlation coefficient. Significance level for the difference in correlation coefficients between the groups (Fisher Z transform): 3 P 쏝 0.001, 4 P 쏝 0.05. 3 PROP, 6-n-propylthiouracil. 2,4

1666

KESKITALO ET AL

less than twice those of the dizygotic correlations, the genetic effects were assumed to be additive and the ACE model was chosen as the starting point for quantitative genetic analysis. Pearson’s correlation coefficients between the phenotypes are shown in Table 2. Many significant correlations were observed between the phenotypes, which suggests that common factors underlie the traits. However, the intensity rating of the sweet solution and PROP were not strongly correlated with any of the other traits.

The correlation (r ҃ 0.23) between liking for the sweet solution and liking for sweet foods was explained solely by genetic factors. The correlations between self-reported measures of sweet taste preference were, in turn, due to both genetic and unshared environmental factors.

DISCUSSION

Genetic and environmental effects on different sweet taste preference measures

Quantitative genetic analysis On the basis of the within-pair correlation patterns, an ACE model was chosen as the starting point of genetic modeling. A comparison of the fits of the twin models with the saturated model showed that the additive genetic and specific environmental (AE) model was found to fit all the variables (poorest change in fit for the use-frequency of sweet foods: ⌬␹ 27 ҃ 12, P ҃ 0.11), and a more complex additive genetic, shared environment, and specific environment (ACE) model did not offer a better fit, which suggests a lack of shared environmental effects. However, for 3 variables (the intensity rating of the sweet solution, intensity rating of PROP, and craving for sweet foods), the CE model provided an equally good fit. The intensity rating of PROP, here used as a positive control, was clearly under genetic influence. The proportional effects of the additive genetic and specific environmental components with their 95% CIs are presented in Figure 1. In the multivariate modeling, we left out the intensity ratings of the sweet solution and the PROP filter paper because they did not correlate with the other measures. The multivariate model built was thus a tetravariate model. Because the variation of all the variables in the model could be explained by additive genetic and unshared environmental factors, we first tested whether the shared environmental factors could be excluded from the model. The removal of these variance components did not significantly worsen the fit of the model compared with the full model (⌬␹ 29 ҃ 9, P ҃ 0.44); thus, we could only consider additive genetic and unshared environmental effects. The additive genetic and unshared environmental correlations that were very low (lower CI ҃ 0) were then dropped from the model if their removal did not significantly worsen the model fit. The final model (⌬␹ 214 ҃ 20, P ҃ 0.14 compared with the full model) calculated using the unstandardized variances of the variables is presented in Figure 2, and the additive genetic and unshared environmental correlations between the traits are shown in Table 3.

Genetic effects clearly contributed to the variation in the sweet taste preference–related traits, ie, liking for sweet solution, liking for sweet foods, and use-frequency of sweet foods. Approximately half of the variation in the latter traits (a2 ҃ 49%, 54%, and 53%, respectively) was explained by additive genetic effects, and shared environmental effects did not contribute to the variation. In our earlier family study (3), we found similar heritability estimates for these traits, but were unable to separate the effects of shared genes and shared environment in the population consisting of families with no members reared apart. The present results thus provide evidence that the heritability of sweet taste preference–related traits is mediated by genetic effects and not by shared environment, eg, by a family’s common dietary habits. Although liking for sweet taste appears to be partly inherited, the intensity perception of the sweetness is only weakly, if at all, inherited. The source of the within-pair correlation of the twins in the intensity ratings of the sweet solution (r ҃ 0.32 for monozygotic and r ҃ 0.23 for dizygotic) could not be determined. When the common variation was modeled to derive from shared genes (additive genetic effects), the heritability estimate was lower than that for the sweet taste preference–related traits, in line with the family study results (3). Although the liking for and the intensity rating of the sweet solution were weakly correlated (r ҃ Ҁ0.20), this study further supports the view that different mechanisms underlie these perceptions. Correlations among the measures of sweet taste preference Liking for sweet solution and for sweet foods was measured with different methods and in different situations, the former by tasting an extremely sweet aqueous sucrose solution at the clinic and the latter by evaluating liking and disliking for listed food

TABLE 2 Pearson correlation coefficients among the phenotypes (n ҃ 663) Liking for sweet solution

Intensity of sweet solution

Intensity of PROP1

Liking for sweet foods

Use-frequency of sweet foods

Craving for sweet foods

1 Ҁ0.202 0.093 0.232 0.132 0.06

1 0.172 0.00 0.03 0.05

1 0.01 0.06 0.05

1 0.552 0.402

1 0.302

1

Liking for sweet solution Intensity of sweet solution Intensity of PROP Liking for sweet foods Use-frequency of sweet foods Craving for sweet foods 1

PROP, 6-n-propylthiouracil. 2 P 쏝 0.001. 3 P 쏝 0.05.

ASSOCIATIONS BETWEEN SWEETNESS PREFERENCE MEASURES

1667

FIGURE 1. Proportion of the variation of traits explained by additive genetic and unshared environmental effects; 95% CIs are shown in parentheses. The variation of the traits in 324 complete twin pairs (149 monozygotic and 175 dizygotic pairs; n ҃ 648) was decomposed with the use of linear structural equation modeling.

names at home. The significant additive genetic correlation (rA ҃ 0.27, 95% CI: 0.15, 0.40) shows that these traits are partly affected by the same genes. Although our study did not reveal which genes are involved, we assume that they are affecting the preference for sweet taste because liking for sweetness is the only obvious explanation for the correlation. Thus, this result suggests that the solution test at least partially reflects the underlying sweet taste preference. In addition, there are probably other factors (eg, cultural factors) that influence the affection for preferred sweetness in specific foods (5, 25).

The instruments used here appear to measure different aspects of sweet taste preference. Liking of the sweet aqueous solution did not correlate significantly with use-frequency or craving for sweet foods. Thus, instruments focusing on sweet preferences may be so different from each other that they do not measure the same issue. In addition, the predictive value of separate sweet taste preference measures on dietary intake has been shown to be limited (11); thus, multiple measures are probably needed to track the best predictor until the best predictors or the role of each instrument is clarified.

FIGURE 2. Path diagram of the tetravariate Cholesky model for the variance in sweet taste preference–related traits. The unstandardized variance of 149 monozygotic and 175 dizygotic pairs (n ҃ 648) was decomposed to additive genetic (A1–A4) and unshared environmental (E1–E4) effects. Each latent (unmeasured) variable, A or E, represents a set of genetic or environmental factors, respectively. An arrow pointing from any given latent variable to 2 traits (ie, observed variables) means that this set of genetic or environmental factors underlies both of these traits. If 2 variables do not correlate significantly, they are not influenced by any common latent variables. The variance and covariance components can be obtained by squaring the path coefficients.

1668

KESKITALO ET AL

TABLE 3 Correlations between additive genetic and unshared environmental factors explaining correlations between different sweet taste preference–related traits in a tetravariate Cholesky model for 149 monozygotic and 175 dizygotic complete twin pairs (n ҃ 648) Additive genetic correlation2

Trait 1

Trait 2

Trait correlation1

r

95% CI

Proportion of trait correlation explained

Specific environmental correlation2

r

95% CI

% Liking for sweet solution Liking for sweet foods Liking for sweet foods Use-frequency of sweet foods 1 2

Liking for sweet foods Use-frequency of sweet foods Craving for sweet foods Craving for sweet foods

0.23 0.55 0.40 0.30

0.27 0.66 0.48 0.31

(0.15, 0.40) (0.54, 0.77) (0.31, 0.64) (0.11, 0.50)

100 72 63 52

Proportion of trait correlation explained %

0 0.36 0.28 0.27

(0.22, 0.48) (0.15, 0.40) (0.13, 0.40)

0 28 37 48

Phenotypic Pearson correlation coefficient. Additive genetic and specifc environmental correlations obtained with the use of linear structural equation modeling.

Correlations among the questionnaire variables Correlations among the traits related to self-reported sweet taste preference were all moderate and positive (r ҃ 0.30 – 0.55, P 쏝 0.001). The correlation among affective ratings and use of sweet foods is in line with earlier studies (6, 7, 26). The scores for craving for sweet foods were correlated with liking and usefrequency of sweet foods, which further validates the use of this scale. Earlier validation studies have shown that the craving scores are associated with pleasantness ratings of chocolate bars and soft drinks (9, 22). The tetravariate Cholesky model showed that the correlations among these traits are due to both additive genetic and unshared environmental effects. The genetic effects explained, on average, 62% of the phenotypic correlation. Shared family environment does not appear to contribute to the covariance of these traits. In addition, 2 separate sets of genes and environmental factors influence the traits, whereas the craving for sweet foods is also affected by genes and environmental factors specific to this trait. Again, our study did not determine which genes or environmental factors are involved. The specific environmental factors, ie, factors making a twin pair dissimilar, may include individual differences in the scale usage or attitudes. The subjects may have misreported either consciously or unconsciously. This misreporting may have resulted from an avoidance of the use of the ends of a scale (27) or from reporting according to expected social desirability (28). Attitudes toward sweetness, eg, concerning the healthiness of sweet foods, have been shown to affect the liking and use-frequency of sweet foods (6). Study limitations As noted earlier, we decomposed the variation to genetic and environmental factors, but were not able to determine the underlying genes or environmental factors. The genetic factors affecting the traits remain to be identified and localized by genemapping experiments and the environmental factors by epidemiologic studies. Another limitation is that the data consist solely of females. Males may prefer higher sweetness (29) and may have more positive attitudes toward sugar (10). In addition, sex differences exist in the variance components of food use (30, 31). Thus, the results of this study may not allow extrapolation to males.

The classic twin design assumes 1) random mating with respect to the traits (in this case, mating of individuals regardless of their sweet taste preferences), 2) that the shared environment affects equally monozygotic and dizygotic pairs, and 3) the absence of gene-environment interactions. The first 2 can be assumed to be true in the case of sweet taste preferences, but gene-environment interactions may occur, which means that individuals with different genotypes respond differently to the environment. This might be expressed as the craving versus the avoidance of sweet foods in stress by individuals with different genetic makeup (32). Conclusions This is the first time that the covariance of different sweet taste preference–related measures has been separated between genetic and environmental factors. The multivariate modeling showed that some of the same genes underlie the liking for a sweet aqueous solution, measured by chemosensory test and the selfreported liking for sweet foods. Thus, an underlying genetic inclination to like sweetness exists and is expressed in both measures. The covariance among scores of the questionnaire phenotypes derives from both genetic and shared environmental factors. Two separate sets of genetic and environmental factors underlie liking for, use-frequency of, and craving for sweet foods. In addition, craving for sweet foods is affected by specific genetic and environmental factors. However, the shared family environment does not appear to influence the variance of separate measures of sweet taste preference or the covariance between them. This suggests that, in adults, liking for sweet foods inherited from the childhood family is mediated through genes rather than through the food habits of the family. However, approximately half of the variation is due to environmental factors and thus modifiable by dietary education. These results broaden our understanding of the background of sweet taste preference. The sweetness preference, which may lead to adverse health effects through the excess use of sugar, derives from multiple separate genetic and environmental factors. Studies of the effect of high carbohydrate intakes on the development of obesity have produced controversial results; although several studies have shown that the consumption of sugar-sweetened drinks is associated with weight gain (for review, see 33), it has also been suggested that a high intake of

ASSOCIATIONS BETWEEN SWEETNESS PREFERENCE MEASURES

sugar may be negatively associated with the indexes of obesity (34). The high use of sugar also increases the risk of dental caries and may increase blood insulin concentrations and the risk of diabetes (35). We showed that commonly used instruments reflecting sweet taste preference may measure different hedonic or behavioral aspects. It is not clear which of the measures best reflect the high intake of refined sugar. The results encourage selection or development of a (set of) sweet taste preference measures that would reveal the most important aspects of the preference and could be universally used to study the effect of taste preferences on the excess use of sugar. Understanding the genetic elements underlying sweetness preference would help to cope with a problem that has major nutritional implications. We thank Kyllikki Kilpi (Finnsugar Ltd) for providing the prepackaged sugar samples and Ursula Perks for excellent technical assistance. The authors’ responsibilities were as follows—KK: drafted the manuscript and analyzed the data; KK, HT, AK, and MP: planned the study; TDS and LFC: collected the data; and KS: assisted in the quantitative genetic analyses. All authors contributed to the interpretation of the results and to the writing of the manuscript and accepted the final version. None of the authors had a conflict of interest.

REFERENCES 1. Desor JA, Maller O, Turner RE. Taste in acceptance of sugars by human infants. J Comp Physiol Psychol 1973;84:496 –501. 2. Pangborn R. Individuality in responses to sensory stimuli. 6th ed. In: Solms J, Hall RL, eds. Criteria of food acceptance. Zu¨rich, Switzerland: Forster Publishing, 1981:177–219. 3. Keskitalo K, Knaapila A, Kallela M, et al. Sweet taste preferences are partly genetically determined; identification of a trait locus on chromosome 16. Am J Clin Nutr 2007;86:55– 63. 4. Department of Health and Human Services and the Department of Agriculture. Dietary Guidelines for Americans, 2005. Released 12 January 2005. Internet: http://www.health.gov/dietaryguidelines (accessed 4 April 2007). 5. Holt SHA, Cobiac L, Beaumont-Smith NE, Easton K, Best DJ. Dietary habits and the perception and liking of sweetness among Australian and Malaysian students: a cross-cultural study. Food Qual Pref 2000;11: 299 –312. 6. La¨hteenma¨ki L, Tuorila H. Attitudes towards sweetness as predictors of liking and use of various sweet foods. Ecol Food Nutr 1994;31:161–70. 7. Drewnowski A, Hann C. Food preferences and reported frequencies of food consumption as predictors of current diet in young women. Am J Clin Nutr 1999;70:28 –36. 8. Benton D, Greenfield K, Morgan M. The development of the attitudes to chocolate questionnaire. Person Individ Diff 1998;24:513–20. 9. Roininen K, La¨hteenma¨ki L, Tuorila H. Quantification of consumer attitudes to health and hedonic characteristics of foods. Appetite 1999; 33:71– 88. 10. Tuorila-Ollikainen H, Mahlama¨ki-Kultanen S. The relationship of attitudes and experiences of Finnish youths to their hedonic responses to sweetness. Appetite 1985;6:115–24. 11. Mattes RD, Mela DJ. Relationships between and among selected measures of sweet-taste preference and dietary intake. Chem Senses 1986; 11:523–39. 12. Spector TD, Williams FM. The UK Adult Twin Registry. Twin Res Hum Genet 2006;9:899 –906.

1669

13. Peeters H, Van Gestel S, Vlietinck R, Derom C, Derom R. Validation of a telephone questionnaire in twins of known zygosity. Behav Genet 1998;28:159 – 63. 14. Martin NG, Martin PG. The inheritance of scholastic abilities in a sample of twins. Ascertainments of the sample and diagnosis of zygosity. Ann Hum Genet 1975;39:213– 8. 15. Laing DG, Prescott J, Bell AG, et al. A cross-cultural study of taste discrimination with Australians and Japanese. Chem Senses 1993;18: 161– 8. 16. Schutz HG, Cardello AV. A labeled affective magnitude (LAM) scale for assessing food liking/disliking. J Sens Stud 2001;16:117–59. 17. Green BG, Shaffer GS, Gilmore MM. Derivation and evaluation of a semantic scale of oral sensation magnitude with apparent ratio properties. Chem Senses 1993;18:683–702. 18. Green BG, Dalton P, Cowart B, Shaffer G, Rankin K, Higgins J. Evaluating the ‘labeled magnitude scale’ for measuring sensations of taste and smell. Chem Senses 1996;21:323–34. 19. Lawless HT, Horne J, Spiers W. Contrast and range effects for category, magnitude and labeled magnitude scales in judgements of sweetness intensity. Chem Senses 2000;25:85–92. 20. Zhao L, Kirkmeyer SV, Tepper BJ. A paper screening test to assess genetic taste sensitivity to 6-n-propylthiouracil. Phys Behav 2003;78: 625–33. 21. Hansen JL, Reed DR, Wright MJ, Martin NG, Breslin PA. Heritability and genetic covariation of sensitivity to PROP, SOA, quinine HCl, and caffeine. Chem Senses 2006;31:403–13. 22. Roininen K, Tuorila H, Zandstra EH, et al. Differences in health and taste attitudes and reported behaviour among Finnish, Dutch and British consumers: a cross-national validation of the Health and Taste Attitudes Scales (HTAS). Appetite 2001;37:33– 45. 23. Neale MC, Boker SM, Xie G, Maes HH. Mx: statistical modeling. 5th ed. Richmond, VA: Department of Psychiatry, Virginia Commonwealth University, 1999. 24. Neale MC, Cardon LR. Methodology for genetic studies of twins and families. Dordrecht, Germany: Kluwer Academic Publishers BV, 1992. 25. Laing DG, Prescott J, Bell GA, et al. Responses of Japanese and Australians to sweetness in the context of different foods. J Sens Stud 1994; 9:131–55. 26. Cardello A, Maller O. Relationships between preferences and food acceptance ratings. J Food Sci 1982;47:1553–7. 27. Lawless HT, Heymann H. Sensory evaluation of food: principles and practices. Gaithenburg, MD: Aspen Publishers, 1999:222–3. 28. Maurer J, Taren DL, Teixeira PJ, et al. The psychosocial and behavioral characteristics related to energy misreporting. Nutr Rev 2006;64:53– 66. 29. Conner MT, Booth DA. Preferred sweetness of a lime drink and preference for sweet over non-sweet foods, related to sex and reported age and body weight. Appetite 1988;10:25–35. 30. van den Bree MB, Eaves LJ, Dwyer JT. Genetic and environmental influences on eating patterns of twins aged 욷 50 y. Am J Clin Nutr 1999;70:456 – 65. 31. Heitmann BL, Harris JR, Lissner L, Pedersen NL. Genetic effects on weight change and food intake in Swedish twins. Am J Clin Nutr 1999; 69:597– 602. 32. Kampov-Polevoy AB, Alterman A, Khalitov E, Garbutt JC. Sweet preference predicts mood altering effect of and impaired control over eating sweet foods. Eat Behav 2006;7:181–7. 33. Malik VS, Schulze MB, Hu FB. Intake of sugar-sweetened beverages and weight gain: a systematic review. Am J Clin Nutr 2006;84:274 – 88. 34. Hill JO, Prentice AM. Sugar and body weight regulation. Am J Clin Nutr 1995;62(suppl):264S–74S. 35. Salmero´n J, Manson JE, Stampfer MJ, Colditz GA, Wing AL, Willet WC. Dietary fiber, glycemic load, and risk of non-insulin-dependent diabetes mellitus in women. JAMA 1997;277:472–7.

Treatment for 2 mo with n⫺3 polyunsaturated fatty acids reduces adiposity and some atherogenic factors but does not improve insulin sensitivity in women with type 2 diabetes: a randomized controlled study1– 4 Morvarid Kabir, Geraldine Skurnik, Nadia Naour, Valeria Pechtner, Emmanuelle Meugnier, Sophie Rome, Annie Quignard-Boulangé, Hubert Vidal, Gérard Slama, Karine Clément, Michèle Guerre-Millo, and Salwa W Rizkalla ABSTRACT Background: Information is lacking on the potential effect of nҀ3 polyunsaturated fatty acids (PUFAs) on the adipose tissue of patients with type 2 diabetes. Objective: We evaluated whether nҀ3 PUFAs have additional effects on adiposity, insulin sensitivity, adipose tissue function (production of adipokines and inflammatory and atherogenic factors), and gene expression in type 2 diabetes. Design: Twenty-seven women with type 2 diabetes without hypertriglyceridemia were randomly allocated in a double-blind parallel design to 2 mo of 3 g/d of either fish oil (1.8 g nҀ3 PUFAs) or placebo (paraffin oil). Results: Although body weight and energy intake measured by use of a food diary were unchanged, total fat mass (P 쏝 0.019) and subcutaneous adipocyte diameter (P 쏝 0.0018) were lower in the fish oil group than in the placebo group. Insulin sensitivity was not significantly different between the 2 groups (measured by homeostasis model assessment in all patients and by euglycemichyperinsulinemic clamp in a subgroup of 5 patients per group). By contrast, atherogenic risk factors, including plasma triacylglycerol (P 쏝 0.03), the ratio of triacylglycerol to HDL cholesterol (atherogenic index, P 쏝 0.03), and plasma plasminogen activator inhibitor-1 (P 쏝 0.01), were lower in the fish oil group than in the placebo group. In addition, a subset of inflammation-related genes was reduced in subcutaneous adipose tissue after the fish oil, but not the placebo, treatment. Conclusions: A moderate dose of nҀ3 PUFAs for 2 mo reduced adiposity and atherogenic markers without deterioration of insulin sensitivity in subjects with type 2 diabetes. Some adipose tissue inflammation-related genes were also reduced. These beneficial effects could be linked to morphologic and inflammatory changes in adipose tissue. This trial was registered at clinicaltrials.gov as NCT0037. Am J Clin Nutr 2007;86:1670 –9. KEY WORDS Adiposity, fish oil, type 2 diabetes, women, adipocyte size, PAI-1, atherogenic index, adipose tissue inflammation-related genes

INTRODUCTION

Since the early 1980s, fish oil consumption has been found to exert beneficial effects on plasma lipids and to be associated with

1670

decreased coronary artery disease in nondiabetic subjects (1). These promising effects of fish oil consumption on health were dampened by the discovery of deleterious effects of fish oil on glucose control in subjects with diabetes (2, 3). These effects, however, were mainly observed in response to high doses of fish oil (5–18 g/d) or in studies comprising patients who did not comply with their treatments (3). Two meta-analyses of randomized controlled trials with moderate doses of fish oil finally concluded that fish oil supplementation decreases plasma triacylglycerol in subjects with type 2 diabetes without adverse effects on plasma glucose control (4, 5). In rodents, additional effects of nҀ3 polyunsaturated fatty acids (PUFAs) on insulin sensitivity and adipose tissue metabolism and gene expression have been reported. Both in rats and in mice, the intake of nҀ3 PUFAs reduces adipose tissue mass, which is preferentially located in visceral depots (6 –9). Moreover, in insulin-resistant rodents, dietary fish oil ameliorates insulin sensitivity (8, 10) and decreases rates of glucose oxidation (6) and lipolysis in isolated adipocytes (7). Gene expression 1

From INSERM, Nutriomique, U872 (team 7), Paris, France (NN, KC, MG-M, and SWR); AP-H, Hôtel-Dieu Hospital, Departments of Diabetes and Nutrition, Paris, France (MK, GS, VP, GS, KC, and SWR); University Pierre and Marie Curie-Paris 6, Center of Research of Cordeliers, Paris, France (NN, KC, MG-M, and SWR); the Center of Research on Human Nutrition (CRNH-Ile de France), Paris, France (KC, MG-M, and SWR); INSERM, U870-INRA U1235, Faculty of Medicine Rene Laennec, Lyon, France (EM, SR, and HV); and INRA-INA-PG, UMR 914, Paris, France (AQ-B). 2 This study was presented in part in the 64th ADA meeting, held in Orlando, FL, 2004, and at the 24th International Symposium on Diabetes and Nutrition DNSG of the EASD, held in Salerno, Italy, 2006. 3 Supported by grants from the National Institute of Health and Medical Research (INSERM); the Pierre and Marie Curie University, Paris 6; the Association of Young Subjects with Diabetes (AJD), Paris; and the National Agency of Research (ANR Obcat) and Programme of Research in Human Nutrition (PRNA), France. The fish oil and placebo capsules were provided by Pierre Fabre Médicament (Castre, France). 4 Reprints not available. Address correspondence SW Rizkalla, Department of Nutrition, CRNH-Ile de France, INSERM U872, Hôtel-Dieu Hospital, 1 Place du Parvis Notre-Dame, 75004 Paris, France. E-mail: [email protected]. Received February 27, 2007. Accepted for publication August 9, 2007.

Am J Clin Nutr 2007;86:1670 –9. Printed in USA. © 2007 American Society for Nutrition

1671

FISH OIL AND ADIPOSE TISSUE IN TYPE 2 DIABETES

studies showed the up-regulation of genes encoding mitochondrial biogenesis and oxidative metabolism proteins in the epididymal adipose tissue of mice fed nҀ3 PUFAs (9). Fish oil may also target adipose tissue secretory factors, because circulating concentrations of leptin and adiponectin and adipose tissue gene expression were modulated in rodents fed nҀ3 PUFAs (8, 11–13). In humans, only one preliminary intervention study, in 6 healthy adults, reported a beneficial effect of fish oil on body fat mass (14). Data are lacking regarding the potential effect of nҀ3 PUFAs on adiposity markers, insulin sensitivity, and adipose tissue gene expression in patients with altered insulin sensitivity or type 2 diabetes. Increasing evidence indicates that metabolic disorders, including type 2 diabetes, are associated with a low-grade inflammatory state. Additionally, adipose tissue has been recognized to contribute to the production of inflammatory factors (15). The expression of a wide panel of inflammation-related factors is increased in the obese state and is reduced after weight loss, a condition associated with marked improvement of insulin sensitivity (16). A series of recent animal and human studies showed that blood-derived macrophages infiltrate the adipose tissue in obesity and therefore contribute to the inflammatory state of adipose tissue (17–20). Experimental and clinical evidence suggest that fish oil exerts immunomodulatory effects in various pathologic situations, including inflammatory joint and bowel diseases (21) and cancer (22). Fish oil may influence monocyte and macrophage functions (23). Given these observations, we hypothesized that nҀ3 PUFA supplementation may modify plasma inflammatory markers and inflammatory gene expression in adipose tissue. Therefore, we conducted the present study to evaluate whether the intake of a moderate dose of nҀ3 PUFAs, which is generally prescribed in France, might be of benefit on 1) insulin sensitivity, 2) adiposity (fat mass and adipocyte diameter), 3) proteins secreted by adipose tissue (plasma concentrations of adipokines, atherogenic factors, and inflammatory factors), and 4) the expression of a subset of inflammation-related genes in adipose tissue in a homogeneous group of women with well-controlled type 2 diabetes. This is a population with a special adipose tissue distribution that differs from that of men. It has been shown that, for a given waist circumference, women tend to have more abdominal subcutaneous and less visceral adipose tissue than do men independent of age (24).

SUBJECTS AND METHODS

Patients and treatments Postmenopausal women with type 2 diabetes were recruited from the Diabetes Department outpatient clinic at the Hôtel-Dieu hospital in Paris. We excluded from the study patients with abnormal renal, hepatic, or thyroid function or gastrointestinal disorders. Twenty-nine patients who met the following criteria were initially included in the study: fasting plasma glucose between 7.7 and 14.0 mmol/L, glycated hemoglobin (HbA1c) of 7–10.5%, age from 40 to 60 y, body mass index (BMI; in kg/m2) between 27 and 40, and plasma triacylglycerol 쏝2.5 mmol/L. The clinical and biological characteristics of the women who participated in the study are shown in Table 1. This sample size was calculated after fixing the probability of type 1 error at 0.05 and that of type 2 error at 0.10 for changes in fat mass (25). During the follow-up,

TABLE 1 Clinical characteristics of the subjects at the time of screening1

Age (y) Body weight (kg) BMI (kg/m2) Fasting glycemia (mmol/L) HbA1c (%) Plasma cholesterol (mmol/L) Plasma triacylglycerol (mmol/L)

Placebo (n ҃ 14)

Fish oil (n ҃ 12)

55 앐 1 81 앐 6 30 앐 2 8.8 앐 0.8 7.8 앐 0.4 5.3 앐 0.2 1.1 앐 0.1

55 앐 2 80 앐 6 30 앐 2 8.2 앐 0.7 7.3 앐 0.2 5.1 앐 0.2 1.2 앐 0.2

1 All values are x៮ 앐 SEM. HbA1c, glycated hemoglobin. There were no significant differences between the 2 groups at the time of screening.

3 patients were excluded from the study, 2 because of acute medical events (severe acute pancreatitis and ischemic cerebral vascular accident) and 1 because of noncompliance with the treatment dose. Three of the remaining subjects were treated by dietary regimen alone, whereas the other 23 patients were taking usual oral hypoglycemic treatments: biguanides alone (7 patients) or bi-therapy with sulfonylureas and biguanides (16 patients). None of the patients had been treated with thiazolidinediones or insulin. Five patients were taking lipid-lowering agents and 6 patients were receiving hormone replacement therapy. The purpose, nature, and potential risks of the study were explained by a physician, and written informed consent was obtained from each patient. The experimental protocol was approved by the ethics committee of the Hôtel-Dieu hospital, Paris. The trial has been registered in the public trials registry at http://www.clinicaltrials. gov with the following identification: clinicaltrials.gov ID NCT0037. Dietary follow-up Before they started the study, the patients were followed on a regular basis. Then, each patient entered a run-in period of 2 mo. Food consumption was individually assessed by a registered dietitian, and each subject received individual counseling. The subjects were asked to follow their usual diet recommendation more strictly with consumption of 55% of their caloric intake as carbohydrates, 15% as protein, and 30% as lipids. Patients were asked to complete a 7-d food diary just before the start of the treatment period. They were recommended to keep their initial caloric intake and nutrient proportions constant throughout the study. To determine compliance with the dietary recommendations, the patients were asked to keep another food diary to be completed the last 7 d of each treatment period. Even if this method (7-d food diary) of measuring food intake might slightly underestimate true calorie intake, the same method was used before and after treatments, and hence the results can be compared. All records were analyzed by a registered dietitian using the computer software program PROFILE DOSSIER V3 (AuditConseil en Informatique Médicale, Bourges, France), with a dietary database made up of 400 foods representative of the French diet as described previously (26). Study design The patients were randomly allocated to 2 mo of 3 g/d of either fish oil (containing 1.8 g nҀ3 PUFAs: 1.08 g eicosapentaenoic acid and 0.72 g docosahexaenoic acid) or placebo (paraffin oil) in a double-blind parallel study design. Fish oil and placebo capsules were provided by Pierre Fabre Médicament (Castre,

1672

KABIR ET AL

France). The plasma fatty acid profile was measured at the beginning and at the end of the treatment periods to assess compliance. At the beginning and at the end of the trial, the subjects were hospitalized for 2 d after they had fasted overnight. During the first day, fasting blood samples were collected to determine plasma concentrations of HbA1c, glucose, insulin, lipids, plasminogen activator inhibitor factor-1 (PAI-1), and several systemic adipokines [leptin, adiponectin, serum amyloid A, interleukin-6, and tumor necrosis factor-␣ (TNF-␣)] Three blood samples were taken while the subjects were in the fasting state, at 5-min intervals, to measure homeostasis model assessment (HOMA). Body lean and fat mass distributions were measured with a total-body dual-energy X-ray absorptiometry scanner (Hologic QDR-2000; Hologic Inc, Waltham, MA) as described previously (27). Subcutaneous and visceral fat areas were determined by a single-slice computerized tomography (CT) scan (Philips Brillance 16; MedImage System Inc, Memphis, TN) at the L4 –L5 disc space level. The second day, clamp studies and a fat biopsy were performed. After the subjects had fasted for 12 h, a sample of abdominal subcutaneous adipose tissue was obtained by needle biopsy with a 14-gauge needle and a 30-mL syringe under local anesthesia with lidocaine 10% without epinephrine. One-half of the biopsy sample was used immediately to measure adipocyte diameter (in a 10-␮L sample) and to perform in vitro adipocyte culture. The other half was immediately frozen at Ҁ80 °C for subsequent RNA extraction and real-time reversetranscriptase polymerase chain reaction (RT-PCR) analysis. Insulin-sensitivity measurements Estimation of pancreatic ␤-cell function (insulin secretion) and insulin sensitivity were calculated from repeated fasting plasma insulin and glucose measurements by using HOMA/ CIGMA software (28). Moreover, to obtain a better estimate of whole-body insulin sensitivity, a euglycemic hyperinsulinemic clamp study was performed before and after each treatment period as described (26). For technical reasons, only 5 patients in each group were studied. After we administered a priming dose of insulin, the infusion rate was maintained at 6 mU 䡠 kgҀ1 䡠 minҀ1 for 180 min. Blood samples were withdrawn every 5 min to adjust the glucose infusion and retain plasma glucose at 5.5 mmol/L, then at 10-min intervals for the last 30 min once a steady state had been obtained. Insulin sensitivity was calculated as the amount of exogenous glucose needed to keep euglycemia during the steady state (mg䡠minҀ1 䡠 kgҀ1). Adipocyte morphology and culture Adipocytes from subcutaneous periumbilical adipose tissue were immediately isolated by collagenase digestion. For cell size measurements, adipocyte suspensions were then visualized under a light microscope attached to a camera and computer interface. Adipocyte diameters were measured by using PERFECT IMAGE software (Numeris, Orsay, France). Mean diameter was defined as the median value for the distribution of adipocyte diameters of 욷250 cells. For cell culture, isolated adipocytes were maintained at a density of 5–10 ҂103 cells/mL in Dulbecco’s modified Eagle’s medium supplemented with 1% fetal calf serum, 2% bovine serum albumin, and antibiotics for 2 d. The medium was changed daily. After 24 and 48 h, the culture medium was aspirated and frozen at Ҁ20 °C for measurement of the different proteins released in the medium.

RNA extraction and amplification Total RNA from subcutaneous adipose tissue (50 – 80 mg of frozen tissue) was obtained by using the Rneasy Lipid Tissue MiniKit (Qiagen, Courtaboeuf, France) according to the manufacturer’s recommendations. RNA quantity and integrity were measured by using an Agilent 2100 Bioanalyzer (Agilent Technologies, Massy, France). For the microarray study, 1 ␮g of total RNA was amplified by using the Message-Amp aRNA kit (Ambion, Austin, TX). This amplification procedure is well validated and it has been shown that it does not distort the relative abundance of individual messenger RNAs within an RNA population (29). Microarray analysis To define new targets of fish oil metabolic effects, a microarray analysis was performed from adipose tissue samples collected before and after fish oil treatment only. Amplified RNA (10 ␮g) was labeled by using the CyScribe Post-Labeling Kit (Amersham Biosciences, Munich, Germany) developed for the generation of Cy3- and Cy5-labeled first-strand cDNA probes by the post-labeling (amino allyl) method. The resulting fluorescent cDNA populations were hybridized to an 800-gene PIQOR cDNA Microarray. These 800 genes were involved in the major metabolic pathways and their transcriptional control (lipogenesis, lipolysis, insulin signaling, fatty acid transport and oxidation, glucose metabolism, thermogenesis, and energy metabolism). Known transcription factors, nuclear receptors, and cofactors involved (or suspected to be involved) in the hormonal and nutritional control of intermediate metabolism were also included. We also included all genes previously found to be regulated by insulin during a 3-h euglycemic hyperinsulinemic clamp (29). Type 1 hybridizations were performed by using the cDNA obtained at baseline (Cy3 labeling) and at 2 mo (Cy5 labeling) in 7 patients in the fish oil group according to the recommended protocol (Internet: http://www.memorec.com/). The 7 slides were scanned with a FLA8000GR microarray scanner (Fuji, Raytest, France). The images were analyzed with GENEPIX PRO 4.1 software (www.moleculardevices.com). Normalization of the signal intensities between Cy3 and Cy5 was performed by using BioConductor (Internet: http://www.bioconductor.org/). Flagged spots and spots with fluorescence intensity below 2.5fold above the background for both dyes were not taken into account. The log2 (Cy5/Cy3) ratio of the other spots was calculated. To compare results from the different subjects, data from each slide were normalized in log-space to have a mean of zero by using CLUSTER 3.0 software (30). Only the spots that were present at 80% were recovered. Genes with significant changes in expression were identified by using the significance analysis of microarrays (SAM) procedure (available at Internet: http:// www-stat.stanford.edu/앑tibs/SAM/) (31). In total, 450 Unigene clusters were also analyzed by use of SAS statistical software (version 8.0; SAS Institute Inc, Cary, NC) to identify differentially expressed genes. Quantification of mRNA by using real-time RT-PCR First-strand cDNA was synthesized from 500 ng of total RNA obtained from adipose tissue biopsy samples with 100 units of Superscript II (Invitrogen, Cergy Pontoise, France) by using

1673

FISH OIL AND ADIPOSE TISSUE IN TYPE 2 DIABETES

oligo (dT) primers and random hexamers (Promega, Charbonnières, France). Real-time RT-PCR was performed on a LightCycler instrument (Internet: http://www.roche-applied-science.com/). The cDNA was amplified by using a TaqMan probe approach in a glass capillary tube in a final volume of 10 ␮L reaction mix containing 2.5 ␮L, 100-fold diluted cDNA, 1x LightCycler-FastStart DNA Master Hybridization probes (Roche Diagnostics, Meylan, France), 3 mmol/L of MgCl2, and the specific forward, reverse, and Taqman-specific probes (TIB Mol-Biol Syntheselabor, Berlin, Germany). PCR was performed in 50 cycles with 48 s at 95 °C and 10 s at 95 °C followed by 40 s at 60 °C and 60 s at 60 °C. The specificity of amplification was determined by melting curve analysis. The quantification of 18S ribosomal RNA was used for sample normalization.

genes measured by real-time RT-PCR, a treatment-by-time interaction was found by using a multivariate analysis of repeated measurements. Consequently, the effect of each treatment was analyzed separately in each group by Student’s t test for paired data. Relations between variables were analyzed by the nonparametric Spearman’s correlation test, and r coefficients are provided. The Kolmogorov-Smirnov test was used to test for a Gaussian distribution. Statistical analysis was performed with XLSTAT (version 2007; Addinsoft, Paris, France), JMP IN (version 5; SAS Institute Inc, Cary, NC), or GraphPad Prism (GraphPad Software Inc, San Diego, CA). A P value 쏝 0.05 was considered significant. Data are expressed as means 앐 SEMs. RESULTS

Biochemical assays Plasma glucose was measured by the glucose oxidase method (Beckman Fullerton, Palo Alto, CA). Plasma insulin was determined by radioimmunoassay (Linco Research, Inc, St Charles, MO). Plasma triacylglycerol and free fatty acids were measured with Biomérieux kits (Marcy l’Etoile, France); total cholesterol, HDL, and LDL cholesterol were measured with Labintest kits (Aix-en-Provence, France). Plasma PAI-1 was measured with Chromolize/PAI-1 kits (Biopool International, Umea, Sweden). Leptin and adiponectin were determined by using radioimmunoassay kits from Linco Research. Interleukin-6 and TNF-␣ were measured by using enzyme-linked immunosorbent assay kits from R&D Systems Inc (Minneapolis, MN). Plasma serum amyloid A was measured by enzyme-linked immunoabsorbent assay Cyto-screen immunoassay kits (BioSource International, Camarilla, CA). Concentrations of fatty acids in plasma phospholipids were chromatographed as methyl esters on a 30-m fused-silica column and were detected by electron-impact ionization mass spectroscopy. Statistical methods The effects of the 2 treatments, fish oil and placebo, were compared by analysis of covariance (ANCOVA) by using the baseline value and treatment as fixed covariates and the 2-mo measurements as the dependent variable. When a significant difference between groups was found, the effect of each treatment in each group between the beginning and the end of treatment was compared further 2 by 2 by use of Student’s paired t tests. Concerning the expression of some inflammatory-related

Patient compliance Patients followed the fish oil and placebo treatments without any reported difficulty. The treatments were well tolerated, and the patients had no complaints or side effects. According to self-report, the subject’s lifestyle was unchanged throughout the study. We evaluated the nҀ3 fatty acid composition of plasma phospholipids. As shown in Table 2, the concentration of both eicosapentaenoic (20:5n-3) and docosahexaenoic (22:6n-3) acids in plasma phospholipids was significantly increased in the fish oil group but not in the placebo group. Accordingly, the nҀ6:nҀ3 ratio was significantly decreased by fish oil treatment compared with placebo. These results reflected acceptable compliance with the fish oil treatment. One patient was excluded at the end of the study because of noncompliance with the fish oil treatment. The results of this patient were excluded. Baseline characteristics There were no detectable significant differences in the clinical and biological variables measured between the patients in the 2 treatment groups at baseline. However, some significant differences were detected in the gene expression values of some inflammation–related markers (measured by real-time RT-PCR). In the 26 overweight women with type 2 diabetes included in the study, significant correlations were found between adipocyte markers (adipocyte diameter and whole fat mass) and the main adipokines (plasma leptin and adiponectin) as well as plasma atherogenic factors (PAI-1, insulin, and triacylglycerol; Table 3). There was no significant correlation

TABLE 2 Concentrations of nҀ3 fatty acids in plasma phospholipids before and after 2 mo of either placebo or fish oil treatment1 Placebo Baseline

2 mo

Fish oil P2

Baseline

NS NS NS

0.20 앐 0.04 2.79 앐 0.38 12.9 앐 1.1

g/100 g 20:5nҀ3 22:6nҀ3 nҀ6:nҀ34

0.28 앐 0.04 2.41 앐 0.47 12.4 앐 1.79

0.26 앐 0.05 2.89 앐 0.43 10.1 앐 1.27

2 mo

P2

P for ANCOVA3

0.05 0.03 0.001

0.051 0.039 0.050

g/100 g 0.50 앐 0.15 3.90 앐 0.51 5.6 앐 0.7

1 All values are x៮ 앐 SEM; n ҃ 14 patients in the placebo group and 12 patients in the fish oil group. There were no significant differences between the groups at baseline. 2 Student’s t test for paired data between baseline and 2-mo values in each group. 3 Treatment effect: ANCOVA with 2-mo values as the dependent variable and baseline values and treatment as covariates. 4 The nҀ6:nҀ3 ratio was calculated by measuring 18:2nҀ6, 20:2nҀ6, 20:4nҀ6, 22:4nҀ6, and 22:5nҀ6 to estimate total nҀ6 fatty acids and 18:3nҀ3, 18:4nҀ3, 20:4nҀ3, 20:5nҀ3, 21:5nҀ3, 22:4nҀ3, 22:5nҀ3, and 22:6nҀ3 to estimate the total nҀ3 fatty acids.

1674

KABIR ET AL

reduction was observed in the fish oil group (P 쏝 0.0003) but not in the placebo group (Table 4).

between adipocyte diameter and the systemic cytokines (plasma TNF-␣ or plasma interleukin-6). As shown in Table 3, adipocyte diameter (and fat mass percentage) was also correlated with some atherogenic (cysteine protease cathepsin S, or CTSS) and inflammation-related genes (the chemoattractant gene plasminogen activator urokinase receptor, or PLAUR; the macrophage surface marker CD11b; and the macrophage phagocytic activity marker CD68).

Glucose homeostasis Fasting plasma glucose, insulin, and HbA1c were not significantly influenced by fish oil treatment compared with placebo (Table 4). Similarly, insulin secretion (HOMA-B) and sensitivity (HOMA-S) determined by HOMA remained unchanged in all subjects (n ҃ 26 patients). There was no significant difference in the glucose disposal rate, as assessed by euglycemic hyperinsulminemic clamp in a subgroup (n ҃ 5 patients per treatment) of subjects studied (Table 4) that could be regarded as a pilot study because of the small n. These data show that dietary fish oil supplementation did not deteriorate or ameliorate plasma glucose control and insulin sensitivity in this group of women with type 2 diabetes.

Effects of intervention Dietary control, body weight, and adiposity There was no significant difference between the groups in daily intake of total energy, intake of macronutrients, or fatty acid composition, as estimated by 7-d dietary records (Table 4). Weekly fish intake did not differ significantly between the 2 treatment groups. Similarly, total body weight remained unchanged. However, whole-body fat mass, evaluated by dualenergy X-ray absorptiometry, was reduced by the fish oil treatment compared with placebo when analyzed by ANCOVA with adjustment for baseline (P ҃ 0.02). This reduction in fat mass was mainly due to a decrease in trunk fat (P ҃ 0.04; Table 4). The trunk region covers the part from the shoulders to the hip joints and represents the whole fat mass minus fat in the limbs (arms and legs) and the head. However, a single-slice CT scan at the L4 –L5 level could not detect a significant difference in the area of subcutaneous or visceral fat at this specific level between the 2 treatments. Adipocyte diameter was also reduced significantly (P ҃ 0.002) by fish oil treatment compared with placebo after adjustment for baseline values according to the ANCOVA. A 6%

Lipid homeostasis Plasma triacylglycerol was significantly lower after 2 mo in the fish oil group than in the placebo group (ANCOVA, P 쏝 0.03; Table 4). A 12% reduction (between baseline and 2 mo, P ҃ 0.03) was found in the fish oil but not the placebo group. Plasma total cholesterol, LDL cholesterol, and HDL cholesterol were not significantly affected by fish oil treatment compared with placebo. The atherogenic index was calculated as the logarithm of the ratio of plasma triacylglycerol to HDL cholesterol (32). The atherogenic index was lowered by 2 mo of fish oil treatment compared with placebo (ANCOVA, P ҃ 0.03; Table 4). Plasma free fatty acids tended to decrease after 2 mo of fish oil treatment compared with placebo, without reaching statistical significance

TABLE 3 Correlations between baseline values of adipose tissue gene expression of some adipokines (leptin and adiponectin) and inflammatory markers on one hand and both adiposity markers and several clinical variables on the other hand1 Plasma concentration Baseline value Baseline value mRNA (AU)3 Leptin Adiponectin CD11b CD68 CTSS PLAUR Adiposity markers10 Adipocyte diameter Fat mass percentage

1.04 앐 0.22 1.07 앐 0.16 1.03 앐 0.17 1.45 앐 0.40 1.15 앐 0.51 1.18 앐 0.27 103.7 앐 2.5 36.0 앐 1.5

Fat mass percentage (%)

Adipocyte diameter (␮m)

Leptin (ng/mL)

Adiponectin (␮g/mL)

PAI-1 (IU/mL)

Insulin (pmol/L)

TG (mmol/L)

36.0 앐 1.52

103.7 앐 2.5

24.0 앐 2.7

6.4 앐 0.5

17.6 앐 2.5

14.4 앐 1.4

1.5 앐 0.11

0.554

0.515

0.576

0.747 0.239 0.716 0.326

0.728 0.767 0.818 0.466

0.554

0.811 1.011

1.011 0.811

0.711 0.7811

0.706

0.386 0.567

0.626 0.319 0.756

0.767

0.488

0.478 0.498

0.666

0.488 0.476 0.736 0.344 0.476

1 AU, arbitary units; CD11b, macrophage surface marker; CD68, macrophage phagocytic activity marker; CTSS, cysteine protease cathepsin S; PLAUR, plasminogen activator urokinase receptor; PAI-1, plasminogen activator inhibitor-1; TG, triacylglycerol. 2 x៮ 앐 SEM (all such values). 3 n ҃ 14. 4 P ҃ 0.06. 5 P ҃ 0.1. 6 P 쏝 0.05. 7 P 쏝 0.0001. 8 P 쏝 0.01. 9 P ҃ 0.07. 10 n ҃ 26. 11 P 쏝 0.005.

1675

FISH OIL AND ADIPOSE TISSUE IN TYPE 2 DIABETES TABLE 4 Dietary record data and metabolic and clinical variables at baseline and after 2 mo of either placebo or fish oil treatment1 Placebo

Body weight and dietary intake Body weight (kg) Energy (kcal/d) Carbohydrate (g/d) Protein (g/d) Fish (g/wk) Fat (g/d) SFA (g/d) MUFA (g/d) PUFA (g/d) AT characteristics Fat mass (%) Trunk fat (kg) SCAT (cm3)4 VAT (cm3)4 Adipocyte diameter (␮m) Glucose homeostasis Fasting glycemia (mmol/L) Fasting insulinemia (pmol/L) HbA1c (%) HOMA S (%) HOMA B (%) Glucose disposal rate (mg 䡠 kgҀ1 䡠 minҀ1)5 Plasma lipid homeostasis Triacylglycerol (mmol/L) Total cholesterol (mmol/L) HDL cholesterol (mmol/L) LDL cholesterol (mmol/L) Free fatty acids (mmol/L) Atherogenic index6 Adipokines and metabolic risk factors Leptin (ng/mL) Adiponectin (␮g/mL) PAI-1 (IU/mL) IL-6 (pg/mL) TNF-␣ (pg/mL) SAA (␮g/mL)

Fish oil

Baseline

2 mo

P2

Baseline

2 mo

P2

P for treatment effect3

81 앐 6 1527 앐 72 149 앐 13 78 앐 3 259 앐 100 67 앐 4 29 앐 2 28 앐 2 12 앐 1

81 앐 7 1573 앐 60 157 앐 9 78 앐 4 251 앐 34 68 앐 5 27 앐 2 30 앐 3 13 앐 1

NS NS NS NS NS NS NS NS NS

80 앐 6 1460 앐 71 153 앐 8 75 앐 5 280 앐 95 60 앐 4 22 앐 2 27 앐 3 12 앐 1

79 앐 6 1412 앐 72 151 앐 8 69 앐 5 295 앐 80 58 앐 4 21 앐 1 23 앐 2 12 앐 1

NS NS NS NS NS NS NS NS NS

NS NS NS NS NS NS NS NS NS

36.9 앐 2.3 15.6 앐 2.3 943 앐 11 363 앐 5 104.8 앐 3.7

36.6 앐 2.3 15.5 앐 2.4 979 앐 115 419 앐 39 106.1 앐 3.6

NS NS NS NS NS

35.0 앐 2.1 14.3 앐 2.1 769 앐 82 329 앐 63 102.3 앐 3.2

33.4 앐 1.9 13.5 앐 1.9 785 앐 66 380 앐 71 95.9 앐 3.5

0.01 0.012 NS NS 0.0003

0.02 0.04 NS NS 0.002

8.9 앐 0.8 14.8 앐 2.1 7.8 앐 0.4 64.7 앐 8.0 55.0 앐 14.6 10.7 앐 0.6

9.2 앐 0.8 15.7 앐 2.1 7.7 앐 0.3 70.6 앐 4.9 53.5 앐 13.6 10.9 앐 0.6

NS NS NS NS NS NS

8.2 앐 0.7 13.9 앐 2.0 7.3 앐 0.2 74.3 앐 10.4 63.1 앐 8.4 7.9 앐 0.8

8.0 앐 0.7 13.2 앐 2.3 7.4 앐 0.3 80.6 앐 8.8 60.6 앐 10.1 8.7 앐 1.1

NS NS NS NS NS NS

NS NS NS NS NS NS

1.05 앐 0.12 5.3 앐 0.2 1.4 앐 0.1 3.6 앐 0.2 0.6 앐 0.1 Ҁ0.14 앐 0.06

1.16 앐 0.13 5.2 앐 0.2 1.5 앐 0.1 3.6 앐 0.2 0.6 앐 0.1 Ҁ0.12 앐 0.06

NS NS NS NS NS NS

1.20 앐 0.19 5.1 앐 0.2 1.4 앐 0.1 3.5 앐 0.2 0.7 앐 0.1 Ҁ0.12 앐 0.07

1.05 앐 0.17 5.2 앐 0.2 1.5 앐 0.1 3.5 앐 0.2 0.6 앐 0.1 Ҁ0.21 앐 0.08

0.03 NS NS NS 0.14 0.01

0.03 NS NS NS 0.09 0.03

24.3 앐 4.0 7.0 앐 0.8 16.9 앐 3.5 2.3 앐 0.5 7.2 앐 5.7 18.3 앐 5.8

26.0 앐 4.8 6.6 앐 0.7 19.2 앐 3.8 2.4 앐 0.3 7.5 앐 6.1 22.8 앐 8.9

NS NS NS NS NS NS

23.8 앐 3.8 5.9 앐 0.4 18.9 앐 3.9 2.8 앐 0.5 5.5 앐 3.8 19.5 앐 6.3

24.0 앐 4.0 6.4 앐 0.7 9.5 앐 2.3 3.1 앐 0.6 6.3 앐 5.0 17.2 앐 4.1

NS NS 0.01 NS NS NS

NS 0.14 0.013 NS NS NS

All values are x៮ 앐 SEM; n ҃ 14 patients in the placebo group and 12 patients in the fish oil group for all variables except when otherwise cited. SFA, saturated fatty acid; MUFA, monounsaturated fatty acid; PUFA, polyunsaturated fatty acid; AT, adipose tissue; SCAT, subcutaneous adipose tissue; VAT, visceral adipose tissue; HbA1c, glycated hemoglobin; HOMA S, sensitivity index calculated by homeostasis model assessment; HOMA B, insulin secretion index measured by HOMA; PAI-1, plasminogen activator inhibitor-1; IL-6, interleukin-6; TNF-␣, tumor necrosis factor-␣; SAA, serum amyloid A. There were no significant differences between the groups at baseline. 2 Student’s t test for paired data between baseline and 2 mo in each group. 3 ANCOVA with 2-mo values as the dependent variable and baseline values and treatment as covariates. 4 The VAT and SCAT measurements were done by a single-slice computerized tomography scan at the L4 –L5 level. 5 n ҃ 5 patients per treatment group. 6 log (Triacylglycerols/HDL cholesterol). 1

(ANCOVA, P ҃ 0.09). Positive correlations were found between adipocyte diameter and fat mass on one hand and plasma triacylglycerol and PAI-1 at the end of 2 mo of fish oil treatment on the other hand (data not shown). Adipokines, atherogenic factors, and inflammatory factors PAI-1 was the circulating factor that showed the greatest change after the fish oil treatment. When 2 mo values were

adjusted by baseline values, PAI-1 was found to be lower after fish oil treatment than after placebo (ANCOVA, P ҃ 0.01). In contrast, plasma concentrations of leptin, interleukin-6, TNF-␣, and serum amyloid A were not significantly changed after 2 mo of fish oil treatment compared with placebo (Table 4). Although fish oil treatment did not significantly increase adiponectin concentrations when compared with placebo treatment (ANCOVA, P ҃ 0.14), plasma adiponectin concentrations were negatively correlated with the atherogenic index values (r ҃ Ҁ0.44, P ҃ 0.015).

1676

KABIR ET AL

TABLE 5 Genes with differential regulation in the subcutaneous fat after 2 mo of fish oil treatment1 Name

UniGene cluster

Fold change2

Function

Adipsin Hѿ transporting mitochondrial F1 Hѿ transporting lysosomal, V1 subunit E1 Hexokinase 2 Enhancer binding protein, ␤ Carboxyl esterase 2 Cyclooxygenase 1 Heterogeneous nuclear ribonucleoprotein D-like Lipoprotein lipase Matrix metallopeptidase 9/gelatinase B Ribosomal phosphoprotein, large, P1 Peroxiredoxin 3

Hs.155597 Hs.81634 Hs.77805 Hs.406266 Hs.99029 Hs.278997 Hs.88474 Hs.372673 Hs.180878 Hs.151738 Hs.356502 Hs.397062

Ҁ1.493 Ҁ1.804 Ҁ1.314 Ҁ1.234 Ҁ1.434 Ҁ1.244 Ҁ1.645 Ҁ1.395 Ҁ1.443 Ҁ1.655 Ҁ1.404 Ҁ1.365

Serine protease, secreted by adipocytes Energy metabolism, mitochondria Acidification of intracellular organelles Glucose metabolism Transcription factor Lipid metabolism Prostaglandin biosynthesis DNA-associated protein Lipid metabolism Proteolysis of the extracellular matrix Ribosomal protein Antioxidant protein

Caveolin 1 Glucokinase Heat shock, protein 8 LDL receptor-related protein associated protein 1

Hs.74034 Hs.1270 Hs.111676 Hs.75140

1.294 1.274 1.273 1.343

SLC25A1

Oxoglutarate/malate carrier

Hs.592122

1.513

MAP2K2 POM121 PCCB PKM2 RAE1 STAT6 STX4A TM4SF2

Mitogen-activated protein kinase 2 POM121 membrane glycoprotein Propionyl CoA carboxylase Pyruvate kinase RNA export 1 homolog (S. pombe) Signal transducer and activator of transcription 6 Syntaxin 4A Transmembrane 4 superfamily member 3

Hs.366546 Hs.488624 Hs.63788 Hs.198281 Hs.371698 Hs.437475 Hs.83734 Hs.439586

1.323 1.665 1.493 1.473 1.293 1.435 1.344 1.525

Gene symbol Down-regulated ADN ATP5F1 ATP6V1E1 HK2 CEBPB CES2 COX1 JKTBP LIPL MMP9 RPLP1 PRDX3 Up-regulated CAV1 GCK E2IG1 LRPAP1

Signaling Glucose metabolism Associated with estrogen action Vesicle-mediated transport, cell proliferation Gluconeogenesis and nitrogen metabolism Signaling Transport Lipid metabolism Glucose metabolism RNA transport Signaling from the leptin receptor Translocation of Glut4 Cell proliferation and cell motility

1

The genes were classified into functional categories by using SOURCE, OMIM, and Gene Ontology annotations to characterize the effect of fish oil on gene expression, an exploratory cDNA microarray analysis was performed in a subset of 7 patients treated with fish oil only to find genes and functions eventually targeted by nҀ3 polyunsaturated fatty acids. Fish oil treatment induced significant down-regulation of 12 genes and up-regulation of 13 genes among 450 genes with significant fluorescent signal on the array. 2 Results represent the fold change between samples after 2 mo of treatment and those at baseline in the fish oil group. 3 P 쏝 0.005. 4 P 쏝 0.05. 5 P 쏝 0.001.

To determine whether fish oil treatment intrinsically alters adipose cell secretory capacity, we measured adipose-secreted factors in a culture medium of adipose cells isolated from adipose tissue biopsies before and after 2 mo of the 2 treatments. Whatever the treatment, levels of PAI-1 activity, leptin, adiponectin, interleukin-6, and TNF-␣ in the media harvested from cells obtained at baseline and after 2 mo were not significantly different (data not shown). Gene expression analysis in adipose tissue The gene expression of a subset of adipose tissue-secreted proteins was analyzed by real-time RT-PCR in the 2 treatment groups before and after 2 mo of fish oil treatment (n ҃ 7) or placebo (n ҃ 7). Leptin and adiponectin gene expression (the main adipocytesecreted hormones) was not significantly affected in either group. Although the PAI-1 circulating concentration was decreased by the fish oil treatment, no significant changes could be detected in adipose tissue PAI-1 gene expression (data not shown). A series of genes encoding inflammation-related factors was also analyzed by real-time RT-PCR. The choice in selecting these genes was based first on the preliminary data of a cDNA

microarray analysis of adipose tissue from fish oil-treated patients only (baseline samples were plotted against 2-mo samples on the same slide). A detailed description of these genes along with their average fold changes are given in Table 5. Among these genes, we selected the gene encoding the matrix metalloprotease MMP9, which was significantly down-regulated by fish oil treatment. The MMPs are involved in matrix modeling and are required for macrophage infiltration. Moreover, the same gene (MM9) and other inflammation-related genes were found to be decreased by 2 mo of dietary fish oil supplementation in longterm insulin-resistant sucrose-fed rats (M Guerre-Millo, N Naour, Y Lombardo, K Clement, and S Rizkalla, unpublished observations, 2007). These genes included the cysteine protease cathepsin S, or CTSS; the chemoattractant gene plasminogen activator urokinase receptor, or PLAUR; the macrophage surface markers CD11b and CD18, which also make up the integrin; and the macrophage phagocytic activity marker CD68. Therefore, we analyzed the same genes by real-time RT-PCR in the 2 treatment groups before and after 2 mo of treatment. A treatment effect (fish oil versus placebo, P ҃ 0.02) was found for MMP9 with a significant decrease when comparing 2 mo with baseline

FISH OIL AND ADIPOSE TISSUE IN TYPE 2 DIABETES

(P 쏝 0.05). Because of a significant (P 쏝 0.05) treatment by time interaction, some genes were analyzed separately in each group and were found to be decreased or to tend to decrease by fish oil (2 mo versus baseline) but not by placebo (Figure 1): CTSS (P ҃ 0.13), PLAUR (P 쏝 0.05), CD11b (P 쏝 0.05), CD18 (P ҃ 0.13), and CD68 (P ҃ 0.11). By contrast, no significant changes in the monocyte chemotactic protein MCP-1 were observed. DISCUSSION

In the women with type 2 diabetes in the present study, 2 mo of a moderate dose of fish oil, which is commonly recommended in France, induced improvements in adiposity markers and atherogenic factors. These effects were seen with a decrease in plasma triacylglycerol even in nonhypertriacylglycerolemic subjects. Previously, however, this decrease had only been shown in hypertriacylglycerolemic subjects (4, 5, 26). The intervention had no significant effect on total cholesterol or LDL cholesterol. By contrast, HDL cholesterol increased in the fish oil group, in agreement with the results of a few previous studies (33–35). The elevated concentrations of HDL cholesterol and lower concentrations of triacylglycerol, and hence the reduced atherogenic index, induced by 2 mo of fish oil might play an important role in protecting patients with type 2 diabetes against the development of premature atherosclerosis (36, 37). Another marker of atherogenic risk, the plasma PAI-1 concentration (38), was also significantly reduced by fish oil treatment. In

FIGURE 1. Effect of 2 mo of treatment with fish oil or placebo on the expression of inflammation-related genes in subcutaneous adipose tissue. MMP9 gene expression showed a treatment effect (fish oil versus placebo, P ҃ 0.02). There was also a significant difference between the groups at baseline for MMP9. Because of the presence of a significant treatment by time interaction (using a multivariate analysis of repeated measurements) for CTSS, CD68, PLAUR, CD11b, and CD18, values between baseline and 2 mo were compared separately in each group by use of Student’s t test for paired data. Data are mean 앐 SEM for baseline (open bars) and 2 mo (black bars). AU, arbitrary units. *P 쏝 0.05, #P ҃ 0.13 for CTSS and 0.11 for CD68 versus baseline.

1677

the literature, the effects of nҀ3 PUFAs on PAI-1 activity or antigen concentrations are divided. In vitro, eicosapentaenoic acid significantly increases PAI-1 secretion from cultured endothelial cells (39).The results of several studies in healthy subjects have been controversial on the favorable or unfavorable effects of nҀ3 fatty acids on PAI-1 activity (40, 41). In contrast with our data, in subjects with type 2 diabetes, another study showed that PAI-1 activity was increased with nҀ3 PUFA supplements (42). This difference might be due to the high doses of nҀ3 PUFAs (3 g) used in the later studies, which resulted in a deterioration in fasting blood glucose concentrations and HbA1c that might in turn have upregulated PAI-1. The decreased plasma PAI-1 concentrations caused by fish oil treatment in our study were not associated with reduced PAI-1 gene expression in subcutaneous adipose tissue. This dissociation between plasma PAI-1 and PAI-1 gene expression agrees with the results of another study showing that a low-calorie diet reduced plasma PAI-1 but not PAI-1 expression in subcutaneous adipose tissue (43). Changes in PAI-1 gene expression in other adipose depots, most notably in visceral adipose tissue, or in other tissues might contribute to the modification of PAI-1 circulating concentrations. Another interesting aspect of the effects of fish oil shown in the present study is its ability to reduce adiposity. This study is the first intervention study to demonstrate a beneficial effect on adiposity markers in diabetic patients. Although body weight did not differ significantly between the 2 treatment groups, both total fat mass and adipocyte size in subcutaneous abdominal adipose tissue were significantly reduced by 3.5% and 6%, respectively, after fish oil treatment compared with placebo. However, visceral and subcutaneous adipose tissue at only one level, L4 –L5, measured with a single-slice CT scan, did not differ significantly between groups. This is not surprising, because the decrease in fat mass after 2 mo of treatment was due mainly to the difference in whole trunk fat and not only to the abdominal fat at this single level. It was recently shown that the use of 8-slice multidetector CT gives better results than does a single-slice CT scan (44). Therefore, the difference between visceral and subcutaneous adipose tissue could not be accurately calculated in our study. The findings of the present study on adiposity agree with studies in rodents. In humans, however, indirect evidence has been found in only 2 studies. One study showed that dietary counseling to substitute dietary saturated fat with polyunsaturated fat in a mixed group of subjects with type 2 diabetes, obese subjects, and healthy subjects resulted in a decrease in subcutaneous, but not abdominal or visceral, fat area (45). Another epidemiologic study showed a correlation between adipocyte size and the nҀ6 and nҀ3 fatty acid content in subcutaneous abdominal adipose tissue in a group of overweight patients who had undergone abdominal surgery (46). The present study points out the direct beneficial role of fish oil to lower adiposity in humans. Several mechanisms can be considered to understand why fish oil treatment compared with placebo induced a selective decrease in fat mass. One possible factor could be decreased energy intake. However, energy intake in the present study did not differ significantly during the 2 treatments. Even if the food diary method often leads to an underestimation of energy intake, the same method was used by the same subjects before and after treatment and thus can be compared. This is strengthened by the absence of a significant change in body weight. Alternatively, decreased fat absorption and increased gastric emptying and gut hormone responses could be implicated in the observed results.

1678

KABIR ET AL

Robertson et al (47) showed that ingestion of a meal rich in nҀ3 fatty acids resulted in rapid gastric emptying in humans. The authors attributed these results to the fact that nҀ3 fatty acids resulted in both a slower release of cholecystokinin and lower postprandial glucagon-like peptide-1, both of which have been implicated in the nutrient-induced slowing of gastric emptying. Unfortunately, we did not measure gastric emptying in our study. Additional factors, however, might be considered, such as increased lipolysis, which has been previously shown in rat models (7). However, the application of these results to our study at this stage (2 mo of fish oil treatment) is doubtful, because the microarray analysis for the fish oil group in the present study did not show significant modifications in the concerned enzymes. Another possible mechanism might be enhanced energy expenditure. Recently, Mostad et al (48) showed that nҀ3 fatty acids alter the proportion of carbohydrate and fat utilization without changing total resting energy expenditure as assessed by indirect calorimetry. There was a relative shift in substrate utilization, because the fish oil group oxidized more fat than carbohydrate after 9 wk than did the placebo group. This hypothesis is likely to be implicated in the results of the present study. Whatever the mechanism, decreasing total adiposity might increase insulin sensitivity and reduce cardiovascular disease risk factors as has been found in rodents (8, 49, 50). In the present patients with type 2 diabetes, insulin sensitivity and plasma glucose control (plasma glucose and Hb1Ac) were neither deteriorated nor ameliorated by moderate doses of nҀ3 PUFAs (1.8 g/d). In contrast, the recent study of Mostad et al (48) showed decreased insulin sensitivity in patients with type 2 diabetes after high doses of nҀ3 PUFAs (5.9 g/d), which agrees with previous results showing the deleterious effects of high doses of fish oil on plasma glucose control (4, 5). Therefore, moderate doses of nҀ3 PUFAs did not reduce insulin sensitivity (mainly measured by HOMA, and in a pilot clamp study) or plasma glucose control in patients with type 2 diabetes. The relation between decreasing adiposity and the beneficial effect of fish oil on cardiovascular disease risks could be mediated by a decrease in the inflammatory markers secreted by adipose tissue (51). The current study was not able to demonstrate any alteration in concentrations of systemic inflammatory factors in response to fish oil treatment in patients with type 2 diabetes. Nor was there an intrinsic influence of fish oil treatment on the secretory capacity of isolated adipocytes in vitro. By contrast, fish oil treatment significantly reduced or tended to reduce the expression of some inflammation-related genes in the subcutaneous adipose tissue of patients with type 2 diabetes. Interestingly, adipose expression of some of the chemotactic genes (PLAUR) and markers for macrophage surface (CD11b) and phagocytic activity (CD68) were positively correlated with adiposity markers (adipocyte size and whole body fat mass percentage) at baseline. The parallelism between the downregulation of these genes and the reduction in adiposity and adipocyte diameter by fish oil treatment suggests a positive relation between adipose cell size and adipose tissue inflammation, which agrees with recent observations in obese subjects (17, 52). The results of the present study in women with type 2 diabetes are strengthened by 2 studies published in rodents during the preparation of the manuscript. Those studies showed that 6 wk of nҀ3 PUFA supplementation prevents adipose tissue inflammation induced by a high-fat diet (53) and that endogenously biosynthesized nҀ3 PUFAs in transgenic mice reduce inflammation

and tissue injury in colitis (54). Additionally, in a recent study from our laboratory, we showed that the presence of fish oil in the diet of long-term insulin-resistant, sucrose-fed rats decreased adipocyte diameter (8) and significantly reduced several inflammation-related genes: MMP9, CTSS, PLAUR, CD11b and TNF-␣ (Guerre-Millo M, Naour N, LombardoY, Clement K, Rizkalla S, unpublished observations, 2007). These studies suggest that reducing adiposity with fish oil could decrease adipose tissue inflammation and macrophage infiltration. This hypothesis warrants further evaluation, in particular, the modification of these biomolecules at the protein level and the histochemical determination of macrophage infiltration of adipose tissue during long-term fish oil treatment. In conclusion, 2 mo of a moderate dose of fish oil in women with controlled type 2 diabetes decreased adiposity and the expression of some inflammation-related genes in adipose tissue, without any beneficial or deleterious effect on whole-body insulin sensitivity. Additional beneficial effects of this treatment on cardiovascular disease risk factors could be attributed to a decreased atherogenic index and PAI-1 concentration in keeping with a triacylglycerol-lowering capacity. The beneficial effects of nҀ3 PUFAs can be linked to local blunting of adipose tissue inflammation. More long-term studies are needed to identify and understand the relation between fish oil and risk factors in diabetes. We thank the medical and nonmedical staff of the Department of Diabetes for their assistance. We are grateful to Aude Rigoire for dietary counseling of the patients and dietary data analysis. We express our gratitude to Nicolas Zamaria for the opportunity to perform the fatty acids profile in plasma phospholipids in his laboratory, Amine Duval for CT scan realization, Josette Boillot for technical assistance during the different stages of this study, and Marmar Kabir and Florence Combes for gene data management. We also thank Pierre Fabre Médicament (Castre, France) for offering the fish oil and placebo capsules. The contributions of the authors were as follows—SWR: conceived the overall study, was responsible for planning the study, and drafted the manuscript; MK: was responsible for adipokine and cytokine analysis, adipocyte culture in vitro, microarray and real-time RT-PCR analysis, and an important part of the manuscript writing; GS and VP: the follow-up of patients, performing the clamp procedures, blood glucose and lipid measurements, and the statistical analysis; NN: the gene expression of the inflammation-related genes and their statistical analysis; HV, EM, and SR: contributed to designing the microarray analysis; MG-M and KC: contributed to evaluating the results, designing the gene expression section, and participating in manuscript writing; AQ-B: determining adipocyte size and advising on adipocyte culture design in vitro. All contributors helped with the revision of the paper. None of the authors had a conflict of interest.

REFERENCES 1. Nestel PJ, Connor WE, Reardon MF, Connor S, Wong S, Boston R. Suppression by diets rich in fish oil of very low density lipoprotein production in man. J Clin Invest 1984;74:82–9. 2. Friday KE, Childs MT, Tsunehara CH, Fujimoto WY, Bierman EL, Ensinck JW. Elevated plasma glucose and lowered triglyceride levels from omega-3 fatty acid supplementation in type II diabetes. Diabetes Care 1989;12:276 – 81. 3. Borkman M, Chisholm D, Furler S. Effects of fish oil supplementation on glucose and lipid metabolism in NIDDM. Diabetes 1989;38:1314 –9. 4. Montori V, Farmer A, Wollan P, Dinneen S. Fish oil supplementation in type 2 diabetes: a quantitative systematic review. Diabetes Care 2000; 23:1407–15. 5. Friedberg C, Janssen M, Heine R, Grobbee D. Fish oil and glycemic control in diabetes. A meta-analysis. Diabetes Care 1998;21:494 –500. 6. Luo J, Rizkalla SW, Boillot J, et al. Dietary (n–3) polyunsaturated fatty acids improve adipocyte insulin action and glucose metabolism in insulin resistant rats: relation to membrane fatty acids. J Nutr 1996;126: 1951– 8.

FISH OIL AND ADIPOSE TISSUE IN TYPE 2 DIABETES 7. Peyron-Caso E, Quignard-Boulangé A, Laromiguière M, et al. Dietary fish oil increases lipid mobilization but does not decrease lipid storagerelated enzyme activities in adipose tissue of insulin-resistant, sucrosefed rats. J Nutr 2003;133:2239 – 43. 8. Rossi A, Lombardo Y, Lacorte J, et al. Dietary fish oil positively regulates plasma leptin and adiponectin levels in sucrose-fed, insulin-resistant rats. Am J Physiol Regul Integr Comp Physiol 2005;289:R486 –94. 9. Flachs P, Horakova O, Brauner P, et al. Polyunsaturated fatty acids of marine origin upregulate mitochondrial biogenesis and induce betaoxidation in white fat. Diabetologia 2005;48:2365–75. 10. Storlien LH, Kraegen EW, Chisholm DJ, Ford GL, Bruce DG, Pascoe WS. Fish oil prevents insulin resistance induced by high fat feeding. Science 1987;237:885– 8. 11. Peyron-Caso E, Taverna M, Guerre-Millo M, et al. Dietary (n-3) polyunsaturated fatty acids up regulate plasma leptin in insulin resistant rats. J Nutr 2002;132:2235– 40. 12. Neschen S, Morino K, Rossbacher J, et al. Fish oil regulates adiponectin secretion by a peroxisome proliferator-activated receptor-gammadependent mechanism in mice. Diabetes 2006;55:924 – 8. 13. Flachs P, Mohamed-Ali V, Horakova O, et al. Polyunsaturated fatty acids of marine origin induce adiponectin in mice fed a high-fat diet. Diabetologia 2006;49:394 –7. 14. Couet C, Delarue J, Ritz P, Antoine J, Lamisse F. Effect of dietary fish oil on body fat mass and basal fat oxidation in healthy adults. Int J Obes Relat Metab Disord 1997;21:637– 43. 15. Bouloumie A, Curat C, Sengenes C, Lolmede K, Miranville A, Busse R. Role of macrophage tissue infiltration in metabolic diseases. Curr Opin Clin Nutr Metab Care 2005;8:347–54. 16. Clement K, Viguerie N, Poitou C, et al. Weight loss regulates inflammation-related genes in white adipose tissue of obese subjects. FASEB J 2004;18:1657– 69. 17. Cancello R, Henegar C, Viguerie N, et al. Reduction of macrophage infiltration and chemoattractant gene expression changes in white adipose tissue of morbidly obese subjects after surgery-induced weight loss. Diabetes 2005;54:2277– 86. 18. Cancello R, Tordjman J, Poitou C, et al. Increased infiltration of macrophages in omental adipose tissue is associated with marked hepatic lesions in morbid human obesity. Diabetes 2006;55:1554 – 61. 19. Curat C, Wegner V, Sengenes C, et al. Macrophages in human visceral adipose tissue: increased accumulation in obesity and a source of resistin and visfatin. Diabetologia 2006;49:744 –7. 20. Xu H, Barnes G, Yang Q, et al. Chronic inflammation in fat plays a crucial role in the development of obesity-related insulin resistance. J Clin Invest 2003;112:1821–30. 21. Stulnig T, Zeyda M. Immunomodulation by polyunsaturated fatty acids: impact on T-cell signaling. Lipids 2004;39:1171–5. 22. Stehr S, Heller A. Omega-3 fatty acid effects on biochemical indices following cancer surgery. Clin Chim Acta 2006;373:1– 8. 23. Hughes D. In vitro and in vivo effects of n-3 polyunsaturated fatty acids on human monocyte function. Proc Nutr Soc 1998;57:521–5. 24. Kuk J, Lee S, Heymsfield S, Ross R. Waist circumference and abdominal adipose tissue distribution: influence of age and sex. Am J Clin Nutr 2005;81:1330 – 4. 25. Piatti PM, Monti LD, Baruffaldi L, et al. Effects of an acute increase in plasma triglyceride levels on glucose metabolism in man. Metabolism 1995;44:883–9. 26. Luo J, Rizkalla SW, Vidal H, et al. Moderate intake of n-3 fatty acids for 2 months has no detrimental effect on glucose metabolism and could ameliorate the lipid profile in type 2 diabetic men. Results of a controlled study. Diabetes Care 1998;21:717–24. 27. Solsman D, Casez J, Pichard C, et al. Assessment of whole-body composition with dual-energy X-ray absorptiometry. Radiology 1992;185:593– 8. 28. Matthews D, Hosker J, Rudenski A, Naylor B, Treacher D, Turner RC. Homeostasis model assessment: insulin resistance and beta-cell function from fasting plasma glucose and insulin concentrations in man. Diabetologia 1985;28:412–9. 29. Rome S, Clement K, Rabasa-Lhoret R, et al. Microarray profiling of human skeletal muscle reveals that insulin regulates approximately 800 genes during a hyperinsulinemic clamp. J Biol Chem 2003;278: 18063– 8. 30. Eisen M, Spellman P, Brown P, Botstein D. Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci U S A 1998; 95:14863– 8. 31. Tusher V, Tibshirani R, Chu G. Significance analysis of microarrays

32.

33. 34. 35.

36. 37. 38. 39. 40. 41. 42. 43. 44.

45. 46.

47. 48.

49. 50. 51. 52. 53. 54.

1679

applied to the ionizing radiation response. Proc Natl Acad Sci U S A 2001;98:5116 –21. Dobiasova M, Frohlich J. The plasma parameter log (TG/HDL-C) as an atherogenic index: correlation with lipoprotein particle size and esterification rate in apoB-lipoprotein-depleted plasma (FER(HDL)). Clin Biochem 2001;34:583– 8. Rillaerts E, Engelmann G, Van camp K, De Leeuw L. Effect of omega-3 fatty acids in diet of type I diabetic subjects on lipid values and hemorheological parameters. Diabetes 1989;38:1412– 6. Kesavulu M, Kameswararao B, Apparao C, Kumar E, Harinarayan C. Effect of omega-3 fatty acids on lipid peroxidation and antioxidant enzyme status in type 2 diabetic patients. Diabetes Metab 2002;28:20 – 6. Sirtori C, Crepaldi G, Manzato E, et al. One-year treatment with ethyl esters of n-3 fatty acids in patients with hypertriglyceridemia and glucose intolerance: reduced triglyceridemia, total cholesterol and increased HDL-C without glycemic alterations. Atherosclerosis 1998;137:419 –27. Gordon D, Probstfield J, Garrison R, et al. High-density lipoprotein cholesterol and cardiovascular disease. Four prospective American studies. Circulation 1989;79:8 –15. Tai E, Emmanuel S, Chew S, Tan B, Tan C. Isolated low HDL cholesterol: an insulin-resistant state only in the presence of fasting hypertriglyceridemia. Diabetes 1999;48:1088 –92. Juhan-Vague I, Alessi M. PAI-1, obesity, insulin resistance and risk of cardiovascular events. Thromb Haemost 1997;78:656 – 60. Nilsson L, Banfi C, Diczfalusy U, Tremoli E, Hamsten A, Eriksson P. Unsaturated fatty acids increase plasminogen activator inhibitor-1 expression in endothelial cells. Arterioscler Thromb Vasc Biol 1998;18:1679 – 85. Emeis J, Van Houwelingen A, Van den Hoogen C, Hornstra G. A moderate fish intake increases plasminogen activator inhibitor type-1 in human volunteers. Blood 1989;74:233–7. Hansen J, Grimsgaard S, Nordoy A, Bonaa K. Dietary supplementation with highly purified eicosapentaenoic acid and docosahexaenoic acid does not influence PAI-1 activity. Thromb Res 2000;98:123–32. Boberg M, Pollare T, Siegbahn A, Vessby B. Supplementation with n-3 fatty acids reduces triglycerides but increases PAI-1 in non-insulindependent diabetes mellitus. Eur J Clin Invest 1992;22:645–50. Bastard J, Vidal H, Jardel C, et al. Subcutaneous adipose tissue expression of plasminogen activator inhibitor-1 gene during very low calorie diet in obese subjects. Int J Obes Relat Metab Disord 2000;24:70 – 4. Maurovich-Horvat P, Massaro J, Fox C, Moselewski F, O’Donnell C, Hoffmann U. Comparison of anthropometric, area- and volume-based assessment of abdominal subcutaneous and visceral adipose tissue volumes using multi-detector computed tomography. Int J Obes (Lond) 2007;31:500 – 6. Summers L, Fielding B, Bradshaw H, et al. Substituting dietary saturated fat with polyunsaturated fat changes abdominal fat distribution and improves insulin sensitivity. Diabetologia 2002;45:369 –77. Garaulet M, Hernandez-Morante J, Lujan J, Tebar FJ, Zamora S. Relationship between fat cell size and number and fatty acid composition in adipose tissue from different fat depots in overweight/obese humans. Int J Obes 2006;30:899 –905. Robertson M, Jackson K, Fielding B, Morgan L, Williams C, Frayn K. Acute ingestion of a meal rich in n-3 polyunsaturated fatty acids results in rapid gastric emptying in humans. Am J Clin Nutr 2002;76:232– 8. Mostad I, Bjerve K, Bjorgaas M, Lydersen S, Grill V. Effects of n-3 fatty acids in subjects with type 2 diabetes: reduction of insulin sensitivity and time-dependent alteration from carbohydrate to fat oxidation. Am J Clin Nutr 2006;84:540 –50. Hill JO, Peters JC, Lin D, Yakubu F, Greene H, Swift L. Lipid accumulation and body fat distribution is influenced by type of dietary fat fed to rats. Int J Obes 1993;17:223–36. Parrish C, Pathy D, Angel A. Dietary fish oils limit adipose tissue hypertrophy in rats. Metabolism 1990;39:217–9. Trayhurn P, Wood I. Signalling role of adipose tissue: adipokines and inflammation in obesity. Biochem Soc Trans 2005;33:1078 – 81. Weisberg S, McCann D, Desai M, Rosenbaum M, Leibel R, Ferrante AJ. Obesity is associated with macrophage accumulation in adipose tissue. J Clin Invest 2003;112:1796 – 808. Todoric J, Loffler M, Huber J, et al. Adipose tissue inflammation induced by high-fat diet in obese diabetic mice is prevented by n-3 polyunsaturated fatty acids. Diabetologia 2006;49:2109 –19. Hudert C, Weylandt K, Lu Y, et al. Transgenic mice rich in endogenous omega-3 fatty acids are protected from colitis. Proc Natl Acad Sci U S A 2006;103:11276 – 81.

Effect of daily or weekly multiple-micronutrient and iron foodlike tablets on body iron stores of Indonesian infants aged 6 –12 mo: a double-blind, randomized, placebo-controlled trial1–5 Maria Wijaya-Erhardt, Juergen G Erhardt, Juliawati Untoro, Elvina Karyadi, Lindawati Wibowo, and Rainer Gross ABSTRACT Background: There is still uncertainty about the best procedure to alleviate iron deficiency. Additionally more reliable methods are needed to assess the effect of iron intervention. Objective: We examined the efficacy of daily iron (10 mg), daily and weekly multiple-micronutrient supplementation (10 and 20 mg Fe, respectively) in improving body iron stores of Indonesian infants. Design: Infants aged 6 –12 mo were randomly allocated to 1 of 4 groups: daily multiple-micronutrients (DMM) foodlike tablets (foodLETs), weekly multiple-micronutrient (WMM) foodLETs, daily iron (DI) foodLETs, or daily placebo. Hemoglobin, ferritin, transferrin receptors, and C-reactive protein data were obtained at baseline and 23 wk. Results: Body iron estimated from the ratio of transferrin receptors to ferritin was analyzed for 244 infants. At baseline, mean iron stores (0.5 앐 4.1 mg/kg) did not differ among the groups, and 45.5% infants had deficits in tissue iron (body iron 쏝 0). At week 23, the group DI had the highest increment in mean body iron (4.0 mg/kg), followed by the DMM group (2.3 mg/kg; P 쏝 0.001 for both). The iron stores in the WMM group did not change, whereas the mean body iron declined in the daily placebo group (Ҁ2.2 mg/kg; P 쏝 0.001). Compared with the daily placebo group, the DMM group gained 4.55 mg Fe/kg, the DI group gained 6.23 mg Fe/kg (both P 쏝 0.001), and the WMM group gained 2.54 mg Fe/kg (P ҃ 0.001). Conclusions: When compliance can be ensured, DI and DMM foodLETs are efficacious in improving and WMM is efficacious in maintaining iron stores among Indonesian infants. Am J Clin Nutr 2007;86:1680 – 6. KEY WORDS Iron deficiency, infants, multiple micronutrients, foodlike tablets, foodLETs, Indonesia, body iron stores INTRODUCTION

Iron deficiency (ID) in the 6 –24-mo-old population impairs the normal mental development of 40 – 60% of the developing world’s infants. ID is highly prevalent in Indonesia, which has an estimated 38% prevalence of iron deficiency anemia (IDA) in children 쏝5 y old (1). Supplementation with iron tablets is the most widely used approach for controlling the global problem of ID and IDA. Because several micronutrients can improve the hemoglobin response to iron, it is reasonable to assume that multiple micronutrients would be more effective in reducing anemia than would

1680

iron alone. One important reason for the relative lack of programmatic success with iron supplements has been the perceived need for these supplements to be taken daily over relatively long periods. For this reason, and because of concern about the potential toxicity and intolerance of daily supplementation, there has been interest in the efficacy of weekly iron supplementation as compared with daily supplementation (2). Anemia measured by hemoglobin concentration is used as a proxy indicator of ID. However, with regard to iron status, this is neither a specific measure— because anemia also may be caused by malaria, intestinal parasites, or other factors (3–5)—nor a sensitive measure— because ID can exist without anemia (4, 5). Although the plasma ferritin (PF) concentration is one of the most widely used indicators of iron status, it is known to be affected by several factors, including infection and inflammation. In contrast, some studies show that serum concentrations of soluble transferrin receptors (TfRs) accurately reflect iron status in the presence of inflammation (6 –12). The ratio of TfR to PF is a more sensitive and specific indicator of iron status than is either measurement alone (5, 8, 11). Quantitative measurements of body iron based on the ratio of TfR to PF were shown to measure absorption of the added iron in trials of iron supplementation in pregnant Jamaican women and iron fortification in Vietnamese women (5). We aimed to evaluate the role of body iron measurements in comparing the efficacy of daily iron (DI) to those of daily and weekly multiple-micronutrient (WMM) foodlike tablets (foodLETs) in improving the iron status of Indonesian infants 6 –12 mo old. 1 From the Southeast Asian Ministers of Education Organization–Tropical Medicine Regional Center for Community Nutrition, University of Indonesia, Jakarta, Indonesia (MW-E, JU, EK, and LW); the Institute of Biological Chemistry and Nutrition, University of Hohenheim, Stuttgart, Germany (JGE); and the Deutsche Gesellschaft fu¨r Technische Zusammenarbeit (German Technical Cooperation), Eschborn, Germany (RG). 2 Any opinion, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the views of the supporting agency. 3 Rainer Gross died on 30 September 2006. 4 Supported by UNICEF (New York, NY). 5 Reprints not available. Address correspondence to M Wijaya-Erhardt, SEAMEO-TROPMED Regional Center for Community Nutrition, University of Indonesia, Jalan Salemba Raya 6, Jakarta 10430, Indonesia. E-mail: [email protected]. Received May 11, 2007. Accepted for publication July 23, 2007.

Am J Clin Nutr 2007;86:1680 – 6. Printed in USA. © 2007 American Society for Nutrition

SUPPLEMENTATION AND BODY IRON STORES OF INFANTS SUBJECTS AND METHODS

The present study was part of a multicenter study of International Research on Infant Supplementation (IRIS) trial that was conducted in populations from 4 quite different countries: Indonesia, Peru, South Africa, and Vietnam. The aim of the IRIS study was to investigate whether multiple micronutrient supplementation can prevent growth faltering, anemia, and micronutrient deficiency during infancy in Indonesia. Study location and population The details about participants, study design, and other results from the Indonesian part of the multicountry study were reported previously (13, 14). Briefly, we conducted the study from June to December 2000 in 12 villages in the Salam subdistrict and in 6 villages in the Ngluwar subdistrict, both of which are located in Magelang district. Magelang lies in the center of Java, the most populous island of Indonesia. We selected the study area according to the requirements of the IRIS protocol—ie, that the prevalence of anemia (hemoglobin: 쏝 110 g/L) and of vitamin A deficiency (serum retinol: 쏝 0.7 ␮mol/L) should be 욷30% in the population 쏝5 y old. Deficiencies of several micronutrients were found in our subjects: 58.1% were anemic, 34.5% had PF concentrations 쏝 12 ␮g/L, 10.4% had plasma zinc concentrations 쏝 10.7 ␮mol/L, and 24.3% had plasma retinol concentrations 쏝 0.7 ␮mol/L (13, 15). Approximately 12% of the subjects were underweight, and 8% were stunted (15). We conducted a screening study in 356 infants 6 –12 mo old from a list of infants in the study area and excluded those with premature birth (쏝37 wk gestation), low birth weight (쏝2500 g), severe wasting (쏝Ҁ3 z scores), severe anemia (hemoglobin: 쏝 80 g/L), or fever (쏜39 °C) on the day of blood sampling. We gave additional WMM foodLETs to the placebo group for 3 mo after the end of the supplementation trial. The purpose and procedures of the study were explained to the parents at enrollment, and only the infants whose parents who gave written informed consent were recruited. The IRIS trial was guided by the ethical considerations recommended by the Council for International Organizations of Medical Sciences (16). The Ethical Committee for Studies on Human Subjects, Faculty of Medicine, University of Indonesia approved the study protocol. Procedures In advance of recruitment, we used a simple computer program (a random number generator) for the randomization process, which was done by household. At baseline, we then enrolled 284 (80%) eligible infants and randomly allocated them to 4 groups as follows: daily multiple micronutrients (DMM; n ҃ 72), WMM (n ҃ 70), DI (n ҃ 72), or daily placebo (DP; n ҃ 70). Through the use of coding, the subjects’ families and the investigators were blinded to the specific composition of the foodLETs assigned. One family had twin boys who fulfilled the requirements to participate in the study; therefore, both boys were included and treated as separate cases, but they were allocated to the same intervention group. The allocation codes 1, 2, 3, and 4 were kept centrally by UNICEF New York and opened at the end of the study, before statistical analyses (13). The multiple-micronutrient supplement and placebo were produced in the form of foodLETs and were provided in blister packs of 7 tablets. The micronutrient content of each foodLET had been formulated according to the daily Recommended Nutrient Intake

1681

for infants 1–2 y old, with the exception of zinc, which was the equivalent of 1 Recommended Nutrient Intake for children 6 –12 mo old: 375 ␮g RE vitamin A (as retinyl acetate), 5 ␮g vitamin D, 6 mg ␣-tocopherol equivalents of vitamin E (as ␣-tocopherol), 10 ␮g vitamin K, 35 mg vitamin C, 0.5 mg thiamine, 0.5 mg riboflavin, 0.5 mg vitamin B-6, 0.9 ␮g vitamin B-12, 6 mg niacin (as niacinamide), 150 ␮g folate, 10 mg iron (as ferrous fumarate), 5 mg zinc (as zinc gluconate), 0.6 mg copper (as cupric gluconate), and 59 ␮g iodine (as potassium iodide). The dose of the WMM (ie, weekly) foodLETs was double that of the DMM. DI foodLETs contained 10 mg iron (as ferrous sulfate), whereas DP foodLETs contained no added micronutrients. The ingredients in all foodLETs were similar (ie, milk solids, confectionery sugar, fructose, flavor, citric acid, calcium carbonate, and magnesium stearate). Milk solids (1100 mg) were added to each foodLET (total weight: 2800 mg) to prevent chemical interactions between the micronutrients. Roche Laboratories (Nutley, NJ) developed the product, and Hersil (Lima, Peru) produced the crushable and dissoluble foodLETs (14, 17, 18). Blisters were labeled with the printed subjects’ names and allocation code only. Each supply of 7 tablets was wrapped in identical coded blister packs, and all foodLETs had the same taste, color, and flavor. The foodLETs were given daily at home— on days 1– 6, the consumption of the foodLET was under the close supervision of a trained fieldworker, and, on day 7, it was under the supervision of the child’s mother. Thus, compliance was recorded directly for the first 6 d, and that on day 7 was recorded on the next day’s visit. Twenty-three infants did not complete the study: during the study period, 12 infants’ families moved to another area (n ҃ 2, 5, 2, and 3 in the DMM, WMM, DMM, and DP groups, respectively), 9 discontinued the trial (n ҃ 4, 3, and 2 in the DMM, WMM, and DP groups, respectively), and 2 refused the second blood collection (n ҃ 1 each in the DMM and WMM groups) and were not included in the data analysis. The losses to follow-up were equally distributed across the 4 treatment groups (13). Infants also received vitamin A as part of a national program; those 욷12 mo old were given 200 000 IU vitamin A, and those 쏝12 mo old were given 100 000 IU. We collected all information by using a structured questionnaire according to survey methods (19). We measured the weight and length of infants monthly and the weight and height of mothers at baseline; the weight was measured to the nearest 0.1 kg with the use of an electronic weighing scale (model 770; SECA, Hamburg, Germany). We measured infant length on a length board (model 210; SECA) and maternal height to the nearest 0.1 cm by using a standing height measurement microtoise. The mothers’ perception of the intervention was evaluated in mid-November 2000 by administering a questionnaire that asked about the mothers’ perceptions of the effects and side effects of the supplementation during the intervention. Blood samples were taken from infants at baseline and at 23 wk of supplementation. Because of a Muslim festival celebrated by most of the subjects studied, we deviated in one instance from the protocol, reducing the study period by 2 wk. However, we continued giving the foodLETs to the infants up to week 25. A 2-mL blood sample was withdrawn via venipuncture into vacuette heparin tubes using a butterflyplus-luer adapter (Greiner, Kremsmuenster, Austria). The tubes were put in a cooled styropore box with coolpacks. We measured the hemoglobin concentration by using the cyanomethemoglobin method (# 4010l; Boehringer, Mannheim, Germany) and centrifuged (Hettich, Tuttlingen, Germany) the whole blood

1682

WIJAYA-ERHARDT ET AL

within 6 h at 3600 rpm for 15 min at room temperature at the Muntilan General Hospital Laboratory, Magelang. The plasma was transferred into two 500-␮L Eppendorf cups and was kept frozen at Ҁ76 °C before being shipped to Germany in December 2000. The blood samples were analyzed in the Micronutrient Laboratory of the Institute of Biological Chemistry and Nutrition at the University of Hohenheim (Stuttgart, Germany). PF and C-reactive protein (CRP) were measured by using an enzymelinked immunosorbent assay (20). Originally, the TfR was not included in the protocol; at the Southeast Asian Ministers of Education Organization–Tropical Medicine Regional Center for Community Nutrition (Jakarta, Indonesia) in 2004, it was analyzed in 259 samples with the use of the same method (20). A hemoglobin concentration 쏝 110 g/L was used as a cutoff for defining anemia, a PF concentration 쏝 12 ␮g/L was used as the cutoff for ID, and a hemoglobin concentration 쏝 110 g/L and a PF 쏝 12 ␮g/L were used as the cutoff for IDA (21, 22). Tissue ID was defined as a TfR concentration 쏜 8.5 mg/L and iron overload with a PF concentration 쏜 400 ␮g/L (22). The cutoff for inflammation was a plasma CRP concentration of 12 mg/L (23). A body iron store was estimated by using the following equation (5, 24):

Body iron (mg/kg) ⫽ ⫺[log(TfR/PF ratio) ⫺ 2.8229]/0.1207 (1) Positive values indicate the amount of iron in stores (body iron 쏜 0), and negative values indicate the deficit in tissue iron (body iron 쏝 0). Statistical analysis We used the Kolmogorov-Smirnov one-sample test to check the normality of the data. Data were reported as means 앐 SDs for normally distributed variables. PF and TfR concentrations were logarithmically transformed and reported as geometric means and 95% CIs. Analysis of variance or Student’s t test was used to examine differences in means, whereas the chi-square test was used to examine differences in proportions. A 2-factor analysis of variance was used to ascertain the time ҂ treatment interaction for PF, TfR, and body iron. If a significant time ҂ treatment interaction was found, comparisons between groups were made by using Scheffe’s multiple comparisons test. To test the difference between baseline and week 23 of the intervention, the paired t test was used for continuous variables. Generalized estimating equations with treatment ҂ time interaction were used to assess the changes in proportion among groups. When treatment ҂ time interaction was significant, we used McNemar’s test for significant changes within groups. Correlation coefficients were calculated by using Pearson’s correlation test. We used SPSS software (version 11.5; SPSS Institute, Chicago, IL) and STATA software (version 9.0; Stata Corp, College Station, TX) for statistical analyses and entered the anthropometry data into the World Health Organization Anthro 2005 database (25). P 쏝 0.05 was considered to be significant (2-tailed). RESULTS

Of the 261 infants who completed the study, the sample from 1 infant was not measured for PF and that from another infant was not measured for TfR (in both cases, because of a low volume of plasma); 1 blood sample was missing. Fifteen infants had plasma

CRP concentrtions 쏜 12 mg/L at baseline or at week 23, which indicated inflammation; those infants were excluded from the statistical analysis for PF and body iron stores. A detailed description of the study profile was presented elsewhere (13). The baseline characteristics of the 4 groups did not differ significantly (Table 1). The average (앐 SD) household size was 5.0 앐 1.6 persons, and the mean number of children 쏝 5 y age was 1.2 앐 0.4. Most (쏜98%) households had electricity, either directly by public supply (173/244) or via a neighbor (67/244). The mean body mass index (in kg/m2) of mothers was normal at 21, and 28 (12%) were overweight (body mass index 쏜 25). Among all infants, 60 (25%) had IDA (hemoglobin: 쏝 110 g/L and PF 쏝 12 ␮g/L); 83 (34%) had both a low PF concentration (쏝12 ␮g/L) and an elevated TfR concentration (쏜8.5 mg/L); 204 (79%) had tissue ID (TfR 쏜 8.5 mg/L). One hundred thirty-three (55%) infants had body iron stores, and the other 111 infants (45.5%) had a mean tissue iron deficit [3.5 and Ҁ3.1 mg/kg, respectively (95% CI for the difference: Ҁ7.3, Ҁ6.1)]. Boys (128/244) on average had a tissue iron deficit, whereas girls (116/244) had body iron stores [Ҁ0.3 and 1.4 mg/kg, respectively (95% CI: Ҁ2.8, Ҁ0.7)]. None of these values were indicative of iron overload (PF 쏜 400 ␮g/L) at the beginning or the end of the study. Before treatment, there were no significant differences among the 4 groups in iron status (PF, TfR, and body iron) (Table 2). The geometric means of PF and TfR were 17.44 ␮g/L (95% CI: 15.38, 19.77 ␮g/L) and 10.08 mg/L (9.81, 10.35 mg/L), respectively, and the mean (앐 SD) of iron stores was 0.5 앐 4.1 mg/kg. During the 23-wk trial, the DMM and DI groups had significantly higher mean PF and significantly lower TfR concentrations than did the WMM and DP groups. The proportion of tissue ID (TfR 쏜 8.5 mg/L) decreased from 72% to 54% and from 75% to 45% in the DMM and DI groups, respectively, and remained unchanged in the WMM and DP groups (from 85% to 80% and from 83% to 85%, respectively). Time ҂ treatment interactions were significant for the proportion of infants with TfR concentrations 쏜 8.5 mg/L (P 쏝 0.001). The mean body iron increased 2.3 mg/kg in the DMM group and 4.0 mg/kg in the DI group (P 쏝 0.001 for both). Thus, the average daily iron gain was 0.015 and 0.025 mg/kg in the DMM and the DI group, respectively. In the WMM group, body iron stores did not change significantly (0.3 mg/kg; P ҃ 0.468). In the DP group, the mean body iron declined 2.2 mg/kg (P 쏝 0.001); this average daily iron loss was 0.014 mg/kg. Compared with the mean iron absorption in the DP group, that in the DMM group over 23 wk was 4.55 mg/kg or 41 mg total body iron, which represented the absorption of 2.5% of the 1610 mg Fe that the DMM foodLET provided; the DI group gained 6.23 mg Fe/kg or 55 mg total body iron, which represented the absorption of 3.4% of the 1610 mg Fe that the DI foodLET provided (P 쏝 0.001 for both). Infants in the WMM group gained 2.54 mg Fe/kg or 22 mg total body iron during 23 wk (P ҃ 0.001), which represented the absorption of 4.8% of the 460 mg Fe that the WMM foodLET provided. The mean iron absorption in the DMM and DI groups was significantly (P 쏝 0.01) higher than that in the WMM group. By the end of the study, the proportion of infants who had a tissue iron deficit (body iron 쏝 0) had decreased significantly in the DMM and DI groups, but the proportion in the WMM group did not change significantly (Table 3). The proportion of infants with body iron deficit in the DP group increased significantly (P

1683

SUPPLEMENTATION AND BODY IRON STORES OF INFANTS TABLE 1 Baseline characteristics1 Treatment groups

Variable Maternal characteristics Age (y) Weight (kg) Height (cm) BMI (kg/m2) 쏝 18.5 [n (%)] Delivery assisted by traditional birth attendant [n (%)] 쏜 3 Antenatal cares during pregnancy [n (%)] Infant characteristics Weight (kg) Length (cm) BMI-for-age z score Received colostrum [n (%)] Currently breastfed [n (%)] IDA [n (%)]4 Tissue iron deficit [n (%)]5

DMM (n ҃ 61)

WMM (n ҃ 58)

DI (n ҃ 65)

DP (n ҃ 60)

All subjects (n ҃ 244)

28.7 앐 5.72 48.4 앐 8.33 150.5 앐 4.9

29.3 앐 6.8 49.0 앐 9.0 149.4 앐 5.3

28.2 앐 4.6 47.0 앐 6.3 150.6 앐 5.1

28.8 앐 5.2 48.5 앐 8.2 149.4 앐 5.5

28.7 앐 5.6 48.2 앐 7.9 150.0 앐 5.2

21.3 앐 3.23 12 (20)3 16 (26) 56 (92)

21.9 앐 3.7 5 (9) 13 (22) 50 (86)

20.7 앐 2.6 13 (20) 8 (12) 61 (94)

21.7 앐 3.1 7 (12) 25 (25) 58 (97)

21.4 앐 3.2 37 (15) 52 (21) 225 (92)

7.9 앐 1.0 69.6 앐 3.1 Ҁ0.51 앐 1.0 53 (91) 49 (85) 12 (21) 24 (41)

7.9 앐 1.0 69.8 앐 2.9 Ҁ0.55 앐 1.2 57 (88) 63 (97) 18 (28) 27 (42)

7.9 앐 1.1 69.0 앐 3.2 Ҁ0.38 앐 0.9 54 (90) 60 (100) 18 (30) 34 (57)

7.9 앐 1.0 69.6 앐 3.1 Ҁ0.48 앐 1.0 218 (89) 229 (94) 60 (25) 111 (46)

8.0 앐 1.0 69.8 앐 3.2 Ҁ0.46 앐 1.0 54 (89) 57 (93) 12 (20) 26 (43)

1 DMM, daily multiple micronutrients; WMM, weekly multiple micronutrients; DI, daily iron; DP, daily placebo; IDA, iron deficiency anemia. There were no significant differences among the groups (ANOVA for continuous variables, Pearson’s chi-square test for categorical variables). 2 x៮ 앐 SD (all such values). 3 n ҃ 60. One pregnant woman is excluded. 4 IDA ҃ hemoglobin 쏝 110 g/L and ferritin concentrations 쏝 12 ␮g/L. 5 Body iron 쏝 0.

҃ 0.02). In anemic infants (hemoglobin: 쏝 110 g/L), the proportion of those with a tissue iron deficit at week 23 decreased significantly (P 쏝 0.001) in the DMM and DI groups. In infants with IDA (hemoglobin 쏝 110 g/L and PF 쏝 12 ␮g/L), the proportion of those with tissue iron deficit decreased in the 3

supplemented groups: P ҃ 0.002, P ҃ 0.031, and P 쏝 0.001 for the DMM, WMM, and DI groups, respectively. The relation between iron status indexes and age was examined by pooling the observations from the DP group (n ҃ 120) at baseline and at week 23 (Table 4). In infants 쏝1 y old, age was

TABLE 2 Iron status indicators in Indonesian infants 6 –12 mo old at baseline and after 23-wk supplementation1 Treatment groups Indicator Ferritin (␮g/L) n Baseline Week 23 Transferrin receptor (mg/L)2 n Baseline Week 23 Body iron (mg/kg)2 n Baseline Week 23

DMM

WMM

DI

DP

61 19.14 (15.35, 23.88)a,3 33.50 (26.79, 41.78)b,4,5

58 17.02 (13.55, 21.38)a 17.62 (14.03, 22.08)a

66 17.26 (13.96, 21.38)a 43.05 (34.83, 53.33)b,5,6

60 16.48 (13.18, 20.61)a 9.59 (7.67, 12.02)b

65 9.65 (9.11, 10.22)a 8.80 (8.43, 9.19)b,5,6

60 10.60 (9.96, 11.28)a 10.02 (9.46, 10.61)b

69 9.92 (9.50, 10.34)a 8.36 (8.05, 8.67)b,5,7

65 10.22 (9.66, 10.81)a 10.99 (10.43, 11.59)b

2

61 1.00 앐 4.48a,8 3.34 앐 2.37b,5,6

58 0.22 앐 4.28a 0.55 앐 3.61a

65 0.52 앐 3.69a 4.54 앐 2.27b,5,7

60 0.24 앐 4.18a Ҁ1.97 앐 3.35b

1 DMM, daily 1 Recommended Nutrient Intake (RNI) multiple micronutrients; WMM, weekly 2 RNI multiple micronutrients; DI, daily iron; DP, daily placebo. Ferritin was presented previously in median and range (13). There were no significant differences among the groups at baseline (ANOVA). Values in the same column with different superscript letters are significantly different, P 쏝 0.05 (paired t test). 2 Significant time ҂ treatment interaction, P 쏝 0.001 (2-factor ANOVA). 3 Geometric x៮ ; 95% CI in parentheses (all such values). 4,6,7 Significantly different from WMM (Scheffe’s test): 4P 쏝 0.05, 6P 쏝 0.01, 7P 쏝 0.001. 5 Significantly different from DP, P 쏝 0.001 (Scheffe’s test). 8 x៮ 앐 SD (all such values).

1684

WIJAYA-ERHARDT ET AL

TABLE 3 Proportion of infants who had tissue iron deficit at baseline and at week 23 by treatment group1 Tissue iron deficit2 Treatment groups Overall DMM (n ҃ 61) WMM (n ҃ 58) DI (n ҃ 65) DP (n ҃ 60) Anemia7 DMM (n ҃ 36) WMM (n ҃ 35) DI (n ҃ 38) DP (n ҃ 32) IDA8 DMM (n ҃ 12) WMM (n ҃ 12) DI (n ҃ 18) DP (n ҃ 18)

Baseline

Week 23

groups would prefer to receive supplements weekly [n ҃ 25 (41%), 28 (48%), and 24 (40%), respectively], whereas the mothers in the DI group did not prefer weekly over daily delivery [n ҃ 23 (35%)]. There was no significant difference among groups in frequency preference (P ҃ 0.82).

P3

DISCUSSION 26 (43) 24 (41) 27 (42) 34 (57)

5

4 (7) 23 (40) 2 (4)5 44 (73)6

쏝0.001

15 (42) 16 (46) 21 (55) 22 (69)

2 (6)5 15 (43) 2 (5)5 26 (81)

쏝0.001

4

1009 100 100 100

2 (17)10 6 (50)6 2 (11)5 17 (94)

0.01

1

DMM, daily multiple micronutrients; WMM, weekly multiple micronutrients; DI, daily iron; DP, daily placebo; IDA, iron deficiency anemia. There were no significant differences among the groups at baseline (Pearson’s chi-square test). 2 Body iron 쏝 0. 3 Calculated by using general estimating equations with treatment ҂ time interaction. 4 n; % in parentheses (all such values). 5,6,10 Significantly different from baseline (McNemar’s test): 5P 쏝 0.001, 6 P 쏝 0.05, 10P 쏝 0.01. 7 Hemoglobin concentrations 쏝 110 g/L. 8 IDA ҃ hemoglobin 쏝 110 g/L and ferritin concentrations 쏝 12 ␮g/L. 9 % (all such values).

negatively correlated with PF and body iron and positively correlated with TfR but was not correlated with hemoglobin. No significant correlation was found between age and iron status indexes in children 쏜1 y old. Children 쏜1 y old had a mean tissue iron deficit (body iron 쏝 0) of Ҁ2.1 mg/kg body wt, whereas those 쏝1 y old had iron stores (body iron 쏜 0) of 0.5 mg/kg body wt (P 쏝 0.001). Of the 38 children with IDA, all had a tissue iron deficit (body iron 쏝 0; data not shown). Similarly, of the 60 infants with IDA in the 4 groups at baseline, all had body iron 쏝 0 (Table 3). The rate of compliance for consuming 욷121 (75%) foodLETs was 54 (89%), 52 (90%), 59 (91%) and 54 (90%) in the DMM, WMM, DI, and DP groups, respectively (P ҃ 0.98). The most common reasons for not taking the foodLETs was illness, such as diarrhea, fever, or stomatitis aphthosa in the DMM group (n ҃ 9) and forgetting to take supplements in the WMM (n ҃ 5), DI (n ҃ 6), and DP (n ҃ 4) groups. When the mothers were asked about their experience of the intervention for their infants, 45 (74%) in the DMM group and 47 (72%) in the DI group reported that their infants had constipation or dark stool after taking the foodLETs, whereas 30 (52%) and 10 (17%) did so in the WMM and DP groups, respectively (P 쏝 0.001). The DMM, WMM, DI, and DP groups did not differ significantly in reported vomiting (39%, 41%, 23%, and 30%, respectively; P ҃ 0.11) or diarrhea (18%, 14%, 20%, and 10%, respectively; P ҃ 0.42). If the intervention were to be repeated, most mothers in the DMM, WMM, and DP

The results from the present study, which featured a doublemasked study design and strict monitoring of compliance, showed that multiple micronutrient and iron foodLETs on a daily basis are efficacious in improving body iron stores and reducing the proportion of a mean tissue iron deficit (body iron 쏝 0) in Indonesian infants. When the change in body iron estimates was used, a difference in each of the 4 groups was significant. In our previous report (13), the change in ferritin in DMM group did not differ significantly from that in the DI and WMM groups, and that in the WMM group did not differ significantly from that in the DP group. Our results agree with those of Cook et al (5), which indicated that body iron measurements showed a difference between the 2 groups given iron supplements that was not identified by the TfR measurement alone. More important, body iron measurements can be used to calculate iron absorption in iron intervention trials (5). In populations in whom the prevalence of ID is high— eg, infants, children, and pregnant women—the serum ferritin concentration usually is close to the iron-deficient range, and it does not accurately portray differences in functional iron (6). The equation for estimating body iron stores from the ratio of TfR to PF was derived from an adult phlebotomy study (5, 24) and has not been validated in children, which could be a limitation with respect to our infant subjects. However, that equation performed well in recent iron trials in schoolchildren (26, 27). A WHO/CDC meeting evaluated the performance of 5 indicators of iron status (ie, hemoglobin, zinc protoporphyrin, mean cell volume, TfR, and serum ferritin) and recommended 3 indicators for predicting changes in and assessing iron status: hemoglobin, serum ferritin, and TfR. If possible, the acute-phase protein CRP or ␣Ҁ1 acid glycoprotein (or both) should also be measured (28). During the 23-wk intervention, the DI and DMM groups had an iron absorption of 3.4% and 2.5%, respectively. With respect to improving the iron status of infants, the DI regimen is more effective than is the DMM regimen. A possible explanation for this effect is that the iron absorption from ferrous fumarate in the TABLE 4 Correlation coefficients between iron status indicators and age in the infant1 Iron status indicators

Age (mo) 6–11 (n ҃ 58) 12–18 (n ҃ 62) 1

Hemoglobin

PF2

Ҁ0.224 Ҁ0.053

Ҁ0.4073 Ҁ0.142

TfR2 0.4243 Ҁ0.027

Body iron Ҁ0.4273 Ҁ0.124

PF, plasma ferritin; TfRs, transferrin receptors. Indicators were examined by using pooled data from the placebo group at baseline and at week 23. 2 Based on log-transformed values. 3 Pearson’s correlation coefficients, P 쏝 0.01.

SUPPLEMENTATION AND BODY IRON STORES OF INFANTS

DMM group was lower than that from ferrous sulfate in the DI group. A previous study showed that the mean iron absorption from ferrous fumarate relative to ferrous sulfate in Bangladeshi children was only 27–35% (29). In Mexican children aged 2– 4 y, iron absorption from complementary food fortified with ferrous sulfate was higher (7.9 앐 9.8%) than that from ferrous fumarate (2.43 앐 2.3%) (30). Another explanation could be that the inclusion of zinc in the DMM supplement may have interfered with the absorption of iron. As was shown, supplementation with iron and zinc (each: 10 mg/d) had a smaller positive effect than did iron alone, but the effect was still significant in improving the iron status in infants (31, 32). However, not all body iron deficits were fully controlled in the DI group infant population after 23 wk of treatment. Global guidelines for iron supplementation have been published by the International Nutritional Anemia Consultative Group/WHO/UNICEF. The recommendations are that, in persons 6 –24 mo old, 12.5 mg Fe/d plus 50 ␮g folic acid/d until the age of 12 mo (2); this dose is slightly higher than the supplementation dose in the present study. In our previous report (13), we showed that DMM was more efficacious than was DI in improving hemoglobin concentration and in decreasing the proportion of anemia in infants. This finding may be due to the fact that anemia is caused not just by iron deficiency but also by a lack of other components that are in the DMM supplement, such as riboflavin, folic acid, vitamin C, and vitamin A, all of which are known to favor iron absorption or hematopoiesis (or both) (15). Several studies in different population groups have shown that weekly dosing with iron alone or iron along with other micronutrient supplementation successfully reduced anemia and improved micronutrient status and that it was as effective as daily administration (33– 40). The currently available evidence suggests that iron supplements should be taken daily to treat anemia, especially by pregnant women who are consuming low amounts of available iron. However, weekly supplements would reduce side effects (36, 39), and the lower costs may improve compliance. The daily foodLETs used in the present trial provided 70 mg Fe when calculated on a weekly basis, whereas the weekly foodLET provided 20 mg Fe. The present study showed that daily dosing of iron alone or with other micronutrient foodLETs is superior to weekly dosing in the treatment of ID in infants, but the WMM supplements had a greater effect in preventing the worsening of iron status than did the DP. At baseline, almost half of respondents had a tissue iron deficit (body iron 쏝 0). Infants receiving DP, who represented the healthy population without intervention, had an average iron loss of 0.01 mg/d and developed a mean deficit in tissue iron over 23 wk, although they were still being breastfed and were given complementary feeding. These findings can be generalized with regard to the importance of providing additional iron to infants. Iron is involved in many central nervous system processes that could affect infant behavior and development. The key liabilities of tissue ID relate to abnormalities in host defense, work performances, and neurologic function (41). When stratified into age groups, we found that infants 쏝1 y old had iron stores (body iron 쏜 0), whereas those 쏜1 y old had a tissue iron deficit (body iron 쏝 0); these findings are consistent with earlier findings showing that infants 쏝1 y old had body iron stores of 0.29 앐 1.04 mg/kg and those 1 y old body iron stores of Ҁ0.50 앐 0.74 mg/kg. This decrease in 1-y-old infants was followed by a linear increase in body iron after the age of 2 y, despite

1685

increasing body size (24). We also found that IDA occurred in infants at an iron stores deficit of 쏝Ҁ3 mg/kg, which is slightly higher than the results of previous studies with a deficit of 쏝Ҁ4 mg/kg in adults and in children (5, 24). The subjects in the present study developed mild anemia (hemoglobin: 100.5 g/L) at an iron deficit of Ҁ1.8 mg/kg, which supports the findings of a study in 6 –10-y-old Moroccan children showing that children develop mild anemia (hemoglobin: 108 앐 4 g/L) at mean iron deficits of Ҁ1.1 mg/kg (26). In conclusion, the present study found that both DMMs and DI foodLETs given for 23 wk increased the body iron stores of Indonesian infants aged 6 –12 mo, and WMM foodLETs prevented the deterioration of iron stores. These findings support the use of the body iron stores calculations suggested by Cook et al (5) to improve the precision of the laboratory diagnosis of iron absorption, detect IDA, and find the most appropriate iron intervention strategy. We thank all the parents and their infants for participating in this study. We also thank Kristiani (head of Salam primary health center) and Nurul (head of Ngluwar primary health center) and their staffs for valuable advice, help, and cooperation in the field. We are grateful for the assistance of fieldworkers, blood analyst team (Muntilan General Hospital), personnel in subdistrict health office (Magelang), and laboratory staffs (Southeast Asian Ministers of Education Organization–Tropical Medicine Regional Center for Community Nutrition–University of Indonesia). The authors’ responsibilities were as follows—MW-E: supervised field activities, entered data, conducted the analysis, and drafted the manuscript; JGE: analyzed the blood samples and assisted in writing the manuscript; JU and EK: were principal investigators and took overall responsibility for the study in Indonesia and writing of the manuscript; LW: supervised field activities, entered data, and conducted the analysis; and RG: conceived of and designed the the multicenter study. None of the authors had a personal or financial conflict of interest.

REFERENCES 1. Adamson P. Vitamin and mineral deficiency. A global progress report. Micronutrient Initiative and UN Children’s Fund. New York, NY: UNICEF, 2004. 2. Allen LH. Iron supplements: scientific issues concerning efficacy and implications for research and programs. J Nutr 2002;132(suppl):813S– 9S. 3. International Nutritional Anemia Consultative Group. Iron deficiency in women. Washington, DC: International Life Sciences Institute Press, 2002. 4. United Nations System Standing Committee on Nutrition. The 5th report on the world nutrition situation. Nutrition for improved development outcomes. Geneva, Switzerland: World Health Organization, 2004. 5. Cook JD, Flowers CH, Skikne BS. The quantitative assessment of body iron. Blood 2003;101:3359 – 64. 6. Skikne BS, Flowers CH, Cook JD. Serum transferrin receptor: a quantitative measure of tissue iron deficiency. Blood 1990;75:1870 – 6. 7. Yeung GS, Zlotkin SH. Percentile estimates for transferrin receptor in normal infants 9 –15 mo of age. Am J Clin Nutr 1997;66:342– 6. 8. Olivares M, Walter T, Cook JD, Hertrampf E, Pizarro F. Usefulness of serum transferrin receptor and serum ferritin in diagnosis of iron deficiency in infancy. Am J Clin Nutr 2000;72:1191–5. 9. Asobayire FS, Adou P, Davidsson L, Cook JD, Hurrell RF. Prevalence of iron deficiency with and without concurrent anemia in population groups with high prevalences of malaria and other infections: a study in Côte d’Ivoire. Am J Clin Nutr 2001;74:776 – 82. 10. Verhoef H, West CE, Ndeto P, Burema J, Beguin Y, Kok FJ. Serum transferrin receptor concentration indicates increased erythropoiesis in Kenyan children with asymptomatic malaria. Am J Clin Nutr 2001;74: 767–75. 11. Garcia OP, Diaz M, Rosado JL, Allen LH. Ascorbic acid from lime juice does not improve the iron status of iron-deficient women in rural Mexico. Am J Clin Nutr 2003;78:267–73.

1686

WIJAYA-ERHARDT ET AL

12. Shell-Duncan B, McDade T. Use of combined measures from capillary blood to assess iron deficiency in rural Kenyan children. J Nutr 2004; 134:384 –7. 13. Untoro J, Karyadi E, Wibowo L, Wijaya-Erhardt M, Gross R. Multiple micronutrient supplements improve micronutrient status and anemia but not growth and morbidity of Indonesian infants: a randomized, doubleblind, placebo-controlled trial. J Nutr 2005;135(suppl):639S– 45S. 14. Smuts CM, Benadé AJ, Berger J, et al. IRIS I: a foodlet-based multiplemicronutrient intervention in 6- to 12-month-old infants at high risk of micronutrient malnutrition in four contrasting populations: description of a multicenter field trial. Food Nutr Bull 2003;3(suppl):27S–33S. 15. Smuts CM, Lombard CJ, Benadé AJ, et al. Efficacy of a foodlet-based multiple micronutrient supplement for preventing growth faltering, anemia, and micronutrient deficiency of infants: the Four Country IRIS trial pooled data analysis. J Nutr 2005;135(suppl):631S– 8S. 16. Council for International Organizations of Medical Sciences. International guidelines for ethical review of epidemiological studies. Geneva, Switzerland: CIOMS, 1990. 17. Lock G. The foodLET vehicle designed for and used in the IRIS 1 intervention. Food Nutr Bull 2003;3(suppl):16S–9S. 18. Gross R, Benadé S, Lopez G. The International Research on Infant Supplementation initiative. J Nutr 2005;135(suppl):628S–30S. 19. Gross R, Kielmann A, Korte R, Schoeneberger H, Schultink W. Nutrition baseline surveys in communities. In: Erhardt J, Gross R, eds. Nutrition surveys and calculations: guidelines, software and additional information. Version current 1997. Internet: http://www.nutrisurvey.de (accessed 7 July 2003). 20. Erhardt JG, Estes JE, Pfeiffer CM, Biesalski HK, Craft NE. Combined measurement of ferritin, soluble transferrin receptor, retinol binding protein, and C-reactive protein by an inexpensive, sensitive, and simple sandwich enzyme-linked immunosorbent assay technique. J Nutr 2004; 134:3127–32. 21. Cook JD, Bothwell TH, Covell AM, Dallman PR, Lynch SR, Worwood MA. Measurements of iron status. A report of the International Nutritional Anemia Consultative Group. Washington, DC: International Life Sciences Institute Press, 1985. 22. Sauberlich HE. Iron. In: Wolinsky I, ed. Laboratory tests for the assessment of nutritional status. 2nd ed. Boca Raton, FL: CRC Press, 1999: 343– 69. 23. Berger J, Dyck JL, Galan P, et al. Effect of daily iron supplementation on iron status, cell-mediated immunity, and incidence of infections in 6 –36 month old Togolese children. Eur J Clin Nutr 2000;54:29 –35. 24. Cook JD, Boy E, Flowers C, Daroca Mdel C. The influence of highaltitude living on body iron. Blood 2005;106:1441– 6. 25. World Health Organization. The WHO child growth standards. WHO Anthro 2005. Version current 17 February 2006. Internet: http://www. who.int/childgrowth/en (accessed 23 May 2006). 26. Zimmermann MB, Chaouki N, Hurrell RF. Iron deficiency due to consumption of a habitual diet low in bioavailable iron: a longitudinal cohort study in Moroccan children. Am J Clin Nutr 2005;81:115–21.

27. Moretti D, Zimmermann MB, Muthayya S, et al. Extruded rice fortified with micronized ground ferric pyrophosphate reduces iron deficiency in Indian schoolchildren: a double-blind randomized controlled trial. Am J Clin Nutr 2006;84:822–9. 28. International Nutritional Anemia Consultative Group. Iron deficiency in early life: challenges and progress. Washington, DC: International Life Sciences Institute Press, 2004. 29. Sarker SA, Davidsson L, Mahmud H, et al. Helicobacter pylori infection, iron absorption, and gastric acid secretion in Bangladeshi children. Am J Clin Nutr 2004;80:149 –53. 30. Pérez-Expo´sito AB, Villalpando S, Rivera JA, Griffin IJ, Abrams SA. Ferrous sulfate is more bioavailable among preschoolers than other forms of iron in a milk-based weaning food distributed by PROGRESA, a national program in Mexico. J Nutr 2005;135:64 –9. 31. Lind T, Lönnerdal B, Stenlund H, et al. A community-based randomized controlled trial of iron and zinc supplementation in Indonesian infants: interactions between iron and zinc. Am J Clin Nutr 2003;77:883–90. 32. Wieringa FT, Berger J, Dijkhuizen MA, et al. Combined iron and zinc supplementation in infants improved iron and zinc status, but interactions reduced efficacy in a multicountry trial in Southeast Asia. J Nutr 2007;137:466 –71. 33. Thu BD, Schultink W, Dillon D, Gross R, Leswara ND, Khoi HH. Effect of daily and weekly micronutrient supplementation on micronutrient deficiencies and growth in young Vietnamese children. Am J Clin Nutr 1999;69:80 – 6. 34. Schultink W, Gross R, Gliwitzki M, Karyadi D, Matulessi P. Effect of daily vs twice weekly iron supplementation in Indonesian preschool children with low iron status. Am J Clin Nutr 1995;61:111–5. 35. Berger J, Aguayo VM, Téllez W, Luja´n C, Traissac P, San Miguel JL. Weekly iron supplementation is as effective as 5 day per week iron supplementation in Bolivian school children living at high altitude. Eur J Clin Nutr 1997;51:381– 6. 36. Angeles-Agdeppa I, Schultink W, Sastroamidjojo S, Gross R, Karyadi D. Weekly micronutrient supplementation to build iron stores in female Indonesian adolescents. Am J Clin Nutr 1997;71:1485–94. 37. Jayatissa R, Piyasena C. Adolescent schoolgirls: daily or weekly iron supplementation. Food Nutr Bull 1999;20:429 –34. 38. Gross R, Schultink W, Juliawati. Treatment of anaemia with weekly iron supplementation. Lancet 1994;334:821. 39. Viteri FE, Ali F, Tujague J. Long-term weekly iron supplementation improves and sustains nonpregnant women’s iron status as well or better than currently recommend short-term daily supplementation. J Nutr 1999;129:2013–20. 40. Muslimatun S, Schmidt MK, Schultink W, et al. Weekly supplementation with iron and vitamin A during pregnancy increases hemoglobin concentration but decreases serum ferritin concentration in Indonesian pregnant women. J Nutr 2001;131:85–90. 41. Lozoff B, Beard J, Connor J, Felt B, Georgieff M, Schallert T. Longlasting neural and behavioral effects of iron deficiency in infancy. Nutr Rev 2006;64(suppl):34S–91S.

Influence of acute phytochemical intake on human urinary metabolomic profiles1–3 Marianne C Walsh, Lorraine Brennan, Estelle Pujos-Guillot, Jean-Louis Sébédio, Augustin Scalbert, Ailís Fagan, Desmond G Higgins, and Michael J Gibney ABSTRACT Background: Diversity in dietary intake contributes to variation in human metabolomic profiles and artifacts from acute dietary intake can affect metabolomics data. Objective: We investigated the role of dietary phytochemicals on shaping human urinary metabolomic profiles. Design: First void urine samples were collected from 21 healthy volunteers (12 women, 9 men) following their normal diet (ND), a 2-d low-phytochemical diet (LPD), or a 2-d standard phytochemical diet (SPD). Nutrient intake was assessed during the study. Urine samples were analyzed by using 1H nuclear magnetic resonance spectroscopy (1H NMR) and mass spectrometry (MS), which was followed by multivariate data analysis. Results: Macronutrient intake did not change throughout the study. Partial least-squares-discriminant analysis indicated a clear distinction between the LPD samples and the ND and SPD samples, relating to creatinine and methylhistidine excretion after the LPD and hippurate excretion after the ND and SPD. The predictive power of the LPD versus the ND model was 74 앐 3% and 82 앐 6% with the 1H NMR and MS data sets, respectively. The predictive power of the LPD versus the SPD model was 83 앐 8% and 69 앐 4% for the 1H NMR and MS data sets respectively. A cross platform comparison of both data sets by co-inertia analysis showed a similar distinction between the LPD and SPD. Conclusions: Acute changes in urinary metabolomic profiles occur after the consumption of dietary phytochemicals. Dietary restrictions in the 24 h before sample collection may reduce diversity in phytochemical intakes and therefore reduce variation and improve data interpretation in metabolomics studies using urine. Am J Clin Nutr 2007;86:1687–93. KEY WORDS Nutritional metabolomics, metabonomics, metabolic profiling, nutrigenomics

may be unique to each group. These factors could lead to misinterpretation of the true differences in endogenous metabolism between the groups. Genetic and environmental variation in human populations contributes to considerable diversity among human metabolomes (1– 6), and diet is one of the key environmental factors because of its dynamic input to metabolism. The role of diet in shaping metabolic profiles is not fully elucidated, but it is clear that diet will have both an acute and chronic effect. Understanding the chronic effects of diet is the most relevant in terms of nutrition research, but in terms of the interpretation of nutritional metabolomics data, an understanding of acute dietary effects is also important. Foods in the human diet are not merely a source of nutrients and are abundant in many other compounds that contribute to their taste, color, aroma, texture, and shelf-life. Plant-derived foods and drinks are a rich source of such compounds, namely phytochemicals, and 5000 – 10 000 are present in human food. It is estimated that the average diet corresponds to a daily dose of 1.5 g phytochemicals (7). The role of these phytochemicals in shaping the metabolic profiles obtained by techniques based on nuclear magnetic resonance (NMR) and mass spectrometry (MS) is not yet clear. Previous studies that investigated the metabolic fate of polyphenols have shown their appearance in plasma as early as 1 h and in urine within 24 h after ingestion (8 –10). Hence, to further our understanding of the role that diet plays in shaping metabolic profiles, it is important to consider these acute dietary influences that contribute to normal variation in metabolic profiles. This is particularly relevant for urine because it accumulates many end products of metabolism. Urine is a complex biofluid with an extensively varying composition, whereas the composition of other biofluids can be controlled more tightly by varying the 1

INTRODUCTION

Metabolomics affords an efficient approach for assessing the overall metabolic profiles of tissues or biofluids and is increasingly being used within a range of scientific disciplines. One of the challenges in adopting a metabolomic approach, particularly in human nutrition research, is the ability to distinguish meaningful responses within metabolic profiles that are produced as a consequence of a specific stimulus rather than as an artifact of daily variation. In addition, when distinct metabolic profiles are identified that differentiate 욷2 sample populations, it is important to understand the contribution of any exogenous factors that

From the Centre for Food and Health (MCW and MJG), the School of Biomolecular and Biomedical Sciences, Conway Institute (LB), and the School of Medicine and Medical Science, Conway Institute (AF and DGH), University College Dublin, Ireland, and the Unite de Nutrition Humaine, Institute National de la Recherche Agronomique, St-Genès-Champanelle, France (EPG, JLS, and AS). 2 Supported by The Irish Research Council for Science, Engineering and Technology. 3 Reprints not available. Address correspondence to MC Walsh, Room 3.02 Nutrition Unit, Centre for Food and Health, School of Agriculture, Food Science and Veterinary Medicine, University College Dublin, Dublin 4, Ireland. E-mail: [email protected]. Received April 19, 2007. Accepted for publication July 26, 2007.

Am J Clin Nutr 2007;86:1687–93. Printed in USA. © 2007 American Society for Nutrition

1687

1688

WALSH ET AL

excretion of many metabolites into urine. In a previous study we reported that dietary standardization for 24 h before sample collection reduced the extent of variation in urinary metabolic profiles, but not in fasting plasma or salivary profiles (6). In an attempt to further our understanding of the effect that acute dietary intakes have on NMR and MS profiles, the current study was designed to investigate the role of phytochemicals in shaping urinary metabolic profiles. In the long term, understanding the influence of acute dietary intakes will lead to an improved interpretation of dietary intervention outcomes.

SUBJECTS AND METHODS

Study design Ethical approval was received from the Faculty of Health Sciences Ethics Committee, Trinity College Dublin, in accordance with the Declaration of Helsinki. Twenty-one healthy, free-living subjects were recruited from the Dublin metropolitan area, and each participant provided written informed consent. Exclusion criteria were a body mass index (in kg/m2) 쏝18.5 or 쏜30.0, being 쏝18 or 쏜35 y of age, use of prescribed medication (oral contraceptive use was permitted), and having had a urinary tract infection within 1 mo of commencing the study. The study duration was 6 d (days 0 –5). Samples of first void urine were collected chilled on study days 1, 3, and 5. Dietary intake was recorded on days 0 – 4. Subjects were instructed to follow their normal diet (ND) on day 0, with the result that the day 1 urine samples represented normal urinary metabolic profiles. On days 1–2, subjects followed a low-phytochemical diet (LPD), with the result that the day 3 urine samples represented urinary metabolic profiles in the absence of phytochemicals. On days 3– 4, subjects continued the LPD with the addition of fruit and vegetable drinks, with the result that the day 5 urine samples represented urinary profiles after a standard fruit and vegetable intake (ie, standardized phytochemical intake). Two days was an adequate time for both the phytochemical washout and dosage period, given that the plasma elimination half-life for major phytochemicals is between 1.1 and 28.1 h (10). For the LPD the subjects were provided with a list of allowed and forbidden foods. There were no restrictions on the quantities of food eaten. All fruit and vegetables were forbidden (this included all products that contain fruit and vegetables, such as jams, sauces, soup, tea, coffee, and chocolate). The allowed list included meat, dairy products, and some low-pigmented plant products, such as white breads, non-whole-grain breakfast cereals, potatoes, white rice, and white pasta. For the standardized phytochemical diet (SPD), the fruit and vegetable source was 4 ҂ 100 mL apple, carrot, and strawberry drinks (Knorr Vie, produced by Unilever; Internet: www.knorr-vie.com). Nutrient intake Nutrient intakes were assessed with the use of 5-d weighed food records. The participants were given detailed instructions regarding the completion of the records and were tutored on the estimation of food portion sizes for instances when the food scales could not be used. The records were reviewed after each sample collection, and clarification of food portions and preparation was made. Total energy, protein, carbohydrate, fat, fiber, vitamin C, and carotene intakes were estimated by using WISP©

(Weighed Intake Software Program; Tinuviel Software, Anglesey, United Kingdom). The WISP database of food composition included the manufacturer’s nutritional information for the standard drinks. Urine collection First void midstream urine samples were collected on the mornings of days 1, 3, and 5. The subjects were given insulated ice packs in which they were asked to store the samples immediately until they were received by the study investigator. On arrival at the laboratory, the samples were centrifuged at 2500 ҂ g for 10 min at 4 °C to remove any solid debris. Fractions (500 ␮L) of the urine supernatants were then stored at Ҁ80 °C until 1H NMR analysis. NMR spectroscopy For NMR spectroscopy the urine samples were buffered with a phosphate buffer (0.2 mol KH2PO4/L, 0.8 mol KH2PO4/L). To each 350 ␮L urine, 180 ␮L phosphate buffer (pH 7.4), TSP (sodium trimethylsilyl [2,2,3,3-2H4] proprionate), and 10% D2O were added. NMR spectra were acquired at 298 K on a 500-MHz DRX NMR spectrometer (Bruker Biospin, Karlsruhe, Germany) using a noesypresat pulse sequence. Spectra were acquired with 32-k data points and 128 scans over a spectral width of 8 kHz. Water suppression was achieved during the relaxation delay (2.5 s) and the mixing time (100 ms). All 1H NMR spectra were referenced to TSP at 0.0 ppm and processed manually with the Bruker software with the use of a line broadening of 0.2 Hz. All spectra were baseline corrected. The spectra were then reduced by integrating into bins across spectral regions of 0.02 ppm with an AMIX (Bruker Biospin). The water region (4.2– 6.0 ppm) was excluded. The data were normalized to the sum of the spectral integral to account for differences between urinary concentrations. Mass spectrometry The urine samples (500 ␮L) were defrosted at room temperature, centrifuged at 7000 ҂ g for 5 min at 4 °C, and then diluted 4-fold with distilled water. Chromatography was performed with a Waters Alliance 2695 HPLC system (Waters Corporation, Manchester, United Kingdom). The HPLC system was coupled to a Waters Qtof-Micro equipped with an electrospray source and a lockmass sprayer. The source temperature was set to 120 °C with a cone gas flow of 50 L/h, a desolvation temperature of 300 °C, and a nebulization gas flow of 400 L/h. The capillary voltage was set at 3000 V and the cone voltage to 30 V. The mass spectrometric data were collected in continuum full-scan mode with a mass-to-charge ratio (m/z) of 100 –1000 from 0 to 10 min, in positive mode. All analyses were acquired by using the lockspray with a frequency of 5 s to ensure accuracy. Leucineenkephalin was used as the lock mass ([MѿH]ѿ m/z 556.2771) at a concentration of 0.5 ng/␮L (in MeOH/water, 50/50 by vol with 0.1% formic acid). To avoid possible differences between sample batches, a Latin square was carried out to obtain a randomized list of samples for analysis. The 2.1 ҂ 150 mm SymmetryShieldRP18 5 ␮m column was injected with 10 ␮L diluted urine at 30 °C. Mobile phase components were A ҃ 1% formic acid and B ҃ acetonitrile with 1% formic acid. The column was eluted with a gradient of 0 –20% B over 0 – 4 min, followed by an increase from 20% to

1689

METABOLOMICS IN HUMAN NUTRITION

95% over 4 – 8 min. The mobile phase was then held at this composition for 1 min and then returned to 100% A at 9 min for 5 min reequilibration. The flow rate was set to 300 ␮L/min. The raw data were transformed to centroid mode and mass corrected before being analyzed with MarkerLynx Applications Manager v1.0. The liquid chromatography–MS data were peakdetected and noise-reduced for both the liquid chromatography and MS components. Each peak in the resulting 3-dimensional data set was represented by retention time m/z and its ion intensity in each sample. The matrix obtained was then exported for statistical analysis.

TABLE 1 Demographic characteristics1 All subjects

Women (n ҃ 12)

Men (n ҃ 9)

Height (m) Weight (kg) BMI (kg/m2) Age (y)

1.63 앐 0.06 60.6 앐 6.9 22.7 앐 2.6 25 앐 2

1.82 앐 0.072 79.0 앐 11.83 23.8 앐 1.8 26 앐 4

All values are x៮ 앐 SD. Significantly different from women (independent t test): 2 P 쏝 0.001, 3 P 쏝 0.05. 1

2,3

Statistical analysis The nutrient intake data were analyzed by using SPSS for WINDOWS (version 12: SPSS Inc, Chicago, IL). One-factor analysis of variance (ANOVA) was used to assess the differences between the ND, LPD, and SPD. Repeated-measures ANOVA was used to check for diet ҂ sex interactions. Differences were considered significant at P 쏝 0.05, and Tamhane’s T2 post hoc tests were carried out to assess the significant differences indicated by the ANOVA results. Multivariate data analyses were applied to both the 1H NMR and MS data by using Simca-Pѿ software (version 10.0; Umetrics, Umeå, Sweden) and the R Statistics package with modules MADE4 (11) and ade4 (12). The data sets were mean centered and Pareto scaled (each variable was weighted according to 1/公SD). Principal component analysis (PCA), an unsupervised pattern recognition technique, was performed initially to assess variation and expose any trends or outlying data. Partial leastsquares-discriminant analysis (PLS-DA) was then performed to define the maximum separation between the LPD versus the ND or the standard SPD. The data were visualized by constructing principal component scores and loadings plots, where each point on the score plot represented an individual urine sample and each point on the loadings plot represented a single 1H NMR spectral region or MS reading. The PLS-DA models were cross-validated by randomly removing each one-third of the data to be used as a test data set, and a training data set was constructed with the remaining two-thirds of the data. The class of each sample in the test data set was then predicted based on the model built from the training set. The quality of all models was judged by the goodness-of-fit parameter (R2) and the predictive ability parameter (Q2), which is calculated by a 7-fold internal cross-validation of the data. Co-inertia analysis (CIA) (13) was applied to the LPD and SPD samples from the 1H NMR and MS data. CIA is a multivariate statistical method used to identify patterns in parallel data sets. In brief, it first carries out a simple ordination such as PCA on each of the data sets. CIA then finds pairs of axes from the 2 data sets that have maximum covariance. The first few axes from the CIA are used to create 2 dimensional plots. On these plots, the relations between the sample data points and variables from both data sets are visible. RESULTS

Twenty-one healthy volunteers (12 women, 9 men) aged 20 –34 y completed the study. The subjects’ demographic characteristics (x៮ 앐 SD) are shown in Table 1. The mean BMI was 23.2 앐 2.3, and the mean age was 25 앐 3 y. BMI and age showed no significant differences between men and women.

Nutrient intake The mean (앐SD) nutrient intake data for all subjects for each of the diets are shown in Table 2. The LPD was followed on days 1 and 2, and the SPD was followed on days 3 and 4; therefore, the mean intakes for both diets are presented. Repeated-measures ANOVA showed no diet-by-sex interactions, but indicated that energy intake in men was significantly higher during the LPD (P 쏝 0.05). No significant differences were found between the 3 diets for total energy, protein, carbohydrate, or fat intakes. Onefactor ANOVA (P 쏝 0.05) with post hoc tests found that fiber intake was significantly lower during the LPD than during the ND (P 쏝 0.05) and the SPD (P 쏝 0.001). In addition, one-factor ANOVA (P 쏝 0.001) with post hoc tests found that vitamin C intake was significantly lower during the LPD than during the ND (P 쏝 0.05) and the SPD (P 쏝 0.001). Finally, one-factor ANOVA (P 쏝 0.001) with post hoc tests found that carotene intake was significantly higher during the SPD than during both the ND (P 쏝 0.001) and the LPD (P 쏝 0.001) and that carotene intake was significantly higher with the ND than with the LPD (P 쏝 0.05). Multivariate data analysis of the 1H NMR urinary profiles Initial PCA of the 1H NMR urinary data showed 5 outlying samples (positioned outside the Hotelling’s T2 elipse on the score plot). NMR spectra of these outlying samples were inspected. Three of these outliers were subsequently removed from the data set, one because of intense signal intensities in the spectral regions 0.87– 0.89, 1.06, 1.22, and 1.37–1.42 ppm, which corresponded to a drug metabolite (drug use was confirmed by the dietary records). Another 2 samples showed enhanced signal intensities in the spectral region 2.17–2.19 ppm, which corresponded to a high p-cresol glucuronide concentration (p-cresol glucuronide originates as p-cresol, a metabolic product of Clostridium difficile in the large intestine and is conjugated with glucuronide in the liver before urinary excretion). The remaining 2 outliers were retained because their spectra presented no unusual features. PCA was then repeated, and the score plot is shown in Figure 1A. The first 2 components accounted for 31% of variation in the data, and the samples from the LPD tended to cluster on the right side of the plot. To probe further the differences between the LPD and the ND, a PLS-DA model was constructed. The first 2 components accounted for 30% (R2X value) of the variation in the model and had a Q2 value of 42%. Interrogation of the loadings plot and the NMR spectra showed that the discrimination between the LPD and the ND samples was

1690

WALSH ET AL

TABLE 2 Nutrient intakes1 All subjects (n ҃ 21)

Energy intake (MJ/d) Men (n ҃ 9) Women (n ҃ 12) Protein (% of total energy) Carbohydrate (% of total energy) Fat (% of total energy) Fiber (g)3 Vitamin C (mg/d) Carotene (␮g/d)

ND

LPD

SPD

8.69 앐 2.44 11.40 앐 3.52 16 앐 5 51 앐 10 36 앐 9 15.7 앐 12.6a 144.5 앐 148.6a 3052 앐 2954a

7.65 앐 1.59 10.15 앐 2.712 18 앐 4 50 앐 10 34 앐 7 6.9 앐 3.2b 42.4 앐 38.4b 425 앐 773b

8.32 앐 1.85 10.09 앐 2.76 17 앐 4 55 앐 8 31 앐 7 12.4 앐 2.9a 155.2 앐 35.1a 7441 앐 130c

All values are x៮ 앐 SD. ND, normal diet; LPD, low-phytochemical diet; SPD, standard phytochemical diet. Means in a row with different superscript letters are significantly different, P 쏝 0.05 (ANOVA with Tamhane’s T2 post hoc tests). A significant diet ҂ sex interaction was observed. 2 Significantly different from energy intake in men, P 쏝 0.05 (repeated-measures ANOVA). 3 Calculated by using the method of the Association of Official Analytical Chemists. 1

mainly dominated by a higher level of hippurate (spectral regions: 3.96 –3.98, 7.54 –7.56, 7.56 –7.58, and 7.84 –7.86 ppm) in the ND samples and a higher level of creatinine (spectral regions: 3.04 –3.06 and 4.06 – 4.08 ppm) and methyl histidine (spectral region: 3.74 –3.76 ppm) in the LPD samples. Validation of these models, as described in Subjects and Methods, indicated that 74 앐 3% of the samples were classified correctly.

20.00

A

To investigate the effect of the controlled addition of phytochemicals to the diet, a PLS-DA model was constructed with the data from the LPD and the SPD (Figure 1B). The first 2 components of the model accounted for 29% (R2X value) of the variation in the data and had a Q2 value of 60%. Inspection of the loadings plot and the NMR spectra showed that the discrimination between the 2 sample groups was dominated

02

05 11

Principal Component 2

10.00

14 11 16

16 17 09

0.00

0304

13 13 14

12

04

05

08 07

07

10 02 11

10

04 21 21

10

0916 08 17 12

06

1509 15 03 13 20 060617 1819 01 07

08

20 18 19

21 20

19

15

01

01

12

-10.00

-20.00 -20.00

-10.00

0.00

10.00

20.00

Principal Component 1

B

2000.00

12

10.00

C 05 01

03

06

05

12

13

14

07

04

11 21

0.00

09 07

18

17 09 01 19 1120

16

06 14

16

10

10

01

02 08 21

18 20

08 17

15 19

Principal Component 2

Principal Component 2

04 13

1000.00 01 07 12

03

04

18

07 0508

03 08

12

18 1114 02 09 13

0.00 16

06

17

15

19

04

09

14

21

16

15

13 11

06

20 17 19

20

-1000.00

15 -10.00

21 -2000.00 10

-10.00

0.00

Principal Component 1

10.00

-2000.00

-1000.00

0.00

1000.00

2000.00

Principal Component 1

FIGURE 1. A: Principal component analysis of nuclear magnetic resonance (NMR) urinary data. B: Partial least-squares-discriminant analysis (PLS-DA) of NMR urinary data. C: PLS-DA of mass spectrometry urinary data. E, normal diet; f, low-phytochemical diet; 䡺, standard phytochemical diet.

METABOLOMICS IN HUMAN NUTRITION

by higher levels of hippurate after the SPD and higher levels of creatinine and methyl histidine after the LPD. Validation of this model indicated that 83 앐 8% of the samples were correctly classified. The PCA and PLS-DA analyses were also carried out separately for men and women to ensure that sex effects did not influence the results. These analyses showed similar results, which indicated that sex did not affect the study findings. Multivariate data analysis of the mass spectrometry urinary profiles Two outlying samples were revealed after initial PCA of the MS urinary data (one of these was also an outlier in the NMR data). The PCA loadings plot and inspection of the individual outputs indicated that these samples had high peak intensities corresponding to masses 313.078 and 447.107, which relate to genistein acetate and genistein glucuronide, respectively. These metabolites result from soy isoflavones, but there was no evidence to confirm the consumption of soy isoflavones before sampling. These samples were subsequently removed from the data set, and PCA was repeated. The first 2 components of this model accounted for 24% of variation in the data, and, although there were no distinct clustering trends, the samples from the ND tended to locate in the bottom half of the plot. A PLS-DA model was constructed to assess any differences between the LPD samples and the ND samples. A 3-component model was generated, accounting for 28% (R2X value) of variation in the data and it had a Q2 value of 59%. Inspection of the PLS-DA loadings plot indicated that discrimination of the samples was mainly dominated by the ions with an m/z of 180.068, 105.028 (both corresponding to hippurate), 312.217, 197.07, and 169.036 (unidentified). These ions were associated with the ND samples, and validation of the model indicated that 82 앐 6% of the samples were classified correctly. Another 3-component PLS-DA model was constructed to assess the differences between the LPD samples and the SPD samples (Figure 1C). This model accounted for 30% (R2X value) of the variation in the data, and the Q2 value was 44%. The corresponding loadings plot indicated that discrimination resulted from high peak intensities associated with the SPD samples, for m/z values of 180.068, 105.028 (both relating to hippurate), 413.045, 312.217, and 169.036 (unidentified). The model was validated, giving correct classification for 69 앐 4% of the samples. Combined analysis of NMR and MS data CIA was applied to the LPD and SPD samples of 1H NMR and MS data to visualize patterns and to identify metabolites with concentrations that changed in each group. It was carried out on 18 matched samples; 3 samples were removed from the analysis as they were considered outliers in either the 1H NMR or MS data or both. The initial ordinations showed that the sample distribution of 1H NMR split reasonably well. The MS data, however, did not separate so clearly (data not shown). CIA was applied to these initial ordinations. The combined sample distribution from the first 2 axes is shown in Figure 2A. Axis 1 and axis 2 from the CIA explain 46% and 20% of the variance, respectively. The base of each arrow represents one 1H NMR sample, and the tip represents the equivalent MS sample. The lengths of the arrows indicate how dissimilar the samples are. Although a perfect separation of

1691

the sample groups was not achieved, the SPD samples (in black) dominate the right side of the plot and the LPD samples (in gray) dominate the left side. The metabolites from the 1H NMR and MS data are plotted in Figures 2B and C. Metabolites that project in the same direction of both 1H NMR and MS plots have similar influence in both sets of data. For example, the top right quadrant of Figure 2B and Figure 2C are dominated by regions that correspond to hippurate. Because the top right quadrant is dominated in the sample plot by SPD, it can be interpreted that the SPD samples have a notable increase in hippurate compared with LPD. DISCUSSION

The application of metabolomics to nutritional studies holds great promise for future research. However, if it is to reach its full potential, we must understand the influence of all aspects of dietary composition on metabolic profiles. In the current study, the effect of acute changes in dietary phytochemicals on human urinary metabolic profiles was investigated. Modulation of the phytochemical intake led to significant changes in the metabolic profiles, and diets with varying phytochemical contents were distinguished and characterized by using pattern recognition analysis. Previous research has shown a notable effect of diet on urinary composition, which was detectable by metabolomic analysis. A recently published study investigated changes in human urinary profiles resulting from dietary intervention. Discrimination between the dietary treatments was dominated by an elevated excretion of creatinine, creatine, trimethylamine-Noxide (TMAO), taurine and 1- and 3-methylhistidine associated with a high meat diet and p-hydroxyphenylacetate associated with a vegetarian diet (14). In another human study, a dietary intervention with soy isoflavones found subtle changes in urinary metabolic profiles that were associated with osmolyte fluctuation and energy metabolism (15). Differentiation between the urinary profiles of populations from distinct geographic locations has also been studied, and some of the differences between the groups have been attributed to diet (2, 4). Dietary intervention studies carried out in animal models have also shown distinct changes resulting from the diet. A crossover metabolomic study in pigs successfully distinguished between a whole-grain diet and a non-whole-grain diet. In urine, differentiation was dominated by betaine and hippurate excretion associated with the whole-grain diet, and creatinine excretion was associated with the non-whole-grain diet (16). In another study, urine samples from rats that received 3 different 24-h dietary treatments (normal, overnight fast, or turkey diet) were discriminated on the basis of their NMR and extractive electrospray ionization mass spectra (17). In addition to the changes that may result in metabolic profiles after a chronic dietary intervention, acute dietary intake can also influence the spectral outputs in metabolomics studies. These influences may be strong enough to produce outlying samples on PCA score plots and may influence the results of some studies if not identified and removed from the data. In a study investigating the metabolic effects of chamomile tea consumption, several outliers were identified after initial PCA analysis because of high urinary TMAO excretion, which resulted from fish consumption within 16 h of sampling (18). Alcohol consumption on the day before sampling has also been reported as a cause of outlying samples (6). The effect of other dietary components, such as

1692

WALSH ET AL

FIGURE 2. A: Co-inertia analysis (CIA) of nuclear magnetic resonance (NMR) and mass spectrometry (MS) urinary data. Arrows are numbered with the subjects’ identification number. The gray arrows represent samples from the low-phytochemical-diet (LPD) group, and the black arrows represent samples from the standard-phytochemical-diet (SPD) group. B: CIA metabolite plot for NMR urinary data. Each closed triangle represents an NMR bin. Hippurate metabolites are highlighted in black on the upper right, and creatinine is highlighted on the lower left. C: CIA metabolite plot for MS urinary data. Each closed triangle represents an MS ion. Again, hippurate ions are highlighted on the upper right.

phytochemicals, is not yet known; therefore, the main focus of the current study was to further our understanding of the effect of phytochemicals on urinary composition. Assessment of nutrient intakes for the 21 volunteers found no significant differences in total energy, protein, carbohydrate, and fat intakes between the 3 dietary groups (ND, LPD, and SPD). Fiber intake was lower during the LPD than during the ND and SPD, but only significantly so for women. This was not unusual because of the removal of all fruit- and vegetable-type foods during the LPD. Fruit and vegetables were forbidden during the LPD and the SPD, but 6 g fiber/d was provided by the standard apple, carrot, and strawberry drinks during the SPD. Vitamin C and carotene intakes for the women were significantly lower during the LPD than during the ND (P 쏝 0.001) and SPD (P 쏝 0.001). Potatoes were permitted during the LPD and may have been the main source of vitamin C when other fruit and vegetables were removed from the diet. Carotene intakes were significantly higher during the SPD than during the ND (men: P 쏝 0.05; women: P 쏝 0.001) and LPD (men: P 쏝 0.001; women: P

쏝 0.001), and it is likely that the standard drinks were the source of carotene. Distinct differences in the urinary metabolic profiles were observed after the removal of fruit and vegetables from the diet (LPD) compared with the ND or the SPD. Consultation of the 1H NMR PLS-DA loadings plots indicated that the separation in both models was associated with hippurate excretion during the SPD and ND and by creatinine and methylhistidine excretion during the LPD. This indicates that hippurate excretion is associated with phytochemical intake and that the appearance of these 3 metabolites in urine is subject to variation based on recent dietary intake. Hippurate is produced by the conjugation of benzoic acid with glycine in the liver, and it is then excreted in urine. There are 2 main dietary routes that yield the production of hippurate. One is by the consumption of foods containing benzoic acid, which may be present naturally or added as a preservative, and the other is through the metabolism of plant phenols by the gut microflora. Phenylpropionic acids are produced by the microbial degradation of polyphenols in the colon, which are

METABOLOMICS IN HUMAN NUTRITION

further metabolized to benzoic acids and finally hippurate in the liver (8). Consumption of polyphenol-rich foods (8) and drinks such as black tea, green tea (19 –20), and chamomile tea (18) have all been associated with an increased excretion of urinary hippurate. Methylhistidine is produced by the catabolism of the myofibrillar proteins actin and myosin. Urinary 3-methylhistidine concentration has been used as a marker of myofibrillar protein turnover in human subjects (21), and, although intestinal smooth muscle and dietary intake also contribute to urinary excretion, skeletal muscle turnover is thought to contribute to 50 – 80% of urinary 3-methylhistidine excretion (22). Urinary creatinine is also related to muscle metabolism because it is produced from the breakdown of creatine phosphate. Creatinine excretion is related to body weight and has been proposed as a predictor of fat-free mass (23). One metabolomics study identified methylhistidine and creatinine as metabolites that contribute to differentiate between a high-meat and vegetarian diet (14). The dietary records in the current study did not indicate that there were any differences in meat consumption, and the nutrient analysis showed that there were no significant differences in protein or energy intakes across the study days. Therefore, it is likely that the only difference between the diets was in their phytochemical content. It is possible that other dietary factors, such as the glycemic load, influenced energy metabolism and muscle proteolysis. Multivariate data analysis of the MS data produced parallel results, although the distinction between the LPD and SPD was not as strong (69 앐 4% for the MS data compared with 83 앐 8% for the NMR data). The MS models showed a distinction between the diet containing phytochemicals and the LPD, and this separation was also dominated by an elevated hippurate excretion after phytochemical consumption. Other ions remain unidentified. The cross platform comparison of the NMR and MS data showed that both methods produced compatible results and identified the same metabolites in the differentiation between the LPD and the SPD as when the data sets were analyzed independently. Use of this type of analysis in the future will provide a powerful means of analyzing data from multiple platforms. In addition, it may aid in the identification of unknowns from different platforms. This study has provided insight into the effects of recent dietary intakes on the outcome of urinary metabolomics analysis. The results indicate that a varying phytochemical consumption can contribute to differences in urinary metabolic profiles. Therefore, dietary standardization or specific dietary restrictions during the 24 h before urine collection may improve data interpretation by reducing the confounding effects caused by diverse dietary intakes among study populations. We offer our sincere thanks to the volunteers for their commitment and patience during the study. The authors’ responsibilities were as follows—MCW: contributed to the study design and responsible for conducting the experiment, data interpretation and writing the manuscript; LB: responsible for conducting the experiment, data interpretation, and manuscript editing; EP-G, J-LS, and AS: responsible for MS analysis and manuscript editing; AF and DGH: responsible for data interpretation and manuscript editing; MJG: responsible for the conception and design of the experiment, data interpretation, and manuscript editing. None of the authors had any conflicts of interest.

1693

REFERENCES 1. Lenz EM, Bright J, Wilson ID, Morgan SR, Nash AFP. A 1H NMR-based metabonomic study of urine and plasma samples obtained from healthy human subjects. J Pharm Biomed Anal 2003;33:1103–15. 2. Lenz EM, Bright J, Wilson ID, et al. Metabonomics, dietary influences and cultural differences: a 1H NMR-based study of urine samples obtained from healthy British and Swedish subjects. J Pharm Biomed Anal 2004;36:841–9. 3. Bollard ME, Stanley EG, Lindon JC, Nicholson JK, Holmes E. NMRbased metabonomic approaches for evaluating physiological influences on biofluid composition. NMR Biomed 2005;18:143– 62. 4. Dumas ME, Maibaum EC, Teague C, et al. Assessment of analytical reproducibility of 1H NMR spectroscopy based metabonomics for largescale epidemiological research: the INTERMAP Study. Anal Chem 2006;78:2199 –208. 5. Kochhar S, Jacobs DM, Ramadan Z, Berruex F, Fuerholz A, Fay LB. Probing gender-specific metabolism differences in humans by nuclear magnetic resonance-based metabonomics. Anal Biochem 2006;352: 274 – 81. 6. Walsh MC, Brennan L, Malthouse JPG, Roche HM, Gibney MJ. Effect of acute dietary standardization on the urinary, plasma, and salivary metabolomic profiles of healthy humans. Am J Clin Nutr 2006;84:531–9. 7. Cassidy A, Dalais FS. Phytochemicals. In: Gibney MJ, Roche HM, ed. Nutrition and metabolism. Oxford, United Kingdom: Blackwell Science, 2003:307–17. 8. Rechner AR, Kuhnle G, Bremmer P, Hubbard GP, Moore KP, RiceEvans CA. The metabolic fate of dietary polyphenols in humans. Free Radic Biol Med 2002;33:220 –35. 9. Ito H, Gonthier MP, Manach C, et al. Polyphenol levels in human urine after intake of six different polyphenol-rich beverages. Br J Nutr 2005; 94:500 –9. 10. Manach C, Williamson G, Morand C, Scalbert A, Rémésy C. Bioavailability and bioefficacy of polyphenols in humans. I. Review of 97 bioavailability studies. Am J Clin Nutr 2005;81(suppl):230S– 42S. 11. Culhane AC, Thioulouse J, Perriere G, Higgins DG. MADE4: an R package for multivariate analysis of gene expression data. Bioinformatics 2005;21:2789 –90. 12. Chessel D, Dufour AB, Thioulouse J. The ade4 package—I: one-table methods. R News 2004;4:5–10. 13. Dolédec S, Chessel D. Co-inertia analysis: an alternative method for studying species— environment relationships. Freshwater Biol 1994; 31:277–94. 14. Stella C, Beckwith-Hall B, Cloarec O, et al. Susceptibility of human metabolic phenotypes to dietary modulation. J Proteome Res 2006;5: 2780 – 8. 15. Solanky KS, Bailey NJ, Beckwith-Hall B, et al. Biofluid 1H NMR-based metabonomic techniques in nutrition research—metabolic effects of dietary isoflavones in humans. J Nutr Biochem 2005;16:236 – 44. 16. Bertram HC, Bach Knudsen KE, Serena A, et al. NMR-based metabonomic studies reveal changes in the biochemical profile of plasma and urine from pigs fed high-fibre rye bread. Br J Nutr 2006;95:955– 62. 17. Gu H, Chen H, Pan Z, et al. Monitoring diet effects via biofluids and their implications for metabolomics studies. Anal Chem 2007;79:89 –97. 18. Wang Y, Tang H, Nicholson JK, Hylands PJ, Sampson J, Holmes E. A metabonomic strategy for the detection of the metabolic effects of chamomile (Matricaria recutita L.) ingestion. J Agric Food Chem 2005; 53:191– 6. 19. Mulder TP, Rietveld AG, van Amelsvoort JM. Consumption of both black tea and green tea results in an increase in the excretion of hippuric acid into urine. Am J Clin Nutr 2005;81(suppl):256S– 60S. 20. Van Dorsten FA, Daykin CA, Mulder TP, Van Duynhoven JP. Metabonomics approach to determine metabolic differences between green tea and black tea consumption. J Agric Food Chem 2006;54:6929 –38. 21. Chinkes DL. Methods for measuring tissue protein breakdown rate in vivo. Curr Opin Clin Nutr Metab Care 2005;8:534 –7. 22. Millward DJ, Bates PC. 3-methylhistidine turnover in the whole body, and the contribution of skeletal muscle and intestine to urinary 3-methylhistidine excretion in the adult rat. Biochem J 1983;214:607– 15. 23. Virgili F, Maiani G, Zahoor ZH, Ciarapica D, Raguzzini A, Ferro-Luzzi A. Relationship between fat-free mass and urinary excretion of creatinine and 3-methylhistidine in adult humans.

Vitamin D insufficiency in children, adolescents, and young adults with cystic fibrosis despite routine oral supplementation1–3 Alisha J Rovner, Virginia A Stallings, Joan I Schall, Mary B Leonard, and Babette S Zemel ABSTRACT Background: Cystic fibrosis (CF) with pancreatic insufficiency is associated with poor absorption of fat and fat-soluble vitamins, including vitamin D. Pancreatic enzyme supplementation does not completely correct fat malabsorption in CF patients. Objective: The objective of the study was to compare the vitamin D status of children, adolescents, and young adults with CF who were treated with routine vitamin D and pancreatic enzyme supplements with the vitamin D status of a healthy reference group from a similar geographic area. Design: Growth, dietary intake, and serum concentrations of 25-hydroxyvitamin D [25(OH)D], 1,25-dihydroxyvitamin D [1,25(OH)2D], and parathyroid hormone (PTH) were measured in 101 white subjects with CF and a reference group of 177 white subjects. Results: The median daily vitamin D supplementation in the CF group was 800 IU. The mean 앐 SD serum concentrations of 25(OH)D were 20.7 앐 6.5 ng/mL in the CF group and 26.2 앐 8.6 ng/mL in the reference group (P 쏝 0.001). Vitamin D deficiency and insufficiency were defined as 25(OH)D concentrations 쏝 11 ng/mL and 쏝 30 ng/mL, respectively. Seven percent of the CF group and 2% of the healthy reference group were vitamin D deficient (P 쏝 0.03). Ninety percent of the CF group and 74% of the healthy reference group were vitamin D insufficient (P 쏝 0.01). Twenty-five percent of the CF group and 9% of the healthy reference group had elevated PTH (P 쏝 0.006). The odds of vitamin D insufficiency in the CF group, compared with the healthy reference group, were 1.2 (95% CI: 1.1, 1.3) after adjustment for season and age. Conclusion: Despite daily vitamin D supplementation, serum 25(OH)D concentrations remain low in children, adolescents, and young adults with CF. Am J Clin Nutr 2007;86:1694 –9. KEY WORDS Cystic fibrosis, vitamin D, fat-soluble vitamins, children, adolescents, young adults INTRODUCTION

Cystic fibrosis (CF), the most common autosomal recessive disease in whites, affects multiple organ systems, including the lungs, the exocrine pancreas, and the hepatobiliary system. Approximately 90% of persons with CF have pancreatic insufficiency (PI), which causes malabsorption of fat (1). Treatment of PI includes supplementation with pancreatic enzymes; however, supplementation does not completely correct the fat malabsorption. In addition, in persons with CF and PI, fat-soluble vitamins (ie, vitamins A, D, E, and K) are malabsorbed. With the recognition that CF patients are at risk of osteopenia and osteoporosis,

1694

attention has been given to optimizing the intakes of vitamins and minerals that are important in bone mineralization, including vitamin D. Current Cystic Fibrosis Foundation (CFF) guidelines on vitamin D recommend supplementation with 욷800 IU/d for children 쏜1 y old(2, 3), an amount that is 4 times the Adequate Intake (4). However, weekly or biweekly doses of up to 12 000 IU (children 쏝5 y old) or 50 000 IU (children 쏜5 y old) may be needed to achieve normal 25(OH)D concentrations, which the CFF defines as between 30 and 60 ng/mL (75–150 nmol/L) (2). The CFF further recommends that 25(OH)D concentrations should be checked in the late autumn or winter when cutaneous synthesis is low. Previous studies in the United States and the United Kingdom have reported low 25(OH)D concentrations in children with CF, despite routine vitamin D supplementation (5– 8). However, not all of those studies included control groups or examined vitamin D status by season. Reports of frequent vitamin D insufficiency in otherwise healthy children have underscored the importance of including in a study a healthy comparison group to improve the understanding of the magnitude of the problem in CF (9 –11). In addition, consideration of seasonal fluctuations in vitamin D concentrations is essential in evaluating the prevalence of hypovitaminosis D. Studies in adults with CF have reported low vitamin D concentrations also, which suggests that the vitamin D concentration is a concern for CF patients throughout life (12, 13). The purpose of the present study was to compare the vitamin D status in children, adolescents, and young adults with CF who were being treated with routine vitamin D and pancreatic enzyme supplements with the status in a healthy, white, reference group from a similar geographic area. 1

From the Divisions of Gastroenterology, Hepatology and Nutrition (AJR, VAS, JIS, and BSZ) and Nephrology (MBL), Department of Pediatrics, The Children’s Hospital of Philadelphia, and the the Department of Biostatistics and Epidemiology, Center for Clinical Epidemiology and Biostatistics (MBL), University of Pennsylvania School of Medicine, Philadelphia, PA. 2 Supported by the Cystic Fibrosis Foundation and the General Clinical Research Center (5-MO1-RR-000240) and The Nutrition Center at The Children’s Hospital of Philadelphia. 3 Reprints not available. Address correspondence to BS Zemel, Division of Gastroenterology, Hepatology and Nutrition, The Children’s Hospital of Philadelphia, CHOP North, Room #1560, 34th and Civic Center Boulevard, Philadelphia, PA 19104. E-mail: [email protected]. Received May 4, 2007. Accepted for publication July 30, 2007.

Am J Clin Nutr 2007;86:1694 –9. Printed in USA. © 2007 American Society for Nutrition

VITAMIN D INSUFFICIENCY IN CYSTIC FIBROSIS SUBJECTS AND METHODS

Subjects Subjects with CF aged 8 –25 y were recruited from 3 CFFaccredited CF centers (2 pediatric and 1 adult) in Pennsylvania (latitude: 39° 43' N to 42° N) between November 2000 and February 2002. A diagnosis of CF was established either by clinical signs confirmed by a sweat test 쏜 60 mEq/L or by a positive genotype analysis of pancreatic insufficiency. Exclusion criteria were forced expiratory volume in 1 s (FEV1) 쏝40% of that predicted, CF-related diabetes, or any other medical illness known to affect growth or bone health. Study-related measurements were taken while the subjects were in their usual state and when they had not had an exacerbation within the previous 2 wk. The reference group included children aged 6 –21 y who were recruited from the greater Philadelphia area between December 2000 and April 2004 as part of a study of skeletal development in healthy children (11). The exclusion criterion for the reference group was the presence of any disease or the use of any medication known to affect growth, nutritional status, or bone health. For the purposes of this study, the reference group was restricted to the 177 white subjects, because most persons with CF are white (1), and vitamin D concentrations vary significantly with race (11). Subjects 쏜18 y old provided written informed consent, as did the parent or guardian of each subject 쏝18 y old; in addition, assent was obtained from children 7–18 y old. The institutional review boards of The Children’s Hospital of Philadelphia and of each participating institution approved the protocol. Anthropometry Body weight was measured with an electronic scale that is accurate to 0.1 kg (Scalatronix Inc, Wheaton, IL), and standing height was measured with a stadiometer that is accurate to 0.1 cm (Holtain, Crymych, United Kingdom); standard research techniques were used (14). Measurements were taken in triplicate by a research anthropometrist, and the average was used. Weight, height, and body mass index (in kg/m2) were compared with the Centers for Disease Control and Prevention 2000 growth charts (15), and age- and sex-specific z scores were calculated. Pulmonary function Pulmonary status was evaluated in the CF group by using standard pulmonary function methods. FEV1 was compared with reference values and reported as the percentage of the predicted value (16). Wang equations were used for the males 쏝18 y old and the females 쏝16 y old, and Hankinson equations were used for the males 쏜18 y old and the females 쏜16 y old (17, 18). Vitamin D metabolites and parathyroid hormone A nonfasting blood sample was drawn between 0800 and 1700 for measurement of serum concentrations of 25-hydroxyvitamin D [25(OH)D], 1,25-dihydroxyvitamin D [1,25(OH)2D], and parathyroid hormone (PTH). Intact PTH was analyzed in the Clinical Laboratory of The Children’s Hospital of Philadelphia. Serum for vitamin D analysis was stored in aliquots at Ҁ70 °C and shipped in batches for analysis. The serum vitamin D concentrations in both groups were analyzed by using the same

1695

radioimmunoassay (DiaSorin, Inc, Stillwater, MN) with a radioiodinated tracer (19). Serum 25(OH)D and 1,25(OH)2D concentrations in the CF group were analyzed by the Bruce W Hollis Laboratory Services (Mount Pleasant, SC), and those in the healthy reference group were analyzed at the Nichols Research Institute (NRI; Quest Diagnostics, San Juan Capistrano, CA). To assess agreement between the 2 laboratories, the 25(OH)D and 1,25(OH)2D concentrations in 81 samples from the same subjects were analyzed at both laboratories. Although the values were highly correlated between laboratories [r ҃ 0.96 for 25(OH)D; r ҃ 0.92 for 1,25(OH)2D], values for both vitamin D assays from the NRI laboratory were systematically higher than those from Bruce W Hollis Laboratory Services. Because the head of that laboratory (Bruce Hollis) developed the radioimmunoassays, the results from that laboratory were used as the reference, and an adjustment was made to 25(OH)D and 1,25(OH)2D results from the NRI laboratory to account for differences. On the basis of regression models, the values from the NRI laboratories were adjusted with the use of the following equations:

Adjusted 25共OH兲2D Quest Diagnostics values ⫽ ⫺ 2.03 ⫹ Quest Diagnostics values 共0.83兲

(1)

and

Adjusted 1,25共OH兲2D Quest Diagnostics values ⫽ 24.73 ⫹ Quest Diagnostics values 共0.30兲

(2)

Dietary and supplemental intakes Dietary intake was collected by using 3-d prospective weighed food records in the CF group and three 24-h recalls in the healthy reference group. All diet records were analyzed with NUTRITION DATA SYSTEM for RESEARCH software (NDS-R, version 4.04; Nutrition Coordinating Center, University of Minnesota, Minneapolis, MN) and vitamin D and calcium intakes were compared with the Dietary Reference Intakes and reported as a percentage of the Adequate Intake (4). Information on the use of vitamin and mineral supplements was obtained by questionnaire. The nutrient content and the dose of each supplement were used to calculate the supplemental calcium and vitamin D intakes. Statistical analysis Continuous variables were expressed as means 앐 SDs if normally distributed or as medians with interquartile (25–75%) ranges if nonnormally distributed. Categorical variables were presented by frequency distributions. Vitamin D deficiency was defined as 25(OH)D concentrations 쏝11 ng/mL, according to the Institute of Medicine’s definition in the Dietary Reference Intakes (4). Vitamin D insufficiency was defined as 쏝30 ng/mL because 1) the most advantageous serum concentrations for multiple health outcomes are 쏜30 ng/mL (20), and 2) the CFF recommends that 25(OH)D concentrations should be maintained at 쏜30 ng/mL (2). Vitamin D sufficiency was defined as 25(OH)D concentrations 욷 30 ng/mL. Vitamin D concentrations were reported by season; the seasons were categorized as winter (December, January, and February), spring (March, April, and May), summer (June, July, and August), and fall (September, October, and November). Normal

1696

ROVNER ET AL

TABLE 1 Characteristics of subjects with cystic fibrosis (CF) and the healthy reference group1 Subjects with CF (n ҃ 101) Females [n (%)] Males [n (%)] Age (y) Weight-for-age z score Height-for-age z score BMI z score FEV1 (% of predicted) 25(OH)D concentration (ng/mL) 1,25(OH)2D concentration (pg/mL)

50 (50) 51 (50) 14.8 앐 4.2 Ҁ0.57 앐 1.00 Ҁ0.53 앐 1.00 Ҁ0.36 앐 0.92 84 앐 19% 20.7 앐 6.5 36.1 앐 8.1

Healthy reference group (n ҃ 177) 104 (58) 73 (42) 12.5 앐 3.52 0.34 앐 0.742 0.25 앐 0.792 0.28 앐 0.792 26.2 앐 8.62 43.0 앐 4.82

1 FEV1, forced expiratory volume in 1 s; 25(OH)D, 25-hydroxyvitamin D; 1,25(OH)2D, 1,25-dihydroxyvitamin D. Student’s t test was used to assess group differences. 2 Significantly different from the CF group, P 쏝 0.001.

PTH concentrations were defined as between 9 and 52 pg/mL. Student’s t test was used to assess group differences in 25(OH)D status for normally distributed variables, and the Mann Whitney U test was used for variables whose distributions were not normal. Logistic regression analysis was used to calculate the odds of vitamin insufficiency in the CF group compared with the reference group after adjustment for season and age. Because vitamin D deficiency is common, we used the method described by Zhang and Yu (21) to adjust the odds ratio for frequent outcomes, in an attempt to avoid overestimation of the relative risk. The relation between 25(OH)D and PTH concentrations was first analyzed by linear regression, and a group interaction was observed. The relation between 25(OH)D and PTH was then analyzed separately for the 2 groups with the use of Spearman correlations. Statistical significance was defined as P 울 0.05. Statistical analyses were performed with STATA software (version 9.0; Stata Corp, College Station, TX). RESULTS

A CF group (n ҃ 101; n ҃ 50 females) and a healthy reference group (n ҃ 177; n ҃ 104 females) participated in the study. Characteristics of the study sample are presented in Table 1. The growth deficits of this sample are consistent with patterns seen in children and young adults with CF in the United States (1). The mean serum concentrations of 25(OH)D were 20.7 앐 6.5 ng/mL in the CF group and 26.2 앐 8.6 ng/mL in the reference group (P 쏝 0.001). There were seasonal fluctuations in 25(OH)D concentrations in both groups; during every season, 25(OH)D concentrations were significantly (P 쏝 0.01) lower in the CF group than in the healthy reference group (Figure 1). Seven percent of subjects with CF and 2% of healthy subjects were vitamin D deficient [ie, 25(OH)D concentrations 쏝 11 ng/mL (P 쏝 0.01)]. Ninety subjects with CF (90%) and 130 subjects in the reference group (74%) were vitamin D insufficient (쏝30 ng/mL). There were no sex differences in vitamin D insufficiency in either group. Mean 1,25(OH)2 D concentrations were significantly (P 쏝 0.001) lower in the CF group than in the healthy subjects (36.1 앐 8.1 and 43.0. 앐 4.8 pg/mL, respectively). There were no significant associations between 25(OH)D concentrations and FEV1, liver function, or growth measurements in

FIGURE 1. Box plots of the distribution of seasonal variations in 25hydroxyvitamin D [25(OH)D] concentrations in subjects with cystic fibrosis (CF) and in a healthy reference group. The bottom line of each box represents the 25th percentile, the middle line represents the median, and the top line represents the 75th percentile. F, an outlier. Seasons were defined as summer (June, July, and August), fall (September, October, and November), winter (December, January, and February), and spring (March, April, and May). Both season and group were significant, P 쏝 0.001. The fall concentrations of 25(OH)D did not differ significantly from those in the summer, but winter and spring concentrations differed significantly (P 쏝 0.01) from those in the summer. The season ҂ group interaction was not significant. Group differences were examined by season with the use of the Mann-Whitney U test; in every season, 25(OH)D concentrations were significantly (P 쏝 0.01) lower in the CF group than in the healthy reference group.

the CF group. Serum 25(OH)D concentrations were negatively associated with age in the healthy reference group but not in the CF group. Logistic regression analysis indicated that the odds of vitamin D insufficiency in the CF group were 1.2 (95% CI: 1.1, 1.3) greater than those in the healthy subjects after adjusting for season and age (Table 2). Serum 25(OH)D concentrations were examined in relation to PTH concentrations (Figure 2). Twenty-five subjects with CF (25%) and 16 subjects in the reference group (9%) had elevated PTH. There was no significant correlation between PTH and 25(OH)D concentrations (r ҃ Ҁ0.16, P ҃ 0.17) in the CF group. However, there was a significant, negative correlation between PTH and 25(OH)D in the healthy reference group (r ҃ Ҁ0.23, P 쏝 0.002). TABLE 2 Multiple logistic regression model for vitamin D insufficiency1 Odds ratio2 (95% CI) Cystic fibrosis Age Season (summer as reference group) Fall Winter Spring

1.2 (1.1, 1.3) 1.0 (0.9, 1.0) 1.1 (0.8, 1.2) 1.3 (1.1, 1.4) 1.0 (1.0, 1.3)

1 Defined as serum 25-hydroxyvitamin D concentrations 쏝 30 ng/mL (2, 20). 2 Logistic regression analysis was used to calculate the odds of vitamin insufficiency. To avoid overestimation of the relative risk for frequent outcomes, the odds ratio was adjusted by the method described by Zhang and Yu (21).

1697

VITAMIN D INSUFFICIENCY IN CYSTIC FIBROSIS

FIGURE 2. Vitamin D and parathyroid hormone (PTH) concentrations in subjects with cystic fibrosis (CF) and in a healthy reference group. The normal range for intact PTH was between 9 and 52 pg/mL (represented by the gray shading). The relation between 25-hydroxyvitamin D [25(OH)D] and PTH concentrations was analyzed by using Spearman’s correlation. There was no significant correlation between PTH and 25(OH)D (r ҃ Ҁ0.16, P ҃ 0.17) in the CF group. However, there was a significant, negative correlation between PTH and 25(OH)D in the healthy reference group (r ҃ Ҁ0.23, P 쏝 0.002).

All CF patients provided labels from their dietary supplements, and 70 (70%) of these patients returned the 3-d diet records. Dietary and supplement data for vitamin D and calcium are shown in Table 3. All CF patients were taking a supplement that contained vitamin D, and 16% were taking one that contained calcium. Nine percent of subjects in the reference group were taking a supplement that contained calcium, and 23% were taking one that contained vitamin D. There was no significant relation between the dietary intake of vitamin D and the serum 25(OH)D concentrations in either group. DISCUSSION

In this sample of children, adolescents, and young adults with CF, PI, and mild-to-moderate lung disease, 7% had vitamin D deficiency and 90% had vitamin D insufficiency, despite routine supplementation with vitamin D and pancreatic enzymes. Vitamin D deficiency and insufficiency occurred most often during the winter. However, vitamin D deficiency also occurred in the spring and fall, and insufficiency occurred in all seasons.

Reports suggested that vitamin D insufficiency is common in healthy children and that it varies by season and ethnicity (9 –11). A study of 11–18-y-olds from an adolescent outpatient center in Boston, MA, found that 42% had vitamin D insufficiency and that 25(OH)D concentrations were 24% lower in the winter than in the summer (10). Another study of participants (12–19 y old) in the third National Health and Nutrition Examination Survey found that 25% of males and 47% of females living at lower latitudes and 21% of males and 28% of females living at higher latitudes had vitamin D insufficiency in the winter (9). The present study, which used a healthy reference group of whose ethnicity and latitude of dwelling were similar to those of the CF group, and which adjusted for seasonal fluctuations, showed 20% increase in vitamin D insufficiency in children, adolescents, and young adults with CF on routine treatment with pancreatic enzyme replacement and vitamin D supplementation. Data from the present study and from previously published studies showing low vitamin D concentrations in CF patients suggested either that this vitamin is inadequately absorbed, even

TABLE 3 Daily intake of energy, vitamin D, and calcium1 Dietary intake

Energy (kcal/d) Calcium (mg/d) (%AI) Vitamin D (IU/d) (%AI) 1 2

Supplemental intake

Cystic fibrosis group

Healthy reference group

Cystic fibrosis group

Healthy reference group

2644 (2056, 4780)

1823 (1525, 2618)2

1423 (1016, 3130) 110 (78, 241)

848 (619, 1685)2 65 (48, 130)2

0 (0, 945) 0 (0, 73)

0 (0, 100) 0 (0, 8)

329 (212, 957) 164 (106, 479)

189 (120, 474)2 95 (60, 237)2

800 (400, 1600) 400 (200, 800)

0 (0, 400)2 0 (0, 200)2

All values are median; 25th and 75th percentiles in parentheses. AI, adequate intake. Wilcoxon-Mann-Whitney test was used to assess group differences. Significantly different from cystic fibrosis group, P 쏝 0.0001.

1698

ROVNER ET AL

in the presence of pancreatic enzyme replacement, or that current doses are not high enough to achieve target 25(OH)D concentrations (5). A recent study of 134 adults with CF (19 – 64 y old) in Baltimore, MD, found that 109 (81%) of the 134 had 25(OH)D concentrations 쏝30 ng/mL (median: 21.5 앐 10.8 ng/mL) (13). Sixty-six of those 109 adults underwent the CFF’s vitamin D– repletion protocol—ie, supplementation with 50 000 IU ergocalciferol (vitamin D2)/wk for 2 mo—and only 8% had serum 25(OH)D concentrations that rose to 쏜30 ng/mL (13). Aris et al (12) compared the absorption of ergocalciferol in 10 adults with CF (18 – 45 y old) and PI with that in 10 healthy, age-, sex-, and race-matched controls. Even though the CF patients took their pancreatic enzymes with the dose of vitamin D, they absorbed less than half the amount of vitamin D that was absorbed by the control subjects. In addition, absorption was quite erratic in the CF patients: 20% absorbed virtually no vitamin D2. It has been shown that cholecalciferol (vitamin D3) increases serum 25(OH)D more efficiently than does ergocalciferol in healthy people (22, 23); therefore, it is possible that absorption would have been greater in the abovementioned studies if cholecalciferol were the form of vitamin D used. To date, no study has examined vitamin D absorption in children with CF. With the increased life expectancy of CF patients, osteopenia and osteoporosis have become major concerns (2). Although the cause of bone disease in CF patients is likely to be multifactorial, one of the modifiable factors that influences bone mass is vitamin D status. Suboptimal vitamin D status may prevent children from reaching their genetic potential for peak bone mass. The compensatory secondary hyperparathyroidism mobilizes calcium from the skeleton and thus reduces bone mass. Therefore, optimizing vitamin D status in CF patients is critical, especially in light of the other risk factors they have for metabolic bone disease, including glucocorticoid therapy, decreased physical activity, reduced body weight, and hypogonadism. There is increasing recognition of the role of vitamin D in muscle function, innate immunity, and the development of cardiovascular disease, diabetes, and some cancers (24). The broader effects of vitamin D insufficiency in CF patients remain to be determined. The CFF recommends supplementation with 800 IU vitamin D/d for children and young adults (2). The subjects in this study reported vitamin D supplementation intakes in keeping with this recommendation. Data from our study, as well as from studies by others, suggest that this dose of vitamin D is too low to maintain the desired 25(OH)D concentrations between 30 and 60 ng/mL in CF patients. Cutaneous synthesis from sunlight exposure is an alternative way of increasing vitamin D concentrations. Unfortunately, for those who live above 앒35° latitude, vitamin D synthesis in the skin does not occur during the winter months (25, 26). Ultraviolet B radiation from a home unit or a tanning bed can be used to synthesize vitamin D. To date, one published Swedish study has examined changes in 25(OH)D concentrations in response to ultraviolet B radiation from fluorescent lamps in 30 children and adults (9 – 40 y old) with CF and mild-to-moderate lung disease (27). Although this was a small study, it provides initial evidence that ultraviolet B lamps may be used to improve serum 25(OH)D concentrations in CF patients, and future largerscale studies are warranted. Our data suggest that vitamin D insufficiency was present in most of the children, adolescents, and young adults with CF in all seasons, despite routine vitamin D supplementation. Future efforts should focus on identifying the optimal dose needed to

maintain 25(OH)D concentrations between 30 and 60 ng/mL. In light of the many risk factors that CF patients have for osteopenia and osteoporosis, careful attention should be given to maintaining adequate vitamin D status. We thank the cystic fibrosis centers at the Children’s Hospital of Philadelphia, The Hospital of the University of Pennsylvania, and the Hershey Medical Center and the study participants and their families. We also thank Rita Herskovitz, the study coordinator, and Kate Temme, the research assistant, for their many contributions to this study. The authors’ responsibilities were as follows—AJR: analysis of the data and writing of the manuscript; VAS: the design and conduct of the study; JIS: collection and management of the data; MBL: provision of the data on laboratory comparisons and assistance in writing the manuscript; and BSZ: the designing of the study, analysis of the data, and writing of the manuscript. None of the authors had a personal or financial conflict of interest.

REFERENCES 1. Cystic Fibrosis Foundation. Cystic Fibrosis Foundation, patient registry 2004. Bethesda, MD: Cystic Fibrosis Foundation, 2005. 2. Aris RM, Merkel PA, Bachrach LK, et al. Guide to bone health and disease in cystic fibrosis. J Clin Endocrinol Metab 2005;90:1888 –96. 3. Borowitz D, Baker RD, Stallings V. Consensus report on nutrition for pediatric patients with cystic fibrosis. J Pediatr Gastroenterol Nutr 2002; 35:246 –59. 4. Institute of Medicine. Dietary reference intakes for calcium, phosphorus, magnesium, vitamin D and fluoride. Washington, DC: National Academy Press, 1997. 5. Chavasse RJ, Francis J, Balfour-Lynn I, Rosenthal M, Bush A. Serum vitamin D levels in children with cystic fibrosis. Pediatr Pulmonol 2004; 38:119 –22. 6. Henderson RC, Lester GE. Vitamin D levels in children with cystic fibrosis. South Med J 1997;90:378 – 83. 7. Feranchak AP, Sontag MK, Wagener JS, Hammond KB, Accurso FJ, Sokol RJ. Prospective, long-term study of fat-soluble vitamin status in children with cystic fibrosis identified by newborn screen. J Pediatr 1999;135:601–10. 8. Mortensen LA, Chan GM, Alder SC, Marshall BC. Bone mineral status in prepubertal children with cystic fibrosis. J Pediatr 2000;136:648 –52. 9. Looker AC, Dawson-Hughes B, Calvo MS, Gunter EW, Sahyoun NR. Serum 25-hydroxyvitamin D status of adolescents and adults in two seasonal subpopulations from NHANES III. Bone 2002;30:771–7. 10. Gordon CM, DePeter KC, Feldman HA, Grace E, Emans SJ. Hypovitaminosis D in healthy adolescents. Arch Pediatr Adolesc Med 2004; 158:531–7. 11. Weng FL, Shults J, Leonard M, Stallings V, Zemel B. Risk factors for vitamin D deficiency in otherwise healthy children. Am J Clin Nutr 2007;86:150 – 8. 12. Aris RM, Lester GE, Dingman S, Ontjes DA. Altered calcium homeostasis in adults with cystic fibrosis. Osteoporos Int 1999;10:102– 8. 13. Boyle MP, Noschese ML Watts SL, et al. Failure of high-dose ergocalciferol to correct vitamin D deficiency in adults with cystic fibrosis. Am J Respir Crit Care Med 2005;172:212–7. 14. Lohman T, Roche AR, Martorell R. Anthropometric standardization reference manual. Champaign, IL: Human Kinetics, 1988. 15. Kuczmarski RJ, Ogden CL, Guo SS, et al. 2000 CDC growth charts for the United States: methods and development. Vital Health Stat 11 2002; 1–190. 16. Morris A, Kanner RE, Crapo R, Gardner RM. Clinical pulmonary function testing: a manual of uniform laboratory procedures. Salt Lake City, UT: Intermountain Thoracic Society, 1984:1–24. 17. Hankinson JL, Odencrantz JR, Fedan KB. Spirometric reference values from a sample of the general U.S. population. Am J Respir Crit Care Med 1999;159:179 – 87. 18. Wang X, Dockery DW, Wypij D, Fay ME, Ferris BG. Pulmonary function between 6 and 18 years of age. Pediatr Pulmonol 1993;15:75– 88. 19. Hollis BW, Kamerud JQ, Selvaag SR, Lorenz JD, Napoli JL. Determination of vitamin D status by radioimmunoassay with an 125I-labeled tracer. Clin Chem 1993;39:529 –33.

VITAMIN D INSUFFICIENCY IN CYSTIC FIBROSIS 20. Bischoff-Ferrari HA, Giovannucci E, Willett WC, Dietrich T, DawsonHughes B. Estimation of optimal serum concentrations of 25hydroxyvitamin D for multiple health outcomes. Am J Clin Nutr 2006; 84:18 –28. 21. Zhang J, Yu K. What’s the relative risk? A method of correcting the odds ratio in cohort studies of common outcomes. JAMA 1998;280: 1690 –1. 22. Trang HM, Cole DE, Rubin LA, Pierratos A, Siu S, Vieth R. Evidence that vitamin D3 increases serum 25-hydroxyvitamin D more efficiently than does vitamin D2. Am J Clin Nutr 1998;68:854 – 8. 23. Houghton LA, Vieth R. The case against ergocalciferol (vitamin D2) as a supplement. Am J Clin Nutr 2006;84:694 –7.

1699

24. Holick MF. The vitamin D epidemic and its health consequences. J Nutr 2005;135(suppl):2739S– 48S. 25. Ladizesky M, Lu Z, Oliveri B, et al. Solar ultraviolet B radiation and photoproduction of vitamin D3 in central and southern areas of Argentina. J Bone Miner Res 1995;10:545–9. 26. Webb AR, Kline L, Holick MF. Influence of season and latitude on the cutaneous synthesis of vitamin D3: exposure to winter sunlight in Boston and Edmonton will not promote vitamin D3 synthesis in human skin. J Clin Endocrinol Metab 1988;67:373– 8. 27. Gronowitz E, Larko O, Gilljam M, et al. Ultraviolet B radiation improves serum levels of vitamin D in patients with cystic fibrosis. Acta Paediatr 2005;94:547–52.

How early dietary factors modify the effect of rapid weight gain in infancy on subsequent body-composition development in term children whose birth weight was appropriate for gestational age1–3 Nadina Karaolis-Danckert, Anke LB Gu¨nther, Anja Kroke, Claudia Hornberg, and Anette E Buyken ABSTRACT Background: It is not clear whether the adverse effects of rapid weight gain in infancy are modified by nutrition during the first 2 y of life in term children whose birth weight was appropriate for gestational age (AGA). Objective: We examined the interaction between rapid weight gain and nutrition in infancy and early childhood and their effect on body fat percentage (BF%) trajectories between 2 and 5 y of age. Design: The study population comprised 249 (51.4% female) term AGA participants of the Dortmund Nutritional and Anthropometric Longitudinally Designed Study, for whom repeated anthropometric measurements until 5 y of age and information on breastfeeding status and on diet at 12 and 18 –24 mo of age were available. Results: Multilevel model analyses showed that, among rapid growers, those who had been fully breastfed for 욷4 mo had a lower BF% at 2 y of age than did those who had not been fully breastfed for 욷4 mo (␤ 앐 SE: Ҁ1.53 앐 0.59%; P ҃ 0.009). This difference persisted until 5 y. Furthermore, those rapid growers who had a consistently high fat intake at both 12 and 18 –24 mo did not show the expected physiologic decrease in BF% between 2 and 5 y seen in those rapid growers with an inconsistent or consistently low fat intake at these time points (0.73 앐 0.26%/y; P ҃ 0.006). Conclusions: Among rapid growers, full breastfeeding for 욷4 mo is protective against a high BF% at 2 y of age, whereas a consistently high fat intake in the second year of life “inhibits” the physiologic decrease in BF% between 2 and 5 y. Am J Clin Nutr 2007;86: 1700 – 8. KEY WORDS Rapid weight gain, appropriate for gestational age, nutrition, breastfeeding, body fat percentage, children, trajectories

and the differences in comparison with normal growers appear to become progressively larger over time (6, 7). In developed countries, where the prevalence of overweight and obesity is on the rise, most children have a birth weight that is AGA. The failure of currently available methods for treating established obesity has resulted in an emphasis on primary prevention and on identifying modifiable risk factors (and susceptible individuals) as early as possible. With respect to rapid weight gain, it remains unclear which factors stimulate, modify, or amplify its effect on growth and body-composition trajectories. Nutrition is an important regulator of growth, and an effect of prenatal nutrition on subsequent growth, metabolism, and obesity risk is well established (8). Several meta-analyses have also emphasized the role of the postnatal diet in infancy, and breastfeeding status in particular, on obesity risk (9, 10). Because breastfed children grow more slowly than do formula-fed children (11), it is conceivable that breastfeeding status could influence or modify the effect of rapid weight gain. On the other hand, the potential role of diet during the complementary feeding (CF) phase has received comparatively little attention, although this is a period when important quantitative and qualitative changes in the diet occur (12). The transition to the family diet is marked by a considerable increase in protein intake and a reduction in fat intake, the speed, degree, and benefits of which are still being debated (12, 13). Nevertheless, this phase may influence growth, and fat mass development in particular, as was suggested by a recent analysis of ours in which high protein intakes in the second year of life were associated with a larger body mass index SD score (BMI SDS) and body fat percentage (BF%) at age 7 y (14). Using data from the Dortmund Nutritional and Anthropometric Longitudinally Designed (DONALD) Study, we sought to 1

INTRODUCTION

The contribution of rapid weight gain in infancy to the development of overweight and adiposity later in life (1, 2), as well as to several other morbidities, including cardiovascular disease (3) and cancer (4), has now been repeatedly reported for a variety of populations. Recent evidence suggests not only that children born small for gestational age (SGA) subsequently compensate for this in utero growth restriction by growing rapidly (5), but also that this phenomenon is seen in some children whose birth weight and length were appropriate for gestational age (AGA) (6). In both cases, rapid weight gain results in a greater fat mass,

1700

From the Research Institute of Child Nutrition, Rheinische FriedrichWilhelms-Universita¨t Bonn, Dortmund, Germany (NK-D, ALBG, and AEB); the Department of Nutrition, Food and Consumer Sciences, Fulda University of Applied Sciences, Fulda, Germany (AK); and the Faculty of Health Sciences, University of Bielefeld, Bielefeld, Germany (CH). 2 The DONALD Study is funded by the Ministry of Science and Research of North Rhine Westphalia, Germany. ALBG also has a research grant from the International Foundation for the Promotion of Nutrition Research and Nutrition Education. 3 Reprints not available. Address correspondence to N Karaolis-Danckert, Nutrition and Health Unit, Research Institute of Child Nutrition, Heinstueck 11, 44225 Dortmund, Germany. E-mail: [email protected]. Received February 21, 2007. Accepted for publication July 31, 2007.

Am J Clin Nutr 2007;86:1700 – 8. Printed in USA. © 2007 American Society for Nutrition

RAPID WEIGHT GAIN AND NUTRITION IN AGA CHILDREN

determine whether nutrition in infancy and early childhood modifies the association between rapid weight gain and BF% and BMI SDS trajectories between 2 and 5 y of age in healthy, AGA children.

SUBJECTS AND METHODS

Study population The DONALD Study is an ongoing, open-cohort study conducted by the Research Institute of Child Nutrition in Dortmund, Germany. This study has been previously described in detail (15). Briefly, since recruitment began in 1985, detailed information on diet, growth, development, and metabolism between infancy and adulthood has been collected from 쏜1100 children. Every year, an average of 40 –50 infants are newly recruited and first examined at the age of 3 mo. Each child returns for 3 more visits in the first year, 2 in the second, and then once annually until early adulthood. The study was approved by the Ethics Committee of the University of Bonn, and all examinations are performed with parental consent. The ages of the children who were initially recruited for the DONALD Study were variable so that information on the first few years of life was not always available. In addition, many children have not yet reached 5 y of age. Therefore, 1) a minimum of anthropometric measurements at 0.5, 2, and 5 y were available for 408 term (37– 42 wk gestation) singletons with a birth weight 쏜2500 g; 2) this number was reduced to those whose birth weights and lengths were AGA (n ҃ 326), ie, all birth weights and lengths lay between the 10th and 90th percentiles of the German sex-specific birth weight–for– gestational age curves (16, 17); and 3) finally, all children had to have complete, plausible dietary records at ages 12 mo and 18 or 24 mo (n ҃ 253 remaining) (18), complete information on breastfeeding, and complete information on maternal characteristics (BMI and educational status) (n ҃ 249 remaining). Hence, the subcohort analyzed here included 249 AGA term singletons (51.4% female). The mean number of measurements per child was 7.95 (range: 7– 8). Anthropometry The DONALD Study participants are measured at each visit according to standard procedures (19). They are dressed in underwear only and are barefoot. The trained nurses who conduct the measurements undergo an annual quality control during which inter- and intraobserver agreements are carefully monitored. Recumbent length in children under 2 y of age is measured to the nearest 0.1 cm with a Harpenden (Crymych, United Kingdom) stadiometer. From the age of 2 y onward, standing height is measured to the nearest 0.1 cm with a digital stadiometer. Weight is measured to the nearest 0.1 kg with an electronic scale (753 E; Seca, Hamburg, Germany). Skinfold thicknesses are measured from the age of 6 mo onward on the right side of the body at the biceps, triceps, subscapular, and suprailiac sites to the nearest 0.1 mm with a Holtain caliper. On their child’s admission to the study, the parents are interviewed by the study pediatrician and are weighed and measured by the study nurses using the same equipment as for children from 2 y onward. Information on birth weight, length and head circumference at birth, gestational age, and maternal weight gain

1701

during pregnancy are abstracted from the Mutterpass, a standardized document given to all pregnant women in Germany. Anthropometric calculations Sex- and age-independent SDS were calculated by using the German reference curves for weight and body mass index [BMI; in kg/m2 (20)] and the Tanner and Whitehouse reference data for subscapular and triceps skinfold-thickness measurements (21). To remove general deviations of our sample from the reference data, these variables were internally standardized. BF% was calculated by using the Deurenberg equations (22). Overweight at age 5 y was defined according to the International Obesity Task Force BMI cutoffs for children, which correspond to an adult BMI of 25 (23). Overfatness at 5 y of age was defined as a BF% higher than the 85th percentile of the sex-specific body fat reference curves by McCarthy et al (24). Rapid weight gain was defined as an increase in weight SDS of 쏜0.67 between birth and 24 mo, as recommended by Monteiro and Victora (1). The value of 0.67 SDS represents the width of each percentile band on standard growth charts, ie, between the 25th and the 50th percentiles, between the 50th and the 75th, and so on, and indicates clinically significant rapid weight gain (25). Diet in infancy: breastfeeding At the infant’s first visit (ie, when the infants are 3 or 6 mo old), the study pediatrician asks the mothers about how long (in weeks) they had fully breastfed their child for. The definition of full breastfeeding (breast milk only without any supplemental solid foods or liquids other than tea or water) is carefully explained. If the mother is still fully breastfeeding, this question is repeated at each subsequent visit (eg, 6, 9, and 12 mo) until complementary feeding is initiated. In addition, for 앒70% of infants, on average, the mothers also keep 3-d weighed dietary records during the first year of life, so that infant feeding can be quantified at 3, 6, 9, or 12 mo. When the study dietitians collect the protocols, they also question the mothers about the introduction of formula or solid foods. Finally, consistency checks to identify possible sources of error are made by comparing the information from the breast milk records with that collected by the pediatricians and the dietitians. Children in this analysis were grouped into those who had been fully breastfed for 욷4 mo (defined as full breastfeeding for 욷17 wk) and those who had not. This cutoff was used because it is the lower limit of the German recommendation for the introduction of complementary foods (26). Diet in early childhood: 12 and 18 –24 mo The dietary intake of DONALD participants is assessed by use of 3-d weighed dietary records. Parents are asked to weigh all foods and beverages consumed by their children, including leftovers, to the nearest 1 g over 3 consecutive days with the use of regularly calibrated electronic food scales [initially Soehnle Digita 8000 (Leifheit AG, Nassau, Germany), now WEDO digi 2000 (Werner Dorsch GmbH, Muenster/Dieburg, Germany)]. The children’s parents are instructed by trained dietitians, and semi-quantitative measures (eg, number of spoons, scoops, etc) are allowed when exact weighing is not possible. Information on recipes and on the types and brands of food items consumed is also requested. The dietary records are analyzed by using the continuously updated in-house nutrient database LEBTAB (27),

1702

KARAOLIS-DANCKERT ET AL

which includes information from standard nutrient tables, product labels, or recipe simulations based on the ingredients and nutrients listed. With regard to those children still breastfeeding, test weighing is performed, ie, the child is weighed before and after each feed to the nearest 10 g with an infant weighing scale (Soehnle Multina 8300). To account for insensible water losses, 5% was added to the test weighing results, as proposed by Reilly et al (28). For this study, absolute intakes of energy (kcal/d), protein (g/d), and fat (g/d) at the ages of 12, 18, and 24 mo were calculated for each participant from the mean of the 3 dietary recording days. A cutoff of 0.97 for the reported energy intake related to the predicted basal metabolic rate was used to identify implausible dietary records (18), which were then excluded (쏝2%). The absolute intakes of energy were log-transformed and internally standardized (mean ҃ 0, SD ҃ 1), and the mean of these values was used to represent the mean intake for the whole second year of life. The residual method was used to adjust macronutrient intakes for total energy intake and sex, separately for each time point and with log-transformed nutrients to improve homoscedasticity (29). The medians of the residuals were used to divide the children into those with low and those with high macronutrient intakes at ages 12 and 18 –24 mo, the latter of which was based on mean individual intakes at these 2 time points. Finally, the children with consistently high energy-adjusted protein or fat intakes (g/d) throughout the second year of life (ie, high intakes at both 12 and 18 –24 mo, or “high-high,” in each case corresponding approximately to an intake 쏜13–14% of energy from protein or 쏜35% of energy from fat, respectively) were grouped together, because we previously showed that DONALD children with such a protein intake pattern had a higher BMI SDS and a higher BF% at age 7 y than did those who did not meet this criterion (ie, those with inconsistent intakes or those with consistently low intakes, who were grouped together as “other”) (14). Statistical analysis Unadjusted associations between the independent variables and rapid weight gain were tested by using chi-square, Student’s t test, or Wilcoxon’s rank-sum test as appropriate. Because there were equal numbers of boys and girls in the 2 growth groups (chi-square ҃ 0.020, P ҃ 0.89), they were pooled together for all statistical analyses. Linear mixed-effects regression models (using PROC MIXED), including both fixed and random effects, were used to construct longitudinal models of BF% and BMI SDS trajectories subsequent to the period of rapid weight gain (between 2 and 5 y of age) and to investigate the effect of rapid weight gain on baseline BMI SDS or BF% status at age 2 and changes over time. The random component of these models accounts for the nested nature of our data (children within families) and the lack of independence between repeated observations on the same person. Initial models included either BF% or BMI SDS measurements between 2 and 5 y inclusive as the dependent continuous variable and rapid weight gain, time (chronological age and age2), and the interaction between rapid weight gain and time as the independent fixed effects. The effect of adding each of the following variables and their interaction with time (used to decide whether a given variable had a significant effect on the change in BF% or BMI SDS over time) to the initial models was then investigated as follows:

1) Birth characteristics: sex, birth at early (weeks 37 or 38) or late (weeks 41 or 42) gestation, parity (nulliparous, parous), and either BF% at 6 mo of age or BMI SDS at birth to adjust for baseline body composition. 2) Maternal characteristics: maternal overweight status (BMI 욷 25) and high educational status (욷12 y of schooling). 3) Nutritional characteristics: breastfeeding status (쏝4 mo, 욷4 mo), mean energy intake during the second year of life, protein intake (high-high, other), and fat intake (high-high, other). Those variables that modified the coefficient of rapid weight gain by 욷10% in the initial models (30); had a significant, independent effect on the outcome variables; or both were included in the series of subsequent multivariable models. These began by considering the nutrition variables, to which the variables related to birth, and then maternal, characteristics were added. Akaike’s Information Criterion was also used to assess model fit (31). A three-way interaction between time, rapid weight gain, and each of the other fixed variables was also included to consider differential effects of rapid weight gain on the BF% or BMI SDS trajectories of children in various subgroups. When the threeway interaction was significant, it remained in the model, along with all lower-order two-way interactions and main effects. An advantage of the MIXED procedure is that it does not delete children from the analysis if they are missing data for a particular time point, but analyzes all of the data available on the assumption that any missing data are missing at random. A P value 쏝 0.05 was considered statistically significant. All statistical analyses were carried out with SAS version 8.2 (SAS Institute, Inc, Cary, NC).

RESULTS

Overall, 28.5% (71/249) of the children in this sample gained weight rapidly between birth and 24 mo. These children were both significantly lighter and shorter at birth, and a larger proportion were born relatively early compared with the normal growers (Table 1). Rapid growers were also more likely to be first-born. With respect to maternal anthropometry and educational status, no significant differences between the 2 growth groups were found. By the age of 5 y, those children who had gained weight rapidly were significantly heavier and taller and had a higher BMI SDS and BF% than did those children who had gained weight normally. Furthermore, a significantly larger proportion of these children were classified as overweight (17% compared with 7%; P ҃ 0.02) or overfat (27% compared with 15%; P ҃ 0.03). The dietary characteristics of the 2 growth groups in infancy and early childhood are shown in Table 2. Neither in terms of breastfeeding status (P ҃ 0.5) nor in terms of macronutrient intake at either 12 or 18 –24 mo did the differences in median intake between the 2 growth groups reach a statistical significance of 쏝 0.05. Furthermore, there was no significant difference in the proportion of rapid or normal growers in each of the high-high macronutrient intake groups for any of the macronutrient variables: protein (35% compared with 33%; P ҃ 0.8), fat (30% compared with 31%; P ҃ 0.8), and carbohydrate (37% compared with 28%; P ҃ 0.2). The results of the linear mixed models analyses for the association between rapid weight gain and BF% trajectories (Table

RAPID WEIGHT GAIN AND NUTRITION IN AGA CHILDREN

1703

TABLE 1 Birth, maternal, and anthropometric characteristics of those DONALD Study children (n ҃ 249) who gained weight rapidly between the ages of 0 and 24 mo and those who did not1 Rate of weight gain Variable n (male/female) Birth characteristics Birth weight (g) Birth length (cm) Gestational age group [% (n)] 37–38 wk 39–40 wk 41–42 wk Primiparous [% (n)] Maternal characteristics [% (n)] Overweight mothers Smoker in the household Mothers 욷12 y schooling Anthropometric characteristics at age 5 y Weight SDS Height SDS BMI SDS Triceps skinfold SDS Subscapular skinfold SDS Body fat percentage (%) Overweight [% (n)]5 Overfat [% (n)]6

Normal

Rapid

178 (86/92)

71 (35/36)

3596 앐 2923 52 (51, 53)4

3238 앐 287 51 (50, 52)

7.3 (13/178) 53.4 (95/178) 39.3 (70/178) 56.7 (101/178)

28.2 (20/71) 54.9 (39/71) 16.9 (12/71) 71.8 (51/71)

21.4 (38/178) 21.4 (38/178) 66.3 (118/178)

25.4 (18/71) 22.5 (16/71) 56.3 (40/71)

Ҁ0.20 앐 0.95 Ҁ0.15 앐 0.96 Ҁ0.16 앐 0.99 Ҁ0.14 앐 0.99 Ҁ0.09 앐 0.96 16.7 (14.7, 19.4) 7.3 (13/178) 15.2 (27/178)

0.50 앐 0.94 0.39 앐 0.98 0.41 앐 0.90 0.34 앐 0.94 0.22 앐 1.08 18.0 (15.4, 20.9) 16.9 (12/71) 26.8 (19/71)

P value2 0.9 쏝0.0001 쏝0.0001 쏝0.0001 0.03 0.5 0.8 0.1 쏝0.0001 쏝0.0001 쏝0.0001 0.0006 0.03 0.02 0.02 0.03

1

SDS, SD score. Student’s t test or Wilcoxon rank-sum test was used for continuous variables; chi-square test was used for categorical variables. 3 x៮ 앐 SD (all such values). 4 Median; quartile 1, quartile 3 in parentheses (all such values). 5 Defined according to the International Obesity Task Force (23). 6 Defined as body fat percentage 쏜85th percentile of the sex-specific body fat reference curves by McCarthy et al (24). 2

3) showed that, even after adjustment for birth, maternal overweight, and educational characteristics (model 2), 2 nutrition variables significantly influenced the relation between rapid weight gain and BF%. First, those rapid growers who were fully breastfed for 욷4 mo had a significantly lower fat mass at age 2 y than did those rapid growers who were not fully breastfed for 욷4 mo (adjusted difference between rapid growth groups: Ҁ1.53 앐 0.59%; P ҃ 0.009; see also Figure 1: solid circles versus solid squares). However, breastfeeding status did not affect the rate of change in BF% between 2 and 5 y of age, ie, there was no three-way interaction between time, rapid weight gain, and breastfeeding status (Figure 1: all symbols). On the contrary, all groups appeared to track at the level of body fat they had reached by 2 y of age. Among normal growers, whether a child had been breastfed for 욷4 mo appeared to have little effect on BF% at age 2 y (Figure 1: open squares and circles). Second, although there was no interaction between a consistently high fat intake at 12 and 18 –24 mo and rapid weight gain at 2 y of age (0.28 앐 0.67%; P ҃ 0.7), those rapid growers who had a consistently high fat intake did not show the physiologic decrease in BF% between 2 and 5 y of age seen in those rapid growers with a consistently low fat intake or an inconsistent fat intake at 12 and 18 –24 mo (adjusted difference between rapid growth groups: 0.73 앐 0.26%/y; P ҃ 0.006). Instead, as shown in Figure 2, rapid growers in the high-high group (solid squares) maintained the level they had reached by 2 y of age, so that the difference in BF% between them and those rapid growers in the

other group (solid circles), whose BF% decreased between 2 and 5 y of age, became progressively larger. The opposite was true of the normal growers. In their case, BF% appeared to decrease faster in those who had had a consistently high fat intake (open squares) than in those with an inconsistent or consistently low fat intake (open circles). Of the other nutrition variables shown in Table 3, none interacted significantly with rapid weight gain at either age 2 or between 2 and 5 y of age. However, some had an independent effect on BF%: a higher mean energy intake during the second year of life tended to be associated with a lower BF% at age 2 y (Ҁ0.32 앐 0.19%; P ҃ 0.08) but with a gain in BF% between 2 and 5 y (0.20 앐 0.07%/y; P ҃ 0.005). A consistently high protein intake, on the other hand, was associated with a higher BF% at age 2 y (0.67 앐 0.31%; P ҃ 0.03), but did not have any effect on change in BF% over time. Separate analyses were also conducted by using the sum of the 4 skinfold-thickness measurements, triceps SDS, or subscapular SDS as the outcome variables, respectively. However, the conclusions about the effect of the interactions between rapid weight gain and the nutrition variables and breastfeeding on BF% trajectories remained unchanged, regardless of the outcome variable used. With respect to BMI SDS (Table 4), none of the nutrition variables considered interacted with rapid weight gain to influence its effect on BMI SDS at 2 y or on the rate of BMI SDS change between 2 and 5 y of age. Nevertheless, a consistently

1704

KARAOLIS-DANCKERT ET AL

TABLE 2 Dietary characteristics during infancy and early childhood of those DONALD Study children (n ҃ 249) who gained weight rapidly between 0 and 24 mo and those who did not1 Rate of weight gain Normal (n ҃ 178)

Variable Infancy Fully breastfed for 욷4 mo [% (n)] Early childhood Energy (kcal/d) 12 mo 18–24 mo Protein (g/d) 12 mo 18–24 mo Protein (% of energy) 12 mo 18–24 mo Fat (g/d) 12 mo 18–24 mo Fat (% of energy) 12 mo 18–24 mo Carbohydrate (g/d) 12 mo 18–24 mo Carbohydrate (% of energy) 12 mo 18–24 mo

65.2 (116/178)

810.8 (726.8, 882.8)2 922.9 (861.4, 1014.2)

Rapid (n ҃ 71) 60.6 (43/71)

848.9 (759.5, 909.9) 939.3 (848.2, 1006.3)

26.0 (22.4, 29.8) 31.6 (28.1, 35.8)

27.4 (23.7, 31.3) 31.5 (28.0, 35.8)

13.0 (11.2, 14.8) 13.7 (12.3, 14.8)

13.4 (11.7, 14.5) 13.6 (12.2, 15.3)

30.7 (27.2, 35.9) 38.0 (32.3, 42.7)

32.4 (27.4, 37.4) 37.5 (32.7, 41.7)

35.2 (31.9, 38.2) 36.9 (32.9, 40.2)

35.1 (30.8, 39.2) 35.5 (32.8, 38.8)

103.6 (92.6, 119.8) 114.7 (103.5, 128.3)

105.5 (93.4, 123.5) 117.3 (103.9, 133.5)

51.3 (48.6, 56.4) 49.6 (45.4, 54.0)

51.6 (47.7, 56.2) 50.7 (46.8, 53.7)

1 There were no significant time (12 or 18 –24 mo) by weight gain group (rapid, normal) interactions for any of the nutrition variables and no significant main effects of time (age) or of weight gain group. 2 Median; quartile 1, quartile 3 in parentheses (all such values).

high protein intake resulted in a higher BMI SDS at age 2 y (0.36 앐 0.13 SDS; P ҃ 0.005). There was also an indication that this difference might decrease between 2 and 5 y (Ҁ0.06 앐 0.03 SDS/y; P ҃ 0.07). A consistently high fat intake during the second year of life, on the other hand, resulted in a lower BMI SDS at age 2 y (Ҁ0.26 앐 0.13 SDS; P ҃ 0.05), but did not affect the change in BMI SDS between the ages of 2 and 5 y.

DISCUSSION

In this study, we showed that the detrimental effect of rapid weight gain on fat mass development in healthy, term AGA children can be modified by dietary factors acting in infancy and early childhood. In particular, being breastfed for 욷4 mo attenuated the effect of rapid weight gain, resulting in a lower BF% at 2 y of age among rapid growers who had been fully breastfed than in those who had not. Furthermore, a consistently high fat intake during the second year of life modified the subsequent effect of rapid weight gain on the longitudinal development of fat mass, thus inhibiting the normal physiologic decrease, thereby resulting in a larger BF% among exposed rapid growers. Several studies have described the differences in growth between breastfed and bottle-fed infants (11, 32); to our knowledge, however, this is the first study to investigate in detail the association between breastfeeding and rapid weight gain. We found that full breastfeeding for 욷4 mo (only) has a protective effect

during the period of rapid weight gain. That is to say, full breastfeeding for 욷4 mo appears to directly influence the extent to which rapid weight gain adversely affects fat mass development in the first 2 y of life, after which the level of BF% reached tracks unchanged. An over-compensatory drive to hyperphagia is one of the mechanisms proposed to play a role in “catch-up fat,” ie, the propensity to gain body fat rather than lean tissue during rapid weight gain (33). It is known that differences in appetite control, feeding frequency, and meal size exist between bottle-fed and breastfed infants (34). This could explain why rapid growers who were not fully breastfed showed a more pronounced gain in fat mass. Although BMI is a convenient proxy measure of adiposity, it cannot differentiate between lean and fat mass (35), so it is perhaps not surprising that the beneficial effect of breastfeeding was only discernible in the models of BF%. In fact, BMI was only moderately correlated with BF% in this sample (r ҃ 0.53– 0.66 in boys and r ҃ 0.60 – 0.71 in girls). Alternatively, the higher protein intakes of bottle-fed children could explain the differences in BF% at 2 y of age. Higher protein intakes may stimulate the secretion of insulin and insulin-like growth factor 1, both of which accelerate growth and increases in fat mass (36, 37). If protein were the major regulatory pathway in the interaction between lack of breastfeeding and rapid weight gain, it would appear that it only has a role to play during a time window in the first year of life. We conclude this because consistently high protein intakes during the second year of life did

1705

RAPID WEIGHT GAIN AND NUTRITION IN AGA CHILDREN

TABLE 3 Linear mixed models of the association between rapid weight gain, nutrition in infancy and early childhood, and baseline percentage body fat (BF%) at 2 y of age and BF% slope between 2 and 5 y of age (n ҃ 249) Risk factor Initial status at 2 y Intercept2 Rapid weight gain Sex (female) Fully breastfed for 욷4 mo Energy (kcal/d) HH protein intake3 HH fat intake3 Rapid weight gain ҂ fully breastfed for 욷4 mo Rapid weight gain ҂ HH protein intake Rapid weight gain ҂ HH fat intake Rate of change 2–5 y Intercept2 Time ҂ rapid weight gain Time ҂ sex Time ҂ fully breastfed for 욷4 mo Time ҂ energy (kcal/d) Time ҂ HH fat intake Time ҂ rapid weight gain ҂ HH fat intake

Model 1 estimate 앐 SE

P

Model 21 estimate 앐 SE

P

19.25 앐 0.40 2.78 앐 0.59 3.13 앐 0.32 0.54 앐 0.37 Ҁ0.39 앐 0.21 0.86 앐 0.35 Ҁ0.32 앐 0.39 Ҁ2.01 앐 0.64 Ҁ0.63 앐 0.65 0.77 앐 0.76

쏝0.0001 쏝0.0001 쏝0.0001 0.1 0.07 0.01 0.4 0.002 0.3 0.3

9.71 앐 1.24 4.16 앐 0.63 1.62 앐 0.34 0.22 앐 0.33 Ҁ0.32 앐 0.19 0.67 앐 0.31 Ҁ0.17 앐 0.35 Ҁ1.53 앐 0.59 Ҁ0.55 앐 0.57 0.28 앐 0.67

쏝0.0001 쏝0.0001 쏝0.0001 0.5 0.08 0.03 0.6 0.009 0.3 0.7

0.06 앐 0.20 Ҁ0.29 앐 0.15 0.17 앐 0.12 Ҁ0.11 앐 0.12 0.20 앐 0.08 Ҁ0.26 앐 0.14 0.76 앐 0.27

0.8 0.05 0.1 0.4 0.008 0.07 0.005

1.01 앐 0.51 Ҁ0.28 앐 0.14 0.33 앐 0.13 Ҁ0.06 앐 0.11 0.20 앐 0.07 Ҁ0.26 앐 0.14 0.73 앐 0.26

0.05 0.05 0.01 0.6 0.005 0.05 0.006

Model 2: as model 1 and additionally adjusted for time ҂ time, BF% at 6 mo, time ҂ BF% at 6 mo, gestational age group, rapid weight gain ҂ gestational age group, time ҂ maternal overweight, maternal schooling, and rapid weight gain ҂ maternal schooling. 2 The multilevel models have 2 intercepts, one for baseline BF% at 2 y of age and another for subsequent linear change in BF% over time. Each represents the mean value of the dependent variable at the baseline time, when all other predictors are 0. 3 HH, high-high, represents the group with consistently high nutrient intakes at both 12 and 18 –24 mo of age on the basis of medians of energy-adjusted intake. 1

not modify the effect of rapid weight gain on BF%, despite a consistently high protein intake being independently associated with a higher BF% and a higher BMI SDS at 2 y of age, and a tendency toward a decreased rate of BMI SDS change between the ages of 2 and 5 y. In this study, among rapid growers, a consistently high fat intake at both 12 and 18 –24 mo inhibited the physiologic decline in fat mass that normally occurs between the end of the first year

FIGURE 1. Predicted mean (앐SEM) percentage body fat trajectory in subgroups of rapid weight gain and breastfeeding status in 249 children: rapid growers who were not breastfed for 욷4 mo (f, n ҃ 28), rapid growers who were breastfed for 욷4 mo (F, n ҃ 43), normal growers who were not breastfed for 욷4 mo (䡺, n ҃ 62), and normal growers who were breastfed for 욷4 mo (E, n ҃ 116). See Table 3 for information on the model used for prediction. The plot shows the 2-way interaction between rapid weight gain and breastfeeding status at age 2 y (P ҃ 0.009 for interaction).

of life and 5– 6 y of age (38). Conversely, among normal growers, a consistently high fat intake at these points in time resulted in a greater decrease in BF% between 2 and 5 y than when fat intakes were inconsistent or consistently low. These apparently contradictory associations were observed for fat intakes consistently exceeding 35% during the second year of life. This level of fat intake lies within the recommendations of both national (39) and

FIGURE 2. Predicted mean (앐SEM) percentage body fat trajectory by subgroups of rapid weight gain and dietary fat consumption at 12 and 18 –24 mo of age in 249 children: rapid growers with a consistently high fat intake (f, n ҃ 21), rapid growers with an inconsistent or consistently low fat intake (F, n ҃ 50), normal growers with a consistently high fat intake (䡺, n ҃ 55), and normal growers with an inconsistent or consistently low fat intake (E, n ҃ 123). See Table 4 for information on the model used for prediction. The plot shows the 3-way interaction between rapid weight gain and fat intake between 2 and 5 y of age (P ҃ 0.006 for interaction).

1706

KARAOLIS-DANCKERT ET AL

TABLE 4 Linear mixed models of the association between rapid weight gain, nutrition in infancy and early childhood, and baseline BMI SD score (SDS) at 2 y of age and BMI SDS change between 2 and 5 y of age (n ҃ 249) Risk factor Initial status at 2 y Intercept2 Rapid weight gain Energy (kcal/d) HH protein intake3 HH fat intake3 Rapid weight gain ҂ HH protein intake Rapid weight gain ҂ HH fat intake Rate of change 2–5 y Intercept2 Time ҂ rapid weight gain Time ҂ HH protein intake Time ҂ HH fat intake Time ҂ rapid weight gain ҂ HH protein intake Time ҂ rapid weight gain ҂ HH fat intake

Model 1 estimate 앐 SE

P

Model 21 estimate 앐 SE

P

Ҁ0.26 앐 0.09 0.81 앐 0.17 0.10 앐 0.07 0.35 앐 0.14 Ҁ0.28 앐 0.14 Ҁ0.25 앐 0.26 0.27 앐 0.26

0.006 쏝0.0001 0.2 0.01 0.05 0.3 0.3

Ҁ0.39 앐 0.10 1.36 앐 0.20 0.07 앐 0.07 0.36 앐 0.13 Ҁ0.26 앐 0.13 Ҁ0.25 앐 0.24 0.20 앐 0.24

0.0005 쏝0.0001 0.3 0.005 0.05 0.3 0.4

0.05 앐 0.02 Ҁ0.13 앐 0.04 Ҁ0.06 앐 0.04 Ҁ0.003 앐 0.04 0.05 앐 0.06 0.09 앐 0.07

0.03 0.001 0.07 0.9 0.4 0.2

0.04 앐 0.03 Ҁ0.23 앐 0.05 Ҁ0.06 앐 0.03 Ҁ0.0004 앐 0.03 0.06 앐 0.06 0.09 앐 0.06

0.1 쏝0.0001 0.07 1.0 0.3 0.2

Model 2: as model 1 and additionally adjusted for BMI SDS at birth, time ҂ BMI SDS, rapid weight gain ҂ BMI SDS, time ҂ rapid weight gain ҂ BMI SDS, gestational age group, time ҂ gestational age group, rapid weight gain ҂ gestational age group, time ҂ rapid weight gain ҂ gestational age group, time ҂ maternal overweight. 2 The multilevel models have 2 intercepts, one for baseline BMI SDS at 2 y of age and another for subsequent linear change in BMI SDS over time. Each represents the mean value of the dependent variable at the baseline time, when all other predictors are 0. 3 HH, high-high, represents the group with consistently high nutrient intakes at both 12 and 18 –24 mo of age on the basis of medians of energy-adjusted intake. 1

international (40) health organizations for these age groups. There seems to be a general consensus that fat intake should only be reduced to 앒30% after the age of 2 y, because there does not appear to be any evidence for a detrimental effect of a high fat intake during the first years of life. On the contrary, it is believed that the rapid transition from breastfeeding (low protein, high fat) to family food (high protein, relatively low fat) results in an inadequately balanced infant diet (13). In a recent review of studies on fat intake in infancy and early childhood and subsequent weight or fatness in infants and children, 11 of 13 studies failed to find an association (41). Those that found a positive association emphasized that the relation was stronger with fat intakes after 2 y. It remains to be ascertained why a consistently high fat intake during the second year of life was detrimental in those children who gained weight rapidly. A Brazilian study showed a significant, positive association between a higher fat intake (% of energy) at baseline and weight-for-height gain over the follow-up period in those 7–11-y-old girls who were mildly stunted (42). Because high-fat diets may promote excessive weight gain and accumulation of adipose tissue (43, 44), the authors questioned whether the high rate of obesity seen in this group of recoveredmalnourished children had to do with “an increased susceptibility to high fat diets.” In his article on the control systems that regulate fat storage, Dulloo (33) proposed that “suppression of thermogenesis,” ie, when glucose spared from oxidation in muscle is directed toward lipogenesis and storage in white adipose tissue, could be the main mechanism by which rapid weight gain leads to increased fat mass in individuals whose system has been programmed by scarcity. It is possible that exposure to a high-fat diet in the second year of life in susceptible individuals, such as AGA children who gain weight rapidly, further exacerbates this mechanism.

Although we did find a tendency toward a higher energy intake among rapid growers at 12 mo, the linear mixed models did not reveal any interaction with rapid weight gain. We therefore believe that the suggestion of differences in energy intake at 12 mo should be interpreted in the context that rapid growers were already taller and heavier at these ages. In their study of formulaor mixed-fed infants, Ong et al (45) found that dietary energy intake at 4 mo predicted greater subsequent weight gain. We were unable to investigate diet during the first year of life because, although DONALD participants provide accurate information on breastfeeding status, few carry out dietary records before 6 mo of age. The DONALD Study participants are characterized by a relatively high socioeconomic and educational status. It is possible that the relative homogeneity of the DONALD sample means that extremes of diet or behavior are not represented. We nevertheless observed effect modification by some nutrition variables despite this homogeneity. Although the DONALD Study sample is also unrepresentative, the dietary data in general, and patterns in infancy, in particular in terms of rates of breastfeeding and choice of breast milk substitutes and commercial weaning foods, are similar to several nationwide studies (46, 47). The considerable strengths of this analysis lie in the DONALD Study’s prospectively collected, repeated assessments of breastfeeding status and diet in early childhood; the repeated anthropometric measurements from as early as 3 mo of age; the Institute’s nutritional database, LEBTAB, which is constantly updated (27); and detailed information on several possible covariates and confounders. We do not know why some AGA infants gain weight rapidly. As far as the nutrition variables considered in this article are concerned, they simply acted as modifiers of the effect of rapid weight gain. However, on a practical level, the findings of this

RAPID WEIGHT GAIN AND NUTRITION IN AGA CHILDREN

study support the continued promotion of breastfeeding. Although our study suggests that principally rapid growers would benefit from being breastfed with regards to fat mass development, at the moment it is not possible to say which newborns will grow rapidly and which will not. Moreover, the multitude of other benefits of breastfeeding speak for themselves (48). Our data also suggest that the current recommendations in favor of a higher fat intake during the complementary feeding period may not necessarily be beneficial for all children. This being an observational study, we cannot clearly separate cause from effect. Nevertheless, individualizing recommendations on the basis of a child’s growth pattern may be worth considering. We acknowledge that we were unable to investigate the qualitative nature of the fats involved, and more information would therefore be required on the type of fat potentially responsible for both the positive and the negative effects observed in this study. In conclusion, the influence of certain nutritional factors early in life on later body composition varies by growth pattern: among rapid growers, full breastfeeding for 욷4 mo exerts a protective effect against high BF%, whereas a consistently high fat intake during the second year of life inhibits the physiologic decrease in BF% between 2 and 5 y of age. The participation of all children and their families in the study is gratefully acknowledged. We also thank Birgit Holtermann, Ute Kahrweg, and Sabine Twenhöven for carrying out the anthropometric measurements. The contributions of the authors were as follows—NK-D and ALBG: conceived the project and performed the initial data analyses; NK-D: conducted further analyses and drafted the manuscript; AEB, AK, and CH: supervised the study. All authors contributed to interpretation of the data and revision of the manuscript. None of the authors had any personal or financial conflicts of interest.

13. 14.

15. 16. 17.

18. 19. 20.

21. 22. 23. 24.

REFERENCES 1. Monteiro PO, Victora CG. Rapid growth in infancy and childhood and obesity in later life—a systematic review. Obes Rev 2005;6:143–54. 2. Reilly JJ, Armstrong J, Dorosty AR, et al. Early life risk factors for obesity in childhood: cohort study. BMJ 2005;330:1357. 3. Eriksson JG, Forsen T, Tuomilehto J, Osmond C, Barker DJ. Early growth and coronary heart disease in later life: longitudinal study. BMJ 2001;322:949 –53. 4. dos Santos Silva I, De Stavola BL, Mann V, Kuh D, Hardy R, Wadsworth ME. Prenatal factors, childhood growth trajectories and age at menarche. Int J Epidemiol 2002;31:405–12. 5. Eriksson J, Forsen T, Tuomilehto J, Osmond C, Barker D. Size at birth, childhood growth and obesity in adult life. Int J Obes Relat Metab Disord 2001;25:735– 40. 6. Karaolis-Danckert N, Buyken AE, Bolzenius K, Perim de Faria C, Lentze MJ, Kroke A. Rapid growth among term children whose birth weight was appropriate for gestational age has a longer lasting effect on body fat percentage than on body mass index. Am J Clin Nutr 2006;84: 1449 –55. 7. Ibanez L, Ong K, Dunger DB, de Zegher F. Early development of adiposity and insulin resistance after catch-up weight gain in small-forgestational-age children. J Clin Endocrinol Metab 2006;91:2153– 8. 8. Lucas A. Programming by early nutrition: an experimental approach. J Nutr 1998;128:401S– 6S. 9. Harder T, Bergmann R, Kallischnigg G, Plagemann A. Duration of breastfeeding and risk of overweight: a meta-analysis. Am J Epidemiol 2005;162:397– 403. 10. Owen CG, Martin RM, Whincup PH, Smith GD, Cook DG. Effect of infant feeding on the risk of obesity across the life course: a quantitative review of published evidence. Pediatrics 2005;115:1367–77. 11. Dewey KG, Heinig MJ, Nommsen LA, Peerson JM, Lonnerdal B. Growth of breast-fed and formula-fed infants from 0 to 18 months: the DARLING Study. Pediatrics 1992;89:1035– 41. 12. Michaelsen KF. What is known? Short-term and long-term effects of

25. 26.

27. 28. 29. 30. 31.

32.

33.

34.

1707

complementary feeding. In: Hernell O, Schmitz J, eds. Feeding during late infancy and early childhood: impact on health. Basel, Switzerland: Nestec Ltd, Vevey/S Karger AG, 2005:185–205. Rolland-Cachera MF, Deheeger M, Maillot M, Bellisle F. Early adiposity rebound: causes and consequences for obesity in children and adults. Int J Obes (Lond) 2006;30(suppl 4):S11–7. Gunther AL, Buyken AE, Kroke A. Protein intake during the period of complementary feeding and early childhood and the association with body mass index and percentage body fat at 7 y of age. Am J Clin Nutr 2007;85:1626 –33. Kroke A, Manz F, Kersting M, et al. The DONALD Study. History, current status and future perspectives. Eur J Nutr 2004;43:45–54. WHO. Physical status: the use and interpretation of anthropometry. Report of a WHO Expert Committee. World Health Organ Tech Rep Ser 1995;854:1– 452. Voigt M, Friese K, Schneider KTM, Jorch G, Hesse V. Kurzmitteilung zu den Perzentilwerten fu¨r die Körperma␤e Neugeborener. [A short communication on the reference values for body weight and height in newborns.] Geburtsh Frauenheilk 2002;62:274 – 6 (in German). Sichert-Hellert W, Kersting M, Schoch G. Underreporting of energy intake in 1 to 18 year old German children and adolescents. Z Ernahrungswiss 1998;37:242–51. Lohmann TG, Roche AF, Martorell R, eds. Anthropometric standardization reference manual. Champaign, IL: Human Kinetics, 1988. Kromeyer-Hauschild K, Wabitsch M, Kunze D, et al. Perzentile fu¨r den Body-mass-Index fu¨r das Kindes- und Jugendalter unter Heranziehung verschiedener deutscher Stichproben. [Percentiles of body mass index in children and adolescents evaluated from different regional German studies.] Monatsschr Kinderheilkd 2001;149:807–18 (in German). Davies PS, Day JM, Cole TJ. Converting Tanner-Whitehouse reference tricep and subscapular skinfold measurements to standard deviation scores. Eur J Clin Nutr 1993;47:559 – 66. Deurenberg P, Pieters JJ, Hautvast JG. The assessment of the body fat percentage by skinfold thickness measurements in childhood and young adolescence. Br J Nutr 1990;63:293–303. Cole TJ, Bellizzi MC, Flegal KM, Dietz WH. Establishing a standard definition for child overweight and obesity worldwide: international survey. BMJ 2000;320:1240 –3. McCarthy HD, Cole TJ, Fry T, Jebb SA, Prentice AM. Body fat reference curves for children. Int J Obes (Lond) 2006;30:598 – 602. Ong KK, Ahmed ML, Emmett PM, Preece MA, Dunger DB. Association between postnatal catch-up growth and obesity in childhood: prospective cohort study. BMJ 2000;320:967–71. National Breastfeeding Committee. Empfehlungen zur Stilldauer. Stellungnahme der Nationalen Stillkommission. [Recommendations for breastfeeding duration. Statement of the National Breastfeeding Committee.] National Breastfeeding Committee, 2004 (in German). Internet: http://www.bfr.bund.de/cm/207/empfehlungen_zur_stilldauer.pdf (accessed 10 January 2007). Sichert-Hellert W, Kersting M, Chahda C, Schafer R, Kroke A. German food composition data base for dietary evaluations in children and adolescents. J Food Comp Anal 2007;20:63–70. Reilly JJ, Ashworth S, Wells JC. Metabolisable energy consumption in the exclusively breast-fed infant aged 3– 6 months from the developed world: a systematic review. Br J Nutr 2005;94:56 – 63. Willett W, Stampfer M. Implications of total energy intake for epidemiologic analyses. In: Willett W, ed. Nutritional epidemiology. 2nd ed. New York, NY: Oxford University Press, 1998:273–301. Maldonado G, Greenland S. Simulation study of confounder-selection strategies. Am J Epidemiol 1993;138:923–36. Akaike H. Information theory and an extension of the maximum likelihood principle. In: Petrov BN, Csaki F, eds. Second International Symposium on Information Theory. Budapest, Hungary: Akademiai Kiado, 1973:267–281. Haschke F, van’t Hof MA. Euro-Growth references for breast-fed boys and girls: influence of breast-feeding and solids on growth until 36 months of age. Euro-Growth Study Group. J Pediatr Gastroenterol Nutr 2000;31(suppl 1):S60 –71. Dulloo AG, Jacquet J, Seydoux J, Montani JP. The thrifty ‘catch-up fat’ phenotype: its impact on insulin sensitivity during growth trajectories to obesity and metabolic syndrome. Int J Obes (Lond) 2006;30(suppl 4): S23–35. Koletzko B. Long-term consequences of early feeding on later obesity risk. Nestle Nutr Workshop Ser Pediatr Program 2006;58:1–18.

1708

KARAOLIS-DANCKERT ET AL

35. Wells JC. A critique of the expression of paediatric body composition data. Arch Dis Child 2001;85:67–72. 36. Karlberg J, Jalil F, Lam B, Low L, Yeung CY. Linear growth retardation in relation to the three phases of growth. Eur J Clin Nutr 1994;48(suppl 1):S25– 43; discussion S43– 4. 37. Hoppe C, Udam TR, Lauritzen L, Molgaard C, Juul A, Michaelsen KF. Animal protein intake, serum insulin-like growth factor I, and growth in healthy 2.5-y-old Danish children. Am J Clin Nutr 2004;80:447–52. 38. Wabitsch M. Molecular and biological factors with emphasis on adipose tissue development. In: Burniat W, Cole T, Lissau I, Poskitt E, eds. Child and adolescent obesity. Causes and consequences, prevention and management. Cambridge, United Kingdom: Cambridge University Press, 2002. 39. German Nutrition Society, Austrian Nutrition Society, Swiss Society for Nutrition Research, Swiss Nutrition Association. Reference values for nutrient intake. Frankfurt am Main, Germany: Umschau/Braus, 2001. 40. FAO/WHO. Fats and oils in human nutrition. Report of a joint expert consultation. Food and Agriculture Organization of the United Nations and the World Health Organization. FAO Food Nutr Pap 1994;57:i-xix, 1–147. 41. Mace K, Shahkhalili Y, Aprikian O, Stan S. Dietary fat and fat types as

42.

43. 44. 45. 46. 47. 48.

early determinants of childhood obesity: a reappraisal. Int J Obes (Lond) 2006;30(suppl 4):S50 –7. Sawaya AL, Grillo LP, Verreschi I, da Silva AC, Roberts SB. Mild stunting is associated with higher susceptibility to the effects of high fat diets: studies in a shantytown population in Sao Paulo, Brazil. J Nutr 1998;128:415S–20S. West DB, York B. Dietary fat, genetic predisposition, and obesity: lessons from animal models. Am J Clin Nutr 1998;67:505S–12S. Klesges RC, Klesges LM, Haddock CK, Eck LH. A longitudinal analysis of the impact of dietary intake and physical activity on weight change in adults. Am J Clin Nutr 1992;55:818 –22. Ong KK, Emmett PM, Noble S, Ness A, Dunger DB. Dietary energy intake at the age of 4 months predicts postnatal weight gain and childhood body mass index. Pediatrics 2006;117:e503– 8. Kersting M, Dulon M. Assessment of breast-feeding promotion in hospitals and follow-up survey of mother-infant pairs in Germany: the SuSe Study. Public Health Nutr 2002;5:547–52. Kersting M, Sichert-Hellert W, Alexy U, Manz F, Schoch G. Macronutrient intake of 1 to 18 year old German children and adolescents. Z Ernahrungswiss 1998;37:252–9. Breastfeeding and the use of human milk. American Academy of Pediatrics. Work Group on Breastfeeding. Pediatrics 1997;100:1035–9.

Portion size effects on daily energy intake in low-income Hispanic and African American children and their mothers1–3 Jennifer O Fisher, Angeles Arreola, Leann L Birch, and Barbara J Rolls ABSTRACT Background: Portion size influences children’s energy intakes at meals, but effects on daily intake are unknown. Objective: Effects of large portions on daily energy intake were tested in 5-y-old Hispanic and African American children from lowincome families. Maternal food intake data were collected to evaluate familial susceptibility to portion size. Design: A within-subjects experimental design with reference and large portion sizes was used in a study of 59 low-income Hispanic and African American preschool-aged children and their mothers. The portion size of 3 entrées (lunch, dinner, and breakfast) and an afternoon snack served during a 24-h period were of a reference size in one condition and doubled in the other condition. Portion sizes of other foods and beverages did not vary across conditions. Weighed food intake, anthropometric measures, and self-reported data were obtained. Results: Doubling the portion size of several entrées and a snack served during a 24-h period increased energy intake from those foods by 23% (180 kcal) among children (P 쏝 0.0001) and by 21% (270 kcal) among mothers (P 쏝 0.0001). Child and maternal energy intakes from other foods for which portion size was not altered did not differ across conditions. Consequently, total energy intakes in the large-portion condition were 12% (P 쏝 0.001) and 6% (P 쏝 0.01) higher in children and mothers, respectively, than in the reference condition. Child and maternal intakes of the portion-manipulated foods were not correlated. Conclusions: Large portions may contribute to obesigenic dietary environments by promoting excess daily intakes among Hispanic and African American children. Am J Clin Nutr 2007;86: 1709 –16. KEY WORDS obesity

Portion size, energy intake, eating behavior,

survey data, however, precludes causal inferences about the effect of portion size on energy consumption. Experimental research has shown that an increase in the entrée serving size at a meal produces elevations in preschool-aged children’s total energy intakes at single meals (12–14). The extent to which large portions promote positive energy balance is contingent on the degree to which increases in meal energy intake are maintained over longer periods. Recent studies of adults have reported sustained portion size effects on energy intake over 2-d (15) and 11-d (16) periods, when all food and beverage portions were increased. Whether portion size has similar effects on daily energy intake among children is unclear because children demonstrate an ability to self-regulate energy intake within (17, 18) and across (19) meals. This study tested portion size effects on food and energy intakes over a 24-h period among low-income 5-y-old Hispanic and African American children—populations known to be disproportionately affected by overweight (2). Portion size effects on maternal intake were also determined to evaluate familial resemblances in the tendency to overconsume large portions. Large portions were hypothesized to promote total energy intake over a 24-h period among children and mothers in both ethnic groups. SUBJECTS AND METHODS

Design The effects of portion size on total energy intake during a 24-h period were tested by using a within-subject design. Each child and mother participated in two 24-h conditions involving a single menu that differed only in the portion sizes of entrées served at 3 separate meals and an afternoon snack. Reference portions of 1

INTRODUCTION

Marked increases in pediatric overweight since the mid-1970s (1, 2) highlight the role of the environment and its effects on behavior (3, 4). Exposure to large portions of energy-dense food may cause excessive energy intakes among children, but empirical evidence is limited (5– 8). Nationally representative surveys have documented increases in average food portion sizes consumed by children in and outside the home since the late 1970s (9). Higher energy intakes among children are associated with larger average food portions consumed per eating occasion and larger meal sizes (10, 11). The cross-sectional nature of these

From the Department of Pediatrics, Baylor College of Medicine, US Department of Agriculture/Agricultural Research Service Children’s Nutrition Research Center, Houston, TX (JOF and AA), and the Department of Human Development and Family Studies (LLB) and the Department of Nutritional Sciences (BJR), The Pennsylvania State University, University Park, PA. 2 Supported by USDA CRIS funds and the National Research Initiative of the USDA Cooperative State Research, Education and Extension Service (grant number 2002-35200-12264). 3 Address reprint requests and correspondence to JO Fisher, USDA Children’s Nutrition Research Center, 1100 Bates Street, Suite 4004, Houston, TX. E-mail: [email protected]. Received March 8, 2007. Accepted for publication July 31, 2007.

Am J Clin Nutr 2007;86:1709 –16. Printed in USA. © 2007 American Society for Nutrition

1709

1710

FISHER ET AL

these foods were served in one condition and were doubled in the other condition. The portion sizes of all other foods offered during the 24-h period were held constant across conditions. This aspect of the design was used to determine whether increasing the portion size of the entrées and snack influenced the intake of other foods for which portion size was not manipulated. The order of condition presentation was randomly assigned to each mother-child pair, and the conditions were separated by 욷2 wk. Three to 4 families were seen on a given day, and all families participated in the same condition. Weighed food intake data were collected. Other measures included in this report were maternal and child body mass index (BMI), family demographics, and food insecurity.

Portion-manipulated foods Foods for which portion size was manipulated were macaroni and cheese, apple juice, graham crackers, chicken, rice (mothers only), and cereal. These foods were familiar (85–95% of children and 95–100% of mothers reported having eaten foods previously) and acceptable to the participants (90 –98% of children and 95–100% of mothers rated foods as being “okay” or “liked”; 22, 23). The reference portions were specified by using previous research (13, 14, 24) and the 50 –75th percentiles of intake for individual foods per eating occasion from the Continuing Survey of Food Intakes of Individuals, 1994 –1996 (25) as guides; the reference portions of several of the foods were increased from the initial specification as the result of pilot testing.

Subjects Participants were 28 African American and 31 Hispanic 5-yolds (n ҃ 35 girls, 24 boys) attending Head Start Programs in the greater metropolitan area of Houston, TX. Head Start is a national program that promotes school readiness through programs serving young children from primarily low-income families (Internet: http://www.acf.hhs.gov/programs/hsb/). Children and their mothers were identified by using flyers and on-site recruiting at Head Start Centers. This investigation focused on 5-y-olds because previous laboratory studies have consistently shown portion size effects on energy consumed at meals among children of this age (12–14). Hispanic and non-Hispanic African American families were recruited to evaluate the generalizability of previous results obtained with non-Hispanic white samples (12– 14). Exclusion criteria were the presence of severe food allergies or chronic illnesses affecting food intake, dislike of 욷2 of the foods for which portion size was manipulated, and self-reported previous diagnosis of maternal depression or eating disorders. Measures Maternal and child BMI scores were calculated on the basis of measured height and weight obtained by trained nurses. Height was measured in triplicate to the nearest 0.1 cm with a stadiometer with the subjects shoeless, and weight was measured in triplicate with the subjects shoeless and in light clothing to the nearest 0.1 kg with an electronic balance. Child BMI percentiles and z scores were calculated with age- and sex-specific reference data (20). Child overweight was defined according to Centers for Disease Control and Prevention guidelines as a BMI 욷95th percentile. Maternal BMI was calculated as weight (kg)/height squared (m). Demographic information was obtained by self-report and included child and maternal race-ethnicity, maternal education, and employment. The 6-item short version of the US Department of Agriculture Food Insecurity Module was used to assess household food insecurity, defined as the limited availability of nutritionally adequate and safe foods or limited or uncertain ability to acquire acceptable foods in socially acceptable ways (21). Items were given an affirmative (1) or negative (0) score, and responses were summed. Three categories of household food insecurity were used: scores of 0 or 1 were categorized to indicate secure households, scores between 2 and 4 indicated low food security, and scores of 5 or 6 indicated very low food security. Experimental menu The experimental menus for the children and mothers are shown in Table 1.

Other foods served during the 24-h period All other foods were served in generous amounts at meals and snacks. The portion sizes of these foods were held constant across conditions to evaluate the effects of a large entrée and snack portions on the intake of other accompanying foods. Energy provided Doubling the portion size of several entrées and an afternoon snack in the large-portion condition provided 47% more total energy to children and 45% more total energy to mothers than in the reference condition. Portion-manipulated foods provided 64% of total energy offered to children and 62% of total energy offered to mothers in the large-portion condition. Total energy offered to children and mothers was compared with participants’ estimated energy requirements (EERs) based on sex, age, measured weight and height, and an assumed low activity level (given the confinement of subjects to the dormitory-like setting for the duration of each 24-h visit) (26). Total energy offered to children was 184% of the mean EER in the reference condition and 270% of the mean EER in the large-portion condition. Similarly, total energy offered to mothers was 180% of the mean EER in the reference condition and 262% of the mean EER in the largeportion condition. Procedures All procedures took place at the Children’s Nutrition Research Center, Houston, TX. Potential participants were screened for inclusion by phone interview. Each mother and child came to the laboratory for an initial visit to obtain informed consent, to familiarize children with the setting, and to obtain preference ratings for menu foods. Mothers provided consent for their own participation and their child’s participation. The mothers were told that the purpose of the study was to evaluate their children’s food preferences and intake patterns and that their own intake patterns would be measured to provide background information. Data collected at the end of the study indicate that mothers generally perceived the child to be the focus of study: less than half of the mothers (28 of 59) made reference to their own eating in describing the study purpose (ie, “to study the eating patterns of children of different ethnicity”), and almost one-third (9 of 28) of those who did believed the study to involve parent-child similarities in food preference (ie, “to observe food preference in children in comparison to the mothers”). The staff did not inform the participating children that their food intakes were being measured.

1711

PORTION SIZE EFFECTS ON CHILDREN'S DAILY ENERGY INTAKE TABLE 1 Experimental menu1 Child Energy density

Reference portions

kcal/g Day 1 Morning snack Grapes Goldfish crackers Lunch Macaroni and cheese2 Carrots Oatmeal cookies Applesauce Milk, 2% fat Afternoon snack Apple juice2 Graham crackers2 Dinner Chicken nuggets2 Chicken strips2 Rice2 Corn Dinner roll Butter Dinner salad Dressing Pears Milk, 2% fat Evening snack Sandwich cookies Vanilla ice cream Chocolate milk Day 2 Breakfast Oat ring cereal2 Bacon Banana Orange juice Milk, 2% fat Total energy provided (kcal) Portion-manipulated foods (kcal) Dietary reference intake (%) 1 2

Mother Large portions

Reference portions

kcal

Large portions kcal

0.71 5.00

46 125

46 125

85 250

85 250

1.51 0.43 4.01 0.43 0.50

453 17 200 47 120

906 17 200 47 120

604 17 300 47 120

1208 17 300 47 120

0.47 4.62

113 185

226 370

158 277

316 554

2.42 1.73 0.80 0.81 2.71 7.17 0.09 4.16 0.71 0.50

368 — — 53 108 — — — 80 120

736 — — 53 108 — — — 80 120

— 346 160 89 217 72 5 67 80 —

— 692 320 89 217 72 5 67 80 —

5.00 1.97 0.83

51 160 —

51 160 —

153 160 199

153 160 199

4.00 5.76 0.92 0.45 0.50

160 35 86 54 120 2727 1279 184

320 35 86 54 120 4006 2558 270

320 69 86 108 120 4109 1865 180

640 69 86 108 120 5974 3730 262

Estimated energy requirements are based on sex, age, activity level (assuming a low active physical activity level), and measured weight and height (26). Portion-manipulated food.

For each of two 24-h periods of observation, the motherchild pairs arrived at the Center at 0930 and left the following morning at the same time. The midmorning admission and discharge times were chosen to allow flexibility in family arrival time. The children and mothers ate meals and snacks separately from one another. Three to 4 children who did not know one another were seated together with a research staff member who facilitated non-food related conversation, ensured that foods were not shared, and accounted for dropped or spilled food. This aspect of the design was consistent with previous studies (13, 14, 24) and avoided the discomfort (and low intake) often observed in the laboratory when children eat alone. Alternatively, mothers ate meals and snacks individually to avoid social influences introduced by eating in the presence of unfamiliar women (27, 28). Participants were informed that they could eat as much or as little as desired

during each meal and snack. Twenty minutes were allotted for each eating occasion. The amount and nature of the structured and unstructured noneating activities occurring during the two 24-h periods were similar. Mothers completed questionnaires (administered in Spanish for 17 of the participants) on a wide range of topics, including demographics, food insecurity, child feeding practices, children’s food preferences, and their own eating behavior. Children participated in structured interviews, group craft activities, board games, and a once daily movie showing. Family free-time periods were offered during the early afternoon (1330 – 1430) and evening (after 1815). Families were compensated for completing the full protocol: two 24-h conditions as well as an initial 3-h visit were used primarily to familiarize families to the setting. All procedures were approved by the Baylor College of Medicine

1712

FISHER ET AL

Institutional Review Board and executed according to its standards. Statistical analyses Data analyses were performed with SAS (version 9.1; SAS Institute, Cary, NC). Descriptive statistics including the mean, median, SD, and range were generated. Data are presented as means 앐 SDs with statistical significance set at P 쏝 0.05, unless otherwise indicated. Child and maternal data were analyzed separately. The primary outcomes of interest were energy consumed from the portion-manipulated foods and cumulative energy intake over the 24-h period. Dependent t tests were used to evaluate, across conditions, differences in energy intake from portionmanipulated foods and other foods at meals and snacks. Bonferroni correction was used to adjust the P value based on the number of tests performed. Sequential dependent t tests with Bonferroni correction were used to identify the first eating occasion at which, across conditions, differences in cumulative energy intake from portion-manipulated foods and other foods were apparent. Potential correlates of changes in food and total energy intake were tested cojointly by analysis of variance: sex, ethnicity, condition order, BMI (z scores used for children), and food insecurity. The number of foods for which 욷95% of the reference portion was consumed was included in the model predicting changes in foods intake. The number of portionmanipulated foods for which intake increased across conditions was also evaluated as a predictor of changes in total energy intake. Correlations were used to evaluate the relations between maternal and child changes in food and energy intake across conditions.

RESULTS

Data from one mother-child pair were excluded from the analyses because the child complained of a toothache and was observed to have a loose tooth for the duration of one of the visits. Data from 58 children and 58 mothers were analyzed. On average, mothers were in their 30’s (age: 30 앐 5 y); half reported being currently employed (55%) and having a high school education or less (53%). A greater number of African American than Hispanic mothers reported being employed (21 of 28 African American mothers compared with 10 of 30 Hispanic mothers; P 쏝 0.01) and having an education beyond high school (20 of 28 African American mothers compared with 7 of 30 Hispanic mothers; P 쏝 0.001). More than one-third of the mothers (8 African Americans, 14 Hispanics) reported low household food security; most of these mothers (4 African Americans, 11 Hispanics) scored in the less extreme category of food insecurity. BMI scores indicated that the mothers were, on average, obese (BMI ҃ 34 앐 9), and their children were of normal weight (BMI percentile ҃ 60 앐 29%). Children’s intake of portion-manipulated foods Among children, doubling the portion size of the entrées and snack increased the energy intake from those foods by an average of 23% (180 kcal) relative to the intake in the reference condition (777 앐 224 compared with 957 앐 306 kcal; P 쏝 0.0001). Significant increases in intake were seen for 2 of the 5 individual foods for which the portion size was doubled (Table 2). Nonparametric analysis showed that most of the 58 children (n ҃ 42)

had some increase (쏜0 g) in their intake of 욷3 of the 5 foods for which portion size was doubled. Portion size effects did not reflect a restriction of food intake in the reference condition. Children consumed, on average, less than two-thirds (63 앐 19%) of the reference portions of the 5 manipulated foods, ranging from 50 앐 28% of the macaroni and cheese to 73 앐 26% of the chicken nuggets. Of 58 children, the number eating 욷95% of the reference portions was as follows: macaroni and cheese (n ҃ 7), apple juice (n ҃ 29), graham crackers (n ҃ 13), chicken nuggets (n ҃ 17), and cereal (n ҃ 23). Across-condition increases in children’s energy intake from the portion-manipulated foods were not associated with the number of foods for which the child consumed 욷95% of the reference portion (P ⫽ 0.74). Across-condition changes in energy intake from the portion-manipulated foods were also unassociated with condition order (P ҃ 0.90), sex (P ҃ 0.17), child ethnicity (P ҃ 0.66), child BMI z score (P ҃ 0.77), or household food insecurity (P ҃ 0.77). Maternal intake of portion-manipulated foods Among mothers, doubling the portion size of the entrées and snack increased energy intake from those foods, on average, by 21% (앒270 kcal) relative to intake in the reference condition (1284 앐 247 compared with 앐 380 kcal; P 쏝 0.0001). Significant increases in consumption were observed for 3 of the 6 individual foods for which portion size was doubled (Table 2). Nonparametric analysis showed that 47 of 58 mothers had some increase (쏜0 g) in the consumption of 욷3 of the 6 foods for which portion size was doubled. Mothers consumed, on average, more than two-thirds (71 앐 13%) of the reference portions, ranging from 60 앐 24% of the macaroni and cheese to 81 앐 22% of the chicken. The number of mothers eating 욷95% of the reference portions was as follows: macaroni and cheese (n ҃ 5), apple juice (n ҃ 26), graham crackers (n ҃ 28), chicken strips (n ҃ 22), rice (n ҃ 17), and cereal (n ҃ 14). Across-condition changes in the amount of energy consumed from the portion-manipulated foods were not associated with the number of foods for which the mother consumed 욷95% of the reference portion (P ⫽ 0.44). Changes in maternal energy intake of portion-manipulated foods across conditions were also unrelated to condition order (P ⫽ 0.59), maternal ethnicity (P ⫽ 0.43), maternal BMI (P ⫽ 0.75), maternal education (P ҃ 0.66; greater than a high school education compared with less), and household food insecurity (P ⫽ 0.33). Association of maternal with child intake changes Maternal and child responses to the portion size manipulations were unrelated. This was the case whether the response was expressed as mean change in energy intake from the portionmanipulated foods (r ҃ Ҁ0.20, P ҃ 0.13) or as the number of foods for which consumption increased when the portion size was doubled (Spearman’s r ҃ Ҁ0.05, P ҃ 0.70). Changes in maternal and child intakes of individual foods were also not correlated (data not shown). Children’s cumulative energy intake over the 24-h period Cumulative energy intakes from portion-manipulated and other foods in the reference and large-portion conditions are shown in Figure 1. Across-condition differences in children’s

1713

PORTION SIZE EFFECTS ON CHILDREN'S DAILY ENERGY INTAKE TABLE 2 Intakes of portion-manipulated foods and other foods1 Child Reference portions

Mother Large portions

Reference portions

kcal Morning snack Portion-manipulated Other Lunch Portion-manipulated Macaroni and cheese Other Afternoon snack Portion-manipulated Apple juice Graham crackers Other Dinner Portion-manipulated Chicken Rice Other Evening snack Portion-manipulated Other Breakfast Portion-manipulated Cereal Other

Large portions kcal

— 101 앐 52

— 103 앐 54

— 242 앐 77

— 221 앐 92

226 앐 125 170 앐 95

239 앐 118 167 앐 86

363 앐 146 289 앐 128

424 앐 1852 251 앐 141

81 앐 37 94 앐 66 —

81 앐 59 115 앐 92 —

120 앐 45 211 앐 79 —

168 앐 873 247 앐 137 —

267 앐 96 — 136 앐 71

357 앐 1433 — 125 앐 62

279 앐 75 109 앐 44 401 앐 114

367 앐 1282 128 앐 63 365 앐 108

— 175 앐 44

— 157 앐 59

— 346 앐 109

— 325 앐 123

108 앐 59 140 앐 50

163 앐 1013 130 앐 47

203 앐 95 256 앐 75

218 앐 103 250 앐 83

All values are x៮ 앐 SD. Paired t tests with Bonferroni corrections were used to compare child (10 comparisons) and maternal (11 comparisons) energy intake from portion-manipulated foods and other foods across reference and large-portion conditions. 2,3 Significantly different from reference portions (Bonferroni adjusted): 2 P 쏝 0.05, 3 P 쏝 0.001. 1

cumulative energy intake from the portion-manipulated foods were not evident until the dinner meal (Figure 1A). By dinner, children had consumed 125 앐 191 kcal more from the largeportion entrées and snack than from the reference portions (P 쏝 0.001). At the end of the 24-h period, children had consumed 180 kcal more from the large entrée and snack portions than from the reference portions of those foods (P 쏝 0.0001). By the end of the 24-h period, children had consumed 41 kcal less from other foods in the large-portion condition than in the reference condition, but this difference was not statistically significant (723 앐 195 kcal in the reference condition compared with 682 앐 169 kcal in the large-portion condition). As a result, serving large portions at multiple meals produced a net increase of 앒140 kcal over the period of observation, which represented an increase in the children’s total energy intake of 12 앐 22% (P 쏝 0.001). The effect of portion size on total energy intake varied widely among children, ranging from a 31% decrease to a 96% increase across reference and large-portion conditions. Change in total energy intake was positively associated with the number of foods for which children showed increased intake when food portion size was doubled (P 쏝 0.001). Alternatively, changes in total energy intake were unassociated with condition order (P ⫽ 0.06), sex (P ҃ 0.73), ethnicity (P ҃ 0.07), child BMI z score (P ҃ 0.43), and household food insecurity (P ҃ 0.49). Total energy intake in the reference condition was 1500 앐 359 kcal, 앒100% of estimated daily requirements calculated for each child based on sex, age, activity (assuming a low active physical

activity level), and measured weight and height (26). Total energy intake in the large-portion condition was 1639 앐 378 kcal, 앒109% of estimated daily requirements. Maternal cumulative energy intake over the 24-h period As depicted in Figure 1B, the effects of large portions on maternal energy intake were evident at the first meal at which large portions were served. Mothers consumed 61 kcal more from the lunch entrée in the large-portion condition than in the reference condition (Table 2). Intake of the portion-manipulated foods was also greater in the large-portion condition than in the reference condition at the afternoon snack and at the dinner meal. By the end of the 24-h period, energy intakes from the large entrées and snack were 267 앐 337 kcal greater than those from the reference condition (P 쏝 0.0001). Decreases in maternal energy intake from nonmanipulated foods were also apparent at the lunch meal (Figure 1B). By the end of the 24-h period, the cumulative reduction in energy intake from the nonmanipulated foods reached 122 앐 243 kcal (1535 앐 298 compared with 1413 앐 320 kcal in the reference and largeportion conditions, respectively; P 쏝 0.001). Maternal intake of nonmanipulated high-energy-density foods (쏜4.0 kcal/g) was lower in the large-portion condition than in the reference condition (473 앐 144 compared with 420 앐 150 kcal; P 쏝 0.001). Maternal energy intake of nonmanipulated low-energy-density foods (울1.0 kcal/g) and medium-energy-density foods (1.1– 4.0 kcal/g), however, did not differ across conditions (data not

1714

FISHER ET AL DISCUSSION

FIGURE 1. Mean (앐SD) cumulative energy intakes from portionmanipulated foods and other foods in the reference and large-portion conditions. Sequential paired t tests with Bonferroni corrections identified eating occasions at which across-condition differences in child (A) and maternal (B) cumulative energy intakes were first apparent. *Across-condition differences in children’s cumulative energy intake from portion-manipulated foods were first apparent by dinner (adjusted P 쏝 0.001) and increased to 180 kcal by the end of the 24-h period (P 쏝 0.0001). Across-condition differences in children’s cumulative energy intake from other foods were not significant at any point during the 24-h period.†Across-condition differences in maternal cumulative energy intake from portion-manipulated foods were first apparent at lunch (adjusted P 쏝 0.05) and increased to 267 kcal by the end of the 24-h period (P 쏝 0.0001). ‡Across-condition differences in maternal cumulative energy intake from other foods were first apparent at lunch (adjusted P 쏝 0.05) and reached Ҁ122 kcal by the end of the 24-h period (P 쏝 0.001).

shown). As a result of decreases in maternal intake of nonmanipulated foods, the net increase in total energy intake across reference and large-portion conditions was 146 kcal, which represented a 6 앐 18% increase (P 쏝 0.01). Change in total energy intake across conditions was positively associated with the number of foods for which mothers had an increased intake when the portion size was doubled (P 쏝 0.0001). Alternatively, changes in total energy intake were unrelated to condition order (P ҃ 0.27), maternal ethnicity (P ҃ 0.54), maternal BMI (P ҃ 0.27), education (P ҃ 0.84; greater than a high school education versus less), and household food insecurity (P ҃ 0.79). Total energy intake in the reference portion condition was 2819 앐 502 kcal, 23% higher than estimated daily requirements based on sex, age, activity (assuming a low physical activity level), weight, and height (26). Intake in the large-portion condition was 2965 앐 616 kcal, 29% higher than estimated needs.

Marketplace trends in food portion size have led to concerns that large portions are contributing to the problem of pediatric overweight (3, 6, 9, 29). Previous studies of children ranging from 2 to 9 y of age have shown 13–16% increases in total energy intake at a single meal when the entrée portion size was doubled (13, 14, 24). The findings of this study provide the first experimental data showing effects of portion size on young children’s energy intake, beyond individual meals, on daily energy intake. Doubling the portion size of several entrée and an afternoon snack served during a 24-h period increased children’s total energy intake by 12%. This effect was observed even though the amount of energy available to children in the reference condition was considerably higher than their estimated energy needs (184%). As in previous studies (13, 14, 24), children’s intake of other foods served at the meals and snacks did not differ across conditions. Taken together, these findings suggest that large food portions contribute to obesigenic dietary environments by increasing children’s daily energy intake. Doubling the entrée and snack portions increased children’s energy intake from those foods by 23% even though children, on average, did not consume the smaller portions in full. Although the portion size manipulations collectively increased children’s energy intake, effects were variable both between individuals and between different types of foods. The 5 foods for which portion size was manipulated were of varied energy density and shape, with some amorphous (eg, macaroni and cheese) and others more clearly defined in units (eg, chicken nuggets). Children showed significant increases in intake of 2 of the mediumenergy-density foods, one a unit food and the other amorphous. In previous research, doubling the portion size of a macaroni and cheese entrée served at a meal increased young children’s food intake by 25– 60% (12–14, 24). Why portion size affected children’s intake of some foods but not others, particularly macaroni and cheese, is not obvious. The macaroni and cheese reference portion was 50 –125 g greater than that used in previous studies (12–14, 24). It is possible that the large size of the reference portions negated effects of further increases to portion size. Although most children were familiar with and liked the portionmanipulated foods, it is also possible that unmeasured aspects of palatability and/or children’s experience with the foods may have been a factor. Portion size effects on adult intake have been shown at single meals in laboratory (30 –37) and naturalistic settings (38) for unit (30, 36, 39) and amorphous (31) foods, beverages (37), foods of varying energy density (34, 35), prepackaged snacks (33), and first-course salads (32). In the present study, the mothers consumed 21% more energy from the larger food portions than from the reference portions. The fact that mothers ate, on average, less than three-quarters of the reference portions suggests that the size of the smaller portions was not artificially limiting. The 3 foods for which maternal intake increased when large portions were served were of low- to medium energy density and included a beverage, an amorphous food, and a unit food. Mothers showed evidence of compensatory decreases in the intake (앒125 kcal) of primarily high-energy-density (쏜4.0 kcal/g) foods served throughout the 24-h period. These adjustments, however, were insufficient to offset energy intake from the large food portions. The net increase of 6% (146 kcal) in total energy intake observed in this study is somewhat smaller than what has been observed in

PORTION SIZE EFFECTS ON CHILDREN'S DAILY ENERGY INTAKE

previous studies. One study observed a 16% (335 kcal) increase in daily energy when all food and beverage portions were increased by 50% during a 2-d period (15). It is important to note that the current study manipulated the portion sizes of only 6 foods, approximately one-fourth of those offered to mothers. That a majority of mothers assigned the highest possible preference rating to each food (with the exception of applesauce; 45% of mothers gave it the highest rating) and rice (62%) suggests that the size of the effects were not likely attributable to a low acceptance of the menu. Finally, the observed 144 kcal increase in maternal daily energy is greater than the 50 –100 daily calories that are thought to separate weight maintenance from weight gain for most adults (4). Previous experimental investigations of portion size among children have involved predominately non-Hispanic white children. The present findings extend that work by demonstrating effects among Hispanic and African American children—2 ethnic groups disproportionately affected by overweight (2). In this study, the effects of portion size on total energy intake did not vary by ethnicity in children or in their mothers. The failure to observe ethnic differences in portion size effects does not imply that cultural influences are irrelevant for understanding the role of portion size in pediatric obesity. Although the effects of portion size on intake may be similar across ethnicities, factors that dictate the extent to which children (and mothers) have routine exposure to large portions may have a cultural component. For instance, children’s consumption of fast food, a noted source of exposure to excessive marketplace portions, varies by ethnicity as well as by income (40, 41). Research is needed to understand cultural and socioeconomic influences on the frequency with which children encounter large food portions at and away from home. Finally, large individual differences were observed in children’s and mothers’ responses to portion size. Consistent with previous experiments in children (13, 14, 24) and adults (31), effects of portion size on intake did not vary by weight status. Furthermore, child and maternal intake of large portions were unrelated. Previous twin and sibpair studies have reported similarities in energy intake among family members (42, 43). In one study of 32 sibpairs aged 3–7 y, total energy intake, but not caloric compensation, showed familial aggregation (42). Why maternal and child scores were unrelated in the present study is not clear. Children ate in groups separately from their mothers to ensure that effects of portion size on daily intake were not biased by adult directives to eat. Given that families were recruited from Head Start programs, we believe that it is not unusual for these children to eat in small groups apart from their mothers. It is possible, however, that this aspect of the design removed child feeding interactions and/or food modeling behavior (44) that might otherwise produce similarities in the amount of food consumed between mother and child. That children ate in social groups while their mothers ate alone prohibited comparison of the data in absolute terms but did not preclude assessment of the relative association between maternal and child scores. As such, these findings do not provide evidence that portion size effects on children’s eating are driven by weight status or a familial-based susceptibility to overconsume large portions. In conclusion, the findings of this research suggest that large portions contribute to obesigenic dietary environments by promoting daily energy intake among low-income Hispanic and African American mothers and their preschool-aged children.

1715

Increases in cumulative energy intake across conditions emerged almost immediately for mothers, were observed by dinner time for children, and persisted over the course of the 24-h period for both. The external validity of this work is supported by the observation that large food portions are pervasive in the market place (41). The generalizability of the findings to other socioeconomic and ethnic groups, however, merits further consideration. Prospective observational studies evaluating the association of children’s exposure to large portions with energy intake and weight gain are needed to inform scientific knowledge of the contribution of large portions to pediatric obesity. We thank Mary Air for her assistance with the statistical analyses. The authors’ responsibilities were as follows—JOF: designed the experiment (primary responsibility), analyzed the data, and wrote the manuscript; AA: collected the data and assisted with the manuscript presentation; and LLB and BJR: provided consultation on the design of the study and assisted with the interpretation of the results and manuscript preparation. None of the authors had a financial or personal interest in organizations sponsoring this research.

REFERENCES 1. Ogden CL, Troiano RP, Briefel RR, Kuczmarski RJ, Flegal KM, Johnson CL. Prevalence of overweight among preschool children in the United States, 1971 through 1994. Pediatrics 1997;99:E1. 2. Ogden CL, Flegal KM, Carroll MD, Johnson CL. Prevalence and trends in overweight among US children and adolescents, 1999 –2000. JAMA 2002;288:1728 –32. 3. Hill JO, Peters JC. Environmental contributions to the obesity epidemic. Science 1998;280:1371– 4. 4. Hill JO, Wyatt HR, Reed GW, Peters JC. Obesity and the environment: where do we go from here? Science 2003;299:853–5. 5. Poston WS, Foreyt JP. Obesity is an environmental issue. Atherosclerosis 1999;146:201–9. 6. French S, Story M, Jeffery R. Environmental influences on eating and physical activity. Ann Rev Public Health 2001;22:309 –35. 7. Ebbeling CB, Pawlak DB, Ludwig DS. Childhood obesity: public-health crisis, common sense cure. Lancet 2002;360:473– 82. 8. Ledikwe JH, Ello-Martin JA, Rolls BJ. Portion sizes and the obesity epidemic. J Nutr 2005;135:905–9. 9. Nielsen SJ, Popkin BM. Patterns and trends in food portion sizes, 1977– 1998. JAMA 2003;289:450 –3. 10. McConahy KL, Smiciklas-Wright H, Mitchell DC, Picciano MF. Portion size of common foods predicts energy intake among preschool-aged children. J Am Diet Assoc 2004;104:975– 6. 11. Huang TT, Howarth NC, Lin BH, Roberts SB, McCrory MA. Energy intake and meal portions: associations with BMI percentile in U.S. children. Obes Res 2004;12:1875– 85. 12. Rolls BJ, Engell D, Birch LL. Serving portion size influences 5-year-old but not 3-year-old children’s food intakes. J Am Diet Assoc 2000;100: 232– 4. 13. Fisher JO, Rolls BJ, Birch LL. Children’s bite size and intake of an entree are greater with large portions than with age-appropriate or self-selected portions. Am J Clin Nutr 2003;77:1164 –70. 14. Fisher JO. Effects of age on children’s intake of large and self-selected portions. Obesity 2007;15:403–12. 15. Rolls BJ, Roe L, Meengs JS. Larger portion sizes lead to a sustained increase in energy intake over 2 days. J Am Diet Assoc 2006;106:543–9. 16. Rolls BJ, Roe L, Meengs JS. The effect of large portion sizes on energy intake is sustained for 11 days. Obesity (Silver Spring) 2007;15:1535– 43. 17. Birch LL, Deysher M. Caloric compensation and sensory specific satiety: evidence for self regulation of food intake by young children. Appetite 1986;7:323–31. 18. Birch LL, Deysher M. Conditioned and unconditioned caloric compensation: evidence for self-regulation of food intake by young children. Learning and Motivation 1985;16:341–55. 19. Birch LL, Johnson SL, Jones MB, Peters JC. Effects of a nonenergy fat substitute on children’s energy and macronutrient intake. Am J Clin Nutr 1993;58:326 –33.

1716

FISHER ET AL

20. Kuczmarski RJ, Ogden C, Grummer-Strawn LM, et al. CDC growth charts: United States. Bethesda, MD: National Center for Health Statistics, 2000. 21. Bickel G, Nord M, Price C, Hamilton W, Cook J. Guide to measuring household food security, revised 2000. Alexandria, VA: Food and Nutrition Service, 2000. 22. Birch LL. Dimensions of preschool children’s food preferences. J Nutr Educ 1979;11:91–5. 23. Birch LL. Preschool children’s preferences and consumption patterns. J Nutr Educ 1979;11:189 –92. 24. Fisher JO, Liu Y, Birch LL, Rolls BJ. Effects of portion size and energy density on young children’s intake at a meal. Am J Clin Nutr 2007;86: 174 –9. 25. Smiciklas-Wright H, Mitchell D, Mickle S, Cook A, Goldman J. Foods commonly eaten in the United States: quantities consumed per eating occasion and in a day, 1994 –96. Houston, TX: US Department of Agriculture, Agriculture Research Service, 2002:252. 26. Food and Nutrition Board. Dietary reference intakes for energy, carbohydrate, fiber, fat, fatty acids, cholesterol, protein, and amino acids (macronutrients). Washington, DC: Institute of Medicine, 2002. 27. Rosenthal B, McSweeney FK. Modeling influences on eating behavior. Addictive Behaviors 1979;4:205–14. 28. Pliner P, Chaiken S. Eating, social motives, and self presentation in men and women. J Exp Psychol 1990;26:240 –54. 29. Young LR, Nestle MS. Portion sizes in dietary assessment: issues and policy implications. Nutr Rev 1995;53:149 –58. 30. Nisbett RE. Determinants of food intake in human obesity. Science 1968;159:1254 –5. 31. Rolls BJ, Morris EL, Roe LS. Portion size of food affects energy intake in normal-weight and overweight men and women. Am J Clin Nutr 2002;76:1207–13.

32. Rolls BJ, Roe L, Meengs J. Salad and satiety: do portion size and energy density of a first course affect lunch intake? Obes Res 2003;11:A22. 33. Rolls BJ, Roe LS, Kral TV, Meengs JS, Wall DE. Increasing the portion size of a packaged snack increases energy intake in men and women. Appetite 2004;42:63–9. 34. Kral TV, Roe LS, Rolls BJ. Combined effects of energy density and portion size on energy intake in women. Am J Clin Nutr 2004;79:962– 8. 35. Kral TV, Rolls BJ. Energy density and portion size: their independent and combined effects on energy intake. Physiol Behav 2004;82:131– 8. 36. Rolls BJ, Roe LS, Meengs JS, Wall DE. Increasing the portion size of a sandwich increases energy intake. J Am Diet Assoc 2004;104:367–72. 37. Flood JE, Roe LS, Rolls BJ. The effect of increased beverage portion size on energy intake at a meal. J Am Diet Assoc 2006;106:1984 –90. 38. Diliberti N, Bordi PL, Conklin MT, Roe LS, Rolls BJ. Increased portion size leads to increased energy intake in a restaurant meal. Obes Res 2004;12:562– 8. 39. Geier AB, Rozin P, Doros G. Unit bias. A new heuristic that helps explain the effect of portion size on food intake. Psychol Sci 2006;17:521–5. 40. Bowman SA, Gortmaker SL, Ebbeling CB, Pereira MA, Ludwig DS. Effects of fast-food consumption on energy intake and diet quality among children in a national household survey. Pediatrics 2004;113: 112– 8. 41. Young LR, Nestle M. Expanding portion sizes in the US marketplace: implications for nutrition counseling. J Am Diet Assoc 2003;103:231– 4. 42. Faith MS, Keller KL, Johnson SL, et al. Familial aggregation of energy intake in children. Am J Clin Nutr 2004;79 844 –50. 43. Faith MS, Rha SS, Neale MC, Allison DB. Evidence for genetic influences on human energy intake: results from a twin study using measured observations. Behav Genet 1999;29:145–54. 44. Hendy HM, Raudenbush B. Effectiveness of teacher modeling to encourage food acceptance in preschool children. Appetite 2000;34:61– 76.

Effects of prolonged and exclusive breastfeeding on child height, weight, adiposity, and blood pressure at age 6.5 y: evidence from a large randomized trial1–3 Michael S Kramer, Lidia Matush, Irina Vanilovich, Robert W Platt, Natalia Bogdanovich, Zinaida Sevkovskaya, Irina Dzikovich, Gyorgy Shishko, Jean-Paul Collet, Richard M Martin, George Davey Smith, Matthew W Gillman, Beverley Chalmers, Ellen Hodnett, and Stanley Shapiro for the Promotion of Breastfeeding Intervention Trial (PROBIT) Study Group ABSTRACT Background: The evidence that breastfeeding protects against obesity and a variety of chronic diseases comes almost entirely from observational studies, which have a potential for bias due to confounding, selection bias, and selective publication. Objective: We assessed whether an intervention designed to promote exclusive and prolonged breastfeeding affects children’s height, weight, adiposity, and blood pressure at age 6.5 y. Design: The Promotion of Breastfeeding Intervention Trial (PROBIT) is a cluster-randomized trial of a breastfeeding promotion intervention based on the WHO/UNICEF Baby-Friendly Hospital Initiative. A total of 17 046 healthy breastfed infants were enrolled from 31 Belarussian maternity hospitals and their affiliated clinics; of those infants, 13 889 (81.5%) were followed up at 6.5 y with duplicate measurements of anthropometric variables and blood pressure. Analysis was based on intention to treat, with statistical adjustment for clustering within hospitals or clinics to permit inferences at the individual level. Results: The experimental intervention led to a much greater prevalence of exclusive breastfeeding at 3 mo in the experimental than in the control group (43.3% and 6.4%, respectively; P 쏝 0.001) and a higher prevalence of any breastfeeding throughout infancy. No significant intervention effects were observed on height, body mass index, waist or hip circumference, triceps or subscapular skinfold thickness, or systolic or diastolic blood pressure. Conclusions: The breastfeeding promotion intervention resulted in substantial increases in the duration and exclusivity of breastfeeding, yet it did not reduce the measures of adiposity, increase stature, or reduce blood pressure at age 6.5 y in the experimental group. Previously reported beneficial effects on these outcomes may be the result of uncontrolled confounding and selection bias. Am J Clin Nutr 2007;86:1717–21. KEY WORDS Breastfeeding, adiposity, obesity, blood pressure, programming

INTRODUCTION

Between 앒3 and 12 mo of age, infants who receive prolonged and exclusive breastfeeding become thinner and shorter than those who are predominantly bottle-fed, although the former

group partially catches up in the second year of life (1). This evidence led the World Health Organization (WHO) to develop new growth standards for infants who follow WHO recommendations for exclusive and prolonged breastfeeding (2). Many studies over the past several decades have suggested that breastfeeding offers long-term protection against obesity and that the degree of protection increases with greater duration and exclusivity of breastfeeding (3). More recent systematic reviews and meta-analyses, however, suggest that inadequate control for confounding differences between breastfed and formula-fed infants may explain the reported differences; these reviews also provide evidence of publication bias—ie, of selective publication of studies that reported a protective effect of breastfeeding (4, 5). Fewer studies have examined long-term effects of breastfeeding on child and adult stature (height), although studies of the Boyd-Orr cohort paradoxically (in view of the negative effects cited above on length-for-age in the first 2 y of life) report an association between breastfeeding and greater stature (6). Recent studies of the Avon Longitudinal Study of Parents and Children cohort in the United Kingdom suggested a long-term, doseresponse effect on serum insulin-like growth factor-I concentrations according to the degree of breastfeeding, which may underlie a positive effect on height (7). 1 From the Departments of Pediatrics (MSK and RWP) and Epidemiology and Biostatistics (MSK, RWP, J-PC, and SS), McGill University Faculty of Medicine, Montreal, PQ, Canada; the National Research and Applied Medicine Mother and Child Centre, Minsk, Belarus (LM, IV, NB, ZS, ID, and GS); the Department of Social Medicine, University of Bristol, Bristol, United Kingdom (RMM and GDS); the Department of Ambulatory Care and Prevention, Harvard Medical School/Harvard Pilgrim Health Care, Boston, MA (MWG); the Department of Epidemiology and Community Health, Queen’s University, Kingston, ON, Canada (BC); and the Faculty of Nursing, University of Toronto, Toronto, ON, Canada (EH). 2 Supported by a grant from the Canadian Institutes of Health Research and by grant no. FOOD-DT-2005-007036 from the European Union’s project on Early Nutrition Programming: Long-term Efficacy and Safety Trials (to RMM and GDS). 3 Reprints not available. Address correspondence to MS Kramer, The Montreal Children’s Hospital, 2300 Tupper Street (Les Tourelles), Montreal, PQ H3H 1P3, Canada. E-mail: [email protected]. Received May 10, 2007. Accepted for publication August 8, 2007.

Am J Clin Nutr 2007;86:1717–21. Printed in USA. © 2007 American Society for Nutrition

1717

1718

KRAMER ET AL

Systematic reviews and meta-analyses suggested that breastfed infants have a mean systolic blood pressure that is slightly lower (pooled estimate: 1.4 mm Hg lower) than that of formulafed infants (8, 9). Studies also suggested that breastfeeding reduces atherosclerosis as compared with formula feeding (10) but does not reduce the risk of coronary heart disease mortality (11). The evidence for breastfeeding’s beneficial effects on adiposity, stature, blood pressure, and other chronic disease outcomes derives almost exclusively from observational studies. As shown in the systematic review by Owen et al (5), residual confounding and publication bias may at least partially explain the published evidence. In addition, studies that compare the growth of infants receiving prolonged and exclusive breastfeeding with the growth of infants with shorter durations and lesser degrees of breastfeeding are susceptible to selection bias. Infants who grow more slowly for genetic or other constitutional reasons may be satisfied with their mothers’ breast-milk output, whereas those who grow more rapidly may require a larger energy intake (12, 13), and mothers may be reluctant (or, in some cases, unable) to increase their milk supply by augmenting the frequency of nursing. To overcome the methodologic limitations inherent in observational studies, we designed a 6.5-y follow-up of children participating in the Promotion of Breastfeeding Intervention Trial (PROBIT), a cluster-randomized trial in the Republic of Belarus (14).

SUBJECTS AND METHODS

Subjects The detailed methods of PROBIT and the results during the first year of follow-up were previously reported (14). The units (clusters) of randomization were maternity hospitals and one affiliated polyclinic (outpatient clinic where the children are followed for well-child and illness care) per hospital, and double randomization was based on both a random numbers table and a coin toss (14). The experimental intervention was based on the Baby-Friendly Hospital Initiative, which was developed by WHO and UNICEF to promote and support breastfeeding, particularly among mothers who have chosen to initiate breastfeeding (15); the control maternity hospitals and polyclinics continued the practices and policies in effect at the time of randomization. The trial results are based on a total of 17 046 healthy breastfed infants from 31 maternity hospitals or polyclinics; all were born at term, weighed 욷2500 g, and were enrolled during their postpartum stay (14). To our knowledge, PROBIT is the largest randomized trial ever undertaken in the area of human lactation. It has been registered as ISRCTN37687716, and it conforms to the Consolidated Standards of Reporting Trials guidelines for the design, analysis, and reporting of cluster-randomized trials (16). As previously reported (14), the 2 randomized groups were similar in baseline sociodemographic and clinical variables, including maternal age, education, number of other children at home, the proportion of mothers who had breastfed a previous child for 욷3 mo, cesarean delivery, maternal smoking during pregnancy, birth weight, gestational age, and 5-min Apgar score. The experimental intervention led to a substantial difference in the duration of breastfeeding, which persisted throughout the first year of follow-up: in the experimental and control groups, respectively, 72.7% and 60.0% of mothers were still breastfeeding at 3 mo, 49.8% and 36.1% were doing so at 6 mo, 36.1% and

24.4% were doing so at 9 mo, and 19.7% and 11.4% were doing so at 12 mo (14). In addition, the prevalence of exclusive breastfeeding in the experimental group at 3 mo was 7 times that in the control group (43.3% and 6.4%, respectively), although it was low in both groups at 6 mo (7.9% and 0.6%, respectively) (14). When breastfeeding was supplemented or discontinued, the primary substitute was locally manufactured infant formula. Details concerning the intakes of other liquids, cereals, and other solid foods during infancy were reported previously (17). Follow-up interviews and examinations at age 6.5 y were performed by 1 pediatrician each at 24 of the 31 polyclinics; in the remaining 7 high-volume clinics, the follow-up visits were shared by 2 pediatricians. All anthropometric measurements were obtained in duplicate and averaged. These included standing and sitting height, measured with a wall-mounted stadiometer for standing height and a uniform, standard wooden bench for sitting height (both: Medtechnika, Pinsk, Belarus); weight, measured on an electronic digital scale (Bella 840; Seca Corporation, Hamburg, Germany); and head, waist, hip, midupper arm, and midthigh circumferences, measured with a nonstretchable cloth tape. Triceps and subscapular skinfold thicknesses were measured (also in duplicate) with the use of Lange skinfold calipers (Beta Technology, Santa Cruz, CA). Systolic and diastolic blood pressures were also measured in duplicate with the use of a digital oscillometric device (M1; Omron Healthcare, Milton Keynes, United Kingdom). When blood pressure was unobtainable with the Omron M1, a standard clinic sphygmomanometer was used. Training and standardization of pediatricians in obtaining the anthropometric and blood pressure measurements were the focus of a week-long training session involving a convenience sample of school-aged children living in a residential facility near Minsk, Belarus. The training included printed materials in Russian and a training video recording used in the National Health and Nutrition Examination Survey (NHANES) and provided by the National Center for Health Statistics. Each participating pediatrician was given a copy of the video, along with detailed written instructions (in Russian) for performing each measurement. Because interpediatrician and intrapediatrician measurement agreement (ie, remeasurement of the same child within a brief interval) was not formally examined, and because blinding of pediatricians to the experimental or control assignment was infeasible, a re-measurement audit was designed to assess the validity and reproducibility of the polyclinic data. For each of the 38 pediatricians, 5 children were randomly selected for the audit, for a total of 190 audited children. So that all children seen in follow-up were eligible for selection, the audit was carried out after primary data collection had been completed, at an average of 17.7 mo (range: 5.3–32.6 mo) after the initial clinic visit. The audit of the anthropometric and blood pressure measurements was carried out by one of the Minsk-based authors (LM, ZS, ID, or GS), all of whom were pediatricians who were blinded to the measures obtained at the initial clinic visit but not to the experimental or control treatment allocation. Because of the time elapsed between the audit and initial polyclinic visits, results were compared with the use of Pearson correlation coefficients. Statistical analysis All statistical analyses were based on intention to treat. Differences in outcome between the experimental and control groups were analyzed by using the MIXED procedure in SAS

BREASTFEEDING: EFFECTS ON ANTHROPOMETRY AND BLOOD PRESSURE

software (version 8.2; SAS Inc, Cary, NC). This statistical procedure accounts for the clustered randomization and thus permits inference at the level of the individual child, rather than at the level of the cluster (maternal hospital and polyclinic). The modeled differences presented below are based on this clusteradjusted model but are very similar to those obtained from a model that also adjusts for stratum-level variables [ie, geographic region (west or east) and urban or rural location] and for the following individual-level covariates: infant age at followup, sex, and birth weight and maternal education, maternal and paternal height (for standing and sitting height), and maternal and paternal BMI (for adiposity measures and blood pressure) (results available on request). We also analyzed mixed models that included terms for the sex of each child and a multiplicative sex ҂ treatment interaction term. In addition to these analyses based on the continuous anthropometric measurements, we compared the proportions of children in the 욷85th and 욷95th percentiles of BMI based on the Centers for Disease Control and Prevention (CDC) 2000 standards (18). The analogous penalized quasilikelihood for generalized linear mixed models [GLIMMIX (19)] procedure in SAS was used to estimate the adjusted odds ratios and 95% CIs for these dichotomous outcomes. Missing data were not imputed. RESULTS

A total of 13 889 children were seen in follow-up for the present study; they represented 81.5% of the 17 046 who had originally been randomly assigned. Of the 3157 (17 046 –13 889) children who were randomly assigned but not followed up, 88 had died, 2938 were lost to follow-up, and 131 were unable or unwilling to attend their visit. Follow-up rates were similar in the experimental (80.2%) and control (82.9%) polyclinic groups, but they varied considerably by polyclinic: from 56.1% at one of the Minsk polyclinics to 94.6% at the small, rural polyclinic at Klimovichi. The mean 앐 SD age at follow-up was 6.6 앐 0.3 y. As shown in Table 1, the children followed up in the experTABLE 1 Baseline comparison of experimental and control groups of children followed up at age 6.5 y

Variable Maternal age (%) 쏝20 y 20–34 y 욷35 y Maternal education (%) Incomplete secondary Complete secondary Advanced secondary or partial university Complete university Older children living in household (%) 0 1 욷2 Maternal smoking during pregnancy (%) Male (%) Birth weight (g)1 1

x¯ 앐 SD.

Experimental group (n ҃ 7108)

Control group (n ҃ 6781)

14.3 81.4 4.3

13.2 82.6 4.2

4.4 34.3 47.8 13.5

3.0 29.7 54.5 12.9

58.8 33.3 7.9 2.6 51.4 3440 앐 418

54.5 36.1 9.4 1.6 52.0 3441 앐 423

1719

TABLE 2 Pearson correlation coefficients (r) (and 95% CIs) for anthropometric and blood pressure measurements at the audit and original polyclinic visits1 Outcome Height Leg length BMI Head circumference Waist circumference Hip circumference Midthigh circumference Midupper arm circumference Triceps skinfold thickness Subscapular skinfold thickness Systolic blood pressure Diastolic blood pressure 1

r (95% CI) 0.84 (0.79, 0.88) 0.74 (0.69, 0.80) 0.89 (0.85, 0.91) 0.65 (0.55, 0.72) 0.84 (0.80, 0.88) 0.72 (0.64, 0.78) 0.55 (0.44, 0.64) 0.85 (0.80, 0.88) 0.59 (0.48, 0.67) 0.65 (0.55, 0.72) 0.55 (0.44, 0.64) 0.45 (0.32, 0.55)

n ҃ 190.

imental and control groups had similar baseline characteristics; the small differences paralleled those seen at randomization and previously reported (14). The children also were virtually identical in mean parental BMI, according to the heights and weights reported by the accompanying parent (usually the mother) at the follow-up visit: BMIs of 24.5 and 25.7 in both groups for the mother and father, respectively. The audit results are summarized in Table 2. The results shown are the Pearson correlation coefficients (and 95% CIs) for comparing the initial clinic visit results and the results at the audit visit. The test-retest correlations were high (쏜0.80) for height, BMI, waist circumference, and midupper arm circumference; intermediate (0.60 – 0.80) for leg length, head circumference, and subscapular skinfold thickness; and low (쏝0.60) for triceps skinfold thickness, midthigh circumference, and systolic (and, particularly, diastolic) blood pressure measurements. The main study results, including the number of children in whom each measure was obtained, the crude subject-based means 앐 SDs in the experimental and control groups, the intraclass correlation coefficients (ICCs) reflecting the degree of within-polyclinic clustering, and the cluster-adjusted differences in means (and 95% CIs), are shown in Table 3. The results showed a high degree of clustering for triceps skinfold thickness and systolic blood pressure. No significant between-group differences were observed in any of the anthropometric or blood pressure results except a slightly higher midthigh circumference in the experimental than in the control group. The CIs were reasonably narrow for those measures with good correlations between measures at the initial clinic visit and at the audit (as shown in Table 2) and with lower ICCs—ie, height, BMI, head circumference, and subscapular skinfold thickness— but were fairly wide for triceps skinfold thickness and both systolic and diastolic blood pressures. Significant (P 쏝 0.05) treatment ҂ sex interactions were observed for head circumference, subscapular skinfold thickness, and midthigh circumference. A significant effect of treatment on head circumference was observed in girls only [cluster-adjusted difference ҃ 0.3 (95% CI: 0.02, 0.5) cm], whereas the effect on midthigh circumference was significant in both sexes, but higher in girls than in boys (1.0 and 0.7 cm, respectively). Effects on subscapular skinfold thicknesses were nonsignificant in both

1720

KRAMER ET AL

TABLE 3 Anthropometric measurements and blood pressure results1 Outcome Height, cm (n ҃ 13 882) Leg length, cm (n ҃ 13 872) BMI, in kg/m2 (n ҃ 13 879) Head circumference, cm (n ҃ 13 882) Waist circumference, cm (n ҃ 13 883) Hip circumference, cm (n ҃ 13 883) Waist:hip ratio (n ҃ 13 883) Midthigh circumference, cm (n ҃ 13 882) Midupper arm circumference, cm (n ҃ 13 882) Triceps skinfold thickness, mm (n ҃ 13 879) Subscapular skinfold thickness, mm (n ҃ 13 878) Systolic blood pressure, mm Hg (n ҃ 13 875) Diastolic blood pressure, mm Hg (n ҃ 13 876) 1 2

Experimental group

Control group

ICC

Cluster-adjusted difference in means (95% CI)

121.1 앐 5.1 55.4 앐 3.8 15.6 앐 1.7 51.9 앐 1.5 54.6 앐 4.3 63.1 앐 4.7 0.87 앐 0.06 33.0 앐 3.3 17.7 앐 1.8 9.9 앐 4.1 5.9 앐 2.3 97.8 앐 9.5 57.3 앐 7.6

120.2 앐 5.2 55.0 앐 3.5 15.6 앐 1.7 51.6 앐 1.6 54.2 앐 4.4 62.2 앐 4.9 0.87 앐 0.06 32.1 앐 3.2 17.9 앐 1.8 10.0 앐 3.6 5.8 앐 2.4 96.7 앐 8.9 57.8 앐 7.4

0.07 0.09 0.03 0.06 0.09 0.07 0.13 0.07 0.08 0.18 0.05 0.21 0.12

0.7 (Ҁ0.3, 1.7) 0.2 (Ҁ0.6, 1.0) 0.1 (Ҁ0.2, 0.3) 0.2 (Ҁ0.03, 0.5) 0.3 (Ҁ0.8, 1.4) 0.6 (Ҁ0.3, 1.4) 0.00 (Ҁ0.02, 0.01) 0.9 (0.2, 1.5)2 Ҁ0.2 (Ҁ0.6, 0.2) Ҁ0.4 (Ҁ1.8, 1.0) 0.0 (Ҁ0.4, 0.5) 0.2 (Ҁ2.9, 3.3) 0.2 (Ҁ1.8, 2.2)

ICC, intraclass correlation coefficient. Significant difference between groups.

sexes but were opposite in direction (ie, positive in boys, negative in girls). The proportions of children with BMIs 욷 85th and 욷 95th percentiles were 13.4% and 5.9% in the experimental group and 12.2% and 5.0% in the control group. The corresponding adjusted odds ratios were 1.1 (95% CI: 0.8, 1.4) and 1.2 (0.8, 1.6). To assess whether we could reproduce the protective associations of prolonged and exclusive breastfeeding reported in previous observational studies, we compared 2 extreme observational infant-feeding groups previously reported (20) to differ widely in their growth trajectories during the first 12 mo of life: 1) those completely weaned within the first month (n ҃ 1136) and 2) those breastfed exclusively for 욷6 mo with continued breastfeeding to any degree until 욷12 mo of age (n ҃ 215). The observational analyses were based on multiple linear regression models containing the same stratum- and individual-level covariates as the expanded mixed models described above. None of the observed differences suggested lower adiposity or blood pressure in the group with prolonged and exclusive breastfeeding than in the group with less breastfeeding. In fact, the former group had significantly higher mean BMIs [cluster-adjusted difference: 0.3 (95% CI: 0.4, 0.5)], triceps skinfold thicknesses [1.3 (0.7, 1.9) mm], and systolic blood pressure [1.5 (0.1, 2.9) mm Hg] than did the latter. DISCUSSION

Our results, the first based on a randomized study of healthy, full-term infants, indicate that prolonged, exclusive breastfeeding provides no apparent beneficial effects on stature, BMI or other measures of adiposity, or blood pressure in 6.5-y-old Belarussian children. Our results are consistent with those of the recently published meta-analysis conducted by Owen et al (5), and they suggest that previous studies reporting differences in these outcomes may have been subject to biases due to residual confounding, subject selection, and selective publication. Residual confounding is of particular concern (21). In the individualpatient meta-analysis by Owen et al, an association between breastfeeding (rather than formula feeding) and mean BMI was virtually abolished after control for socioeconomic status, parental BMI, and maternal smoking (5). A recent analysis of

the large Nurses’ Health Study II also found no association between the duration or exclusivity of breastfeeding and obesity in adulthood (22). Despite the large sample size and the high rate of follow-up in the present study, the precision of the observed differences in those anthropometric blood pressure measures that were more observer (ie, pediatrician) dependent was only modest: in particular, for triceps skinfold thickness and blood pressure. Wide CIs resulted not only from random measurement error among children examined by individual pediatricians, but also from systematic differences among the 38 participating pediatricians, which led to systematic differences in the means of these anthropometric and blood pressure measurements among the 31 polyclinics. These systematic measurement differences constituted a source of clustering within polyclinics and resulted in high intraclass correlations, ie, a tendency for children within a single polyclinic than for those from different polyclinics to have similar measurements. Because randomization was also clustered within polyclinics, this “double clustering” substantially reduced the precision (ie, yielded wider CIs) of the observed differences between the experimental and control groups. Unfortunately, for both geographic and economic reasons, it was infeasible for pediatricians who worked at one polyclinic to examine children enrolled in other polyclinics. Thus, the present study cannot exclude small differences that may be due to the intervention, especially with respect to triceps skinfold thickness and blood pressure. It is important to emphasize that the experimental intervention was designed to increase the degree and duration of breastfeeding, not to increase its initiation. Thus, our findings may not apply to comparisons of breastfeeding versus formula feeding (the comparisons most often emphasized in previous studies). Nonetheless, many previous studies have reported graded, doseresponse associations between the degree or duration (or both) of breastfeeding and adiposity measures. Our findings clearly do not support those associations. Although Belarus is a developed country that has experienced lower rates of infant mortality and that has more-favorable health indicators than have other former Soviet countries, we are aware of no published data on the extent to which the obesity epidemic

BREASTFEEDING: EFFECTS ON ANTHROPOMETRY AND BLOOD PRESSURE

currently gripping North America and Western Europe is affecting Belarus. The fact that the observed proportions of children with BMIs 욷 85th (13%) and 욷 95th (5%) percentiles are far lower than those recently reported from the United States (23) suggests a far less severe problem among Belarussian children. Caution is therefore advised in generalizing our results to settings with a much higher prevalence of child obesity. Nonetheless, in an examination of reported protective effects of breastfeeding on obesity over the several decades that this association has been studied, no apparent differences are evident, nor are we aware of any biological mechanisms that would modify the potential protective effect of breastfeeding in different settings. Continued follow-up of the PROBIT cohort, along with attempts to provide more direct measurements of body fat and to reduce variations in the measurement of blood pressure, should help in detecting effects of smaller magnitude (which could nonetheless have an important public health effect), as well as those that may develop later in childhood or adulthood with increased deposition of body fat. Until that time, however, it seems unwise to depend on current efforts to promote exclusive and prolonged breastfeeding as an effective population health strategy for stemming the current obesity epidemic or reducing the risk of future hypertension. The additional contributing members of the PROBIT Study Group are named. National Research and Applied Medicine Mother and Child Centre (Minsk, Belarus): Natalia Bazulko, Olga Gritsenko, Lidia Ovchinikova, and Julia Rizkovskaya. Polyclinic pediatricians: Natalia Andreeva (Rogachev), Tatiana Avdeichuk (Brest), Elena Avsiuk (Vitebsk), Irina Baikevich (Slonim), Zinaida Bisucova (Zlobin), Irina Bujko (Oshmiany), Tamara Galushkina (Volkovysk), Marina Gotovchi (Brest), Danuta Iodkovskaya (Berestovitsa), Galina Ivanova (Mogilev), Larisa Kebikova (Minsk), Galina Kluchnikova (Ostrovets), Maria Kotliarovich (Soligorsk), Galina Kovalevskaya (Lepel), Natalia Krokas (Mosty), Nadezda Kushkova (Rechitsa), Afanasia Lazarenko (Klimovichi), Ludmila Lazuta (Minsk), Zinaida Liamkina (Borisov), Raisa Lisiura (Stolin), Tamara Nabedo (Novolukoml), Svetlana Pleskach (Baranovichi), Oksana Potapenko (Soligorsk), Svetlana Pridhodoskaya (Bereuza), Valentina Rahotskaya (Oshmiany), Irina Rogach (Mstislavl), Ludmila Rutkovskaya (Kobrin), Natalia Senchuk (Rechitsa), Elena Seraia (Baranovichi), Ludmila Sheveleva (Kobrin), Vera Shota (Svisloch), Anna Silvanovich (Shuchin), Lilia Smolskaya (Glubokoe), Valentina Solovey (Volkovysk), Zoya Solovyova (Dokshitsy), Natalia Tsarik (Svetlogorsk), Nadezda Turkovskaya (Zlobin), and Oxana Zarodova (Minsk Region). The authors’ responsibilities were as follows—MSK, RWP, J-PC, SS, EH, and BC: contributed to obtaining funding for this project and to the design, analysis, interpretation, and writing or revision (or both) of the manuscript; LM, IV, NB, ZS, ID, and GS: contributed to the design of the study and to the planning, implementation, and monitoring of the field work in Belarus; RMM, GDS, and MWG: contributed to the selection of, and measurement methods for, the anthropometric and blood pressure outcomes and to the content of the manuscript. MSK is Senior Investigator of the Canadian Institutes of Health Research; RWP is a Monat-McPherson Career Investigator of McGill University and a career investigator (chercheur-boursier) of the Fonds de la recherche en santé du Québec.

REFERENCES 1. Dewey K, Peerson J, Brown K, et al. Growth of breast-fed infants deviates from current reference data: a pooled analysis of US, Canadian, and European data sets. Pediatrics 1995;96:495–503.

1721

2. de Onis M, Garza C, Victora CG, Bhan MK, Norum KR. The WHO Multicentre Growth Reference Study (MGRS): rationale, planning, and implementation. Forensic Sci Int 2004;25(suppl 1):S3– 84. 3. Harder T, Bergmann R, Kallischnigg G, Plagemann A. Duration of breastfeeding and risk of overweight: a meta-analysis. Am J Epidemiol 2005;162:397– 403. 4. Owen CG, Martin RM, Whincup PH, Davey Smith G, Cook DG. Effect of infant feeding on the risk of obesity across the life course: a quantitative review of published evidence. Pediatrics 2005;115:1367–77. 5. Owen CG, Martin RM, Whincup PH, Davey-Smith G, Gillman MW, Cook DG. The effect of breastfeeding on mean body mass index throughout life: a quantitative review of published and unpublished observational evidence. Am J Clin Nutr 2005;82:1298 –307. 6. Martin RM, Davey Smith G, Mangtani P, Frankel S, Gunnell D. Association between breast feeding and growth: the Boyd-Orr cohort study. Arch Dis Child Fetal Neonatal Ed 2002;87:F193–201. 7. Martin RM, Holy JMP, Davey Smith G, et al. Could associations between breastfeeding and insulin-like growth factors underlie associations of breastfeeding with adult chronic disease? The Avon Longitudinal Study of Parents and Children. Clin Endocrinol 2005;62: 728 –37. 8. Owen CG, Whincup PH, Gilg JA, Cook DG. Effect of breast feeding in infancy on blood pressure in later life: systematic review and metaanalysis. BMJ 2003;327:1–7. 9. Martin RM, Gunnell D, Davey Smith G. Breastfeeding in infancy and blood pressure in later life: Systematic review and meta-analysis. Am J Epidemiol 2005;161:15–26. 10. Martin RM, Ebrahim S, Griffin M, et al. Breastfeeding and atherosclerosis: intima-media thickness and plaques at 65-year follow-up of the Boyd Orr cohort. Arterioscler Thromb Vasc Biol 2005;25:1–7. 11. Martin RM, Davey Smith G, Mangtani P, TYilling K, Frankel S, Gunnell D. Breastfeeding and cardiovascular mortality: the Boyd Orr cohort and a systematic review with meta-analysis. Eur Heart J 2004;25:778 – 86. 12. Hill A. A short textbook of medical statistics. London, United Kingdom: Hodder & Stoughton, 1977. 13. Sauls H. Potential effect of demographic and other variables in studies comparing morbidity of breast-fed and bottle-fed infants. Pediatrics 1979;64:523–7. 14. Kramer MS, Chalmers B, Hodnett ED, et al. Promotion of breastfeeding intervention trial (PROBIT): a randomized trial in the Republic of Belarus. JAMA 2001;285:413–20. 15. WHO/UNICEF. Protecting, promoting and supporting breastfeeding: the special role of maternity services. Geneva, Switzerland: World Health Organization, 1989. 16. Campbell MK, Elbourne DR, Altman DG, for the CONSORT Group. CONSORT statement: extension to cluster randomised trials. BMJ 2004;328:702– 8. 17. Kramer MS, Guo T, Platt RW, et al. Feeding effects on growth during infancy. J Pediatr 2004;145:600 –5. 18. Kuczmarski RJ, Ogden CL, Grummer-Strawn LM, et al. CDC growth charts: United States. Adv Data 2000;314:1–27. 19. Breslow NE, Clayton DG. Approximate inference in generalized linear mixed models. J Am Stat Assoc 1993;88:9 –25. 20. Kramer MS, Guo T, Platt RW, et al. Breastfeeding and infant growth: biology or bias? Pediatrics 2002;110:343–7. 21. Davey Smith G, Ebrahim S. Data dredging, bias, or confounding. BMJ 2002;325:1437– 8. 22. Michels KB, Willett WC, Graubard BI, et al. A longitudinal study of infant feeding and obesity throughout life course. Int J Obes (Lond) 2007;31:1078 – 85. 23. Ogden CL, Carroll MD, Curtin LR, McDowell MA, Tabak CJ, Flegal KM. Prevalence of overweight and obesity in the United States, 1999 – 2004. JAMA 2006;295:1549 –55.

Childhood dairy intake and adult cancer risk: 65-y follow-up of the Boyd Orr cohort1–3 Jolieke C van der Pols, Chris Bain, David Gunnell, George Davey Smith, Clare Frobisher, and Richard M Martin ABSTRACT Background: Dairy consumption affects biological pathways associated with carcinogenesis. Evidence for a link between cancer risk and dairy consumption in adulthood is increasing, but associations with childhood dairy consumption have not been studied adequately. Objective: We investigated whether dairy consumption in childhood is associated with cancer incidence and mortality in adulthood. Design: From 1937 through 1939, some 4999 children living in England and Scotland participated in a study of family food consumption, assessed from 7-d household food inventories. The National Health Service central register was used to ascertain cancer registrations and deaths between 1948 and 2005 in the 4383 traced cohort members. Per capita household intake estimates for dairy products and calcium were used as proxy for individual intake. Results: During the follow-up period, 770 cancer registrations or cancer deaths occurred. High childhood total dairy intake was associated with a near-tripling in the odds of colorectal cancer [multivariate odds ratio: 2.90 (95% CI: 1.26, 6.65); 2-sided P for trend ҃ 0.005] compared with low intake, independent of meat, fruit, and vegetable intakes and socioeconomic indicators. Milk intake showed a similar association with colorectal cancer risk. High milk intake was weakly inversely associated with prostate cancer risk (P for trend ҃ 0.11). Childhood dairy intake was not associated with breast and stomach cancer risk; a positive association with lung cancer risk was confounded by smoking behavior during adulthood. Conclusions: A family diet rich in dairy products during childhood is associated with a greater risk of colorectal cancer in adulthood. Confirmation of possible underlying biological mechanisms is needed. Am J Clin Nutr 2007;86:1722–9. KEY WORDS Dairy products, neoplasms, life-course analysis, incidence, mortality, United Kingdom

INTRODUCTION

The exact relation between dairy consumption and cancer risk remains unclear and is likely to vary by cancer type and the timing of exposure. The main biological pathways through which a person’s dairy intake is suspected to modify cancer risk include greater circulation of insulin-like growth factor I (IGF-I) (1, 2), modification of vitamin D status (3, 4), a greater intake of conjugated linolenic acid (5, 6), and exposure to contaminants such as polychlorinated biphenyls (7–9). IGF-I plays a central role in the regulation of prenatal and postnatal growth, but it also exerts a growth-promoting effect in

1722

adulthood through low apoptosis, high cell proliferation, and angiogenesis (10). High concentrations of IGF-I are associated with a greater risk of prostate, colorectal, and premenopausal breast cancer (11, 12). Milk intake is positively associated with plasma IGF-I concentrations in cross-sectional (13–16) and experimental (17) studies of adults and children (2, 18). In contrast, inverse associations between milk intake in childhood and IGF-I concentrations in adulthood have been reported in a randomized controlled trial of prenatal and postnatal milk supplementation (19). Similar findings have been reported in a recent analysis of the Boyd Orr cohort (20). A long-term programming effect of early diet on the IGF-I axis and cancer risk has been proposed (21, 22). It has been proposed that high calcium intakes, including calcium of dairy origin, may increase prostate cancer risk by lowering circulating 1,25-dihydroxyvitamin D [1,25(OH)2D] (4, 23, 24), which regulates growth and differentiation in multiple normal and malignant cell types (25). In contrast, dietary calcium may help prevent colorectal cancer through the binding of bile acids in the small intestine and its involvement in the vitamin D pathway (26). The intake of dairy products thus may affect a mixture of pathways associated with carcinogenesis, and the existence of an early-life programming effect of dairy consumption remains largely unknown. To date, most studies of dairy consumption and cancer risk have been based on adult dairy intake estimates. A comprehensive review concluded that adult milk consumption does not appear to have a strong association with breast cancer risk (27), but a study in Norwegian women showed that the combination of high childhood and high adult milk consumption is a significant predictor of lower breast cancer risk (28). Two US studies also found that milk intake in childhood (29) and milk-fat 1

From the Longitudinal Studies Unit, Division of Epidemiology and Social Medicine, School of Population Health, University of Queensland, Brisbane, Australia (JCvdP and CB); the Department of Social Medicine, University of Bristol, Bristol, United Kingdom (DG, GDS, and RMM); and the Department of Public Health and Epidemiology, The University of Birmingham, Birmingham, United Kingdom (CF). 2 Supported by the World Cancer Research Fund (follow-up of the Boyd Orr cohort) and by Capacity Building in Population Health Research grant no. 252834 from the National Health and Medical Research Council of Australia (to JCvdP). 3 Reprints not available. Address correspondence to JC van der Pols, Longitudinal Studies Unit, Division of Epidemiology and Social Medicine, School of Population Health, University of Queensland, Brisbane, QLD 4006, Australia. E-mail: [email protected]. Received April 22, 2007. Accepted for publication July 24, 2007.

Am J Clin Nutr 2007;86:1722–9. Printed in USA. © 2007 American Society for Nutrition

CHILDHOOD DAIRY INTAKE AND ADULT CANCER RISK

consumption in adolescence (30) were associated with a lower risk of breast cancer, in which both the calcium-vitamin D pathway (31) and the IGF axis, through their effect on breast density (32), were suspected to play a role. Adult dairy consumption has fairly consistently been found to have a small positive association with prostate cancer risk (23, 33) and to be inversely associated with colorectal cancer risk (34 –37), but the effect of childhood exposure to dairy products on cancer risk has not been evaluated. SUBJECTS AND METHODS

The Boyd Orr Cohort The establishment of the Boyd Orr cohort has been described in detail elsewhere (38). In brief, the data forming the basis of these analyses were obtained from the original records of the Carnegie (Boyd Orr) survey of diet and health in pre-World War II (prewar) Great Britain (39). The survey was carried out from 1937 through 1939 in 1343 families, mainly working-class, who were living in 16 rural and urban areas of England and Scotland. Detailed measurements were made of household diet (see below) and of the health, growth, and living conditions of the children in the households. The name, age, and address of the children (mean age: 8 y; interquartile range: 4 –11 y) in the surveyed families were obtained from the original records and used to trace them through the National Health Service Central Register (NHSCR). Of the 4999 children identified, 4383 (87.7%) have been successfully traced and are included in this analysis; of the remaining 616 participants, 424 could not be identified through the NHSCR, 171 were censored before 1948, and 21 were identified but had been lost to follow-up by the NHSCR. As a result of further searches of archived records, contacts with surviving study members, and additional notifications from the NHSCR, the trace rate has increased slightly since earlier publications. The representativeness of those traced was described previously (38). Traced survey participants were almost 1 y younger than their untraced counterparts (P 쏝 0.0001) but did not differ in terms of childhood energy intake, food expenditure, or socioeconomic status (SES). The NHSCR was used to follow up survey members for mortality and cancer registration (38). Cause of death is ascertained from death certificates; partly with the use of cancer registrations, the cause of death was classified according to the International Classification of Diseases, 9th (ICD-9) and 10th Revisions (ICD-10). This analysis is based on traced cohort members who were resident in Great Britain on 1 January 1948 and on cancer deaths and registrations occurring up to 31 July 2005. It is limited to the 4374 traced participants (2159 men and 2215 women) for whom full data are available; 9 traced participants were excluded because some of their dietary data were missing. Ethical approval for the revitalization of the Boyd Orr Study was provided by the local Research Ethics Committee of the United Bristol Healthcare Trust. Dietary assessment Dietary data in the original Carnegie survey were obtained by using a 7-d household inventory method (39, 40). A weighed inventory of all foods in the household was recorded in a diary at the beginning of the survey period. A weighed record of all subsequent food brought into the home was made, and a second

1723

inventory was carried out at the end of the survey period. Data from the diaries were then transcribed onto separate summary sheets for each household. Reanalysis of the food records was necessary to include nutrients not measured in the original study and to make use of advances in analytic techniques for foods whose composition is unlikely to have changed from the 1930s to today. Recoding of the foods for the present study was carried out by using a diet-in, data-out program (DIDO software; Medical Research Council Human Nutrition Research Center, Cambridge, United Kingdom) (41). Total fruit and vegetable (excluding potato) consumption was reanalyzed with the use of programs based on McCance and Widdowson’s The Composition of Foods (42) and supplements to that publication. Prewar food tables were used to adapt the database if the composition of 1930s foods was very different from that of foods today, or if there was no modern-day equivalent. Per capita food and nutrient intakes were calculated by dividing daily total intake by the total number of household members, taking into account meals missed by family members and meals consumed by visitors. The food category “dairy products” included all liquid milks (predominantly, whole milk), infant formulas, cream, cheese, ice creams, and milk puddings. Statistical analysis A composite outcome was derived from the presence of a cancer code anywhere on the death certificate or from the first cancer that was registered. Cancer sites were defined as all cancers—ie, all malignant neoplasms (ICD-9 codes: 140.0 –208.9; ICD-10 code: C0 –C97), breast cancer (ICD-9 codes: 174.0 – 174.9; ICD-10 codes: C50 –C509), lung cancer (ICD-9 codes: 162.0 –162.9; ICD-10 codes: C33–C349), colorectal cancer (ICD-9 codes: 153.0 –154.9, excluding 154.2 and 154.3; ICD-10 codes: C18 –C20), prostate cancer (ICD-9 code: 185; ICD-10 code: C61), and stomach cancer (ICD-9 code: 151; ICD-10 code: C16). Intake estimates of total dairy products, dairy subgroups (ie, milk, cheese, cream, milk pudding, and ice cream), and calcium were categorized into 4 equal groups according to their distribution in the study population. Odds ratios (ORs) for each cancer type and all cancer types combined were calculated by comparing each of the higher exposure groups with the lowest group by logistic regression analysis. To estimate associations between dairy products and cancer risk, we used logistic regression models, rather than Cox or Poisson regression, because the timings of cancer registrations and death in relation to cancer diagnosis are different, and thus the timings of the composite outcomes are not comparable. To investigate the robustness of this approach— and, in particular, to investigate whether competing causes of death would affect our results—we used Cox regression modeling to estimate hazard ratios (HRs) for comparison with ORs. The proportional hazards assumption was investigated both graphically and by formally testing that the log HR was constant over time for covariates in each model. In models in which the HR for a covariate was not constant over time, we included an interaction term of the covariate and time in the model. The choice of regression model made little difference to our conclusions. Clustering effects may have arisen because most participants in the cohort belonged to families that included other cohort

1724

VAN DER

POLS ET AL

TABLE 1 Baseline characteristics by total dairy intake group in 4374 children from the Boyd Orr Study cohort (1948 –2005)1 Total dairy intake Characteristic Age at survey (y) Male (%) Height (z score)4 BMI (z score)4 High household food expenditure (%) High socioeconomic status (%)5 Townsend score4 Smoking status in adulthood (% current smokers)4,6 Dietary intake Milk (% of total dairy) Fruit (g/d) Vegetables (g/d) Eggs and egg dishes (g/d) Calcium (mg/d)7 Fat (g/d)7 Energy (kJ)

Group 1

Group 2

Group 3

Group 4

P for trend2

7.9 앐 5.03 48 Ҁ0.19 앐 0.95 0.00 앐 0.95 5 32 0.37 앐 4.30 27

7.5 앐 4.9 49 Ҁ0.14 앐 0.94 0.01 앐 1.01 6 35 0.90 앐 4.69 27

7.3 앐 4.7 52 Ҁ0.11 앐 0.94 Ҁ0.07 앐 0.97 15 42 0.67 앐 5.00 17

7.7 앐 4.6 49 0.38 앐 0.96 0.09 앐 1.09 39 76 Ҁ0.81 앐 2.95 12

0.004 0.46 쏝0.0001 0.14 쏝0.0001 쏝0.0001 쏝0.0001 쏝0.0001

90 앐 11 18 앐 22 58 앐 32 13 앐 11 406 앐 123 76 앐 11 8447 앐 1877

93 앐 8 24 앐 27 62 앐 34 16 앐 14 481 앐 129 80 앐 14 9148 앐 2061

96 앐 4 37 앐 37 71 앐 37 18 앐 14 560 앐 121 82 앐 14 9302 앐 1995

97 앐 5 55 앐 64 65 앐 52 24 앐 15 743 앐 201 78 앐 22 11 201 앐 2362

쏝0.0001 쏝0.0001 0.003 쏝0.0001 쏝0.0001 0.004 쏝0.0001

Total median dairy intake was 89, 163, 255, and 471 g/d in group 1, 2, 3, and 4, respectively. n ҃ 1086 in group 1 and 1096 in groups 2– 4. P values from Mantel-Haenszel chi-square test (categorical variables) or linear regression (continuous variables). 3 x៮ 앐 SD (all such values). 4 z Scores were available only for a proportion of study participants: height, n ҃ 2632 (60%); BMI, n ҃ 2620 (60%); Townsend score, n ҃ 4069 (93%); smoking status, n ҃ 1626 (37%). 5 Socioeconomic categories: I, professional and higher management; II, intermediate; III, skilled worker. 6 Smoking status as reported in 1997 follow-up data collection. 7 Energy adjusted to 9200 kJ (앒2200 kcal) by using the residual method (47). Height, BMI, household food expenditure, socioeconomic status, and dietary intake were measured in childhood; Townsend score and smoking status were measured in adulthood. 1 2

members, and, therefore, the cohort participants shared childhood conditions and possible genetic effects on cancer. We calculated robust SEs by using the “repeated” option for PROC GENMOD and the “id” option for PROC PHREG in SAS to allow for a between-family component of variation. Because age is a strong determinant of mortality risk and because energy intake was associated with dairy intake and cancer mortality in this population (43), we controlled for age, sex, and energy intake (continuous) in a basic model. In an expanded multivariate model, we explored the confounding effect of the following factors measured in childhood: fruit, vegetable, and fat intakes; weight and height; district and season of the survey; SES (determined from the occupation of the head of the household); and per capita food expenditure of the household. Townsend scores (an area-based measure of deprivation, with high positive values indicating high levels of socioeconomic deprivation) based on the British Health Authority area of residence at the time of death, emigration, or participation in follow-up study in 1998 were used as a proxy for SES in adulthood (44). These covariates were retained in the model if they changed the OR estimate of the highest compared with the lowest intake group by 쏜10%. Because dairy intake varied by survey district, all models were also evaluated with stratification for this variable. We analyzed associations between total dairy intake and cancer risk and associations between individual dairy food groups and cancer risk to determine whether particular dairy foods could explain the associations. We also repeated the multivariate model with additional inclusion of calcium and protein intake to investigate whether associations between dairy intake and cancer risk could be explained by the calcium or protein content of dairy foods. Associations between calcium intake and cancer risk were

assessed after correction for confounders to determine whether calcium was associated with cancer risk independent of dairy intake. Associations between dairy intake and colorectal cancer were further adjusted for childhood meat intake, because high meat consumption is a known risk factor for that disease (45). Fat and calcium were included in the models as unadjusted nutrients, because the correlation between these nutrients and energy intake was corrected for by inclusion of energy in all multivariate models. To assess whether associations between childhood dairy intake and adult cancer risk may be mediated by IGF-I concentration, we used linear regression analysis to examine the association between childhood dairy intake and adult height, which is a marker for childhood IGF concentrations (46). To assess whether associations between childhood dairy intake and adult cancer risk were confounded by smoking status, we repeated the analyses in the subgroup of 1626 participants who provided information about their smoking behavior (current, past, or never) in a follow-up data collection in 1997. A test for linear trend was obtained by modeling ordinal numbers ranging from 1 to 4 (for lowest to highest intake group, respectively) as a continuous term in the regression model. All analyses were performed with SAS statistical software (version 9.1; SAS Institute Inc, Cary, NC). All reported P values are 2-sided.

RESULTS

The baseline characteristics of the study population by total dairy intake group are shown in Table 1. The median daily intake of dairy products ranged from 89 g/d in the lowest group to 471

1725

CHILDHOOD DAIRY INTAKE AND ADULT CANCER RISK TABLE 2 Odds ratios (OR) (and 95% CIs) of cancer incidence and mortality by total dairy intake group in 4374 participants from the Boyd Orr Study cohort (1948 –2005)1 Total dairy intake Cancer site All Cases (n) Basic OR2 Multivariate4 Multivariate plus calcium4 Breast Cases (n) Basic OR2 Multivariate4 Multivariate plus calcium5 Colorectal Cases (n) Basic OR2 Multivariate4 Multivariate plus calcium5 Prostate Cases (n) Basic OR2 Multivariate4 Multivariate plus calcium5 Lung Cases (n) Basic OR2 Multivariate4 Multivariate plus calcium5 Stomach Cases (n) Basic OR2 Multivariate4 Multivariate plus calcium5

Group 1

Group 2

Group 3

Group 4

P for trend

198 1.00 1.00 1.00

215 1.09 (0.86, 1.38)3 1.10 (0.87, 1.39) 1.11 (0.87, 1.42)

178 0.87 (0.69, 1.10) 0.91 (0.72, 1.16) 0.94 (0.72, 1.21)

179 0.78 (0.60, 1.02) 0.84 (0.64, 1.10) 0.89 (0.61, 1.29)

0.02 0.09 0.40

26 1.00 1.00 1.00

27 1.03 (0.57, 1.87) 1.02 (0.56, 1.86) 1.13 (0.61, 2.09)

18 0.69 (0.37, 1.28) 0.67 (0.35, 1.25) 0.80 (0.41, 1.58)

26 0.93 (0.47, 1.82) 0.89 (0.45, 1.75) 1.35 (0.54, 3.39)

0.59 0.49 0.79

12 1.00 1.00 1.00

17 1.45 (0.68, 3.07) 1.47 (0.69, 3.12) 1.60 (0.73, 3.50)

17 1.46 (0.70, 3.07) 1.59 (0.75, 3.35) 1.89 (0.82, 4.36)

30 2.62 (1.15, 5.98) 2.90 (1.26, 6.65) 4.31 (1.30, 14.22)

0.01 0.005 0.007

11 1.00 1.00 1.00

10 0.83 (0.34, 2.00) 0.83 (0.34, 1.99) 0.75 (0.30, 1.84)

11 0.91 (0.39, 2.12) 0.89 (0.38, 2.11) 0.74 (0.29, 1.88)

9 0.56 (0.22, 1.45) 0.55 (0.21, 1.42) 0.34 (0.11, 1.04)

0.31 0.30 0.20

44 1.00 1.00 1.00

47 1.06 (0.68, 1.64) 1.09 (0.70, 1.69) 0.98 (0.62, 1.53)

35 0.77 (0.49, 1.22) 0.90 (0.57, 1.44) 0.73 (0.45, 1.19)

27 0.56 (0.33, 0.93) 0.66 (0.39, 1.10) 0.38 (0.19, 0.75)

0.01 0.09 0.01

8 1.00 1.00 1.00

9 1.00 (0.35, 2.85) 1.01 (0.36, 2.85) 1.17 (0.39, 3.47)

10 1.08 (0.39, 3.04) 1.11 (0.40, 3.10) 1.46 (0.44, 4.89)

5 0.40 (0.10, 1.58) 0.42 (0.10, 1.71) 0.81 (0.09, 7.34)

0.23 0.26 0.79

1

Total median dairy intake was 89, 163, 255, and 471 g/d in group 1, 2, 3, and 4, respectively. All ORs were from logistic regression analysis. Adjusted for age, sex, and energy intake. 3 OR; 95% CI in parentheses (all such values). 4 Adjusted for age, sex, and energy and fruit intakes. 5 Adjusted for age, sex, and energy, fruit, and calcium intakes. 2

g/d in the highest group. On average, 94% of dairy intake was due to milk consumption. Children in the highest dairy intake group were taller, more likely to be in a household with high food expenditure during childhood, and less likely to live in a deprived area as an adult than were those in the lowest intake group (P for trend 쏝0.0001 for all). The former group also had higher intakes of fruit, eggs and egg dishes, calcium, and total energy (P for trend 쏝0.0001); moreover, milk consumption contributed more to the total dairy intake of this group than it contributed to the total dairy intake of the lowest intake group (P for trend 쏝0.0001). Group differences in age and in vegetable and fat intakes did not show clear patterns. Dairy intake was highly correlated with total calcium intake (Spearman correlation coefficient: 0.91). In total, 273 incident cancers (registrations) and 497 deaths from cancer were identified in 4374 participants in the follow-up period. Most prominent among the cancers were breast cancer (53 registrations, 45 deaths), colorectal cancer (35 registrations, 41 deaths), prostate cancer (17 registrations, 24 deaths), lung cancer (9 registrations, 147 deaths), and stomach cancer (3 registrations, 29 deaths). One breast cancer death and 3 lung cancer

deaths were secondary to earlier registration of a different cancer type, and these cases were excluded from the analyses. Total dairy intake The ORs for each cancer type and all cancers combined across levels of total dairy intake are shown in Table 2. Adjustment for fruit intake attenuated a weak inverse association between overall cancer risk and dairy consumption (multivariate OR for the highest versus the lowest intake group: 0.84; 95% CI: 0.64, 1.10; P for trend ҃ 0.09); further adjustment for calcium intake abolished the association (P ҃ 0.40). There was no evidence of confounding by weight; height; season of survey; vegetable, egg, or fat intake; or socioeconomic indicators in childhood or adulthood. Stratification by survey district did not alter the risk estimates. Dairy intake was not associated with breast or stomach cancer risk. High dairy intake was associated with a near-tripling in the odds of colorectal cancer (multivariate OR: 2.90; 95% CI: 1.26, 6.65; P for trend ҃ 0.005) compared with low dairy intake. There

1726

VAN DER

may have been negative confounding by calcium intake, because, after additional correction for this covariate, there was an even stronger association between dairy intake and colorectal cancer risk (multivariate- plus calcium-adjusted OR for highest versus lowest dairy group: 4.31; 95% CI: 1.30, 14.22; P for trend ҃ 0.007). Calcium intake did not have a clear independent association with colorectal cancer risk in this population (multivariateadjusted OR for highest versus lowest intake: 1.91; 95% CI: 0.84, 4.32; P for trend ҃ 0.18). Analysis using Cox regression produced results similar to those obtained from logistic regression: the multivariate- plus calcium-adjusted HR for the highest versus lowest dairy group was 3.98 (95% CI: 1.50, 10.64; P for trend ҃ 0.01). Further adjustment for protein and meat intakes, household food expenditure (in childhood), or Townsend deprivation score (in adulthood) did not appreciably change these estimates. The risk of prostate cancer was considerably lower in those with high dairy intakes, and it showed a dose-response effect. However, because of the small number of cases (n ҃ 41), evidence against the null hypothesis of no association was weak (multivariate- plus calcium-adjusted OR for highest versus lowest dairy group: 0.34; 95% CI: 0.11, 1.04; P for trend ҃ 0.20). Further adjustment for protein intake or Townsend deprivation score did not appreciably change this estimate, but the association was abolished after adjustment for household food expenditure in childhood (multivariate- plus calcium-adjusted OR for highest versus lowest dairy group: 0.41; 95% CI: 0.14, 1.26; P for trend ҃ 0.32). Cox regression analysis showed similar results (not shown). Lung cancer risk had a negative association with dairy intake after adjustment for calcium and other confounders; similar results were obtained by Cox regression analysis (multivariateplus calcium-adjusted HR for highest versus lowest dairy group: 0.44; 95% CI: 0.21, 0.92; P for trend ҃ 0.04). We investigated this association further in the subgroup of 1626 participants who provided information about their smoking behavior (current, past, or never) in a follow-up data collection in 1997. These analyses indicated that the association between childhood dairy intake and lung cancer was confounded by smoking status in adulthood. For example, the multivariate- and calcium-adjusted OR for the highest dairy group compared with the lowest for lung cancer in participants with follow-up data changed from 0.42 (95% CI: 0.12, 1.54) to 0.67 (95% CI: 0.17, 2.62) after additional adjustment for smoking status. This confounding is due to the fact that childhood dairy intake was significantly (P 쏝 0.001) lower in persons who, at the follow-up in adulthood were current smokers (226 g/d) than in those who were past (298 g/d) or never (300 g/d) smokers. Smoking status did not confound the associations between childhood dairy intake and adult risk of breast, colorectal, prostate, or stomach cancer or all cancer types combined. Milk intake Associations between childhood milk intake and cancer risks were very similar to those shown above (Table 3). A modest inverse association with overall cancer risk was abolished after adjustment for calcium and protein intake. A positive doseresponse relation with colorectal cancer risk remained after a similar adjustment (P for trend ҃ 0.02), with OR estimates close to those for total dairy intake. Compared with participants in the lowest milk intake group (쏝0.5 cups/d), those in the highest milk intake group (욷1.2 cups/d) had a significantly lower risk of prostate cancer (multivariate- plus calcium-adjusted OR: 0.24;

POLS ET AL

95% CI: 0.08, 0.70), although there was no clear dose-response relation. This inverse association was not confounded by household food expenditure in childhood. Associations for lung cancer with milk intake were the same as those for dairy intake. Milk consumption was not associated with breast or stomach cancer. Further adjustment for protein intake did not appreciably change these estimates. Other dairy products and dietary factors Participants who consumed the greatest amount of cream (쏜7 tablespoons/wk) had a multivariate- and calcium-adjusted OR of 3.46 (95% CI: 1.19, 10.06; P for trend ҃ 0.005) for all cancers combined versus participants consuming 쏝1.5 teaspoons cream/ wk. High intake of cream was also associated with a greater risk of lung cancer, but CIs were very wide (data not shown). Cream intake showed no relation to breast and colorectal cancer; estimates for prostate and stomach cancer are unavailable because the small numbers in the study made the regression model unstable. There were no associations between childhood consumption of cheese, milk pudding, or ice cream and adult cancer risk (data not shown). There were no independent associations between total calcium intake and the risk of any of the cancer types (data not shown), apart from weak evidence of an inverse association with prostate cancer risk (multivariate- plus dairy intake-adjusted OR for highest versus lowest group: 0.35; 95% CI: 0.08, 1.47; P for trend ҃ 0.08). Children in the highest dairy intake group also achieved taller heights in adulthood than did those in lower dairy intake groups; in a follow-up study of 1604 Boyd Orr study participants in 1997, mean self-reported age-adjusted heights were 165.4 cm in those who had been in the lowest and 166.7 cm in those who had been in the highest childhood dairy intake group (P for trend ҃ 0.02). However, this association disappeared after adjustment for household food expenditure or SES in childhood. DISCUSSION

The results of our 65-y follow-up study of children born in the 1920s or 1930s suggest that a family diet rich in dairy foods during childhood is associated with a substantially increased risk of colorectal cancer risk in adulthood. The association was independent of childhood meat, fruit, and vegetable intakes; SES in childhood and adulthood; and other risk factors. This increased risk appears to be in sharp contrast with the steadily increasing pool of evidence of a protective effect of dairy consumption on colorectal cancer risk in adult populations (34 –37). Our finding that high childhood intake of milk is associated with lower prostate cancer risk is also contrary to findings in adult intake studies (23, 33). The association of a greater risk of colorectal cancer with high childhood dairy intake seen in this study was strengthened after further adjustment for calcium intake, which suggested that there was negative confounding by calcium intake. Calcium intake did not have a clear independent association with colorectal cancer risk in this population, but, because of the high correlation between calcium and dairy intake, it is difficult to fully determine the independent effects of these dietary exposures in our data. Our study did not show any associations between childhood dairy consumption and breast cancer risk, and thus it conflicts with the findings of Hjartaker et al (28). A recent meta-analysis

1727

CHILDHOOD DAIRY INTAKE AND ADULT CANCER RISK TABLE 3 Odds ratios (OR) (and 95% CIs) of cancer incidence and mortality by milk intake group in 4374 participants from the Boyd Orr Study cohort (1948 –2005)1 Milk intake group Cancer site All Cases (n) Multivariate2 Multivariate plus calcium4 Breast Cases (n) Multivariate Multivariate plus calcium4 Colorectal Cases (n) Multivariate Multivariate plus calcium Prostate Cases (n) Multivariate Multivariate plus calcium4 Lung Cases (n) Multivariate Multivariate plus calcium4 Stomach Cases (n) Multivariate Multivariate plus calcium4

Group 1

Group 2

Group 3

Group 4

P for trend

206 1.00 1.00

207 1.02 (0.81, 1.29)3 1.02 (0.81, 1.30)

180 0.88 (0.69, 1.12) 0.89 (0.69, 1.14)

177 0.80 (0.61, 1.04) 0.82 (0.57, 1.16)

0.04 0.18

28 1.00 1.00

24 0.85 (0.46, 1.57) 0.91 (0.49, 1.72)

19 0.65 (0.34, 1.24) 0.76 (0.38, 1.49)

27 0.83 (0.41, 1.69) 1.21 (0.49, 2.99)

0.45 0.92

14 1.00 1.00

14 1.05 (0.48, 2.26) 1.11 (0.51, 2.45)

18 1.43 (0.71, 2.90) 1.62 (0.75, 3.50)

30 2.45 (1.11, 5.41) 3.32 (1.10, 10.00)

0.007 0.01

13 1.00 1.00

7 0.50 (0.19, 1.28) 0.45 (0.17, 1.19)

13 0.90 (0.40, 1.98) 0.74 (0.33, 1.69)

8 0.41 (0.16, 1.05) 0.24 (0.08, 0.70)

0.16 0.11

45 1.00 1.00

47 1.11 (0.72, 1.72) 1.02 (0.65, 1.58)

34 0.89 (0.56, 1.41) 0.74 (0.46, 1.19)

27 0.65 (0.40, 1.08) 0.39 (0.21, 0.73)

0.07 0.006

8 1.00 1.00

9 1.03 (0.37, 2.88) 1.17 (0.41, 3.34)

10 1.12 (0.40, 3.09) 1.40 (0.46, 4.26)

5 0.43 (0.11, 1.69) 0.79 (0.11, 5.73)

0.27 0.85

Daily milk intake was 쏝0.5 cup (쏝118 mL), 0.5– 0.8 cup (118 –188 mL), 쏜0.8 –쏝1.2 cups (쏜188 –282 mL), and 욷1.2 cups (쏜282 mL) in group 1, 2, 3, and 4, respectively. 1 cup milk ҃ 앒235 mL. All ORs were from logistic regression analysis. 2 Adjusted for age, sex, and energy and fruit intakes. 3 OR; 95% CI in parentheses (all such values). 4 Adjusted for age, sex, and energy, fruit, and calcium intakes. 1

showed that adult IGF-I concentrations are associated with premenopausal but not with postmenopausal breast cancer (12). We were not able to distinguish premenopausal from postmenopausal cases of breast cancer in our study population, and agespecific analyses were not possible because of the small numbers. Thus, a dilution of subtle associations with premenopausal breast cancer may have occurred. Possible mechanisms Although the biological mechanisms that may underlie our findings cannot be determined from our study, we propose that the associations between dairy intake in childhood and the risk of colorectal and prostate cancer in adulthood may have to do with the programming effects of early-life nutrition on the IGF system or other pathways. Our observation that a high intake of dairy foods in childhood is associated with a substantially greater risk of colorectal cancer in adulthood may indicate that it is the effect of childhood dairy intake on childhood (rather than on adult) concentrations of IGF-I that is the important mediator of future risk. This may have to do with the positive association between dairy intake and IGF-I concentrations in childhood (18), the inverse association between childhood dairy intake and adult IGF-I concentrations (20), and the positive link between adult IGF-I concentrations and colorectal cancer risk (11). An alternative biological mechanism may be early-life programming effects, possibly involving

the vitamin D pathway. The process of colonic formation of 1,25(OH)2D, which helps colon cells avoid hyperproliferation and prevents their progression into malignancy (48), is probably under epigenetic regulation (49). The early postnatal period is a period of physiologic plasticity, and diets in early childhood can affect disease risk in adulthood (50)— eg, through epigenetic imprinting of genes (51). There is some indirect evidence that epigenetic regulatory mechanisms in the gastrointestinal tract continue to develop in the postnatal period (52), although we are not aware of specific effects of dairy consumption on these processes. Our finding that the highest milk intake in childhood is associated with a lower risk of prostate cancer risk is contrary to findings in adult intake studies (23, 33) but is in keeping with the possible long-term programming effect of childhood nutrition on adult IGF-I (19, 21, 22) and with the inverse association between childhood dairy intake and adult IGF- I concentrations observed in the population of the present study (20). Recent genetic analyses have confirmed a role for IGF-I in prostate cancer risk by showing that inherited variations in IGF-I may play a role in the risk of prostate cancer (53). Study strengths and limitations This study has several strengths. First, diet was measured in childhood long before the occurrence of disease, which averted the problem of recall bias that is encountered in studies based on

1728

VAN DER

remembering childhood diet (28, 30, 54). Second, all foods consumed in the home were assessed, which allowed the consideration of other dietary factors— eg, fruit and vegetable intakes—as potential confounders of the relation between dairy intake and cancer risk. Third, our estimates of family milk consumption were reliable: in a study of 195 families who kept 2 inventories between 3 and 15 mo apart, repeatability was high for milk intake (intraclass correlation coefficient: 0.85) (55). It is important, however, to interpret these results in light of the study limitations. First, childhood diet was based on household rather than individual consumption. Nevertheless, misclassification resulting from imprecise diet measurements would likely have been nonsystematic, attenuating rather than explaining the associations observed. Moreover, our previous finding in this population—that energy intake in childhood was positively associated with cancer risk (43)—was consistent with other evidence (56) and thus provided support for the reliability of our dietary estimates. We expect that, more so than other foods, dairy products were consumed by the children of the families in the Boyd Orr Study; thus, we have greater confidence in attributing this exposure to our cohort members. The analyses are based on the traced participants only. Study participants who were excluded from the present analysis were older, were more likely to have lower SES, and had higher intakes of fat and total energy in childhood than did those who were included (P 쏝 0.05, data not shown). However, it is unlikely that the small group of excluded participants (13% of the original cohort) would have substantially altered the estimates presented here. Because repeat assessments of diet during the life course are not available for most of the participants, we are unable to discern whether some of these associations are confounded by dietary habits in adulthood, which may have been correlated with childhood diet (57), but which could not be accounted for in these analyses. Data from the 1997 follow-up study in a subgroup of the study participants showed that the correlation between milk intake in childhood and that in adulthood was very low (Pearson correlation coefficient: 0.07), and thus these results probably cannot be explained by similar dairy consumption patterns in adulthood, and they likely point to a childhood-specific effect. The long-term and probably subtle effects of childhood diet on adult cancer risk may have been negated by the perhaps much stronger effects of the longer-term adult dietary habits. The strongest associations with cancer risk in this study are seen with total dairy and milk intakes in the highest intake groups. The 2 highest groups of milk intake equated to a median milk intake of 1 cup/d and 1.9 cups/d, respectively. These milk consumption levels are similar to current estimates for US children [average consumption: 앒1.6 servings milk/d (58)]. Thus, the associations in our data occur at intakes that are similar to current “normal ” intakes. Conclusions In conclusion, we found some evidence of a greater risk of colorectal cancer and weak evidence of a lower risk of prostate cancer associated with a family diet rich in dairy foods during childhood. There was no evidence for an independent association between adult cancer risk and a family diet rich in calcium during childhood. These results are in the opposite direction of associations between adult dairy consumption and the risks of colorectal and prostate cancer. The possibility that early-life programming effects are underlying these associations cannot be

POLS ET AL

confirmed from our study, but that hypothesis warrants further investigation. Dairy products are important contributors to children’s intake of protein, vitamins, and minerals, and they play an important role in the maintenance of bone health. More evidence from studies including a more complete life-course assessment of dairy intake and related dietary and lifestyle factors is needed before any firm conclusions can be drawn. We acknowledge Peter Morgan, director of The Rowett Research Institute, for the use of the archive and in particular Walter Duncan, honorary archivist to the Rowett Institute, for providing access to and checking of the study records. We acknowledge the staff at the National Health Service Central Register at Southport and Edinburgh and John Pemberton for providing information on the conduct of the original survey. We acknowledge all of the participants and research workers in the original survey of 1937 through 1939. Clare Frobisher, Pauline Emmett, and Maria Maynard undertook the re-analyses of the childhood household diet diaries. The authors’ responsibilities were as follows—JCvdP: the data analysis and writing of the manuscript; CB: preparation of the food diaries for data analysis and contributions to the analysis plan and writing of the manuscript; DG and RMM: study design, analysis plan, and writing of the paper; GDS: establishment of the Boyd Orr cohort and writing of the manuscript; and CF: preparation of the food diaries data for analysis and contributions to writing of the manuscript. None of the authors had any personal or financial conflict of interest.

REFERENCES 1. Voskuil DW, Vrieling A, van’t Veer LJ, Kampman E, Rookus MA. The insulin-like growth factor system in cancer prevention: potential of dietary intervention strategies. Cancer Epidemiol Biomarkers Prev 2005; 14:195–203. 2. Hoppe C, Udam TR, Lauritzen L, Molgaard C, Juul A, Michaelsen KF. Animal protein intake, serum insulin-like growth factor I, and growth in healthy 2.5-y-old Danish children. Am J Clin Nutr 2004;80:447–52. 3. Tseng M, Breslow RA, Graubard BI, Ziegler RG. Dairy, calcium, and vitamin D intakes and prostate cancer risk in the National Health and Nutrition Examination Epidemiologic Follow-up Study cohort. Am J Clin Nutr 2005;81:1147–54. 4. Giovannucci E. Dietary influences of 1,25(OH)2 vitamin D in relation to prostate cancer: a hypothesis. Cancer Causes Control 1998;9:567– 82. 5. Larsson SC, Bergkvist L, Wolk A. High-fat dairy food and conjugated linoleic acid intakes in relation to colorectal cancer incidence in the Swedish Mammography Cohort. Am J Clin Nutr 2005;82:894 –900. 6. Belury MA. Inhibition of carcinogenesis by conjugated linoleic acid: potential mechanisms of action. J Nutr 2002;132:2995– 8. 7. Li Y, Millikan RC, Bell DA, et al. Polychlorinated biphenyls, cytochrome P450 1A1 (CYP1A1) polymorphisms, and breast cancer risk among African American women and white women in North Carolina: a population-based case-control study. Breast Cancer Res 2005;7:R12– 8. 8. Laden F, Ishibe N, Hankinson SE, et al. Polychlorinated biphenyls, cytochrome P450 1A1, and breast cancer risk in the Nurses’ Health Study. Cancer Epidemiol Biomarkers Prev 2002;11:1560 –5. 9. Schaum J, Schuda L, Wu C, Sears R, Ferrario J, Andrews K. A national survey of persistent, bioaccumulative, and toxic (PBT) pollutants in the United States milk supply. J Expo Anal Environ Epidemiol 2003;13: 177– 86. 10. Khandwala HM, McCutcheon IE, Flyvbjerg A, Friend KE. The effects of insulin-like growth factors on tumorigenesis and neoplastic growth. Endocr Rev 2000;21:215– 44. 11. Renehan AG, Zwahlen M, Minder C, O’Dwyer ST, Shalet SM, Egger M. Insulin-like growth factor (IGF)-I, IGF binding protein-3, and cancer risk: systematic review and meta-regression analysis. Lancet 2004;363: 1346 –53. 12. Renehan AG, Harvie M, Howell A. Insulin-like growth factor (IGF)-I, IGF binding protein-3, and breast cancer risk: eight years on. Endocr Relat Cancer 2006;13:273– 8. 13. Morimoto LM, Newcomb PA, White E, Bigler J, Potter JD. Variation in plasma insulin-like growth factor-1 and insulin-like growth factor binding protein-3: personal and lifestyle factors (United States). Cancer Causes Control 2005;16:917–27.

CHILDHOOD DAIRY INTAKE AND ADULT CANCER RISK 14. Holmes MD, Pollak MN, Willett WC, Hankinson SE. Dietary correlates of plasma insulin-like growth factor I and insulin-like growth factor binding protein 3 concentrations. Cancer Epidemiol Biomarkers Prev 2002;11:852– 61. 15. Giovannucci E, Pollak M, Liu Y, et al. Nutritional predictors of insulinlike growth factor I and their relationships to cancer in men. Cancer Epidemiol Biomarkers Prev 2003;12:84 –9. 16. Gunnell D, Oliver SE, Peters TJ, et al. Are diet-prostate cancer associations mediated by the IGF axis? A cross-sectional analysis of diet, IGF-I and IGFBP-3 in healthy middle-aged men. Br J Cancer 2003;88:1682– 6. 17. Heaney RP, McCarron DA, Dawson-Hughes B, et al. Dietary changes favorably affect bone remodeling in older adults. J Am Diet Assoc 1999;99:1228 –33. 18. Rogers I, Emmett P, Gunnell D, Dunger D, Holly J. Milk as a food for growth? The insulin-like growth factors link. Public Health Nutr 2006; 9:359 – 68. 19. Ben-Shlomo Y, Holly J, McCarthy A, Savage P, Davies D, Davey Smith G. Prenatal and postnatal milk supplementation and adult insulin-like growth factor I: long-term follow-up of a randomized controlled trial. Cancer Epidemiol Biomarkers Prev 2005;14:1336 –9. 20. Martin RM, Holly JMP, Middleton N, Davey Smith G, Gunnell D. Childhood diet and insulin-like growth factors in adulthood: 65-year follow-up of the Boyd Orr Cohort. Eur J Clin Nutr 2007 Feb 7 [Epub ahead of print]. 21. Symonds ME, Budge H, Stephenson T, Gardner DS. Experimental evidence for long-term programming effects of early diet. Adv Exp Med Biol 2005;569:24 –32. 22. Martin RM, Holly JM, Smith GD, et al. Could associations between breastfeeding and insulin-like growth factors underlie associations of breastfeeding with adult chronic disease? The Avon Longitudinal Study of Parents and Children. Clin Endocrinol (Oxf) 2005;62:728 –37. 23. Gao X, LaValley MP, Tucker KL. Prospective studies of dairy product and calcium intakes and prostate cancer risk: a meta-analysis. J Natl Cancer Inst 2005;97:1768 –77. 24. Giovannucci E, Liu Y, Stampfer MJ, Willett WC. A prospective study of calcium intake and incident and fatal prostate cancer. Cancer Epidemiol Biomarkers Prev 2006;15:203–10. 25. Peterlik M, Cross HS. Vitamin D and calcium deficits predispose for multiple chronic diseases. Eur J Clin Invest 2005;35:290 –304. 26. Jacobs ET, Haussler MR, Martinez ME. Vitamin D activity and colorectal neoplasia: a pathway approach to epidemiologic studies. Cancer Epidemiol Biomarkers Prev 2005;14:2061–3. 27. Moorman PG, Terry PD. Consumption of dairy products and the risk of breast cancer: a review of the literature. Am J Clin Nutr 2004;80:5–14. 28. Hjartaker A, Laake P, Lund E. Childhood and adult milk consumption and risk of premenopausal breast cancer in a cohort of 48 844 women— the Norwegian Women and Cancer Study. Int J Cancer 2001;93:888 –93. 29. Michels KB, Rosner BA, Chumlea WC, Colditz GA, Willett WC. Preschool diet and adult risk of breast cancer. Int J Cancer 2006;118:749 – 54. 30. Pryor M, Slattery ML, Robison LM, Egger M. Adolescent diet and breast cancer in Utah. Cancer Res 1989;49:2161–7. 31. Shin MH, Holmes MD, Hankinson SE, Wu K, Colditz GA, Willett WC. Intake of dairy products, calcium, and vitamin d and risk of breast cancer. J Natl Cancer Inst 2002;94:1301–11. 32. Diorio C, Berube S, Byrne C, et al. Influence of insulin-like growth factors on the strength of the relation of vitamin D and calcium intakes to mammographic breast density. Cancer Res 2006;66:588 –97. 33. Gao X, La Valley M, Tucker KL. Response to Re: prospective studies of dairy product and calcium intakes and prostate cancer risk: a metaanalysis. J Natl Cancer Inst 2006;98:795 (letter). 34. Kesse E, Boutron-Ruault MC, Norat T, Riboli E, Clavel-Chapelon F. Dietary calcium, phosphorus, vitamin D, dairy products and the risk of colorectal adenoma and cancer among French women of the E3N-EPIC prospective study. Int J Cancer 2005;117:137– 44. 35. Larsson SC, Bergkvist L, Rutegard J, Giovannucci E, Wolk A. Calcium

36. 37. 38.

39. 40. 41. 42. 43. 44. 45. 46. 47. 48. 49. 50. 51. 52. 53. 54. 55. 56. 57. 58.

1729

and dairy food intakes are inversely associated with colorectal cancer risk in the Cohort of Swedish Men. Am J Clin Nutr 2006;83:667–73. Cho E, Smith-Warner SA, Spiegelman D, et al. Dairy foods, calcium, and colorectal cancer: a pooled analysis of 10 cohort studies. J Natl Cancer Inst 2004;96:1015–22. Norat T, Riboli E. Dairy products and colorectal cancer. A review of possible mechanisms and epidemiological evidence. Eur J Clin Nutr 2003;57:1–17. Gunnell DJ, Frankel S, Nanchahal K, Braddon FE, Smith GD. Lifecourse exposure and later disease: a follow-up study based on a survey of family diet and health in pre-war Britain (1937–1939). Public Health 1996;110:85–94. Rowett Research Institute. Family diet and health in pre-war Britain. Edinburgh, United Kingdom: Carnegie United Kingdom Trust, 1955. Maynard M, Gunnell D, Emmett P, Frankel S, Davey Smith G. Fruit, vegetables, and antioxidants in childhood and risk of adult cancer: the Boyd Orr cohort. J Epidemiol Community Health 2003;57:218 –25. Price G, Paul A, Key F, et al. Measurement of diet in a large national survey: comparison of computerised and manual coding of records in household measures. J Hum Nutr Diet 1995;8:417–28. The Royal Society of Chemistry and Ministry of Agriculture, Fisheries, and Food (MAFF). McCance and Widdowson’s the composition of foods. London: HMSO, 1991. Frankel S, Gunnell DJ, Peters TJ, Maynard M, Davey Smith G. Childhood energy intake and adult mortality from cancer: the Boyd Orr Cohort Study. BMJ 1998;316:499 –504. Townsend P, Phillimore P, Beattie A. Indicator of health and deprivation: inequality in the north. London, United Kingdom: Croom Helm, 1988:30 – 40. Shin A, Shrubsole MJ, Ness RM, et al. Meat and meat-mutagen intake, doneness preference and the risk of colorectal polyps: the Tennessee Colorectal Polyp Study. Int J Cancer 2007;121:136 – 42. Smith GD, Gunnell D, Holly J. Cancer and insulin-like growth factor-I. A potential mechanism linking the environment with cancer risk. BMJ 2000;321:847– 8. Willett WC. Nutritional epidemiology. 2nd ed. New York, NY: Oxford University Press, 1998. Cross HS. Commentary: from epidemiology to molecular biology— vitamin D and colorectal cancer prevention. Int J Epidemiol 2006;35: 225–7. Cross HS, Lipkin M, Kallay E. Nutrients regulate the colonic vitamin D system in mice: relevance for human colon malignancy. J Nutr 2006; 136:561– 4. Buckley AJ, Jaquiery AL, Harding JE. Nutritional programming of adult disease. Cell Tissue Res 2005;322:73–9. Waterland RA, Lin JR, Smith CA, Jirtle RL. Post-weaning diet affects genomic imprinting at the insulin-like growth factor 2 (Igf2) locus. Hum Mol Genet 2006;15:705–16. Waterland RA. Epigenetic mechanisms and gastrointestinal development. J Pediatr 2006;149:S137– 42. Cheng I, Stram DO, Penney KL, et al. Common genetic variation in IGF1 and prostate cancer risk in the Multiethnic Cohort. J Natl Cancer Inst 2006;98:123–34. Potischman N, Weiss HA, Swanson CA, et al. Diet during adolescence and risk of breast cancer among young women. J Natl Cancer Inst 1998;90:226 –33. Frobisher C, Tilling K, Emmett PM, et al. Reproducibility measures and their effect on diet-cancer associations in the Boyd Orr Cohort. J Epidemiol Commun Health 2007;61:434 – 40. Ross MH, Bras G. Lasting influence of early caloric restriction on prevalence of neoplasms in the rat. J Natl Cancer Inst 1971;47:1095–113. Lake AA, Mathers JC, Rugg-Gunn AJ, Adamson AJ. Longitudinal change in food habits between adolescence (11–12 years) and adulthood (32–33 years): the ASH30 Study. J Public Health (Oxf) 2006;28:10 – 6. Fiorito LM, Mitchell DC, Smiciklas-Wright H, Birch LL. Dairy and dairy-related nutrient intake during middle childhood. J Am Diet Assoc 2006;106:534 – 42.

Association between dietary fiber and endometrial cancer: a dose-response meta-analysis1–3 Elisa V Bandera, Lawrence H Kushi, Dirk F Moore, Dina M Gifkins, and Marjorie L McCullough ABSTRACT Background: Endometrial cancer is the most common female gynecologic cancer in the United States. Excessive and prolonged exposure of the endometrium to estrogens unopposed by progesterone and a high body mass are well-established risk factors for endometrial cancer. Although dietary fiber has been shown to beneficially reduce estrogen concentrations and prevent obesity, its role in endometrial cancer has received relatively little attention. Objective: The objective was to summarize and quantify the current evidence of a role of dietary fiber consumption in endometrial cancer risk and to identify research gaps in this field. Design: We conducted a systematic literature review of articles published through February 2007 to summarize the current evidence of a relation between dietary fiber consumption and endometrial cancer risk and to quantify the magnitude of the association by conducting a dose-response meta-analysis. Results: Ten articles representing 1 case-cohort study and 9 casecontrol studies that evaluated several aspects of fiber consumption and endometrial cancer risk were identified through searches in various databases. On the basis of 7 case-control studies, the random-effects summary risk estimate was 0.82 (95% CI: 0.75, 0.90) per 5 g/1000 kcal dietary fiber, with no evidence of heterogeneity (I2: 0%, P for heterogeneity: 0.55). The random-effects summary estimate was 0.71 (95% CI: 0.59, 0.85) for the comparison of the highest with the lowest dietary fiber intake in 8 case-control studies, with little evidence of heterogeneity (I2: 20.8%, P for heterogeneity: 0.26). In contrast, the only prospective study that evaluated this association did not find an association. Conclusions: Although the current evidence, based on data from case-control studies, supports an inverse association between dietary fiber and endometrial cancer, additional population-based studies, particularly cohort studies, are needed before definitive conclusions can be drawn. Am J Clin Nutr 2007;86:1730 –7. KEY WORDS Endometrial carcinoma, diet, fiber, metaanalysis, systematic literature review

INTRODUCTION

Endometrial cancer is the most common female gynecologic cancer in the United States, ranking fourth among all cancers in women in age-adjusted incidence (1). Excess body mass and excessive and prolonged exposure to estrogens unopposed by progesterone are well-established and strong risk factors for endometrial cancer (2). Dietary fiber has been shown to influence estrogen absorption, metabolism, and bioavailability (3) and

1730

weight loss (4). Thus, it is reasonable to expect an inverse association of dietary fiber intake with endometrial cancer. However, surprisingly, the relation between fiber intake and endometrial cancer risk has received relatively little attention. In support of the Second World Cancer Research Fund (WCRF)/American Institute for Cancer Research (AICR) Report on Food, Nutrition, Physical Activity and the Prevention of Cancer (Internet: www.wcrf.org), and commissioned by the WCRF, we conducted a systematic and comprehensive literature review of diet, nutrition, physical activity, and endometrial cancer (5) to enhance and update the previous 1997 report (6). In the 1997 report, the possible role of fiber on endometrial cancer was not mentioned. In fact, to our knowledge, this is the first systematic literature review and meta-analysis on this topic. The objective of this article was to summarize and quantify the current evidence for a role of dietary fiber consumption on endometrial cancer risk and to identify research gaps in this field.

METHODS

In general, we followed the WCRF Specification Manual, available online at www.wcrf.org. The methods used in this article diverge from the WCRF instructions in that we followed our own criteria for inclusion of studies and used our own methods for data tabulation and analysis and interpretation of the evidence. Although WCRF required the inclusion of all studies regardless of quality, this systematic review and meta-analysis was limited to case-control and cohort studies. Randomized trials would have been included, but none were found. Ecologic and cross-sectional studies were excluded. For the purposes of this 1

From The Cancer Institute of New Jersey, Robert Wood Johnson Medical School, New Brunswick, NJ (EVB and DMG); the Division of Research, Kaiser Permanente, Oakland, CA (LHK); the School of Public Health, University of Medicine and Dentistry of New Jersey, Piscataway, NJ (EVB, DFM, and DMG); and Epidemiology and Surveillance Research, American Cancer Society, Atlanta, GA (MLM). 2 Supported by World Cancer Research Fund International and the National Cancer Institute (NIH-K07 CA095666 to EVB). However, interpretation of the evidence may not represent the views of WCRF or the NCI, and our conclusions may differ from those in the 2007 WCRF report summarizing evidence related to food, nutrition, physical activity, and cancer risk. 3 Reprints not available. Address correspondence to EV Bandera, The Cancer Institute of New Jersey, 195 Little Albany Street, 5568; New Brunswick, NJ 08903. E-mail: [email protected]. Received April 9, 2007. Accepted for publication July 30, 2007.

Am J Clin Nutr 2007;86:1730 –7. Printed in USA. © 2007 American Society for Nutrition

DIETARY FIBER AND ENDOMETRIAL CANCER RISK

study, we conducted additional analyses, such as sensitivity analyses to evaluate the impact of excluding studies that did not meet certain a priori criteria (ie, population-based studies with 쏜200 cases, known hysterectomy status among controls, and adjustment for total energy intake and body mass). Interpretation of the evidence does not necessarily represent the views of WCRF. Search strategy Searches were conducted in July 2003, October 2004, and December 2005. The databases searched included Medline, ISI Web, Embase, Biosis, Ingenta, CINAHL, Science Direct, LILACS, Pascal, ExtraMed, and Allied CompMEd. On the basis of results from the 2003 searches, we excluded databases that did not produce any new results in subsequent searches. Bibliographic searches were complemented with manual searches of references in published articles. Translations were provided by WCRF when necessary. For the purposes of this study we also monitored the literature using PubMed Alerts for all new papers on endometrial cancer from January 2006 through February 2007. Exposure terms for PubMed were provided by the WCRF and can be found in the Specification Manual and in the Appendix of another article (7). General terms included diet[tiab] OR diets[tiab] OR dietetic[tiab] OR dietary[tiab] OR eating[tiab] OR intake[tiab] OR nutrient*[tiab] OR nutrition[tiab] OR vegetarian*[tiab] OR vegan*[tiab] OR “seventh day adventist”[tiab] OR macrobiotic[tiab] OR food and beverages[MeSH Terms]. Specifically for dietary fiber, terms included fiber[tiab] OR fiber[tiab] OR polysaccharide*[tiab]. Per WCRF instructions, our searches included endometrial hyperplasia because this includes precancerous lesions. However, we only found a few papers evaluating the role of diet and nutrition on endometrial hyperplasia, and none of them evaluated dietary fiber. Outcomes terms included: 1) endometrial neoplasm [MeSH]; 2) malign* [tiab] OR cancer*[tiab] OR carcinoma*[tiab] OR tumor*[tiab] OR tumor*[tiab]; 3) endometr* [tiab] OR corpus uteri [tiab] OR uterine [tiab]; 4): #2 AND #3; and 5): #3 AND hyperplasia [tiab]. Article selection and data extraction Overall search results and the criteria for selection were described elsewhere (8). In brief, citations identified from these searches were reviewed independently by 2 of us (LHK and EVB) for relevance. For the overall systematic literature review, we included peer-reviewed manuscripts containing original data in any language that evaluated the relation between any aspect of diet, nutrition, physical activity (ie, what we classified as “relevant exposures” for the overall systematic literature review) and endometrial cancer or endometrial hyperplasia published through June 2006. Studies that evaluated the influence of these factors on survival after cancer were excluded. Following WCRF instructions, no article was excluded on the basis of quality, except when the data provided were not useful (eg, only P values were presented for a given relevant factor). For the overall review we identified 522 possible relevant citations. Five articles could not be located but they did not appear to contain primary data. Of the remaining 517 articles, we included 285 and excluded 232. Reasons for excluding manuscripts after a full review included the following: published only as an abstract (n ҃ 5), case series (n ҃ 76), no relevant exposure data (n ҃ 83), no relevant outcome (n ҃ 12), survival data only (n ҃ 1), combined outcomes

1731

(eg, hyperplasia and cancer) (n ҃ 3), non-peer reviewed publication (n ҃ 9), reviews (n ҃ 39), and no useful data, such as only P values (n ҃ 4). Of the 285 included papers, 9 mentioned dietary fiber (9 –17) and all were written in English. Through monitoring the endometrial cancer literature through February 2007 using PubMed Alerts for “endometrial cancer,” one additional article was identified (18). Data on study characteristics and results were extracted by trained research personnel using an Access program developed by Leeds University under WCRF sponsorship. Each entry was reviewed by at least one of us. Assessment of study quality and sensitivity analyses We decided not to grade individual articles according to a quality score for 2 reasons: 1) no widely accepted tool for quality assessment of epidemiologic studies is available, and 2) we had few articles in which dietary fiber was evaluated, which limited these types of analyses. Instead, we present evidence from all case-control studies that examined fiber intake and then repeated certain analyses, excluding studies that did not meet certain a priori quality criteria. In the forest plots, studies excluded are marked and the specific reasons for exclusion are indicated in the footnotes. The a priori quality criteria were as follows: 1) population-based studies as the appropriateness of hospital controls in diet and cancer studies is controversial, 2) sample size of 욷200 cases for more optimal statistical power, 3) exclusion of hysterectomies from the control group, and 4) adjustment for important confounders such as total energy and body mass. Statistical analysis To conduct dose-response meta-analyses, published results were transformed into a common scale. For example, results related to dietary fiber were reported as units g/d or g 䡠 1000 kcalҀ1 䡠 dҀ1. The average daily energy intake from study to study varied considerably, and much of this variance was probably attributable to dietary assessment methods rather than to true differences in food intake. For example, the median daily energy intake in the study by Potischman et al (11) was 1248 kcal, whereas in the study by McCann et al (13) it was 2102 kcal. Both of these studies were conducted in the United States, which made it unlikely that cultural or lifestyle differences were a major contributor to these differences in reported intake. Thus, for dose-response meta-analyses, we elected to convert reported intakes into nutrient density measures expressed as g/1000 kcal. To convert reported category intakes or medians into nutrient density measures, it was necessary to have a reported value for median (or mean) kcal/d intake for the study population; this was available in all studies, except for one hospital-based study (14). For studies reporting only categorical analyses, an estimate of mean intake for each category was computed following the methodology developed by Cheˆne and Thompson (19). The iterative method described in Greenland and Longnecker (20) was used to estimate a single logistic regression parameter per study. This method imputes expected numbers of cases and controls (or cases for a prospective study) and computes the logistic regression slope parameter (which may be interpreted as the log relative risk) and SE. Finally, we estimated fixed-effects and randomeffects pooled logistic regression coefficients across studies by study design. We used the random-effects models in forest plots

1732

BANDERA ET AL

and for interpretation of the evidence, because it uses a combination of within-study variance and between-study variance for computing weights. The Cheˆne and Thompson (19) and Greenland and Longnecker (20) algorithms described above were implemented in the statistical language R (R: A Language and Environment for Statistical Computing, version 2.4.1, 2006; R Development Core Team, R Foundation for Statistical Computing, Vienna, Austria). Fixed- and random-effects pooled estimates and forest plots were produced by using the “metan” package of STATA/SE version 8.2 (StataCorp LP, College Station, TX). Heterogeneity was assessed by conducting Q tests (ie, testing for the presence or absence of heterogeneity) and quantifying the degree of heterogeneity by estimating the I2 index (21).

RESULTS

Our systematic literature review included 10 articles representing 1 cohort study and 9 case-control studies that evaluated various aspects of fiber intake (9 –18). In Table 1, the characteristics of these studies and the variables evaluated are listed. As shown in the table, these studies were conducted in several countries and varied considerably concerning the quality of dietary assessment and in the evaluation and classification of dietary fiber. The case-control study by Salazar-Martinez et al (14) was not included in the dose-response meta-analyses because it did not present an estimate of total energy intake; therefore, transformation into a nutrient density measure was not possible.

As shown in Table 2, 7 (10 –13, 15, 16, 18) of the 8 casecontrol studies that evaluated dietary fiber and endometrial cancer risk suggested an inverse relation, with odds ratios ranging from 0.5 to 0.71 for comparisons of the highest with the lowest category of intake. In contrast, a small hospital-based casecontrol study in Mexico (14) and a case-cohort study in Canada (9) reported risk estimates 쏜1, but the CIs included 1. Other fiber-related exposures examined in case-control studies of endometrial cancer included crude fiber (12, 17), soluble and insoluble fiber (9), cellulose and noncellulosic polysaccharides (12), and fiber categorized by food group: cereal or grain fiber, fruit fiber, and vegetable fiber (9, 12, 15). Jain et al (9) examined the role of soluble and insoluble fiber in a case-cohort study and found limited evidence of an association. Insoluble fiber was also unrelated to endometrial cancer risk in a population-based case-control study conducted by the same group (15). Goodman et al (12) conducted the most thorough evaluation of fiber subtypes and found similar results for cellulose (adjusted OR: 0.59; 95% CI: 0.32, 1.07) and noncellulosic polysaccharides (OR: 0.62; 95% CI: 0.34, 1.15). Other results for fiber subtypes are shown in Table 3. Two studies evaluated crude fiber, one study found a suggestion of an inverse association (12), whereas no evidence of an association was found in the other study (17). Fiber intake by food source was evaluated in 4 studies, with inconsistent results. Goodman et al (12). found a stronger relation for fruit and cereal fiber intakes than for vegetable fiber. In contrast, Jain et al (15) found a relation only for vegetable fiber

TABLE 1 Characteristics of observational studies that evaluated dietary fiber and endometrial cancer risk1 Reference

Country

Cases/controls or cohort size

Age

Dietary assessment

Time frame of dietary assessment

Variables evaluated

Case-cohort study Jain et al, 2000 (9)

Canada

221/3697

40–59 y

FFQ (86 items)

1 mo prior

Dietary fiber, insoluble fiber, soluble fiber, cereal fiber, fruit fiber, vegetable fiber

Population-based case-control studies Shu et al, 1993 (17) Potischman et al, 1993 (11)

China United States

268/268 399/296

18–74 y 20–74 y

10 y Past few years

Crude fiber Fiber

Goodman et al, 1997 (12)

United States

332/511

18–84 y

FFQ (63 items) FFQ (Block, 60 items) Dietary history (250 items)

1y

Jain et al, 2000 (15)

Canada

552/562

30–79 y

Dietary history (unknown items)

1y

McCann et al, 2000 (13) Littman et al, 2001 (16)

United States United States

232/639 679/944

40–85 y 45–74 y

2y 5y

China

1204/1212

30–69 y

FFQ (172 items) FFQ (modified Block, 98 items) FFQ (71 foods)

Crude fiber, nonstarch polysaccharides, cellulose, noncellulosic polysaccharides, dietary fiber, cereal fiber, vegetable fiber, fruit fiber Dietary fiber, insoluble fiber, cereal fiber, fruit fiber, vegetable fiber Dietary fiber Total fiber

5y

Dietary fiber

United States

103/236

1y

Dietary fiber

Mexico

85/629

18–81 y

FFQ (Willett, 116 items) FFQ (116 items)

1y

Dietary fiber

Xu et al, 2007 (18) Hospital-based case-control studies Barbone et al, 1993 (10) Salazar-Martinez et al, 2005 (14) 1

FFQ, food-frequency questionnaire.

United States Mexico 18–81 y

30–69 y

103/236 85/629

1204/1212

232/639 679/944

399/296 332/511 552/562

221/3697

Cases/controls or total cohort

Dietary fiber Dietary fiber

Dietary fiber

Dietary fiber Total fiber

Dietary fiber Dietary fiber Dietary fiber

Dietary fiber

Exposure evaluated

쏜21.5 vs 쏝16.6 g/d 쏜24 vs 쏝13 g/d

0.7 0.47 0.71 0.89 0.5 0.68

쏜13.6 vs 쏝7.7 g/d 쏜23.9 vs 쏝12.3 g/d 쏜27.5 vs 쏝17.2 g/d Per 10 g 쏜32 vs 쏝20 g/d 쏜10.7 vs 쏝5.6 g 䡠 1000 kcalҀ1 䡠 dҀ1 8.1 vs 4.8 g 䡠 1000 kcalҀ1 䡠 dҀ1 0.6 1.46

0.8

1.24

RR/OR

쏜23.2 vs 쏝15.1 g/d

Contrast

(0.3, 1.1) (0.76, 2.79)

(0.6, 1.0)

(0.4, 1.3) (0.25, 0.86) (0.49, 1.03) (0.75, 1.05) (0.3, 1) (0.47, 0.99)

(0.82, 1.87)

95% CI

0.43 0.33

0.002 0.16

0.02 0.14

NS

P for trend

1 1

1

1 (1) 1 1 1 1

1

A

1 1

1

1 1 1 1 1 1

1

B

1 1

1

–3 1 1 1 1 1

1

E

1

1 1 1 1

1

1

S

1

1 1 1 1 1 1

1

H

Covariates considered2

3 1

1

1 1 2 2 3

2

R

2

1

OR, odds ratio; RR, relative risk; NS, P 쏜 0.05. The numbers in each column refer to the number of covariates adjusted for under that grouping. A, age; B, BMI/weight; E, total energy; S, smoking; H, hormone replacement therapy or estrogen replacement therapy; R, reproductive factors; (1), matched for age. 3 Adjusted for noncarbohydrate calories.

Hospital-based case-control studies Barbone et al, 1993 (10) Salazar-Martinez et al, 2005 (14)

China

Xu et al, 2007 (18)

40–85 y 45–74 y

20–74 y 18–84 y 30–79 y

United States United States Canada

United States United States

40–59 y

Age

Canada

Country

McCann et al, 2000 (13) Littman et al, 2001 (16)

Case-cohort study Jain et al, 2000 (9) Population-based case-control studies Potischman et al, 1993 (11) Goodman et al, 1997 (12) Jain et al, 2000 (15)

Reference

TABLE 2 Studies that evaluated fiber intake and endometrial cancer risk1

DIETARY FIBER AND ENDOMETRIAL CANCER RISK

1733

40–59 y 18–84 y 30–79 y

Canada United States Canada

Canada United States Canada

Fruit fiber Jain et al, 2000 (9) Goodman et al, 1997 (12) Jain et al, 2000 (15) 221/3697 332/511 552/562

221/3697 332/511 552/562

221/3697 332/511 552/562

268/268 332/511

Case cohort Population-based CC Population based CC

Case cohort Population-based CC Population based CC

Case cohort Population-based CC Population based CC

Population-based CC Population-based CC

Type of study

Fruit fiber Fruit fiber Fruit fiber

Vegetable fiber Vegetable fiber Vegetable fiber

Cereal fiber Cereal fiber Cereal fiber

Crude fiber Crude fiber

Exposure

0.9 (0.61, 1.34) 0.71 (0.4, 1.23) 0.64 (0.44, 0.91) 0.87 (0.75, 1.01)

쏜9.5 vs 쏝5.5 g/d 쏜6.25 vs 쏝2.94 g/d 쏜12.8 vs 쏝6.6 g/d Per 6.2 g

1.08 (0.73, 1.61) 0.54 (0.32, 0.92) 1.34 (0.92, 1.95) 1.0 (0.86, 1.18)

1.07 (0.73, 1.58) 0.55 (0.33, 0.92) 1.03 (0.72, 1.47)

쏜4.8 vs 쏝2.7 g/d 쏜2.51 vs 쏝0.01 g/d 쏜10.5 vs 쏝4.8 g/d

쏜5.7 vs 쏝2.2 g/d 쏜6.27 vs 쏝2.21 g/d 쏜8.9 vs 쏝3.4 g/d Per 5.5 g

1.1 0.6 (0.33, 1.09)

쏜4.68 vs 쏝2.65 g/d 쏜6.04 vs 쏝3.01 g/d

Contrast

RR/OR (95% CI)

0.22 0.31

0.22 0.02

0.001 0.83

0.7 0.31

P for trend

1 (1) 1 1

1 (1) 1 1

1 (1) 1

1 (1)

A

1 1 1 1

1 1 1 1

1 1 1

1 1

B

1 1 1 1

1 1 1 1

1 1 1

1 1

E

1 1

1

1 1

1

1

1

S

1 1 1 1

1 1 1 1

1 1 1

1

H

Covariates considered2

2 1 2 2

2 1 2 2

2 1 2

1 1

R

2

OR, odds ratio; RR, relative risk; CC, case-control. The numbers in each column refer to the number of covariates adjusted for under that grouping. A, age; B, BMI/weight; E, total energy; S, smoking; H, hormone replacement therapy or estrogen replacement therapy; R, reproductive factors; (1), matched for age.

1

40–59 y 18–84 y 30–79 y

Canada United States Canada

40–59 y 18–84 y 30–79 y

18–74 y 18–84 y

Age

China United States

Country

Crude fiber Shu et al, 1993 (17) Goodman et al, 1997 (12) Cereal fiber Jain et al, 2000 (9) Goodman et al, 1997 (12) Jain et al, 2000 (15) Vegetable fiber Jain et al, 2000 (9) Goodman et al, 1997 (12) Jain et al, 2000 (15)

Reference

Cases/controls or total cohort

TABLE 3 Studies that evaluated fiber subtypes and endometrial cancer risk1

1734 BANDERA ET AL

1735

DIETARY FIBER AND ENDOMETRIAL CANCER RISK

(Table 3). Littman et al (16) found an association only from fiber from fruit and vegetables combined (OR: 0.67; 95% CI: 0.47, 0.96), whereas there was no association with fiber from grains or fiber from beans (data not shown). As shown in Table 3, the case-cohort study (9) failed to find an association for any of these fiber subtypes. Meta-analysis The data only allowed for a meta-analysis of total dietary fiber, because there were only a few studies that evaluated fiber subtypes, and the results tended to be inconsistent. As shown in Figure 1, 7 case-control studies were included in the doseresponse meta-analyses (10 –13, 15, 16, 18). A continuous relative risk for the same increment was also computed for the casecohort study for comparison purposes. The study by SalazarMartinez et al (14) was not included in the dose-response metaanalyses because it did not present an estimate of total energy intake and, therefore, transformation into a nutrient density measure was not possible. There was no evidence of heterogeneity among case-control studies (I2: 0.0%, P for heterogeneity: 0.55). On the basis of data from case-control studies, we estimated an 18% reduction in endometrial cancer risk per 5 g/1000 kcal total fiber intake (random- and fixed-effects pooled OR: 0.82; 95% CI: 0.75, 0.90). In contrast, the case-cohort study showed no association, with a derived RR of 1.15 (95% CI: 0.89, 1.49) per 5 g/1000 kcal fiber intake. Exclusion of the one case-control study that did not meet our quality criteria (10) essentially did not change results. A forest plot showing results of the comparison between the highest and lowest categories of fiber intake is shown in Figure 2. There was little evidence of heterogeneity among case-control studies, and, overall, we estimated a reduction of 앒30% in endometrial cancer risk associated with high fiber intake. However, as mentioned earlier, the only prospective study that evaluated this association failed to find an association. DISCUSSION

This meta-analysis of 8 case-control studies (10 –13, 15, 16, 18) suggested an inverse association between dietary fiber intake and endometrial cancer risk, with little evidence of heterogeneity among studies. We estimated a reduction in risk of endometrial

FIGURE 1. Random-effects meta-analysis of studies that evaluated dietary fiber and endometrial cancer risk (per 5 g/1000 kcal). All studies excluded hysterectomies from the control group and adjusted for BMI (weight) and total energy intake [Potischman et al (11), adjusted for noncarbohydrate calories]. **Excluded studies for the following reasons: *hospital based, †쏝200 cases. OR, odds ratio; RR, relative risk.

Study

RR/OR (95% CI)

Case-cohort Jain et al, 2000 (9) Case-control Barbone et al, 1993 (10) Potischman et al, 1993 (11) Goodman et al, 1997 (12) Jain et al, 2000 (9) McCann et al, 2000 (13) Littman et al, 200 (16) Salazar-Martinez et al, 2005 (14) Xu et al, (18) Summary OR1 Summary OR2

0.5

1 Relative Risk

2

% Weight

1.24 (0.82, 1.87)

100.0

0.60 0.70 0.47 0.71 0.50 0.68 1.46 0.80 0.71 0.70

7.0 8.3 7.6 17.3 8.0 17.2 7.0 27.8 100.0

(0.31, 1.15) (0.39, 1.26) (0.25, 0.87) (0.49, 1.03) (0.27, 0.91) (0.47, 0.99) (0.76, 2.80) (0.62, 1.03) (0.59, 0.85) (0.59, 0.82)

3

Test for heterogeneity Pooled OR1: P value for heterogeneity: 0.26; I 2=20.8% Pooled OR2: P value for heterogeneity: 0.57; I 2=0.0%

FIGURE 2. Random-effects meta-analysis of studies that evaluated dietary fiber and endometrial cancer risk: highest compared with lowest category. All studies excluded hysterectomies from the control group and adjusted for BMI (weight) and total energy intake [Potischman et al (11), adjusted for noncarbohydrate calories]. **Excluded studies for the following reasons: *hospital based, †쏝200 cases.

cancer of 앒30% for women with the highest fiber consumption compared with the lowest consumption and a reduction in risk of 앒20% per 5 g/1000 kcal fiber intake. However, the only prospective study that evaluated dietary fiber and endometrial cancer suggested, if anything, a positive association (9). Because there were only a few studies that evaluated this exposure, we had limited power to assess publication bias through funnel plots or to conduct sensitivity analyses. We repeated some analyses excluding studies that did not meet our quality criteria, with essentially no change in these findings. Results for fiber subtypes were limited and inconsistent, thus precluding meta-analysis and any conclusions based on the current evidence. It is generally thought in epidemiology that cohort studies provide stronger evidence regarding an association than casecontrol studies because they are less prone to differential recall of dietary habits or selection bias. However, when considering the available data for the relation between dietary fiber and endometrial cancer one must question whether the whole body of evidence from case-control studies should be discarded based on the results of one single case-cohort study, particularly considering that there are reasonable biological mechanisms to suggest that dietary fiber may decrease endometrial cancer risk. Dietary fiber in broad terms is defined as the endogenous components of plant foods that are resistant to digestion in the intestinal tract (22). However, its classification and definition are not standardized (23). Although there is a relatively large body of literature evaluating the role of dietary fiber with other cancers, particularly with colorectal cancer, the overall evidence for a role of fiber on cancer is weak (24). The chemical complexity of fiber, inconsistencies in its definition, and the limitations of currently available dietary assessment tools have been postulated as possible explanations for the inconsistent findings stemming from studies of fiber and cancer (22). Several lines of evidence support a role of dietary fiber in the reduction of endometrial cancer risk. Dietary fiber intake may result in reduced exposure to endogenous estrogens. Through mechanical effects on the gut, it decreases transit time and therefore may result in less reabsorption of bile acid, metabolites of cholesterol, which is itself a precursor of endogenous synthesis of estrogens (25). Dietary fiber also can bind bile acids and may

1736

BANDERA ET AL

prevent the deconjugation of sterols that are excreted into the gut, thereby preventing their reabsorption (26). Diets high in dietary fiber have a relatively lower glycemic load, and these characteristics are associated with favorable effects on glucose and insulin metabolism and insulin resistance (4), which have been implicated in endometrial cancer etiology (2). Epidemiologic studies have also shown an inverse association between fiber intake and weight (4). Lignans, a class of compounds that have weak estrogenic effects, are found in foods that are high in dietary fiber, so a diet high in fiber may also provide high phytoestrogen exposure (27). High dietary fiber intake is also characteristic of diets high in whole grains, which contain other compounds, including phenolic compounds and antioxidants, which may also lower the risk of cancers in general (28). Dietary fiber consumption has been shown to result in lower blood pressure levels (29) and diabetes risk (30, 31). Both hypertension and diabetes are risk factors for endometrial cancer (32). High dietary fiber intake may also be a marker for a generally “healthier” dietary pattern and lifestyle, and various associated factors, such as increased vegetable consumption, lower fat intake, increased physical activity, and lower adiposity, which are associated with lower endometrial cancer risk (5). Thus, there is strong indirect evidence to support an inverse association of dietary fiber intake on endometrial cancer risk, based on its effects on weight maintenance, estrogen concentrations, glucose and insulin metabolism, diabetes, and hypertension. The question remains as to whether the evidence from casecontrol studies should be outweighed by the null findings from a single prospective study (9). However, several issues should be taken into account when interpreting the findings from this prospective study. First, the number of cases included in the analyses was relatively small; therefore, the study may not have had sufficient statistical power to detect an association. Virtually all CIs reported in the article for dietary variables include the null value. Second, women with a history of hysterectomy were excluded from baseline, but women undergoing hysterectomies during the follow-up period were not censored and, as the authors acknowledged, this may have led to an overestimation of person-years at risk. Although analyses were controlled for all relevant confounders, energy intake was adjusted for by simply adding it as continuous variable to multivariate analyses, rather than using

other, more robust methods, such as the residual method (33). An additional possible limitation is that the study population consisted of Canadian women undergoing mammographic screening, which may have resulted in a relatively homogeneous population with a narrower range and less variability in nutrient intake compared with other populations. The demographic characteristics of this cohort were not reported in the article. However, quartile cutoffs confirm that the range of intake for the Jain population was narrower than that in most of the other studies (Table 4), including the case-control study by the same authors (15). This case-control study (15) was also conducted using a different, more detailed dietary assessment method than used in the cohort study. This narrower range of intake may have affected the magnitude of the association that could be detected. The range of intakes in the 8 studies that examined the association of dietary fiber with endometrial cancer and for which the range of fiber intake in g/1000 kcal was available or could be estimated is shown in Table 4. The narrowest range of intake was in the study by Barbone et al (10), but that study used tertiles rather than quartiles (or quintiles) of intake. Otherwise, with the exception of the study by Xu et al (18), the range of intake was narrower in the case-cohort study (9) than in any of the other studies. The last column in Table 4, which presents the absolute value of the natural log of the relative risk estimate, provides a measure of the magnitude of the departure from the null on a common scale, regardless of the direction of the association. As illustrated in Table 4, studies with the largest range of intake tended to report the associations of larger magnitude. Overall, the correlation between the range of intake and the magnitude of the association was 0.58, including the Jain et al (case-cohort) study and 0.55 excluding that study. To our knowledge this is the first systematic literature review and meta-analysis of the role of dietary fiber intake on endometrial cancer risk. The 1997 WCRF/AICR Report, based on a narrative (and not comprehensive) review of the association of diet and endometrial cancer, did not mention a possible role of dietary fiber intake, even though 3 studies (10, 11, 17) that evaluated this relation had been published before 1997. In conclusion, our meta-analysis of case-control studies suggests an inverse association between dietary fiber intake and endometrial cancer risk. This finding is strongly supported by

TABLE 4 Range of dietary fiber intake (g/1000 kcal) and magnitude of relative risk (RR) estimate in studies that investigated associations with endometrial cancer risk1

Study

Jain et al, 2000, case-cohort study Barbone et al, 1993 (10) Xu et al, 2007 (18) Potischman et al, 1993 (11) Littman et al, 2001 (16) Jain et al, 2000, case-control study (15) McCann et al, 2000 (13) Goodman et al, 1997 (12) 1

Fiber intake categories

4 3 5 4 5 4 4 4

Lower cutoff 2

Upper cutoff 2

g 䡠 1000 kcal⫺1 䡠 d⫺1

g 䡠 1000 kcal⫺1 䡠 d⫺1

7.649 10.024 4.8 6.170 5.6 9.529 9.515 6.856

11.753 12.983 8.1 10.897 10.7 15.235 15.224 13.322

Difference (estimate of range of intake)

High vs low RR estimate

Ln(RR), absolute value3

4.103 2.959 3.3 4.728 5.1 5.706 5.709 6.466

1.24 0.6 0.8 0.7 0.68 0.71 0.5 0.47

0.215 0.511 0.223 0.357 0.386 0.342 0.693 0.755

The correlation between range of intake and absolute value of ln(RR) is 0.58, including Jain et al (9) and 0.55 excluding Jain et al (9). Values were estimated from reported g/d and total energy intake, except for Littman et al (16) and Xu et al (18). 3 Provides a common measure of departure of odds ratio/RR from the null value, regardless of direction of effect. 2

DIETARY FIBER AND ENDOMETRIAL CANCER RISK

several indirect lines of evidence. Because only one prospective cohort study examined this association, additional cohort studies are needed before definitive conclusions can be drawn regarding the role of dietary fiber on endometrial cancer prevention. Future studies should address the effects of different fiber characteristics, such as fiber solubility and food source, and should take into account the role of several important confounders, such as body mass index, total energy intake, and physical activity. Furthermore, the effects by body mass, exogenous estrogen use, and menopausal status should be evaluated. Overall, the consistent findings from case-control studies are intriguing and deserve further study. We thank James Thomas for his valuable help with the data extraction program Access. The authors’ responsibilities were as follows—EVB and LHK: study design and implementation; EVB: study management; DFM, LHK, and EVB: data analysis; DMG, EVB, and MLM: bibliographic searches; EVB and DMG: data extraction and tabulation; EVB, LHK, and MLM: interpretation of the evidence; and LHK, DFM, DMG, and MLM: critical revision of the article. None of the authors declared any conflicts of interest.

REFERENCES 1. Khan M, Mori M, Sakauchi F, et al. Risk of endometrial cancer mortality by ever-use of sex hormones and other factors in Japan. Asian Pac J Cancer Prev 2006;7:260 – 6. 2. Kaaks R, Lukanova A, Kurzer MS. Obesity, endogenous hormones, and endometrial cancer risk: a synthetic review. Cancer Epidemiol Biomarkers Prev 2002;11:1531– 43. 3. Sowers MR, Crawford S, McConnell DS, et al. Selected diet and lifestyle factors are associated with estrogen metabolites in a multiracial/ethnic population of women. J Nutr 2006;136:1588 –95. 4. Slavin JL. Dietary fiber and body weight. Nutrition 2005;21:411– 8. 5. Bandera EV, Kushi LH, Moore DF, Gifkins DM, McCullough ML. The association between food, nutrition, and physical activity and the risk of endometrial cancer and underlying mechanisms. In: Second Report on Food, Nutrition, Physical Activity and the Prevention of Cancer: World Cancer Research Fund International/American Institute for Cancer Research. 2007. Internet: www.wcrf.org (accessed 1 November 2007). 6. World Cancer Research Fund, American Institute for Cancer Research. Food, nutrition, and the prevention of cancer: a global perspective. Washington, DC: American Institute for Cancer Research, 1997. 7. Bandera EV, Kushi LH, Moore DF, Gifkins DM, McCullough ML. Consumption of animal foods and endometrial cancer risk: a systematic literature review and meta-analysis. Cancer Causes Control 2007;18: 967– 88. 8. Bandera EV, Kushi LH, Moore DF, Gifkins DM, McCullough ML. Fruits and vegetables and endometrial cancer risk: a systematic literature review and meta-analysis. Nutr Cancer 2007;58:6 –21. 9. Jain MG, Rohan TE, Howe GR, Miller AB. A cohort study of nutritional factors and endometrial cancer. Eur J Epidemiol 2000;16:899 –905. 10. Barbone F, Austin H, Partridge EE. Diet and endometrial cancer: a case-control study. Am J Epidemiol 1993;137:393– 403. 11. Potischman N, Swanson CA, Brinton LA, et al. Dietary associations in a case-control study of endometrial cancer. Cancer Causes Control 1993; 4:239 –50. 12. Goodman MT, Wilkens LR, Hankin JH, Lyu LC, Wu AH, Kolonel LN.

13.

14.

15. 16.

17.

18.

19.

20.

21.

22.

23.

24. 25. 26. 27. 28. 29.

30.

31.

32.

33.

1737

Association of soy and fiber consumption with the risk of endometrial cancer. Am J Epidemiol 1997;146:294 –306. McCann SE, Freudenheim JL, Marshall JR, Brasure JR, Swanson MK, Graham S. Diet in the epidemiology of endometrial cancer in western New York (United States). Cancer Causes Control 2000;11:965–74. Salazar-Martinez E, Lazcano-Ponce E, Sanchez-Zamorano LM, Gonzalez-Lira G, Escudero de los Rios P, Hernandez-Avila M. Dietary factors and endometrial cancer risk. Results of a case-control study in Mexico. Int J Gynecol Cancer 2005;15:938 – 45. Jain MG, Howe GR, Rohan TE. Nutritional factors and endometrial cancer in Ontario, Canada. Cancer Control 2000;7:288 –96. Littman AJ, Beresford SA, White E. The association of dietary fat and plant foods with endometrial cancer (United States). Cancer Causes Control 2001;12:691–702. Shu XO, Zheng W, Potischman N, et al. A population-based case-control study of dietary factors and endometrial cancer in Shanghai, People’s Republic of China. Am J Epidemiol 1993;137:155– 65. Xu WH, Dai Q, Xiang YB, et al. Nutritional factors in relation to endometrial cancer: a report from a population-based case-control study in Shanghai, China. Int J Cancer 2007;120:1776 – 81. Cheˆne G, Thompson SG. Methods for summarizing the risk associations of quantitative variables in epidemiologic studies in a consistent form. Am J Epidemiol 1996;144:610 –21. Greenland S, Longnecker MP. Methods for trend estimation from summarized dose-response data, with applications to meta-analysis. Am J Epidemiol 1992;135:1301–9. Huedo-Medina TB, Sanchez-Meca J, Marin-Martinez F, Botella J. Assessing heterogeneity in meta-analysis. Psychol Methods 2006;11:193– 206. Martinez ME, Jacobs ET. Dietary fiber and carbohydrates. In: Heber D, Blackburn GL, Go VLW, Milner J, eds. Nutritional oncology. 2nd ed. San Diego, CA: Academic Press, 2006:521–530. Lupton JR, Turner ND. Dietary fiber. In: Stipanuk MH, ed. Biochemical, physiological, and molecular aspects of human nutrition. 2nd ed. St Louis, MO: Elsevier, 2006:240 –1. McCullough ML, Giovannucci EL. Diet and cancer prevention. Oncogene 2004;23:6349 – 64. Weisburger JH, Wynder EL. Dietary fat intake and cancer. Hematol Oncol Clin North Am 1991;5:7–23. Gorbach SL, Goldin BR. Diet and the excretion and enterohepatic cycling of estrogens. Prev Med 1987;16:525–31. Lampe JW. Isoflavonoid and lignan phytoestrogens as dietary biomarkers. J Nutr 2003;133(suppl 3):956S– 64S. Slavin JL. Mechanisms for the impact of whole grain foods on cancer risk. J Am Coll Nutr 2000;19(suppl):300S–7S. Streppel MT, Arends LR, van ’t Veer P, Grobbee DE, Geleijnse JM. Dietary fiber and blood pressure: a meta-analysis of randomized placebo-controlled trials. Arch Intern Med 2005;165:150 – 6. Salmeron J, Manson JE, Stampfer MJ, Colditz GA, Wing AL, Willett WC. Dietary fiber, glycemic load, and risk of non-insulin-dependent diabetes mellitus in women. JAMA 1997;277:472–7. Meyer KA, Kushi LH, Jacobs DR Jr, Slavin J, Sellers TA, Folsom AR. Carbohydrates, dietary fiber, and incident type 2 diabetes in older women. Am J Clin Nutr 2000;71:921–30. Persson I, Adami H-O. Endometrial cancer. In: Adami H-O, Hunter D, Trichopoulos D, eds. Textbook of cancer epidemiology. New York, NY: Oxford University Press, 2002. Willett WC, Howe GR, Kushi LH. Adjustment for total energy intake in epidemiologic studies. Am J Clin Nutr 1997;65(suppl):1220S– 8S; discussion 1229S–31S.

L-Carnitine

treatment reduces severity of physical and mental fatigue and increases cognitive functions in centenarians: a randomized and controlled clinical trial1–3 Mariano Malaguarnera, Lisa Cammalleri, Maria Pia Gargante, Marco Vacante, Valentina Colonna, and Massimo Motta ABSTRACT Background: Centenarians are characterized by weakness, decreasing mental health, impaired mobility, and poor endurance. L-Carnitine is an important contributor to cellular energy metabolism. Objective: This study evaluated the efficacy of L-carnitine on physical and mental fatigue and on cognitive functions of centenarians. Design: This was a placebo-controlled, randomized, double-blind, 2-phase study. Sixty-six centenarians with onset of fatigue after even slight physical activity were recruited to the study. The 2 groups received either 2 g levocarnitine once daily (n ҃ 32) or placebo (n ҃ 34). Efficacy measures included changes in total fat mass, total muscle mass, serum triacylglycerol, total cholesterol, HDL cholesterol, LDL cholesterol, Mini-Mental State Examination (MMSE), Activities of Daily Living, and a 6-min walking corridor test. Results: At the end of the study period, the levocarnitine-treated centenarians, compared with the placebo group, showed significant improvements in the following markers: total fat mass (Ҁ1.80 compared with 0.6 kg; P 쏝 0.01), total muscle mass (3.80 compared with 0.8 kg; P 쏝 0.01), plasma concentrations of total carnitine (12.60 compared with Ҁ1.70 ␮mol; P 쏝 0.05), plasma long-chain acylcarnitine (1.50 compared with Ҁ0.1 ␮mol; P 쏝 0.001), and plasma short-chain acylcarnitine (6.0 compared with Ҁ1.50 ␮mol; P 쏝 0.001). Significant differences were also found in physical fatigue (Ҁ4.10 compared with Ҁ1.10; P 쏝 0.01), mental fatigue (Ҁ2.70 compared with 0.30; P 쏝 0.001), fatigue severity (Ҁ23.60 compared with 1.90; P 쏝 0.001), and MMSE (4.1 compared with 0.6; P 쏝 0.001). Conclusions: Our study indicates that oral administration of levocarnitine produces a reduction of total fat mass, increases total muscular mass, and facilitates an increased capacity for physical and cognitive activity by reducing fatigue and improving cognitive functions. Am J Clin Nutr 2007;86:1738 – 44. KEY WORDS L-Carnitine, L-acetyl-carnitine, centenarians, fatigue, cognitive functions

INTRODUCTION

SUBJECTS AND METHODS

Subjects A total of 70 centenarians, aged from 100 to 106 y (24 men, 46 women) were recruited to the study. The centenarians were treated with L-carnitine for 6 mo. 1

Aging is characterized by a slow decline of the physiologic functions, with a progressive deterioration of various organs and consequently of the organism until death, and appears to be associated with a substantial loss of the ability to regulate energy balance (1, 2). Mitochondria, because of their critical importance for energy production, have attracted the attention of scientists interested in unraveling the complex changes associated with aging and age-related diseases (3, 4). Because mitochondria are

1738

responsible for most cellular energy conversion, these associations are consistent with the hypothesis that DNA damage in mitochondria may contribute to the age-associated decrease in energy expenditure for physical activity (5). L-Carnitine is an endogenous molecule and is an important contributor to cellular energy metabolism. It is present ubiquitously in the organism, and the main concentrations are found in the most active metabolic tissue, such as the myocardium and skeletal muscle. L-Carnitine is indispensable for the transport of long-chain fatty acids across the inner mitochondrial membrane to their site of oxidation and the production of energy in the form of ATP (6, 7). Among all the substances whose concentration decreases with age, L-carnitine diminution is fundamentally important, given its function in the production of energy. One of the most important consequences of carnitine deficiency is therefore manifested in the alteration of the metabolic pathways that lead to the production of energy. In our previous study, the treatment with exogenous levocarnitine in elderly subjects showed a progressive increase in total muscle mass and a significant reduction in muscle fatigue compared with placebo (8). In the elderly variations are found in the plasma concentration of L-carnitine even if its causes are not known. In fact, the concentration of carnitine actually increases with the advancement of age until 앒70 y, subsequently tending to diminish the parallel in the reduction in body mass index (in kg/m2) and muscle mass (9). The aim of this study was to evaluate the efficacy of L-carnitine in physical and mental fatigue and on the cognitive functions of centenarians.

From the Department of Senescence, Urological, and Neurological Sciences, University of Catania, Catania, Italy. 2 Supported by a grant from MURST (Ministero dell’Universita` e Ricerca Scientifica e Tecnologica). 3 Reprints not available. Address correspondence to M Malaguarnera, Department of Senescence, Urological, and Neurological Sciences, Ospedale Cannizzaro, Viale Messina, 829 – 95125 Catania, Italy. E-mail: [email protected]. Received May 14, 2007. Accepted for publication August 6, 2007.

Am J Clin Nutr 2007;86:1738 – 44. Printed in USA. © 2007 American Society for Nutrition

L-CARNITINE

TREATMENT IN CENTENARIANS

Wessely’s test and Powell’s test were used to examine fatigue, both mental and physical, and the severity was expressed with the Krupp’s test. The Wessely and Powell score consists of 2 scales measuring physical fatigue [8 items scored from 0 (no fatigue) to 2 (highest possible fatigue); total score range: 0 –16] and mental fatigue (5 items; total score range: 0 –16) (10). We also used the Fatigue Severity Scale, composed of 9 items (11). Here, the total score ranges from 9 to 63 and is directly related to the severity observed. Mini-Mental State Examination (MMSE) was used to assess cognitive function (12, 13, 14). The MMSE score ranges between 0 and 30. Subjects were excluded if they had experienced any of the following: a significant medical or surgical event within the previous 3 mo, significant cardiac failure (New York Heart Association class III or IV) (15), acute or chronic renal failure, severe respiratory disorders, severe digestive disorders, severe cognitive disorders, diabetes mellitus, or other endocrine diseases. Patients taking corticosteroids or diuretics were also excluded from the trial. This study was designed and conducted in compliance with the ethical principles of Good Clinical Practice and the Declaration of Helsinki (16). The study protocol was approved by the ethics committee of the Cannizzaro Hospital (Catania, Italy). Informed consent was obtained from centenarians or from their relatives (in cases resulting from illiteracy or difficulty in vision or hearing) before any study procedures were initiated. Study design This was a randomized, double-blind, placebo-controlled study. The study was conducted between 1999 and 2002, and the study participants, living in Sicily, were recruited through the Registry office. The 70 centenarians were randomly assigned by a computer-generated randomization schedule to receive a 6-mo supply of either levocarnitine or placebo. The treatment was for 6 mo. The follow-up was for another 6 mo after treatment to evaluate survival. The measurements were made every month, both for efficacy tests and for tolerability. Prerandomization phase The subjects or the caregivers were required to document all caloric intake with the use of a diary, completed every 2 d. This prerandomization period was designed to nullify the effects of dietary changes on metabolic markers. During the initial 2-wk phase, subjects (nurses or caregivers) were instructed by a dietitian to follow an ad libitum diet as classified by the National Cholesterol Education Program (Step 2). Subjects underwent weekly visits throughout the treatment period to assess adherence to the study protocol, to measure blood pressure and cognitive functions, and to record adverse events. Randomization phase Throughout the trial, L-carnitine was supplied in vials with 2 g carnitine (Sigma Tau, Rome, Italy) taken orally once a day. All drugs and placebos were identical in appearance, and neither investigators nor patients were informed of the selected agent until the end of the study phase. Dosing instructions were provided with each patient pack. All trial medication was instructed to be taken as prescribed. Subjects were considered compliant if the number of returned vials

1739

was between 80% and 120% of the planned treatment regimen. For the duration of the trial any concomitant drugs were administered at the lowest possible therapeutic dosage and, as far as possible, were not changed. L-Carnitine

determination

Patients were studied in the morning between 0800 and 1000 after an overnight fast. The patients were first asked to empty their bladder. Then, venous blood samples were drawn into tubes containing EDTA or heparin, and serum or plasma was obtained by centrifugation (2000 ҂ g for 5 min at 25 °C). A spot urine sample was obtained 10 min after the collection of the blood sample. Serum was measured immediately; plasma and urine were stored at Ҁ20 °C until analysis. The L-carnitine concentration in plasma and urine was measured by a method described by Cederblad and Lindstedt (17) and modified by Brass and Hoppel (18). Plasma was treated with perchloric acid (final concentration of 3% vol:vol) and centrifuged for 2 min at 10 000 ҂ g and at 25 °C. Long-chain acylcarnitines (LCACs) were measured in the pellet after alkaline hydrolysis, and free and short-chain acylcarnitines (SCACs) were measured in the supernatant fluid. The interassay CVs were 3.8%, 3.9%, and 4.1%, respectively; the intraassay CVs were 5.4%, 5.8%, and 6.4%, respectively. Addition of LCACs, together with the free and SCACs yields the total acyl-carnitine. In urine no perchloric acid precipitation was performed (LCACs are normally not found in urine). Free carnitine and SCACs were measured in plasma. The within-assay CVs were 4.1% and 3.9%, respectively; the between-assay CVs were 5.2% and 5.8%, respectively. The creatinine concentrations were measured by a kinetic colorimetric reaction in the same samples as used for the measurement of the L-carnitine concentrations. The within- and betweenassay CVs were 3.9% and 5.4%, respectively. Seventy samples of centenarians were used to calculate the intraassay and interassay CVs. Efficacy assessment Throughout the randomization phase of the study, thriceweekly alimentary diary cards were used to collect efficacy data. The primary efficacy measures were changes in total fat mass, total muscle mass, triacylglycerols, total cholesterol, HDL cholesterol, and LDL cholesterol. Measurements were made at the beginning and at the end of the study period. Data were collected in the morning, after an overnight fast. Anthropometric data were measured at baseline and at the end of the study period. The body mass index was calculated from body weight and body height. To measure total fat mass and total muscle mass, bioelectrical impedance analysis was used. Before measurement, subjects were instructed to refrain from physical activity for 12 h and liquids for 4 h and were asked to urinate 30 min before examination. For the 5 min leading up to the measurement period, subjects were told to adopt a supine position with their legs apart. After the skin was cleaned with 70% alcohol, 4 adhesive electrodes (3 M Red Dot T; 3 M Health Care, Borken, Germany) were placed on the surface of the right hand and right foot, according to the manufacturer’s guidelines. We used a BioZ2 generator (Spengler, Paris, France). The interobserver and interday variability was 0.003 kg for fat-free mass (95% CI: Ҁ0.2, 0.2).

1740

MALAGUARNERA ET AL

Physical and mental fatigue, the severity of fatigue, and the MMSE were assessed before and after treatment. Functional assessment was performed with the use of the Katz Index of Activities of Daily Living (ADLs; range: 0 – 6). ADL questions included walking, feeding, bathing, using the toilet, and dressing (19). Tolerability assessment Laboratory assessments were monitored at baseline and monthly until the end of the trial. These data included blood glucose, total cholesterol, HDL cholesterol, triacylglycerols, creatine phosphokinase, lactate dehydrogenase, aspartate aminotransferase, alanineaminotransferase, alkaline phosphatase, creatinine, and blood urea nitrogen. Fasting plasma glucose was measured by the glucose-oxidase method. Serum total cholesterol and triacylglycerol concentrations were measured by the enzymatic method. HDL cholesterol was measured by the heparin-calcium method. Serum creatinine was measured by Jaffe reaction. Electrocardiogram and blood pressure were monitored with the use of standard techniques. Physical activity was evaluated weekly with 6-min walking test (6MWT). The centenarians walked in a corridor of known length for 6 min. The 6MWT was performed in the morning after an overnight fast in a quiet room at a constant temperature of 22 앐 2 °C. The walking distance was the distance in meters walked by the centenarians in 6 min. Statistical analysis For all nonparametric data, discrete and continuous variables were compared with the use of either the Student’s t test or Wilcoxon’s Mann-Whitney test. Categorical variables were compared with the use of either the chi-square test or Fisher’s exact test. STATISTICAL ANALYSIS SYSTEM software (version 6.11; SAS Institute, Cary, NC) was used for all analysis. A repeated-measures 2-factor analysis was done. All P values were 2-sided, with the use of ␣ ҃ 0.05 as the reference standard for determining the significance of the principal outcomes. The primary population for statistical analysis was an intention-totreat population of all subjects randomly assigned. RESULTS

Baseline characteristics The baseline characteristics were evenly distributed across the 2 groups of enrolled patients. No significant differences between the 2 groups were observed before treatment (Table 1). In the comparison between the groups, we observed that the proportion in the local rest homes was 87.5% in the L-carnitine group compared with 79.4% in the placebo group, for those residing in the community was 12.5% compared with 20.6%, for those who were illiterate was 46.9% compared with 52.9%, for those with bad eyesight was 78.1% compared with 64.7%, for those with hearing loss 87.5% compared with 58.8%, for those with cognitive deterioration was 50% compared with 44.1%, in current or past smokers was 62.5% compared with 52.9%, respectively. Plasma and clinical markers Total cholesterol showed a significant decrease during the treatment with L-carnitine (P 쏝 0.01; 95% CI, 0.17, 1.21) compared with placebo (Table 2).

TABLE 1 Baseline characteristics and basal plasma variables of L-carnitine and placebo cohorts at randomization1 L-Carnitine

Age (y) Sex (n) Men Women SBP (mm Hg) DBP (mm Hg) Heart rate (bpm) BMI (kg/m2) BUN (mmol/L) Plasma creatinine (␮mol/L) Blood glucose (mmol/L) Total cholesterol (mmol/L) HDL cholesterol (mmol/L) Triacylglycerols (mmol/L) CPK (IU/L) LDH (IU/L)

(n ҃ 32)

Placebo (n ҃ 34)

101 앐 1.32

101 앐 1.4

10 22 155 앐 24.2 86.2 앐 10.1 88 앐 10 22.2 앐 4.7 7.42 앐 3.14 95.47 앐 49.50 4.44 앐 1.45 4.87 앐 1.01 1.27 앐 0.21 1.51 앐 0.63 45.4 앐 18.2 341.2 앐 44.6

11 23 154 앐 25.1 84.7 앐 10.3 87 앐 12 22.6 앐 4.1 7.01 앐 3.50 84.86 앐 61.88 4.68 앐 1.13 4.79 앐 1.04 1.22 앐 0.21 1.47 앐 0.64 44 앐 20.1 356 앐 40.2

1 SBP, systolic blood pressure; DBP, diastolic blood pressure; bpm, beats per minute; BUN, blood urea nitrogen; CPK, creatine phosphokinase; LDH, lactate dehydrogenase. P was not significant between the 2 treatment groups. 2 x៮ 앐 SD (all such values).

L-Carnitine

in urine and plasma

In the levocarnitine group, significant differences were observed in the following markers after treatment compared with baseline: plasma concentrations of total L-carnitine (12.6 ␮mol/ L), plasma LCAC (1.5 ␮mol/L), and SCAC (6.0 ␮mol/L). No significant differences of levocarnitine concentrations were observed in the urine. In the placebo group the plasma concentrations of free L-carnitine and LCAC and the urinary excretion of free L-carnitine and SCAC did not show significant differences compared with baseline. At the end of the study period, compared with placebo, the levocarnitine-treated centenarians showed significant improvements in the following markers: plasma concentrations of total L-carnitine (12.60 compared with Ҁ1.70 ␮mol/L), LCAC (1.50 compared with Ҁ0.1 ␮mol/L), and SCAC (6.0 compared with Ҁ1.50 ␮mol/L). No significant differences of levocarnitine concentrations were observed in the urine (Table 3). Physical performance and daily activity In the levocarnitine group, significant differences were observed in the following markers after treatment compared with baseline. The total muscle mass increased by 3.8 kg. The total fat mass decreased by 1.8 kg, the ADLs increased by 0.5 points, the 6MWT increased by 4.4 m. In the placebo group the physical performance and daily activity did not show a significant difference from baseline. At the end of the study period, compared with placebo, the levocarnitine-treated centenarians showed significant improvements in the following markers: total fat mass (Ҁ1.80 compared with 0.6 kg), total muscle mass (3.80 compared with 0.8 kg), plasma concentrations of total L-carnitine (12.60 compared with Ҁ1.70 ␮mol), LCAC (1.50 compared with Ҁ0.1 ␮mol), SCAC (6.0 compared with Ҁ1.50 ␮mol), ADLs (0.5 compared with 0.1), and 6MWT (4.4 compared with 0.4 m) (Table 4).

L-CARNITINE

1741

TREATMENT IN CENTENARIANS

TABLE 2 Comparison of the plasma and clinical markers between treatment groups1 L-Carnitine

SBP (mm Hg) DBP (mm Hg) Heart rate (bpm) BMI (kg/m2) Total cholesterol (mmol/L) HDL cholesterol (mmol/L) Triacylglycerols (mmol/L) Plasma creatinine (mmol/L) BUN (mmol/L) Glucose (mmol/L) CPK (IU/L) LDH (IU/L)

treatment (n ҃ 32)

Placebo treatment (n ҃ 34)

Before treatment

After treatment

Before treatment

After treatment

158.2 앐 22.1 81.4 앐 11.1 85 앐 9 22.3 앐 4.6 4.79 앐 1.06 1.24 앐 0.10 1.59 앐 0.44 93.70 앐 44.2 7.65 앐 2.89 4.65 앐 1.02 45.1 앐 19.3 358.1 앐 36.71

156.1 앐 21.8 80.2 앐 12.8 86 앐 8 23.4 앐 4.2 4.10 앐 1.02 1.28 앐 0.11 1.50 앐 0.38 90.17 앐 53.92 7.47 앐 2.97 4.39 앐 1.04 40.2 앐 18.2 354.2 앐 38.2

151.4 앐 23.8 83.2 앐 10.7 84 앐 10 22.8 앐 4.7 4.84 앐 1.04 1.27 앐 0.21 1.61 앐 0.45 84.86 앐 60.11 7.19 앐 3.34 4.67 앐 1.08 43.8 앐 19.4 354.1 앐 36.82

152.1 앐 24.7 82.8 앐 11.9 87 앐 8 22.8 앐 43.1 4.69 앐 1.043 1.27 앐 0.23 1.60 앐 0.46 79.56 앐 62.76 7.27 앐 3.25 4.65 앐 1.01 42.2 앐 18.7 351.6 앐 39.6

All values are x៮ 앐 SD. SBP, systolic blood pressure; DBP, diastolic blood pressure; bpm, beats per minute; BUN, blood urea nitrogen; CPK, creatine phosphokinase; LDH, lactate dehydrogenase. 2 Significantly different from before treatment, P 쏝 0.01 (ANOVA). 3 Significantly different from the L-carnitine group, P 쏝 0.05 (ANOVA). 1

Fatigue In the levocarnitine group significant differences were observed in the following markers after treatment compared with baseline: the score for the physical fatigue component of the Wessely and Powell Scale decreased by 4.10 points, whereas the mental score decreased by 2.70. The fatigue severity score decreased by 23.60. In the placebo group no significant differences were observed compared with baseline. At the end of the study period, compared with placebo, the levocarnitine-treated centenarians showed significant differences in the following markers: physical fatigue (Ҁ4.10 compared with Ҁ1.10), mental fatigue (Ҁ2.70 compared with 0.30), and fatigue severity (Ҁ23.60 compared with 1.90) (Table 4). Cognitive function In the levocarnitine group significant differences were observed in the following marker after treatment compared with baseline: the MMSE score increased by 4.10. In the placebo

group there were no significant differences compared with baseline. At the end of the study period, compared with placebo, the levocarnitine-treated centenarians showed significant improvements in the MMSE score (4.10 compared with 0.60) (Table 4). Tolerability Of the 70 patients randomly assigned, 3 withdrew their consent and 1 died; 66 received the treatment. In the group treated with L-carnitine, 5 subjects withdrew, 3 died (1 after 66 d of beginning treatment, 1 after 121 d, 1 after 156 d), 1 for side effects, and 1 withdrew consent (Figure 1). In the group treated with placebo, 7 subjects withdrew, 5 died (1 after 36 d, 1 after 64 d, 1 after 85 d, 1 after 94 d, 1 after 110 d), 1 for side effects, and 1 withdrew consent. The 2 groups were homogeneous for baseline characteristics and clinical markers. In the group treated with L-carnitine, 1 patient decided against continuing the treatment after diarrhea (diarrhea was a consequence of the treatment with levocarnitine). In the group treated

TABLE 3 Comparison of plasma and urinary concentrations of L-carnitine between treatment groups1 L-Carnitine

Free plasma carnitine (␮mol/L) Plasma SCAC (␮mol/L) Plasma LCAC (␮mol/L) Total plasma carnitine (␮mol/L) Free urinary carnitine (␮mol/L) Urinary SCAC (␮mol/L) 1

treatment (n ҃ 32)

Before treatment

After treatment

Before treatment

After treatment

P for time2

P for group ҂ time2

41.8 앐 7.73 10.3 앐 5.1 2.8 앐 0.7 55.2 앐 9.9 15.8 앐 9.6 12.8 앐 7.1

49.2 앐 17.6 16.3 앐 13.04 4.3 앐 1.84 67.8 앐 29.94 16.7 앐 8.7 13.2 앐 7.4

40.3 앐 8.4 10.1 앐 5.4 3.0 앐 0.8 53.4 앐 12.5 14.9 앐 10.6 13.1 앐 9.4

43.3 앐 10.7 8.6 앐 2.95 3.1 앐 0.95 55.1 앐 12.05 13.2 앐 10.1 12.0 앐 9.0

쏝0.05 쏝0.01 쏝0.001 쏝0.05 NS NS

NS 쏝0.001 쏝0.001 쏝0.05 NS NS

SCAC, short-chain acylcarnitine; LCAC, long-chain acylcarnitine. Determined with ANOVA. 3 x៮ 앐 SD (all such values). 4 Significantly different from before treatment. 5 Significantly different from the L-carnitine group. 2

Placebo treatment (n ҃ 34)

1742

MALAGUARNERA ET AL

TABLE 4 Comparison of physical and mental markers and clinical characteristics between treatment groups1 L-Carnitine

Total fat mass (kg) Total fat-free mass (kg) Physical fatigue score (0–16) Mental fatigue score (0–16) Fatigue Severity Scale (9–63) MMSE (0–30 points) Activity Index of Daily Living (score) Walking distance (m)

treatment (n ҃ 32)

Placebo treatment (n ҃ 34)

Before treatment

After treatment

Before treatment

After treatment

P for time2

P for group ҂ time2

21.4 앐 3.83 35.1 앐 3.2 12.9 앐 2.6 7.5 앐 2.1 54.2 앐 5.6 16.4 앐 3.6 3.1 앐 0.4 10.2 앐 3.8

19.6 앐 3.94 38.9 앐 3.94 8.8 앐 2.44 4.8 앐 1.74 30.6 앐 9.44 20.5 앐 2.94 3.6 앐 0.54 14.6 앐 3.94

21.2 앐 3.6 35.4 앐 3.5 12.7 앐 2.4 7.4 앐 2.3 53.8 앐 5.2 16.6 앐 2.9 2.9 앐 0.6 10.8 앐 3.4

21.8 앐 3.45 36.2 앐 3.95 11.6 앐 2.55 7.1 앐 2.05 51.9 앐 7.65 17.2 앐 2.85 3.0 앐 0.45 11.2 앐 3.45

NS 쏝0.001 쏝0.001 쏝0.001 쏝0.001 쏝0.001 쏝0.001 쏝0.001

쏝0.01 쏝0.01 쏝0.001 쏝0.001 쏝0.001 쏝0.001 쏝0.01 쏝0.001

1

MMSE, Mini Mental State Examination. Determined with ANOVA. 3 x៮ 앐 SD (all such values). 4 Significantly different from before treatment. 5 Significantly different from the L-carnitine group. 2

with placebo a patient had bronchopneumopathy chronic obstructive, and for this reason treatment was discontinued. In the other subjects the repeated administration of levocarnitine was well tolerated with good compliance. DISCUSSION

Centenarians are characterized by a general weakness, decreasing mental health, impaired mobility and balance, and poor endurance. In the centenarians the L-carnitine treatment improved total L-carnitine, SCACs, and LCACs, which dynamically interact with multiple coenzyme A– dependent biochemical pathways. L-Carnitine is required for mitochondrial long-chain fatty acid oxidation (6), a main source of energy during exercise (20). The increase in L-carnitine content might increase the rate

of fatty acid oxidation, permitting a reduction of glucose utilization, preserving muscle glycogen content, and ensuring maximal rates of oxidative ATP production (21, 22, 23). Under normal nutritional conditions and in healthy persons, L-carnitine availability is not a limiting step in ␤-oxidation; however, in centenarians its supplementation may provide beneficial effects. In fact, L-carnitine was shown to improve total muscle mass, reduce total fat mass with weight gain, and improve walking capacity. L-Carnitine enhances exercise performance through SCAC production. The bioenergetic demands of exercise place stresses on the metabolic machinery of the muscle (24, 25). L-Carnitine treatment showed an improvement of performance levels (a measure of patient’s daily activity) in these subjects, suggesting that abnormality in mitochondrial homeostasis was the basis of some

70 Centenarians randomly assigned (3 withdrew consent; 1 died)

66 Centenarians receiving blinded study drug

Allcocated to L- carnitine: 32

Withdrawn: 5

Allocated to placebo: 34

Wit hd rawn: 7

3 died 1 for adverse events 1 withdrew consent

5 died 1 for adverse events 1 withdrew consent

FIGURE 1. Trial profile of L-carnitine treatment.

L-CARNITINE

TREATMENT IN CENTENARIANS

symptoms, such as fatigue, depression, and sarcopenia, observed in the elderly and centenarians (26). When administered orally, L-carnitine enhances the performance efficiency of highintensity muscular exercise. The beneficial effects of L-carnitine treatment were observed not only in muscle metabolism but also in the myocardium (27, 28). In fact, supplementation of the myocardium with L-carnitine results in an increased tissue L-carnitine content that restores L-carnitine losses and lessens the severity of ischemic injury (29, 30). In brain tissue, the L-carnitine shuttle mediates translocation of the acetyl moiety from mitochondria into the cytosol and thus contributes to the synthesis of acetylcholine and of acetylcarnitine (31, 32). The neurobiologic effects of acetyl carnitine include modulation of brain energy and phospholipids metabolism, cellular macromolecules (such neurotrophic factors and neurohormones), synaptic morphology, and synaptic transmission of multiple neurotransmitters (33, 34). Our study has several limitations. Our analyses were not adjusted for traditional risk factors, such as dyslipidemia and smoking, which we did not believe would be relevant in subjects aged 욷100 y and who represent a clinical end stage rather than a population at risk. Another limitation is the inclusion of subjects with mild cognitive deficits, bad eyesight, or hearing loss or who were illiterate. This would severely limit their ability to provide accurate answers on questionnaires about the primary outcomes. Furthermore, the centenarians were always assisted by relatives, nurses, or caregivers. In our study, compared with baseline, we observed changes not only in both physical and mental fatigue but also in cognitive deterioration. In our previous study, we found that treatment with exogenous levocarnitine in elderly subjects was associated with an increase in total muscle mass and a significant reduction in muscle fatigue compared with placebo (8). The beneficial effects of L-carnitine on heart function recovery from ischemia cannot be justified by these drugs, stimulating fatty acid oxidation only. L-Carnitine treatment improves not only the total L-carnitine serum concentrations but also acetylcarnitine serum concentrations. Moreover, the action of L-carnitine on the central nervous system can slow cognitive deterioration that occurs as a result of the normal physiologic aging of nervous cells. It also improves cardiac output and muscular function (26, 29). In centenarians we detected stable urinary excretion of total, free, and acyl carnitine. This could be related to the difference between tissues about the uptake of and synthetic capacity for L-carnitine (35). The variation of storage and metabolism of L-carnitine between different tissues (hepatic, renal, cardiac, skeletal muscle, brain, and pancreatic) and possible unstable excretion may explain the lack of correlation between blood and urinary L-carnitine in our patients. The administration of L-carnitine improves cardiac output and muscular function and reduces cognitive deterioration (31, 36). Our study indicates that oral administration of levocarnitine evokes a reduction of total fat mass, increases total muscular mass, and facilitates an increased capacity for physical and cognitive activity, by reducing fatigue and improving cognitive functions. We thank Ashraf Virmani for excellent technical assistance and linguistic advice and Marcella Malaguarnera for statistical advices.

1743

The author’s responsibilities were as follows—M Malaguarnera contributed to the study design, data analysis, and the drafting of the manuscript; LC, MPG, VC, M Motta have contributed to enrollment of patients and data interpretation; MV helped with statistical analysis and data interpretation. None of the authors had any relevant personal or financial conflicts of interest.

REFERENCES 1. Malaguarnera M, Pistone G, Motta M. Mythology in medicine: the elderly and quality of life. Br Med J 1995;311:1136. 2. Motta M, Bennati E, Ferlito L, Malaguarnera M, Motta L. Italian Multicenter Study on Centenarians (IMUSCE). Successful aging in centenarians: myths and reality. Arch Gerontol Geriatr 2005;40:241–51. 3. Liu J, Head E, Kuratsune H, Cotman CW, Ames BN. Comparison of the effects of L-carnitine and acetyl-L-carnitine on carnitine levels, ambulatory activity, and oxidative stress biomarkers in the brain of old rats. Ann N Y Acad Sci 2004;1033:117–31. 4. Hagen TM, Ingersoll RT, Wehr CM, et al. Acetyl-L-carnitine fed to old rats partially restores mitochondrial function and ambulatory activity. Proc Natl Acad Sci U S A 1998;95:9562– 6. 5. Wallace DC. Mitochondrial paradigm of metabolic and degenerative diseases, aging, and cancer: a dawn for evolutionary medicine. Annu Rev Genet 2005;39:359 – 407. 6. Brenner RR. Essential fatty acids: its transformations and functions Arch Latinoam Nutr 1983;33:735– 47. 7. Long CS, Haller RG, Foster DW, McGarry JD. Kinetics of carnitinedependent fatty acid oxidation: implications for human carnitine deficiency. Neurology 1982;32:663– 6. 8. Pistone G, Marino A, Leotta C, Dell’arte S, Finocchiaro G, Malaguarnera M. Levocarnitine administration in elderly subjects with rapid muscle fatigue. Drugs Aging 2003;20:761–7. 9. Malaguarnera M, Pistone G, Receputo G et al. Serum carnitine levels in centenarians. Clin Drug Invest 1999;17:321–7. 10. Wessely S, Powell R. Fatigue syndromes: a comparison of chronic “postviral” fatigue with neuromuscular and affective disorders. J Neurol Neurosurg Psych 1989;52:940 – 8. 11. Krupp LB, La Rocca NG, Muir-Nash J, Steinberg AD. The fatigue severity scale. Application to patients with multiple sclerosis and systemic lupus erythematosus. Arch. Neurol 1989;46:1121–3. 12. Folstein MF, Folstein SE, McHugh PR. “Mini Mental State” a practical method for grading the cognitive state of patients for the cognitive state of patients for the clinician. J Psychiatr Research 1975;12:189 –98. 13. Maugeri D, Santangelo A, Abbate S, et al. Correlation between the bone mass, psychometric performances and the levels of autonomy and autosufficiency in an elderly Italian population above 80 years of age. Arch Gerontol Ger 2001;33:265271. 14. Malaguarnera M, Pistone G, Motta M, Lo Manto PC, Di Fazio I. Assessment of self-sufficiency in ultraoctogenarians. Arch Gerontol Geriatr 1996;22(suppl):505– 8. 15. Hurst JW, Morris DC, Alexander RW. The use of the New York Heart Association’s classification of cardiovascular disease as a part of the patient’s complete problem list. Clin Cardiol 1999;22:385–90. 16. World Medical Association Declaration of Helsinki. Recommendations guiding physicians in biomedical research involving human subjects. JAMA 1997;277:925– 6. 17. Cederblad G, Lindstedt S. A method for the determination of carnitine in the picomole range. Clin Chim Acta 1972;37:235– 43. 18. Brass EP, Hoppel CL. Carnitine metabolism in the fasting rat. J Biol Chem 1978;253:2688 –93. 19. Katz S, Ford AB, Moskowitz RW, Jackson BA, Jafee MW. Studies of illness in the aged: the index of ADL: the standardized measure of biological and psychosocial function. JAMA 1963;185:914 –9. 20. Hiatt WR, Regensteiner JG, Wolfel EE, Ruff L, Brass EP. Carnitine and acylcarnitine metabolism during exercise in humans. Dependence on skeletal muscle metabolic state. J Clin Invest 1989;84:1167–73. 21. Brass EP, Hoppel CL, Hiatt WR. Effect of intravenous L-carnitine on carnitine homeostasis and fuel metabolism during exercise in humans. Clin Pharmacol Ther 1994;55:681–92. 22. Brass EP. Overview of coenzyme A metabolism and its role in cellular toxicity. Chem Biol Interact 1994;90:203–14.

1744

MALAGUARNERA ET AL

23. Brass EP, Hiatt WR. Carnitine metabolism during exercise. Life Sci 1994;54:1383–93. 24. Paradies G, Ruggiero FM, Petrosillo G, Gadaleta MN, Quagliariello E. Effect of aging and acetyl-L-carnitine on the activity of cytochrome oxidase and adenine nucleotide translocase in rat heart mitochondria. FEBS Lett 1994;350:213–5. 25. Paradies G, Petrosillo G, Gadaleta MN, Ruggiero FM. The effect of aging and acetyl-L-carnitine on the pyruvate transport and oxidation in rat heart mitochondria. FEBS Lett 1999;454:207–9. 26. Malaguarnera M, Di Mauro A, Gargante PM, Rampello L. L-carnitine reduces severity of physical and mental fatigue and improves daily activities in the elderly. South Med J 2006;99:315– 6. 27. Siliprandi N, Di Lisa F, Menabo R, Ciman M, Sartorelli L. Transport and functions of carnitine in muscles. J Clin Chem Clin Biochem 1990;28: 303– 6. 28. Vecchiet L, Di Lisa F, Pieralisi G, et al. Influence of L-carnitine administration on maximal physical exercise. Eur J Appl Physiol Occup Physiol 1990;61:486 –90. 29. Brevetti G, Angelini C, Rosa M, et al. Muscle carnitine deficiency in patients with severe peripheral vascular disease. Circulation 1991; 84:1490 –5. 30. Liu B, El Alaoui-Talibi Z, Clanachan AS, Schulz R, Lopaschuk GD.

31.

32. 33.

34. 35. 36.

Uncoupling of contractile function from mitochondrial TCA cycle activity and MVO2 during reperfusion of ischemic hearts. Am J Physiol 1996;270:H72– 80. Montgomery SA, Thal LJ, Amrein R. Meta-analysis of double blind randomized controlled clinical trials of acetyl-L-carnitine versus placebo in the treatment of mild cognitive impairment and mild Alzheimer’s disease. Int Clin Psychopharmacol 2003;18:61–71. Nalecz KA, Nalecz MJ. Carnitine–a known compound, a novel function in neural cells. Acta Neurobiol Exp (Wars) 1996;56:597– 609. Pettegrew JW, Levine J, McClure RJ. Acetyl-L-carnitine physicalchemical, metabolic, and therapeutic properties: relevance for its mode of action in Alzheimer’s disease and geriatric depression. Mol Psychiatry 2000;5:616 –32. Virmani A, Binienda Z. Role of carnitine esters in brain neuropathology. Mol Aspects Med 2004;25:533– 49. Rodrigues B, Xiang H, McNeill JH. Effect of L-carnitine treatment on lipid metabolism and cardiac performance in chronically diabetic rats. Diabetes 1988;37:1358 – 64. Liu J, Head E, Gharib AM, et al. Memory loss in old rats is associated with brain mitochondrial decay and RNA/DNA oxidation partial reversal by feeding acetyl-L-carnitine and/or R-␣-lipoic acid. Proc Natl Acad Sci U S A 2002;99:2356 – 61.

Intake of whole grains, refined grains, and cereal fiber measured with 7-d diet records and associations with risk factors for chronic disease1–3 PK Newby, Janice Maras, Peter Bakun, Denis Muller, Luigi Ferrucci, and Katherine L Tucker ABSTRACT Background: Research studies examining foods are important, because they account for biological interactions that might otherwise be lost in the analysis of individual nutrients. Single-nutrient studies are also needed to explore the mechanisms by which foods may be protective. Objective: Our objective was to examine associations between whole grains, refined grains, and cereal fiber and chronic disease risk factors. Design: In a cross-sectional analysis of participants in the Baltimore Longitudinal Study of Aging, associations between dietary intakes and risk factors were examined with multivariate linear regression analysis. Dietary intakes were assessed with 7-d dietary records and quantified in g/d. Results: Compared with subjects in the lowest quintile (Q1) of whole-grain intake, subjects in the highest quintile (Q5) had lower body mass index (BMI; in kg/m2; Q1: 25.5; Q5: 24.8; P for trend 쏝0.0001) and weight (Q1: 75.0 kg; Q5: 72.4 kg; P for trend ҃ 0.004) and smaller waist circumference (Q1: 87.4 cm; Q5: 85.0 cm; P for trend ҃ 0.002). Whole grains were also inversely associated with total cholesterol (P for trend ҃ 0.02), LDL cholesterol (P for trend ҃ 0.04), and 2-h glucose (P for trend ҃ 0.0006). Associations between cereal fiber and anthropometrics and plasma lipids were similar. In subgroup analyses, refined grains were positively associated with fasting insulin among women (P for trend ҃ 0.002). Conclusions: Similar associations of whole grains and cereal fiber with weight, BMI, waist circumference, plasma cholesterol, and 2-h glucose were observed, suggesting that cereal fiber and its constituents may in part mediate these relations. Refined grains were associated with fasting insulin among women but not men. Additional research should explore potential interaction effects with BMI, sex, age, and genes. Am J Clin Nutr 2007;86:1745–53. KEY WORDS Whole grains, refined grains, fiber, diet records, risk factors INTRODUCTION

Carbohydrate nutrition, including food sources, chemical structures, and physiologic properties, is an important area of research. The role of nondigestible polysaccharides (ie, fiber) in a healthy diet has been appreciated for many decades (1), leading to statements by both the American Dietetic Association (2) and the Council on Scientific Affairs (3) that fiber consumption is related to decreased risk of several diseases, including colon

cancer, heart disease, diabetes, diverticulosis, and obesity. More recently, significant associations were observed between wholegrain intakes and cardiovascular disease and stroke (4), cancer (5, 6), diabetes (7, 8), obesity (9), and the metabolic syndrome (10, 11). Refined grains are the counterpart to whole grains, but evidence is conflicting about associations with metabolic and anthropometric variables (10 –13), although some evidence shows that consumption of refined grains is associated with a risk of cancer (14, 15). Whole grains contain many bioactive components that might be responsible for their protective effect, including fiber, resistant starch, and oligosaccharides, as well as vitamins, minerals, phytate, phytoestrogens, and phytosterols (16). Research studies examining food groups such as whole grains are important, because they account for biological interactions that might otherwise be lost in an analysis of individual nutrients. However, studies examining single nutrients such as cereal fiber are also needed to increase our understanding of the mechanisms by which whole grains may be protective. Additional research is also needed to further examine whether intake of refined grains is associated with risk factors for chronic disease. The objective of this study was to examine associations of the intakes of whole grains, refined grains, and cereal fiber measured with 7-d dietary records and quantified in gram weights with selected risk factors for chronic disease among adults participating in the Baltimore Longitudinal Study on Aging (BLSA). An additional goal was to explore whether associations with risk factors were modified by sex or body mass index (BMI; in kg/m2). 1 From the Department of Pediatrics, Boston University School of Medicine, Boston Medical Center, and the Department of Epidemiology, Boston University School of Public Health, Boston, MA (PKN); the Jean Mayer US Department of Agriculture Human Nutrition Research Center on Aging at Tufts University, Boston, MA (JM, PB, and KLT); the National Institute on Aging, National Institutes of Health, Baltimore, MD (DM and LF). 2 Supported by the USDA contract 58-1950-7-707; the Intramural Research Program of the NIH, National Institute on Aging; and General Mills Bell Institute of Health and Nutrition. Data for these analyses were obtained from the Baltimore Longitudinal Study of Aging, a study performed by the National Institute on Aging. 3 No reprints available. Address correspondence to PK Newby, Department of Pediatrics, Boston University School of Medicine, Boston, MA 02118. E-mail: [email protected] ([email protected]). Received May 8, 2007. Accepted for publication August 1, 2007.

Am J Clin Nutr 2007;86:1745–53. Printed in USA. © 2007 American Society for Nutrition

1745

1746

NEWBY ET AL

SUBJECTS AND METHODS

Study population The BLSA is an open prospective cohort study that began in 1958 with the goals of studying the physical, mental, and emotional effects of aging among healthy, active persons; the original study design and data collection have been described in detail elsewhere (17). Briefly, initial study participants were white male community-dwelling volunteers 27– 88 y of age living in Baltimore, Maryland. The study protocol was expanded in 1978 to include women and minorities. Participants return approximately every 12–24 mo for repeated measurements (eg, height, weight, body composition analysis). Subjects were also invited to participate in the (optional) dietary assessment portion of the study, in which 7-d diet records were used to assess dietary intakes. The Institutional Review Boards of the Johns Hopkins Bayview Medical Research Center and the Gerontology Center approved the BLSA protocol, and all subjects gave written informed consent for their participation. Initially, 1572 persons completed at least one 7-d diet record during the course of the study. Of those, we excluded subjects who had not completed 욷4 d of the record, as well as those with implausible energy intakes or statistical outliers for total energy intake (쏝2510 or 쏜16 736 kJ/d for women; 쏝3347 and 쏜17 572 kJ/d for men) or who were missing information on age (n ҃ 53). Our baseline population therefore included 1516 participants. Because of the limited number of persons with both repeated dietary data and outcome measures, this analysis only included participants from the first visit at which participants had complete dietary and outcome data. Thus, for each outcome, our sample size varied somewhat, as follows: BMI and weight (n ҃ 1502), waist circumference (n ҃ 1404), total cholesterol (n ҃ 1444), HDL cholesterol (n ҃ 1029), LDL cholesterol (n ҃ 1025), triacylglycerols (n ҃ 1430), diastolic and systolic blood pressures (n ҃ 1464), fasting glucose (n ҃ 1324), 2-h glucose (n ҃ 882), fasting insulin (n ҃ 460), and 2-h insulin (n ҃ 455). Dietary assessment Dietary intake was assessed by 7-d dietary records, and subjects were instructed by trained dietitians how to assess portion size, weigh foods, and complete the records. Reports that detail the dietary collection methods and dietary intake in the BLSA population were published previously (18, 19). Food records were completed at home by the participant and sent back to the study center. Before 1993, subjects were given food models and a booklet of food pictures to help them assess portion size. Since 1993, subjects were given a portable scale to weigh food portions. Participants were contacted by telephone with any questions about their records. Dietary records were originally coded and entered into a nutrient database maintained by the BLSA, whereas diet records completed since 1993 were coded and entered into the Minnesota Nutrient Database at Tufts University. Dietary data collected before 1993 were then reentered in the Minnesota Nutrient Database, and nutrient intakes were back adjusted to correct for changes in the food supply (eg, nutrient content because of fortification of cereals) with data from the US Department of Agriculture to correspond to appropriate time intervals (20, 21).

Measurements of whole grains, refined grains, and cereal fiber For this study, we used all available dietary data from the BLSA to create a whole-grains database. In brief, all foods containing grains or mixed dishes with foods containing grains, either whole or refined, were identified and then assigned the best-matched pyramid code from the Pyramid Servings Database, a reference database of servings per 100 g from 30 food groups, including 3 grain groups (total grain, whole grain, and nonwhole grain). We then used the Continuing Survey of Food Intakes by Individuals 1994 –1996 recipe database to disaggregate mixed dishes that contained grains (either whole or refined) to obtain gram weights of individual ingredients. Some foods could not be disaggregated to the ingredient level, in which case the reference gram amount of grains was determined by multiplying by the pyramid definition of a serving (eg, a grain serving of one slice of bread contains 16 g of flour and a grain serving of one serving of cereal contains 28 g of flour). Once the reference database was completed, we calculated the absolute amount of grains consumed by multiplying the gram amount eaten by the reference values/100 for each day of the dietary record, and values for all days were summed and divided by the total number of days of dietary records to obtain an average intake. Cereal fiber was measured by summing the fiber content from cereal foods, which includes the starchy grains produced from grass plants. Thus, our quantification of cereal fiber included fiber from whole grains (eg, wheat, rice, corn, oats, etc) and all foods made from grains, including ready-to-eat and hot cereals, bread, pasta, crackers, sweet baked goods, and salty snacks. Whole grains, refined grains, and cereal fiber were adjusted for total energy intake with the nutrient residual method. Outcome assessment Our outcome variables included various risk factors for chronic disease. Anthropometric and clinical measurements were obtained by following standardized procedures (22) that have been fully described elsewhere (23). In summary, weight and height were measured for each subject at each visit, from which BMI was calculated. Waist circumference was measured with the use of an inelastic tape at the narrowest part of the torso at the end of expiration (23), roughly equivalent to the bottom of the ribcage for most persons. Blood pressure was measured from a sitting position with the use of a standard protocol (24). For lipid measurements, an antecubital venous blood sample was drawn from study subjects after an overnight fast. As previously described (25, 26), concentrations of triacylglycerols and total cholesterol were measured by enzymatic method (Abbott Laboratories ABA-200 ATC Biochromatic Analyzer; Irving, TX). HDL cholesterol was measured by the dextran sulfate– magnesium precipitation procedure (27), and LDL cholesterol was estimated by the Friedewald formula (28). As described previously, fasting plasma glucose and 2-h glucose concentrations were obtained with the use of an oral glucose tolerance test after overnight fasting, in which the glucose dose was 40 g/m2 body surface area, corresponding to an average dose of 78 g in men and 68 g in women (29). Plasma glucose was measured by the glucose oxidase method (Beckman Instruments, Fullerton, CA). Plasma insulin was measured in duplicate by radioimmunoassay (30).

GRAINS, CEREAL FIBER, CHRONIC DISEASE RISK FACTORS

Covariate assessment Race-ethnicity, physical activity, smoking, education, and vitamin supplement use were determined by questionnaire at the time dietary records were collected. Physical activity was measured by an adapted version of the Harvard Alumni questionnaire, which asked participants about all daily activities (eg, activities at home, work, and during recreation or sports). The amount of time spent for each activity was summed across all activities to determine the daily energy output per body weight (in kJ/kg) and was described previously (18, 32). Statistical analysis We first examined the relations between the intakes of whole grains, refined grains, and cereal fiber and the sample characteristics at the time of the first visit, as well as associations with nutrient intakes. We also explored associations between the 3 dietary variables with the use of Spearman correlation coefficients. Our main analyses used multivariate linear regression to estimate separately the relation between whole grains, refined grains, and cereal fiber with each of our outcome variables. As such, we built separate regression models for each dietary variable, and each was divided into quintiles of intake. For example, we fit models estimating the relation between whole grains (in quintiles) and each risk factor (BMI, weight, waist circumference, total cholesterol, HDL cholesterol, LDL cholesterol, diastolic blood pressure, systolic blood pressure, fasting glucose, 2-h glucose, fasting insulin, and 2-h insulin). Similar models were fit for refined grains and cereal fiber. We used a generalized linear model to estimate the least-squares means for each outcome, and each model was adjusted for age, sex, race, education, decade of visit, vitamin supplement use, total energy, percentage of energy from saturated fat, and alcohol. The whole-grain models were further adjusted for intakes of refined grains, and the refined-grain models were further adjusted for intakes of whole grains. Models that predicted lipid outcomes were further adjusted for BMI, use of lipid-lowering medication, and diagnosis of hypercholesterolemia. Models that predicted blood pressure were further adjusted for BMI, use of blood pressure–lowering medication, and diagnosis of hypertension. Models that predicted glucose and insulin outcomes were further adjusted for BMI, use of oral hypoglycemic medication, and diagnosis of diabetes. We also repeated the analyses for the above outcomes excluding subjects rather than adjusting for these variables in the analysis (ie, subjects with a diagnosis of diabetes were excluded from the glucose and insulin analyses) to remove the possibility of residual confounding. We created interaction terms to test whether the effects of the intakes of whole grains, refined grains, and cereal fiber on risk factors were modified by sex in all models and whether the effects were modified by BMI in our models that estimated fasting glucose, 2-h glucose, fasting insulin, and 2-h insulin. We performed a number of additional analyses to test the robustness of our models. First, we tested models for each relation with further adjustment for physical activity on a limited subset of subjects for whom such data were available. In addition, we tested all associations limiting our dataset to subjects entering the study in 1980 or later to reduce the potential of a cohort effect. We also added a quadratic term for age to our models to determine whether it improved model fit. All analyses were performed

1747

with the use of SAS for WINDOWS, version 9.1 (SAS Institute, Cary, NC). RESULTS

Age was positively related to the intakes of whole grains and cereal fiber (P for trend 쏝 0.0001) and inversely related to the intake of refined grains (P for trend ҃ 0.004) (Table 1). BMI was inversely related to whole-grain intake (P for trend ҃ 0.001), and subjects in the highest quintile of whole-grain intake had the lowest prevalence of overweight (39% compared with 54% in the lowest quintile; P ҃ 0.0003); similar associations were seen for cereal fiber intake. The highest percentages of women were found in the highest quintile of whole-grain intake (40% compared with 15% in the lowest quintile; P 쏝 0.0001) and cereal fiber intake (38% compared with 21% in the lowest quintile; P 쏝 0.0001) and in the lowest quintile of refined-grain intake (56% compared with 18% in the highest quintile; P 쏝 0.0001). Intake of whole grains was positively related to intakes of total energy, percentage of energy from carbohydrates, total fiber, folate, magnesium, vitamin E, and vitamin B-6 (P 쏝 0.0001 for all) and inversely related to percentage of energy from total and saturated fats, alcohol, and cholesterol (P 쏝 0.0001 for all); similar associations were observed with cereal fiber (P 쏝 0.0001 for all), with the exception of vitamin B-6 (P ҃ 0.20) (Table 2). Intake of refined grains was positively related to total energy intake and to percentage of energy from carbohydrates (P 쏝 0.0001 for both) and inversely related to percentage of energy from protein, total fat, saturated fat, fiber, alcohol, and magnesium (P 쏝 0.05 for all). In addition, whole grains were positively correlated with cereal fiber (r ҃ 0.77, P 쏝 0.0001) and negatively correlated with refined grains (r ҃ Ҁ0.18, P 쏝 0.0001) (data not shown). Whole grains were significantly related to each of the anthropometric outcomes in multivariate adjusted models (Table 3). Compared with subjects in the lowest quintile (Q1) of intake, subjects in the highest quintile (Q5) had lower BMI (Q1: 25.5 kg/m2; Q5: 24.8 kg/m2; P for trend 쏝0.0001) and weight (Q1: 75.0 kg; Q5: 72.6 kg; P for trend ҃ 0.004) and smaller waist circumference (Q1: 87.4 cm, Q5:҃ 85.0 cm; P for trend ҃ 0.002). Whole grains were also inversely associated with total cholesterol and LDL cholesterol (P for trend ҃ 0.02 and P for trend ҃ 0.04), as well as with 2-h glucose (P for trend ҃ 0.0006). Associations with blood pressure, insulin, and fasting glucose variables were not significant. No interaction was observed between sex and whole grains in any of our models, nor was an interaction observed between BMI and whole grains in our models for fasting glucose, 2-h glucose, fasting insulin, or 2-h insulin (P 쏜 0.05 for all interaction terms). Associations remained similar when further adjusted for physical activity. A significant interaction was observed between intake of refined grains and sex in relation to fasting insulin (P ҃ 0.04). When a stratified analysis was performed with the use of quartiles of refined-grain intake (rather than quintiles, to preserve statistical power), a positive association was observed among women (quartile 1: 57.0; quartile 5: 81.1; P for trend ҃ 0.002) (Figure 1), and the association remained significant and of similar magnitude when further adjusted for cereal fiber and physical activity, and when women with a diagnosis of diabetes were excluded from the analysis (data not shown). No association was seen among men (quartile 1: 73.9; quartile 4; 76.0; P for trend ҃

1748

NEWBY ET AL

TABLE 1 Sample characteristics of 1516 men and women participating in the Baltimore Longitudinal Study on Aging at the time of the first visit at which participants had complete dietary data1 Whole grains2 Sample characteristic Age (y)6 BMI (kg/m2)8 Weight (kg)8 Waist circumference (cm)8 Physical activity (kcal/kg)8 BMI 욷 25 (%) Hypertension (%) Female (%) White (%) Vitamin users (%) Current smokers (%) College graduate (%)

Refined grains3

Cereal fiber4

Q1

Q5

P5

Q1

Q5

P5

Q1

Q5

P5

52.4 앐 1.07 25.5 앐 0.2 74.9 앐 0.6 86.0 앐 0.7 13.8 앐 0.3 54 10 15 89 14 26 65

63.3 앐 0.9 24.5 앐 0.2 72.9 앐 0.7 83.5 앐 0.6 14.3 앐 0.4 39 10 40 86 38 4 71

쏝0.0001 0.001 0.01 0.0006 0.13 0.0003 0.35 쏝0.0001 0.26 쏝0.0001 쏝0.0001 0.37

59.4 앐 1.0 24.9 앐 0.2 73.5 앐 0.7 84.1 앐 0.6 13.3 앐 0.3 45 7 56 90 32 10 67

55.4 앐 1.0 25.2 앐 0.2 73.9 앐 0.7 85.6 앐 0.7 15.0 앐 0.3 48 7 18 87 31 12 70

0.004 0.51 0.78 0.10 0.0005 0.89 0.62 쏝0.0001 0.22 0.30 0.23 0.03

53.3 앐 1.0 25.7 앐 0.2 75.5 앐 0.7 86.1 앐 0.6 13.4 앐 0.3 55 9 21 87 22 22 68

63.8 앐 0.9 24.3 앐 0.2 72.2 앐 0.7 82.9 앐 0.6 14.7 앐 0.3 38 7 38 88 37 4 71

쏝0.0001 쏝0.0001 0.001 쏝0.0001 0.007 0.0003 0.06 쏝0.0001 0.06 0.001 쏝0.0001 0.004

Sample sizes differed because of missing data, as follows: BMI and weight (n ҃ 1500), waist circumference (n ҃ 1106), vitamin use (n ҃ 1185), education (n ҃ 1500), and physical activity (n ҃ 1028). Nutrients are adjusted for total energy intake with the use of the residual approach. Q, quintile. 2 Median intake: Q1, 0.68 g/d; Q5, 45.8 g/d. 3 Median intake: Q1, 39.0 g/d; Q5, 102.7 g/d. 4 Median intake: Q1, 2.2 g/d; Q5, 9.5 g/d. 5 Associations with continuous variables were examined with linear regression analysis (generalized linear model) and a test for trend in which subjects in each quintile were assigned the median value of intake. Associations with categorical variables were examined with a chi-square analysis. 6 Adjusted for sex. 7 x៮ 앐 SE (all such values). 8 Adjusted for age and sex. 1

0.48). A significant interaction with sex was also detected in the analysis of refined grains and 2-h insulin (P ҃ 0.03). However, associations among women (quartile 1: 289 mmol/L; quartile 5: 367 mmol/L; P for trend ҃ 0.11) and men (quartile 1: 439 mmol/L; quartile 5: 528 mmol/L; P for trend ҃ 0.29) were not

significant in stratified analyses. No significant relations between refined grains and any of the other risk factors (BMI, weight, waist circumference, total cholesterol, HDL cholesterol, LDL cholesterol, triacylglycerols, systolic blood pressure, diastolic blood pressure, fasting glucose, 2-h glucose, fasting insu-

TABLE 2 Dietary intakes of 1516 men and women participating in the Baltimore Longitudinal Study on Aging at the time of the first visit at which participants had complete dietary data1 Whole grains2 Nutrient intake Energy (kJ)6 Carbohydrate (% of energy)8 Protein (% of energy)8 Total fat (% of energy)8 Saturated fat (% of energy)8 Total fiber (g)8 Alcohol (g)8 Cholesterol (mg)8 Folate (mg)8 Magnesium (mg)8 Vitamin E (mg)8 Vitamin B-6 (mg)8

Q1

Q5

7932 앐 1117 41.4 앐 0.5 16.1 앐 0.2 38.9 앐 0.4 13.7 앐 0.2 13.9 앐 0.5 15.6 앐 1.0 224 앐 2 315 앐 134 243 앐 5 24.8 앐 6.1 3.0 앐 0.4

9125 앐 109 51.7 앐 0.5 16.2 앐 0.2 32.2 앐 0.4 10.8 앐 0.2 24.7 앐 0.5 8.1 앐 1.1 206 앐 2 525 앐 13 382 앐 5 63.1 앐 5.4 6.2 앐 0.6

Refined grains3 P5

Q1

쏝0.0001 7306 앐 102 쏝0.0001 43.1 앐 0.65 0.29 16.7 앐 0.2 쏝0.0001 36.8 앐 0.5 쏝0.0001 12.7 앐 0.2 쏝0.0001 19.8 앐 0.7 쏝0.0001 17.3 앐 1.2 쏝0.0001 218 앐 3 쏝0.0001 435 앐 17 쏝0.0001 322 앐 7 쏝0.0001 39.7 앐 5.5 쏝0.0001 3.7 앐 0.6

Cereal fiber4

Q5

P5

Q1

Q5

P5

9868 앐 102 50.5 앐 0.5 15.2 앐 0.2 34.3 앐 0.4 11.9 앐 0.2 18.6 앐 0.5 11.2 앐 0.9 218 앐 2 380 앐 14 291 앐 6 30.1 앐 5.4 3.1 앐 0.6

쏝0.0001 쏝0.0001 쏝0.0001 쏝0.0001 0.001 0.03 쏝0.0001 0.08 0.33 쏝0.001 0.54 0.89

8879 앐 102 40.1 앐 0.5 16.6 앐 0.2 39.2 앐 0.4 13.8 앐 0.2 14.5 앐 0.5 17.5 앐 0.9 227 앐 2 339 앐 13 254 앐 5 26.9 앐 5.5 3.0 앐 0.6

9537 앐 103 52.5 앐 0.5 16.2 앐 0.2 31.8 앐 0.4 10.8 앐 0.2 25.9 앐 0.5 7.4 앐 0.9 203 앐 2 531 앐 13 397 앐 5 60.9 앐 5.5 3.9 앐 0.6

쏝0.0001 쏝0.0001 0.84 쏝0.0001 쏝0.0001 쏝0.0001 쏝0.0001 쏝0.0001 쏝0.0001 쏝0.0001 쏝0.0001 0.20

n ҃ 230 or 231 per quintile (Q). Median intake: Q1, 0.68 g/d; Q5, 45.8 g/d. 3 Median intake: Q1, 39.7 g/d; Q5, 102.7 g/d. 4 Median intake: Q1, 2.2 g/d; Q5, 9.5 g/d. 5 Associations were examined with linear regression analysis (generalized linear model) and a test for trend in which subjects in each quintile were assigned the median value of intake. 6 Adjusted for age and sex. 7 x៮ 앐 SE (all such values). 8 Adjusted for age, sex, and total energy. Intakes from dietary supplements are not included. 1 2

1749

GRAINS, CEREAL FIBER, CHRONIC DISEASE RISK FACTORS

TABLE 3 Chronic disease risk factors according to quintile (Q) of whole-grain intakes among men and women participating in the Baltimore Longitudinal Study of Aging Whole-grain intake1 Risk factor BMI (kg/m2) [n ҃ 1502] Simple adjusted2 Multivariate adjusted4 Weight (kg) [n ҃ 1502] Simple adjusted2 Multivariate adjusted4 Waist circumference (cm) [n ҃ 1404] Simple adjusted2 Multivariate adjusted4 Total cholesterol (mmol/L) [n ҃ 1444] Simple adjusted2 Multivariate adjusted4,5 HDL cholesterol (mmol/L) [n ҃ 1029] Simple adjusted2 Multivariate adjusted4,5 LDL cholesterol (mmol/L) [n ҃ 1025] Simple adjusted2 Multivariate adjusted4,5 Triacylglycerols (mmol/L) [n ҃ 1430] Simple adjusted2 Multivariate adjusted4,5 Diastolic blood pressure (mm Hg) [n ҃ 1464] Simple adjusted2 Multivariate adjusted4,6 Systolic blood pressure (mm Hg) [n ҃ 1464] Simple adjusted2 Multivariate adjusted4,6 Fasting glucose (mmol/L) [n ҃ 1324] Simple adjusted2 Multivariate adjusted4,7 2-h Glucose (mmol/L) [n ҃ 882] Simple adjusted2 Multivariate adjusted4,7 Fasting insulin (mmol/L) [n ҃ 460] Simple adjusted2 Multivariate adjusted4,7 2-h Insulin (mmol/L) [n ҃ 455] Simple adjusted2 Multivariate adjusted4,7

Q1

Q2

Q3

Q4

Q5

P for trend

25.7 앐 0.23 25.5 앐 0.2

25.7 앐 0.2 25.7 앐 0.2

25.0 앐 0.2 25.0 앐 0.2

24.7 앐 0.2 24.7 앐 0.2

24.2 앐 0.2 24.8 앐 0.2

쏝0.0001 쏝0.0001

75.4 앐 0.7 75.0 앐 0.7

75.0 앐 0.6 75.1 앐 0.6

73.8 앐 0.6 73.7 앐 0.6

72.0 앐 0.6 73.0 앐 0.6

72.2 앐 0.7 72.6 앐 0.7

0.0002 0.004

87.8 앐 0.6 87.4 앐 0.6

87.7 앐 0.6 87.6 앐 0.6

85.8 앐 0.6 85.7 앐 0.6

85.6 앐 0.6 85.7 앐 0.6

84.4 앐 0.6 85.0 앐 0.6

쏝0.0001 0.002

5.70 앐 0.06 5.71 앐 0.06

5.64 앐 0.06 5.64 앐 0.05

5.52 앐 0.06 5.52 앐 0.05

5.51 앐 0.06 5.50 앐 0.05

5.49 앐 0.06 5.49 앐 0.06

0.01 0.02

1.26 앐 0.02 1.27 앐 0.02

1.27 앐 0.02 1.28 앐 0.02

1.29 앐 0.02 1.29 앐 0.02

1.25 앐 0.02 1.25 앐 0.02

1.23 앐 0.02 1.22 앐 0.02

0.25 0.07

3.15 앐 0.06 3.16 앐 0.06

3.00 앐 0.06 3.02 앐 0.06

2.98 앐 0.06 2.99 앐 0.06

2.97 앐 0.06 2.98 앐 0.06

3.00 앐 0.06 2.96 앐 0.06

0.14 0.04

1.26 앐 0.04 1.23 앐 0.05

1.30 앐 0.04 1.25 앐 0.05

1.20 앐 0.04 1.21 앐 0.05

1.15 앐 0.04 1.18 앐 0.05

1.12 앐 0.05 1.16 앐 0.05

0.005 0.22

79.8 앐 0.6 79.8 앐 0.6

80.8 앐 0.5 80.8 앐 0.6

78.2 앐 0.5 78.2 앐 0.6

80.6 앐 0.5 80.6 앐 0.6

79.1 앐 0.6 79.2 앐 0.7

0.04 0.42

130.9 앐 1.1 129.2 앐 1.0

130.7 앐 1.1 130.0 앐 0.9

125.6 앐 1.1 126.9 앐 0.9

130.6 앐 1.1 131.1 앐 0.9

127.8 앐 1.1 128.3 앐 1.0

0.12 0.79

5.52 앐 0.06 5.49 앐 0.06

5.62 앐 0.06 5.58 앐 0.06

5.51 앐 0.06 5.51 앐 0.06

5.45 앐 0.06 5.51 앐 0.06

5.48 앐 0.06 5.49 앐 0.06

0.21 0.57

8.34 앐 0.18 8.24 앐 0.17

7.62 앐 0.18 7.59 앐 0.16

7.66 앐 0.18 7.63 앐 0.16

7.79 앐 0.18 7.94 앐 0.16

7.32 앐 0.18 7.32 앐 0.17

0.002 0.006

74.5 앐 4.0 71.6 앐 3.9

70.5 앐 3.9 70.0 앐 3.7

72.2 앐 3.9 73.4 앐 3.7

72.1 앐 3.9 71.3 앐 3.7

68.9 앐 4.0 71.8 앐 4.1

0.41 0.90

534 앐 39.2 479 앐 38.0

399 앐 38.8 416 앐 36.0

357 앐 38.8 360 앐 35.7

417 앐 38.8 421 앐 35.7

381 앐 39.2 414 앐 39.1

0.04 0.43

1 Quintiles were developed separately for each outcome because of differences in sample sizes. Medan values for Q1 and Q5 were as follows: BMI and weight, 0.65 and 46.0 g/d, respectively; waist circumference, 0.94 and 49.3 g/d, respectively; total cholesterol, 0.63 and 45.6 g/d, respectively; HDL cholesterol, 4.1 and 54.0 g/d, respectively; LDL cholesterol and triacylglycerols, 3.9 and 54.8 g/d, respectively; blood pressure, 0.62 and 45.4 g/d, respectively; fasting glucose, 0.56 and 45.4 g/d, respectively; 2-h glucose, 1.1 and 50.6 g/d, respectively; fasting insulin, 2.2 and 51.5 g/d, respectively; 2-h insulin, 2.4 and 51.7 g/d, respectively. 2 Adjusted for age, age2, sex, total energy, and decade of visit. 3 x៮ 앐 SEM (all such values) estimated with least-squares means from a linear regression analysis (generalized linear model). 4 Adjusted for age, age2, sex, total energy, decade of visit, race, education, vitamin supplement use, smoking, and percentage of energy from saturated fat, alcohol, and refined grains. 5 With additional adjustment for BMI, use of lipid-lowering medication, and hypercholesterolemia. 6 With additional adjustment for BMI, use of blood pressure–lowering medication, and hypertension. 7 With additional adjustment for BMI, use of oral hypoglycemic medication, and diabetes.

lin, and 2-h insulin) were observed (P 쏜 0.05 for all; data not shown). Similar to the findings with whole grains, cereal fiber was inversely associated with BMI (P for trend 쏝0.0001), weight, (P for trend ҃ 0.0004), waist circumference (P for trend 쏝0.0001), and total cholesterol (P for trend ҃ 0.005) (Table 4), and the

associations remained significant when further adjusted for physical activity (data not shown). Cereal fiber was also inversely associated with 2-h glucose (Q1: 8.05 mmol/L; Q5: 6.48 mmol/L; P for trend ҃ 0.02) in the multivariate-adjusted model but was no longer significant when further adjusted for physical activity (P for trend ҃ 0.09). No interaction with sex or BMI was

1750

NEWBY ET AL 100

Women Men

90

80

Fasting insulin (mmol/L)

70

60

50

40

30

20 P for interaction = 0.04 Women: P for trend = 0.002 Men: P for trend = 0.48

10

0 Q1

Q2

Q3

Q4

Quartile (Q) of refined grain intake

FIGURE 1. Association between quartiles of intake of refined grains (in g/d) and fasting insulin (in mmol/L) among 183 women and 277 men participating in the Baltimore Longitudinal Study of Aging. Because of differences in intakes, intake quartiles (Q1–Q4) were derived separately for women and men, as follows: women (Q1: 32.5; Q2: 46.7; Q3: 56.1; Q4: 73.7) and men (Q1: 49.4; Q2: 64.8; Q3; 80.8; Q4: 102.0). Values are x¯ 앐 SEM, estimated with the least-squares means from a linear regression analysis (generalized linear model). Models are adjusted for age, age2, total energy, decade of visit, race, education, vitamin supplement use, BMI, smoking, percentage of energy from saturated fat, alcohol, whole-grain intake, and diabetes diagnosis.

observed in any of the models. A quadratic term for age improved model fit (P 쏝 0.001) and was retained in all models. DISCUSSION

One of the main objectives of this study was to examine independent associations between whole grains and cereal fiber with risk factors for chronic disease. We found that whole grains were significantly inversely associated with anthropometric variables (weight, BMI, and waist circumference), plasma lipid measures (total cholesterol and LDL cholesterol), and 2-h glucose. Cereal fiber was also related to these variables, and the magnitude of effects was similar. The aleurone layer and cell walls of whole grains, largely composed of nonstarch polysaccharides (32), also include many additional bioactive components such as phytochemicals (eg, phytic and phenolic acids), phytoestrogens (eg, lignans and isoflavones), antioxidants, vitamins, and minerals (eg, potassium, magnesium, and selenium) (33). Indeed, many of these bioactive constituents are insoluble and bound to cell wall materials (34), and they act independently and synergistically to confer many protective health effects (16, 35–38). Our findings suggest that cereal fiber, its bioactive components, or both may mediate the associations with whole grains examined in this study. Several other studies have shown similar inverse associations with anthropometric variables for whole grains (11, 12, 39) and cereal fiber (40, 41). Fiber contributes to satiation, satiety, and the secretion of gut hormones (9), thereby affecting body weight and body composition through its effect on energy intake. McKeown et al (10) also observed comparable associations for whole

grains (odds ratio: 0.67; 95% CI: 0.48, 0.91) and cereal fiber (odds ratio: 0.62; 95% CI: 0.45, 0.86) with risk of metabolic syndrome. Unlike other studies (11, 39, 41), our study showed significant inverse associations between whole grains and cereal fibers with total cholesterol, and whole grains were also inversely associated with LDL cholesterol. Jensen et al (42) also reported a significant inverse association between whole grains and total cholesterol (P ҃ 0.02). The relation between dietary fiber and cholesterol is thought to be mainly due to soluble fibers rather than insoluble fibers (43, 44), but phytosterols (16) and phytoestrogens (35) may also affect cholesterol metabolism. No significant associations were observed between whole grains and blood pressure in our study, as seen in 2 other cross-sectional studies (11, 39), but a small intervention study observed a decline in blood pressure after a whole-grain diet (45). Cereal fiber intake was also unrelated to blood pressure in our study. Although 2 meta-analyses reported a small but significant inverse association between total fiber and blood pressure (46, 47), neither specifically addressed the role of cereal fiber, so whether cereal fiber specifically is associated with blood pressure requires more research. Antioxidants and minerals present in whole grains and cereal fiber may contribute to insulin sensitivity (9, 48), and fiber may also be related to insulin sensitivity through effects on delayed gastric emptying. We observed a significant inverse association between whole grains and cereal fiber and 2-h glucose, and the magnitude of the effects was quite similar for both variables. McKeown et al (39) observed null associations between whole grains and 2-h glucose and 2-h insulin in a cross-sectional study of persons in the Framingham Offspring Study, and whole grains

1751

GRAINS, CEREAL FIBER, CHRONIC DISEASE RISK FACTORS TABLE 4 Chronic disease risk factors according to quintile (Q) of cereal fiber intake at the time of the first visit among men and women participating in the Baltimore Longitudinal Study of Aging Cereal fiber intake1 Risk factor BMI (kg/m2) [n ҃ 1502] Simple adjusted2 Multivariate adjusted4 Weight (kg) [n ҃ 1502] Simple adjusted2 Multivariate adjusted4 Waist circumference (cm) [n ҃ 1404] Simple adjusted2 Multivariate adjusted4 Total cholesterol (mmol/L) [n ҃ 1444] Simple adjusted2 Multivariate adjusted4,5 HDL cholesterol (mmol/L) [n ҃ 1029] Simple adjusted2 Multivariate adjusted4,5 LDL cholesterol (mmol/L) [n ҃ 1025] Simple adjusted2 Multivariate adjusted4,5 Triacylglycerols (mmol/L) [n ҃ 1430] Simple adjusted2 Multivariate adjusted4,5 Diastolic blood pressure (mm Hg) [n ҃ 1464] Simple adjusted2 Multivariate adjusted4,6 Systolic blood pressure (mm Hg) [n ҃ 1464] Simple adjusted2 Multivariate adjusted4,6 Fasting glucose (mmol/L) [n ҃ 1324] Simple adjusted2 Multivariate adjusted4,7 2-h Glucose (mmol/L) [n ҃ 882] Simple adjusted2 Multivariate adjusted4,7 Fasting insulin (mmol/L) [n ҃ 460] Simple adjusted2 Multivariate adjusted4,7 2-h Insulin (mmol/L) [n ҃ 455] Simple adjusted2 Multivariate adjusted4,7

Q1

Q2

Q3

Q4

Q5

P for trend

25.9 앐 0.23 25.7 앐 0.2

25.3 앐 0.2 25.2 앐 0.2

24.4 앐 0.2 24.3 앐 0.2

24.8 앐 0.2 24.8 앐 0.2

24.1 앐 0.2 24.3 앐 0.2

쏝0.0001 쏝0.0001

76.0 앐 0.6 75.6 앐 0.7

74.3 앐 0.6 74.1 앐 0.6

74.4 앐 0.6 74.3 앐 0.6

73.4 앐 0.6 73.0 앐 0.6

71.4 앐 0.7 71.4 앐 0.8

쏝0.0001 0.0004

87.9 앐 0.6 87.4 앐 0.6

87.2 앐 0.6 86.9 앐 0.6

87.1 앐 0.6 87.0 앐 0.6

85.7 앐 0.6 86.0 앐 0.6

83.5 앐 0.6 84.1 앐 0.6

쏝0.0001 쏝0.0001

5.73 앐 0.06 5.73 앐 0.06

5.61 앐 0.06 5.60 앐 0.06

5.53 앐 0.06 5.51 앐 0.06

5.57 앐 0.06 5.57 앐 0.06

5.43 앐 0.07 5.44 앐 0.06

0.001 0.005

1.23 앐 0.02 1.23 앐 0.02

1.28 앐 0.02 1.30 앐 0.02

1.28 앐 0.02 1.27 앐 0.02

1.25 앐 0.02 1.25 앐 0.02

1.26 앐 0.02 1.25 앐 0.02

0.91 0.59

3.11 앐 0.06 3.13 앐 0.06

3.07 앐 0.06 3.07 앐 0.06

3.00 앐 0.06 3.01 앐 0.06

2.91 앐 0.06 2.90 앐 0.06

3.00 앐 0.06 2.99 앐 0.06

0.09 0.07

1.27 앐 0.05 1.24 앐 0.05

1.28 앐 0.04 1.26 앐 0.04

1.18 앐 0.04 1.16 앐 0.04

1.20 앐 0.04 1.23 앐 0.04

1.11 앐 0.05 1.15 앐 0.05

0.04 0.12

80.8 앐 0.7 79.5 앐 0.8

80.7 앐 0.7 79.8 앐 0.6

79.4 앐 0.7 79.1 앐 0.7

79.6 앐 0.7 80.8 앐 0.7

77.8 앐 0.8 79.2 앐 0.8

0.009 0.90

130.8 앐 1.1 128.7 앐 1.0

130.1 앐 1.1 129.0 앐 0.9

128.1 앐 1.0 127.7 앐 0.9

128.9 앐 1.1 130.6 앐 0.9

127.6 앐 1.1 129.7 앐 1.0

0.06 0.27

5.64 앐 0.06 5.55 앐 0.05

5.49 앐 0.06 5.48 앐 0.05

5.54 앐 0.06 5.53 앐 0.05

5.44 앐 0.06 5.49 앐 0.05

5.47 앐 0.06 5.52 앐 0.05

0.12 0.95

8.23 앐 0.18 8.05 앐 0.21

8.03 앐 0.18 7.94 앐 0.20

7.63 앐 0.18 7.72 앐 0.19

7.44 앐 0.18 7.55 앐 0.20

7.40 앐 0.18 6.48 앐 0.21

0.0008 0.02

70.1 앐 4.0 68.9 앐 4.

73.6 앐 3.9 72.2 앐 3.8

72.4 앐 3.9 73.0 앐 3.7

71.5 앐 4.0 71.3 앐 3.8

70.6 앐 4.0 73.0 앐 4.0

0.60 0.68

492 앐 39.5 438 앐 38.8

469 앐 38.9 477 앐 36.2

387 앐 38.7 404 앐 35.8

366 앐 39.3 356 앐 36.3

374 앐 39.1 413 앐 38.2

0.02 0.33

1 Quintiles were developed separately for each outcome because of differences in sample sizes. Values for Q1 and Q5 were as follows: BMI and weight, 2.19 and 9.51 g/d, respectively; waist circumference, 2.31 and 9.74 g/d, respectively; total cholesterol, 2.20 and 9.56 g/d, respectively; HDL and LDL cholesterol, 2.90 and 11.49 g/d, respectively; triacylglycerols, 2.20 and 9.58 g/d, respectively; blood pressure, 2.19 and 9.43 g/d, respectively; fasting glucose, 2.27 and 9.67 g/d, respectively; 2-h glucose, 2.39 and 10.42 g/d, respectively; fasting insulin and 2-h insulin, 2.54 and 11.63 g/d, respectively. 2 Adjusted for age, age2, sex, total energy, and decade of visit. 3 x៮ 앐 SEM (all such values) estimated with least-squares means from a linear regression analysis (generalized linear model). 4 Adjusted for age, age2, sex, total energy, decade of visit, race, education, vitamin supplement use, smoking, and percentage of energy from saturated fat and alcohol. 5 With additional adjustment for BMI, use of lipid-lowering medication, and hypercholesterolemia. 6 With additional adjustment for BMI, use of blood pressure–lowering medication, and hypertension. 7 With additional adjustment for BMI, use of oral hypoglycemic medication, and diabetes.

were also unrelated to fasting insulin in 2 large cohort studies (42). Although Lairon et al (41) saw no significant association between whole grains and fasting glucose, Sahyoun et al (11) showed an inverse dose-response relation (P for trend ҃ 0.01). A small crossover study among overweight persons reported improved insulin sensitivity after an intervention with cereal fiber

(49), and several other studies have shown significant associations between whole grains and plasma glucose (39) and insulin sensitivity (50, 51) among subjects with higher BMI, suggesting effect modification. Inconsistent findings on associations of whole grains or cereal fiber in relation to insulin or glucose measures may be due to differences in study design and dietary

1752

NEWBY ET AL

assessment methods, inadequate testing of interaction effects, or chance. We observed a significant positive association between refined grains and fasting insulin among women but not men. Interaction effects are more likely to occur by chance and often are not reproducible (52), so our findings need to be interpreted with caution. However, our result is supported by recent results from the Hypertension Genetic Epidemiology Network study, which found a quantitative trait locus influencing fasting insulin in female, but not male, subjects (53). Refined grains are major contributors to glycemic load, and several reviews support a positive association between glycemic load and insulin resistance or glycemic control (54 –56). That the percentage of energy from total carbohydrate in the highest quintile of refined grains (50.5%) and whole grains (51.7%) is so similar underscores the importance of differentiating between whole and refined foods when making dietary choices. Indeed, a small randomized crossover feeding study among overweight subjects with hyperinsulinemia observed 10% lower fasting insulin when consumption with refined grains were replaced with whole grains (57), as seen in a trial that included a reduction in refined grains as part of a comprehensive dietary intervention (58, 59). Note also that we observed a significant interaction between refined grains and sex in relation to 2-h insulin. Although the observed effects were in the expected direction, we likely had inadequate power to achieve statistical significance in stratified analyses. Refined grains were not significantly associated with any other risk factor in this study. As mentioned, findings about refined grains and risk factors for chronic disease are inconsistent. Decreasing intake of refined grains was associated with a smaller weight change in one study (12), whereas increasing intake of refined-grain breakfast cereals was inversely related to weight gain and BMI in another (13). Refined grains were positively associated with fasting glucose among older adults (11) and hypertriglyceridemic waist phenotype (60). Refined grains were also positively associated with risk of metabolic syndrome in 2 cross-sectional studies (10, 11), but neither showed associations with individual risk factors. Those results and ours suggest that the effect of refined grains on measures of insulin and glucose may be modified by age (11), BMI (4, 33), and sex, probably due to differences in glucose tolerance and insulin sensitivity. There are several limitations to our study. First, our study is cross-sectional; thus, our results do not allow us to make inferences about the directions of our associations and may be obfuscated by reverse causation. Although a longitudinal study would be stronger, our data are limited by the small numbers of subjects who have repeated measures of both dietary and risk factor data. An additional limitation is that our study sample, which is selected from an open cohort study, includes subjects across several decades. However, a cohort effect was not observed in an earlier study in this population (19). In addition, our models were adjusted for decade of the visit, and results remained similar when we limited our study to data from 1980 or later. Our study has a small sample size for some of our outcome variables; thus, we probably had insufficient power to detect and explore potential interactions. Finally, our study did not include markers of genetic variability, which are increasingly understood to modify the effect of diet on disease risk factors such as plasma lipids (61). In conclusion, our study shows significant and similar associations of whole grains and cereal fiber with weight, BMI, waist

circumference, and total cholesterol. Refined grains were positively associated with fasting insulin among women but not men. Longitudinal studies are needed to reproduce these findings, and special attention should be dedicated to exploring potential interactions with BMI, sex, age, and genes. We thank Nicola McKeown, PhD, for helpful comments on this manuscript. The author’s responsibilities were as follows—PKN: was responsible for the design and analysis for this report and drafted the manuscript; JM and PB: created the database for this study; DM and LF: contributed to data collection for the Baltimore Longitudinal Study of Aging; KLT: contributed to the database development and oversaw the HNRCA collaboration with the BLSA. All authors made critical comments during the preparation of the manuscript and fully accept responsibility for the work. None of the authors had a financial interest or professional or personal affiliation that compromises the scientific integrity of this work.

REFERENCES 1. Bing FC. Dietary fiber–in historical perspective. J Am Diet Assoc 1976; 69:498 –505. 2. Marlett JA, McBurney MI, Slavin JL. Position of the American Dietetic Association: health implications of dietary fiber. J Am Diet Assoc 2002; 102:993–1000. 3. Dietary fiber and health. Council on Scientific Affairs. JAMA 1989; 262:542– 6. 4. Flight I, Clifton P. Cereal grains and legumes in the prevention of coronary heart disease and stroke: a review of the literature. Eur J Clin Nutr 2006;60:1145–59. 5. Jacobs DR Jr, Marquart L, Slavin J, Kushi LH. Whole-grain intake and cancer: an expanded review and meta-analysis. Nutr Cancer 1998;30: 85–96. 6. Chatenoud L, Tavani A, La Vecchia C, et al. Whole grain food intake and cancer risk. Int J Cancer 1998;77:24 – 8. 7. Venn BJ, Mann JI. Cereal grains, legumes and diabetes. Eur J Clin Nutr 2004;58:1443– 61. 8. Murtaugh MA, Jacobs DR Jr, Jacob B, Steffen LM, Marquart L. Epidemiological support for the protection of whole grains against diabetes. Proc Nutr Soc 2003;62:143–9. 9. Koh-Banerjee P, Rimm EB. Whole grain consumption and weight gain: a review of the epidemiological evidence, potential mechanisms and opportunities for future research. Proc Nutr Soc 2003;62:25–9. 10. McKeown NM, Meigs JB, Liu S, Saltzman E, Wilson PW, Jacques PF. Carbohydrate nutrition, insulin resistance, and the prevalence of the metabolic syndrome in the Framingham Offspring Cohort. Diabetes Care 2004;27:538 – 46. 11. Sahyoun NR, Jacques PF, Zhang XL, Juan W, McKeown NM. Wholegrain intake is inversely associated with the metabolic syndrome and mortality in older adults. Am J Clin Nutr 2006;83:124 –31. 12. Liu S, Willett WC, Manson JE, Hu FB, Rosner B, Colditz G. Relation between changes in intakes of dietary fiber and grain products and changes in weight and development of obesity among middle-aged women. Am J Clin Nutr 2003;78:920 –7. 13. Bazzano LA, Song Y, Bubes V, Good CK, Manson JE, Liu S. Dietary intake of whole and refined grain breakfast cereals and weight gain in men. Obes Res 2005;13:1952– 60. 14. Slattery ML, Potter JD, Coates A, et al. Plant foods and colon cancer: an assessment of specific foods and their related nutrients (United States). Cancer Causes Control 1997;8:575–90. 15. Levi F, Pasche C, Lucchini F, Chatenoud L, Jacobs DR Jr, La Vecchia C. Refined and whole grain cereals and the risk of oral, oesophageal and laryngeal cancer. Eur J Clin Nutr 2000;54:487–9. 16. Slavin J. Why whole grains are protective: biological mechanisms. Proc Nutr Soc 2003;62:129 –34. 17. Shock N, Greulich R, Andres R, et al. Normal human aging: the Baltimore Longitudinal Study of Aging. Washington, DC: US Government Printing Office, 1984. 18. McGandy R, Barrows C, Spanias A, Meredith A, Stone J, Norris A. Nutrient intakes and energy expenditure in men of different ages. J Gerontol 1966;21:581–7. 19. Elahi VK, Elahi D, Andres R, Tobin JD, Butler MG, Norris AH. A

GRAINS, CEREAL FIBER, CHRONIC DISEASE RISK FACTORS

20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39.

40.

longitudinal study of nutritional intake in men. J Gerontol 1983;38:162– 80. US Department of Agriculture. Nutrient Database for Standard Reference, releases 10 and 11. Washington, DC: US Government Printing Office, 1993. Watt B, Merrill A. Composition of foods: raw, processed, prepared. Washington, DC: US Department of Agriculture, 1963. Lohman TG, Roche AF, Martorell R. Anthropometric standardization reference manual. Champaign, IL: Human Kinetics Books, 1988. Shimokata H, Tobin JD, Muller DC, Elahi D, Coon PJ, Andres R. Studies in the distribution of body fat: I. Effects of age, sex, and obesity. J Gerontol 1989;44:M66 –73. Pearson JD, Morrell CH, Brant LJ, Landis PK, Fleg JL. Age-associated changes in blood pressure in a longitudinal study of healthy men and women. J Gerontol A Biol Sci Med Sci 1997;52:M177– 83. Iwao S, Iwao N, Muller DC, Elahi D, Shimokata H, Andres R. Effect of aging on the relationship between multiple risk factors and waist circumference. J Am Geriatr Soc 2000;48:788 –94. Iwao N, Iwao S, Muller DC, Elahi D, Shimokata H, Andres R. A test of recently proposed BMI standards with respect to old age. Aging (Milano) 2000;12:461–9. Warnick GR, Benderson J, Albers JJ. Dextran sulfate-Mg2ѿ precipitation procedure for quantitation of high-density-lipoprotein cholesterol. Clin Chem 1982;28:1379 – 88. Friedewald WT, Levy RI, Fredrickson DS. Estimation of the concentration of low-density lipoprotein cholesterol in plasma, without use of the preparative ultracentrifuge. Clin Chem 1972;18:499 –502. Rodriguez A, Muller DC, Engelhardt M, Andres R. Contribution of impaired glucose tolerance in subjects with the metabolic syndrome: Baltimore Longitudinal Study of Aging. Metabolism 2005;54:542–7. Soeldner JS, Slone D. Critical variables in the radioimmunoassay of serum insulin using the double antibody technique. Diabetes 1965;14: 771–9. Shimokata H, Muller D, Fleg J, Sorkin J, Ziemba A, Andres R. Age as independent determinant of glucose tolerance. Diabetes 1991;40:44 –51. Selvendran RR. The plant cell wall as a source of dietary fiber: chemistry and structure. Am J Clin Nutr 1984;39:320 –37. Truswell AS. Cereal grains and coronary heart disease. Eur J Clin Nutr 2002;56:1–14. Adom KK, Liu RH. Antioxidant activity of grains. J Agric Food Chem 2002;50:6182–7. Setchell KD, Cassidy A. Dietary isoflavones: biological effects and relevance to human health. J Nutr 1999;129(suppl):758S– 67S. Slavin JL, Jacobs D, Marquart L, Wiemer K. The role of whole grains in disease prevention. J Am Diet Assoc 2001;101:780 –5. Thompson LU. Antioxidants and hormone-mediated health benefits of whole grains. Crit Rev Food Sci Nutr 1994;34:473–97. Flight I, Clifton P. Cereal grains and legumes in the prevention of coronary heart disease and stroke: a review of the literature. Eur J Clin Nutr 2006;60:1145–59. McKeown NM, Meigs JB, Liu S, Wilson PW, Jacques PF. Whole-grain intake is favorably associated with metabolic risk factors for type 2 diabetes and cardiovascular disease in the Framingham Offspring Study. Am J Clin Nutr 2002;76:390 – 8. Koh-Banerjee P, Chu NF, Spiegelman D, et al. Prospective study of the association of changes in dietary intake, physical activity, alcohol consumption, and smoking with 9-y gain in waist circumference among 16 587 US men. Am J Clin Nutr 2003;78:719 –27.

1753

41. Lairon D, Arnault N, Bertrais S, et al. Dietary fiber intake and risk factors for cardiovascular disease in French adults. Am J Clin Nutr 2005;82: 1185–94. 42. Jensen MK, Koh-Banerjee P, Franz M, Sampson L, Gronbaek M, Rimm EB. Whole grains, bran, and germ in relation to homocysteine and markers of glycemic control, lipids, and inflammation 1. Am J Clin Nutr 2006;83:275– 83. 43. Glore SR, Van Treeck D, Knehans AW, Guild M. Soluble fiber and serum lipids: a literature review. J Am Diet Assoc 1994;94:425–36. 44. Kay RM. Dietary fiber. J Lipid Res 1982;23:221– 42. 45. Behall KM, Scholfield DJ, Hallfrisch J. Whole-grain diets reduce blood pressure in mildly hypercholesterolemic men and women. J Am Diet Assoc 2006;106:1445–9. 46. Whelton SP, Hyre AD, Pedersen B, Yi Y, Whelton PK, He J. Effect of dietary fiber intake on blood pressure: a meta-analysis of randomized, controlled clinical trials. J Hypertens 2005;23:475– 81. 47. Streppel MT, Arends LR, van ’t Veer P, Grobbee DE, Geleijnse JM. Dietary fiber and blood pressure: a meta-analysis of randomized placebo-controlled trials. Arch Intern Med 2005;165:150 – 6. 48. Slavin J, Jacobs D, Marquart L. Whole-grain consumption and chronic disease: protective mechanisms. Nutr Cancer 1997;27:14 –21. 49. Weickert MO, Mohlig M, Schofl C, et al. Cereal fiber improves wholebody insulin sensitivity in overweight and obese women. Diabetes Care 2006;29:775– 80. 50. Steffen LM, Jacobs DR Jr, Murtaugh MA, et al. Whole grain intake is associated with lower body mass and greater insulin sensitivity among adolescents. Am J Epidemiol 2003;158:243–50. 51. Liese AD, Roach AK, Sparks KC, Marquart L, D’Agostino RB Jr, Mayer-Davis EJ. Whole-grain intake and insulin sensitivity: the Insulin Resistance Atherosclerosis Study. Am J Clin Nutr 2003;78:965–71. 52. Willett WC. Nutritional epidemiology. 2nd ed. New York, NY: Oxford University Press, 1998. 53. North KE, Franceschini N, Borecki IB, et al. Genotype-by-sex interaction on fasting insulin concentration: the HyperGEN study. Diabetes 2007;56:137– 42. 54. Liu S, Willett WC. Dietary glycemic load and atherothrombotic risk. Curr Atheroscler Rep 2002;4:454 – 61. 55. Willett W, Manson J, Liu S. Glycemic index, glycemic load, and risk of type 2 diabetes. Am J Clin Nutr 2002;76(suppl):274S– 80S. 56. Wolever TM, Miller JB. Sugars and blood glucose control. Am J Clin Nutr 1995;62(suppl):212S–21S. 57. Pereira MA, Jacobs DR Jr, Pins JJ, et al. Effect of whole grains on insulin sensitivity in overweight hyperinsulinemic adults. Am J Clin Nutr 2002; 75:848 –55. 58. Kaaks R, Bellati C, Venturelli E, et al. Effects of dietary intervention on IGF-I and IGF-binding proteins, and related alterations in sex steroid metabolism: the Diet and Androgens (DIANA) Randomised Trial. Eur J Clin Nutr 2003;57:1079 – 88. 59. Berrino F, Bellati C, Secreto G, et al. Reducing bioavailable sex hormones through a comprehensive change in diet: the diet and androgens (DIANA) randomized trial. Cancer Epidemiol Biomarkers Prev 2001; 10:25–33. 60. Esmaillzadeh A, Mirmiran P, Azizi F. Whole-grain intake and the prevalence of hypertriglyceridemic waist phenotype in Tehranian adults. Am J Clin Nutr 2005;81:55– 63. 61. Masson LF, McNeill G, Avenell A. Genetic variation and the lipid response to dietary intervention: a systematic review. Am J Clin Nutr 2003;77:1098 –111.

Fruit and vegetable intake and prevalence of colorectal adenoma in a cancer screening trial1–3 Amy E Millen, Amy F Subar, Barry I Graubard, Ulrike Peters, Richard B Hayes, Joel L Weissfeld, Lance A Yokochi, and Regina G Ziegler for the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial Project Team ABSTRACT Background: Research on the association between fruit and vegetable intake and risk of colorectal adenoma is inconclusive. Objective: We studied whether intake of fruit, vegetables, or their subgroups is associated with a lower risk of prevalent colorectal adenoma. Design: In men and women (aged 55–74 y) who were screened for colorectal cancer in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO) (1993–2001), we compared 3057 cases with at least one prevalent histologically verified adenoma of the distal large bowel with 29 413 control subjects. Using a foodfrequency questionnaire, we quantified intake of fruit and vegetables in the 12 mo before screening as energy-adjusted pyramid servings/d (ps/d). Adjusted odds ratios (ORs) and 95% CIs were estimated by logistic regression. Results: Risk of distal adenoma was significantly lower among subjects in high (앒5.7 ps/d) versus low (앒1.2 ps/d) quintiles of total fruit intake (OR: 0.75; 95% CI: 0.66, 0.86, P for trend 쏝0.001), which was not completely explained by dietary folate or fiber intake. Inverse associations between adenoma and total fruit intake were observed regardless of adenoma histopathology and multiplicity. However, the protective effect was seen only for colon and not rectal adenoma. Total vegetable intake was not significantly associated with reduced risk of adenoma. ORs for colorectal adenoma among persons with high versus low intakes of deep-yellow vegetables, dark-green vegetables, and onions and garlic were significantly related to lower risk of adenoma, although the P for trend for darkgreen vegetables was not significant. Conclusion: Diets rich in fruit and deep-yellow vegetables, darkgreen vegetables, and onions and garlic are modestly associated with reduced risk of colorectal adenoma, a precursor of colorectal cancer. Am J Clin Nutr 2007;86:1754 – 64. KEY WORDS Fruit, vegetables, epidemiology, colorectal neoplasms, adenoma, diet

INTRODUCTION

Since the mid 1980s, numerous studies have investigated the relation between colorectal cancer and consumption of fruit and vegetables. The hypotheses as to how fruit and vegetable intake may reduce the risk of colon or rectal cancer are numerous and involve independently or additively the many

1754

potential anticarcinogenic compounds found in fruit and vegetables (eg, fiber, carotenoids, vitamin C, folate, glucosinolates, and allium compounds) (1–3). In 1997 the American Institute for Cancer Research reviewed the literature (4 prospective and 22 case-control studies) on diet and risk of all types of cancer (4), and concluded that the “evidence that diets rich in vegetables protect against cancers of the colon and rectum is convincing. The data on fruits are more limited and inconsistent; no judgment is possible.” (4). Since the American Institute for Cancer Research published its review, at least 14 additional prospective studies, which assessed diet before development of disease and thus were not subject to recall bias in diet reports, investigated the role of fruit and vegetables in colorectal cancer prevention (5–18). Among the largest of these prospective studies, 2 (11, 13) observed no reduction in risk of colorectal cancer with intake of fruit and vegetables, and 1 (18) observed a reduction in risk for colorectal cancer with high compared with low intake of total vegetables in men but not women. Additionally, the International Agency for Research on Cancer’s 2003 review of the literature suggested no association between fruit and vegetable consumption and risk of colorectal cancer 1

From the University at Buffalo, School of Public Health and Health Professions, Department of Social and Preventive Medicine, Buffalo, NY (AEM); the National Cancer Institute, Division of Cancer Control and Population Sciences, Applied Research Program, Risk Factor Monitoring and Methods Branch, Bethesda, MD (AFS); the National Cancer Institute, Division of Cancer Epidemiology and Genetics, Epidemiology and Biostatistics Program, Biostatistics Branch, Bethesda, MD (BIG); the Fred Hutchinson Cancer Research Center, Public Health Sciences, Cancer Prevention, Seattle, WA (UP); the National Cancer Institute, Division of Cancer Epidemiology and Genetics, Epidemiology and Biostatistics Program, Occupational and Environmental Epidemiology Branch, Bethesda, MD (RBH); the University of Pittsburgh, Department of Epidemiology, Pittsburgh, PA (JLW); the Pacific Health Research Institute, Honolulu, HI (LAY); and the National Cancer Institute, Division of Cancer Epidemiology and Genetics, Epidemiology and Biostatistics Program, Office of the Director, Bethesda, MD (RGZ). 2 The Prostate Lung Colorectal and Ovarian Cancer Screening Trial is fully funded by the National Cancer Institute, National Institutes for Health, and the Department of Health and Human Services of the US government. 3 Address reprint requests and correspondence to AE Millen, University at Buffalo, School of Public Health and Health Professions, Department of Social and Preventive Medicine, Farber Hall, Room 270, Buffalo, NY, 14214-8001. E-mail: [email protected]. Received April 17, 2007. Accepted for publication August 3, 2007.

Am J Clin Nutr 2007;86:1754 – 64. Printed in USA. © 2007 American Society for Nutrition

FRUIT AND VEGETABLE INTAKE AND COLORECTAL ADENOMA

(19), and recent results from the Women’s Health Initiative Randomized Controlled Dietary Modification Trial (20) did not support a protective role of fruit and vegetable consumption on colorectal cancer risk in postmenopausal women. Concurrently, the literature on the role of fruit and vegetable intake with respect to colorectal adenoma, a precursor to colorectal cancer (21), includes 23 studies (22– 44); however, only 2 of these were cohort studies with prospective designs in which diet was assessed before the development of disease (32, 40). The prospective cohort studies suggest that intake of fruit and specific vegetables may also reduce the risk of colorectal adenoma, suggesting protective effects early in the carcinogenic process. However, in 2000 a randomized clinical trial of a high-fiber, low-fat diet enriched with fruit and vegetables did not show any influence of these dietary components on the risk of colorectal adenoma recurrence (35). Because of the remaining uncertainty about the influences of fruit and vegetable intake on colorectal adenoma, we evaluated these risks in the Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial. The present study is substantially larger than any previously published cohort study of colorectal adenoma (n ҃ 1720 cases) (40). Among the 58 000 men and women randomly assigned to the PLCO screening arm by 2001, a histologically verified distal adenoma was diagnosed at baseline in 쏜3000 participants by means of endoscopy. We analyzed associations between intake of fruit and vegetables, which were determined from a baseline food-frequency questionnaire (FFQ), and risk of adenoma in the distal large bowl and by adenoma subtype (histologic type, number, and location in the bowel). We hypothesized that intake of fruit and vegetables would be associated with decreased risk of colorectal adenoma.

SUBJECTS AND METHODS

Study design and population This study was conducted as part of the PLCO Cancer Screening Trial (45, 46), a multisite investigation (Birmingham, AL; Denver, CO; Detroit, MI; Honolulu, HI; Marshfield, WI; Minneapolis, MN; Pittsburgh, PA: Salt Lake City, UT; St. Louis, MO; and Washington, DC) of the effectiveness of early detection of prostate, lung, colorectal, and ovarian cancers. After approval by the institutional review boards of the US National Cancer Institute and the participating centers, each eligible participant provided informed consent. PLCO trial participants randomly assigned to the screening arm of the trial were offered a 60-cm flexible sigmoidoscopic exam of the distal colon (including the descending colon, sigmoid colon, and the rectum) at trial entry. Those identified with lesions suspect for colorectal neoplasia (ie, sigmoidoscopically visualized polypoid lesions or masses) were referred to their medical care providers for further diagnostic work-up. All available medical and pathological reports on all lesions removed during the diagnostic and related surgical procedures were obtained and coded by trained medical abstractors. Between November 1993 and December 2001, 57 560 men and women aged 55–74 y had an initial successful sigmoidoscopic screening exam, which was described in detail elsewhere (47). Of these, 52 102 participants (90.5%) completed the baseline risk factor questionnaire and FFQ. After excluding a total of 7570 participants for self-reported history of cancer other than

1755

nonmelanoma skin cancer (n ҃ 2360); self-reported history of colorectal polyps, ulcerative colitis, Crohn disease, familial polyposis, or Gardner syndrome (n ҃ 4796); extremely high or low energy intake (highest and lowest 1% in sex-specific energy intake; n ҃ 998); or 쏜7 items missing on the FFQ (n ҃ 440), 44 532 participants remained in the study. The participants (n ҃ 5972) were ineligible for this study if they had hyperplastic polyps only (n ҃ 1545), benign lesions not further specified (n ҃ 297), colorectal lesions (polyps or cancer) of unknown location (n ҃ 277), polyps of uncertain histology or cancer (n ҃ 1241), indeterminate screening results (n ҃ 13), or positive screening results but no follow-up endoscopy (n ҃ 2599, of which 1887, or 72.6%, had a polyp on the screening exam 쏝5 mm in the distal colon). Some participants were excluded for more than one reason. The FFQ was completed before (n ҃ 10 838; 28%), on (n ҃ 21 632; 56%), or after (n ҃ 6079; 16%) the day of the sigmoidoscopy among the remaining 38 560 participants (excluding those missing data on the timing of FFQ completion, n ҃ 11). The protocols given to the study centers instructed that, on the day of screening, the participants should complete the risk factor and dietary questionnaire before the screening procedures began. Because of the potential for dietary recall bias, individuals who filled out their FFQs after the screening, and those with missing data, were excluded from these analyses; this left 32 470 participants. All cases included in this study had a positive result on the screening exam followed by histologic verification of colorectal adenoma of the distal colon (n ҃ 3057). Comparison controls screened negative for polyps of the distal colon (n ҃ 27 966) or screened positive but were found on follow-up diagnostic exam not to have polyps of the distal colon (n ҃ 1447). Of the 3057 cases, 1164 (38.1%) were considered advanced on the basis of the presence of at least one adenoma 욷1 cm in size or with high-grade dysplasia (including cancer in situ) or villous elements (including tubulovillous adenomas). Adenomas were also categorized according to number, with 735 cases having 욷2 adenomas. The data for this analysis were last updated in April 2004. Baseline questionnaires At the time of random assignment, all study subjects were asked to complete a self-administered baseline questionnaire that included questions on demographic factors, medical history, and health-related behaviors. All participants randomly assigned to the PLCO screening arm were given a 137-item FFQ, which was designed to be self-administered and to characterize usual dietary intake over the past 12 mo (Internet: http://www.cancer.gov/ prevention/plco/DQX.pdf). This FFQ was modeled after 3 wellestablished and validated questionnaires: the National Cancer Institute’s Diet History Questionnaire (48), the Block FFQ (49), and the Willett FFQ (50), and incorporated elements of both cognitive (48, 51, 52) and database (53) research. Descriptive data for calculating nutrients and food groups were derived from the two 24-h recalls administered in the 1994 –1996 US Department of Agriculture’s Continuing Survey of Food Intake by Individuals (CSFII) (53), a nationally representative survey conducted during the period when the FFQ was being used. The FFQ also queried for use of supplements taken since 25 y of age, including multivitamins [one-a-day type (100% RDA; Bayer Corp, Pittsburgh, PA), therapeutic or high-dose type

1756

MILLEN ET AL

(쏜100% RDA, such as Theragran; Bristol-Myers Squibb, New York, NY), Stresstabs (B-complex ѿ vitamin C; Inverness Medical Inc, Waltham, MA), B-complex, and other] and singlenutrient supplements (vitamin A, ␤-carotene, vitamin C, vitamin E, calcium, and vitamin D). Data were obtained about current and past supplement use (use 2 and 5 y ago), duration of use (y), and daily or weekly dose. Nutrient intake from supplements was estimated from intake of both multiple and single supplements, except for folic acid, which was estimated from intake of multivitamins only (54). Fruit and vegetable frequency measures and pyramid servings A methodologic component of these analyses was to compare the study results when fruit and vegetable intake was measured by using frequencies compared with pyramid servings. Food frequencies are the reported amount of times a food is eaten in a specific time period. The fruit and vegetable intake responses on the PLCO FFQ were converted into frequencies/d (f/d) of consumption of specific food groups by summing the daily frequency reports of FFQ line items included in the food group. Frequencies do not incorporate portion size measurements or food intake from hidden sources, such as mixed dishes. Pyramid servings/d (ps/d) of fruit and vegetable food groups were estimated from the fruit and vegetable intake responses on the FFQ by using a Pyramid Servings Database (55) developed from the 1994 –1996 CSFII data (28). The Pyramid Servings Database uses a recipe file to disaggregate food mixtures into their component ingredients and assigns the components to appropriate food groups and also calculates standardized quantitative estimates of food group intake, called pyramid servings. Pyramid servings reflect the full, continuous range of reported food intake, including small amounts of foods from all sources, and are standardized to portion sizes from the 1992 Food Guide Pyramid (56). The Diet History Questionnaire, one of the FFQs that the PLCO FFQ was modeled after, was previously validated for pyramid servings (57), and a more detailed description of pyramid servings is provided elsewhere (57). The fruit and vegetable food groups for which ps/d were created are listed in Appendix A. Statistical analyses Fruit and vegetable intakes were energy-adjusted by using the residual method (58). Log-transformed fruit and vegetable intakes were regressed on log-transformed total energy intake to compute residuals. The mean of the log-transformed fruit and vegetable intake was added to each residual, and the antilogarithm of this sum was taken. Further addition of total calories to the models by using fruit and vegetable intake adjusted by the residual method or by adjusting for energy by using the nutrient density method did not significantly alter the results. The intake of fruit and vegetable food groups, as ps/d and f/d, was categorized by sex-specific quintiles, with quintile 1 being the lowest intake and quintile 5 the highest. Odds ratios (ORs) and 95% CIs (95% CIs) for colorectal adenoma were calculated by using logistic regression models for quintiles of fruit and vegetable intake, with the lowest quintile level as the reference category. Linear trends (P for trend) were assessed by assigning the quintiles the values 1, 2, 3, 4, or 5. Two models are presented in the tables. The study design model was adjusted for age at screening, sex, and study center.

The multivariate model was additionally adjusted for a priori nondietary risk factors for colorectal adenoma and colorectal cancer: ethnic origin (non-Hispanic white, non-Hispanic black, Hispanic, Asian, Pacific Islander, and American Indian or Alaskan native), educational attainment (쏝8 y, 8 –11 y, 12 y or high school equivalent, post-high school other than college, some college, college graduate, or postgraduate), family history of colon cancer [yes (sibling, parent, or child with colon cancer) or no], smoking (never smoked, only smoked cigars or pipe, quit cigarette smoking 욷20 y ago and 울1 pack cigarettes/d, quit 욷20 y ago and 쏜1 pack/d, quit 쏝20 y ago and 울1 pack/d, or quit 쏝20 y ago and 쏜1 pack/d), alcohol use (쏝1 g/d, 욷1–15 g/d, 쏜15–30 g/d, or 쏜30 g/d), use of aspirin and ibuprofen (no regular use, 쏝2 times/mo, 2–3 times/mo, 1 time/wk, 2 times/wk, 3– 4 times/wk, 1 time/d, or 2 times/d), use of hormone replacement therapy (never used, former user 울5 y, former user 욷6 y, current user 울5 y, current user 욷6 y), physical activity (no, 쏝1 h, 1 h, 2 h, 3 h, 욷4 h of vigorous activity/wk), and current body mass index (BMI; in kg/m2). Because substantial confounding, primarily due to smoking, was observed for fruit and vegetable subgroups, we chose to focus on the estimate of risk obtained from the multivariate model. Relations of colorectal adenoma with fruit and vegetable intake were examined by levels of potential effect modifiers, both dietary (animal fat intake, calcium intake, or red meat consumption) and nondietary (age, sex, BMI, ever having smoked, family history of colon cancer, regular ibuprofen use, regular aspirin use, or multivitamin use), by including interaction terms between intake (as a 5-level, ordinal variable) and the effect modifier. A Wald chi-square test with a P value of 쏝 0.05 for the overall interaction term was considered statistically significant. All analyses were conducted by using SAS version 8.2 software (SAS Institute Inc, Cary, NC). All statistical tests were two-sided with a significance level of 0.05.

RESULTS

Men and women in the highest quintile of fruit or vegetables reported consuming on average 앒5 ps/d more than did those in the lowest quintile of intake, and more vegetables were reported consumed than fruit (Table 1). For both fruit and vegetables, high consumers were more likely than low consumers to be older, more highly educated, leaner, more physically active, neversmokers, and infrequent drinkers. In the overall population of men and women combined, there was a statistically significantly decreased risk of colorectal adenoma among participants who consumed high versus low ps/d of total fruit (OR comparing extreme quintiles ҃ 0.58; 95% CI: 0.52, 0.66; P for trend 쏝 0.001) and total vegetables (OR comparing extreme quintiles ҃ 0.81; 95% CI ҃ 0.72, 0.92; P for trend 쏝0.001), after adjustment for age at screening, sex, and study center (Table 2, study design model). After further adjustment for a priori, nondietary risk factors, the statistically significant association remained for total fruit (OR comparing extreme quintiles ҃ 0.75; 95% CI ҃ 0.66, 0.86; P for trend 쏝0.001), but not for total vegetables (OR comparing extreme quintiles ҃ 0.94; 95% CI ҃ 0.83, 1.06,;P for trend ҃ 0.24) (Table 2, multivariate model). Additionally, in the multivariate model, a reduction in risk of adenoma was observed when extreme quintiles of total fruit and vegetable intake combined (ps/d) were compared (OR

1757

FRUIT AND VEGETABLE INTAKE AND COLORECTAL ADENOMA TABLE 1 Characteristics of the study participants in quintiles 1 and 5 of fruit and vegetable intake: the PLCO Cancer Screening Trial1 Fruit Characteristic Total fruit (energy-adjusted pyramid servings/d) Men Women Total vegetables (energy-adjusted pyramid servings/d) Men Women Age at screening (y) Sex (% women) Ethnic origin (%) Non-Hispanic white Non-Hispanic black Hispanic Asian Pacific Islander American Indian or Alaskan Native Some college-level education (%) BMI (kg/m2) 욷1 h recent strenuous physical activity/wk (%) Smoking (% ever) Recent alcohol intake, 쏜15 g/d (%) Sibling or parent with colon cancer (% yes) Hormone replacement therapy (women; % ever use) Ibuprofen use (% ever regular use) Aspirin use (% ever regular use)

Vegetables

Quintile 1

Quintile 5

Quintile 1

Quintile 5

0.98 앐 0.022 1.5 앐 0.02

5.7 앐 0.013 6.7 앐 0.013

2.6 앐 0.03 3.3 앐 0.04

3.7 앐 0.033 4.4 앐 0.043

4.1 앐 0.03 4.6 앐 0.04 61.4 앐 0.07 46.8

5.4 앐 0.033 5.9 앐 0.033 63.7 앐 0.073 47.8

2.6 앐 0.01 2.9 앐 0.01 62.1 앐 0.07 46.9

7.4 앐 0.013 8.1 앐 0.013 63.4 앐 0.073 47.8

91.3 2.3 1.4 4.1 0.6 0.3 64.3 27.4 앐 0.06 54.8 64.7 18.3 8.7 66.4 32.5 41.2

88.13 5.4 1.5 4.2 0.6 0.1 74.03 26.5 앐 0.063 78.13 45.73 6.73 9.4 69.0 30.6 42.9

88.9 4.8 1.4 4.0 0.7 0.2 65.5 27.3 앐 0.06 58.0 57.6 14.2 9.3 66.9 33.7 42.1

90.63 2.7 1.5 4.5 0.5 0.2 76.53 26.7 앐 0.063 77.63 51.13 11.03 8.8 68.94 30.24 41.3

1 n ҃ 32 470. Quintiles were calculated separately for men and women. Least-squares (LS) means for continuous variables were calculated for all 5 quintiles and were compared across all 5 quintiles by using ANOVA models with adjustment for age at screening, sex, and study center. Proportions for all 5 quintiles were directly standardized for age at screening, sex, and center and were compared across all 5 quintiles by using the Cochran-Mantel-Haenszel chi-square test for general associations. 2 LS x៮ 앐 SE (all such values). 3 P 쏝0.001 for differences across all 5 quintiles. 4 P 쏝 0.05 for differences across all 5 quintiles.

comparing extreme quintiles ҃ 0.82; 95% CI ҃ 0.72, 0.93; P for trend 쏝0.001). When potatoes were removed from the total vegetable food group, a modest but not statistically significant relation between total vegetable intake (minus potatoes) and adenoma was observed in the multivariate model (OR comparing extreme quintiles ҃ 0.90; 95% CI ҃ 0.80, 1.02; P for trend ҃ 0.08; Table 2). Also observed in the multivariate model was a decreased risk of adenoma among participants with high versus low intake, in ps/d, of fruit (without juice); fruit juice; citrus, melon, and berry fruit; deep-yellow vegetables; dark-green vegetables; and onions and garlic. However, the P for trend for dark-green vegetables was 쏜0.05. When these analyses were repeated using f/d rather than ps/d, the results were comparable. However, no relation was observed between colorectal adenoma and intake of dark-green vegetables or onions and garlic (data not shown). For total fruit intake and total vegetable intake, measured in f/d, the multivariate ORs and 95% CIs for adenoma in quintile 5 versus 1 were 0.75 (95% CI ҃ 0.66, 0.86; P for trend 쏝0.001) and 0.91 (95% CI ҃ 0.80, 1.02; P for trend ҃ 0.10), respectively. The relation between risk of adenoma and total fruit intake also did not vary significantly by age at screening, BMI, family history of colon cancer, regular ibuprofen use, regular aspirin use, multivitamin use, animal fat intake, calcium intake, or red meat

consumption, but did differ statistically significantly by ever having smoked (P for interaction ҃ 0.001) and sex (P for interaction 쏝 0.001). The relation between risk of adenoma and total vegetable intake did not vary significantly by any of these factors. The risk of adenoma among participants in quintile 5 compared with 1 for intake of total fruit was stronger among those who had ever smoked (OR ҃ 0.68; 95% CI ҃ 0.57, 0.80; P for trend 쏝 0.001) than among those who had never smoked (OR ҃ 0.92; 95% CI ҃ 0.74, 1.14; P for trend ҃ 0.29) and in men (OR ҃ 0.62; 95% CI ҃ 0.53, 0.73; P for trend 쏝 0.001) than in women (OR ҃ 0.86; 95% CI ҃ 0.70, 1.06; P for trend ҃ 0.27). It is possible that the observed associations in Table 2 were confounded by intake of specific nutrients or foods hypothesized to be related to adenoma risk. Further adjustment of the multivariate model for current intake of folate from supplements, calcium from diet and supplements combined, vitamin D from diet and supplements combined, and red meat modestly weakened the risk reduction for all fruit and vegetables listed in Table 3. The inverse associations, ORs comparing extreme quintiles, and the P for trend for total fruit, fruit without juice, and deepyellow vegetables remained significant (P 쏝 0.05). The P for trend but not the ORs comparing extreme quintiles remained statistically significant (P 쏝 0.05) for citrus, melon, and berry fruit and onions and garlic. The statistically significant protective

1758

MILLEN ET AL

TABLE 2 Odds ratios (ORs) and 95% CIs for prevalent colorectal adenoma by fruit and vegetable intake, in energy-adjusted pyramid servings/d in the overall population of men and women: the PLCO Cancer Screening Trial1 OR (95% CI)

Food group Total fruit Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend5 Total vegetables Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend Total fruit and vegetables Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend Fruit and vegetable subgroups Fruit without juice Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend Fruit juice Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend Citrus, melon, berry Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend Total vegetables, without potatoes Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend Total vegetables, without potatoes, beans, and starchy vegetables Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend

Quintile median2

Cases

Noncases

Study design model3

Multivariate model4

1.2 2.2 3.0 4.0 5.7

730 657 606 569 495

5763 5837 5889 5925 5999

1.00 0.85 (0.76, 0.95) 0.75 (0.67, 0.85) 0.69 (0.61, 0.77) 0.58 (0.52, 0.66) 쏝0.001

1.00 0.95 (0.85, 1.06) 0.88 (0.78, 0.99) 0.84 (0.74, 0.95) 0.75 (0.66, 0.86) 쏝0.001

2.8 3.9 4.7 5.7 7.3

639 651 597 615 555

5854 5843 5898 5879 5939

1.00 1.00 (0.89, 1.12) 0.90 (0.80, 1.01) 0.92 (0.81, 1.03) 0.81 (0.72, 0.92) 쏝0.001

1.00 1.05 (0.93, 1.18) 0.97 (0.86, 1.10) 1.01 (0.90, 1.14) 0.94 (0.83, 1.06) 0.24

4.7 6.5 7.9 9.5 12.2

685 677 625 568 502

5808 5817 5870 5926 5992

1.00 0.95 (0.85, 1.07) 0.85 (0.75, 0.95) 0.75 (0.67, 0.84) 0.64 (0.57, 0.73) 쏝0.001

1.00 1.04 (0.93, 1.17) 0.97 (0.86, 1.09) 0.91 (0.80, 1.02) 0.82 (0.72, 0.93) 쏝0.001

0.7 1.4 2.1 2.8 4.2

732 647 591 594 493

5761 5847 5904 5900 6001

1.00 0.82 (0.74, 0.92) 0.73 (0.65, 0.82) 0.71 (0.64, 0.80) 0.56 (0.50, 0.64) 쏝0.001

1.00 0.91 (0.81, 1.02) 0.85 (0.75, 0.95) 0.86 (0.76, 0.97) 0.72 (0.63, 0.82) 쏝0.001

0.06 0.3 0.7 1.2 2.2

708 602 604 591 552

5785 5892 5891 5903 5942

1.00 0.83 (0.74, 0.93) 0.82 (0.73, 0.92) 0.79 (0.70, 0.89) 0.75 (0.66, 0.84) 쏝0.001

1.00 0.89 (0.79, 1.00) 0.90 (0.80, 1.01) 0.90 (0.80, 1.01) 0.87 (0.77, 0.98) 0.04

0.4 0.9 1.3 1.8 2.8

701 652 609 548 547

5792 5842 5886 5946 5947

1.00 0.89 (0.79, 1.00) 0.80 (0.72, 0.90) 0.70 (0.63, 0.79) 0.70 (0.62, 0.79) 쏝0.001

1.00 0.98 (0.87, 1.10) 0.91 (0.81, 1.03) 0.83 (0.73, 0.93) 0.86 (0.76, 0.97) 쏝0.001

2.0 2.9 3.7 4.6 6.3

669 633 612 594 549

5824 5861 5883 5900 5945

1.00 0.93 (0.83, 1.04) 0.88 (0.79, 0.99) 0.84 (0.75, 0.95) 0.77 (0.68, 0.87) 쏝0.001

1.00 0.98 (0.87, 1.11) 0.97 (0.87, 1.10) 0.94 (0.84, 1.06) 0.90 (0.80, 1.02) 0.08

1.7 2.5 3.2 4.0 5.5

658 634 610 604 551

5835 5860 5885 5890 5943

1.00 0.95 (0.85, 1.07) 0.90 (0.80, 1.01) 0.88 (0.78, 0.99) 0.79 (0.70, 0.89) 쏝0.001

1.00 1.00 (0.89, 1.13) 0.98 (0.87, 1.11) 0.98 (0.87, 1.10) 0.92 (0.81, 1.04) 0.16 (Continued)

1759

FRUIT AND VEGETABLE INTAKE AND COLORECTAL ADENOMA TABLE 2 (Continued) OR (95% CI)

Food group Cruciferous vegetables Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend Deep-yellow vegetables Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend Dark-green vegetables Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend Tomatoes Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend White potatoes Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend Dry beans Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend Onions and garlic Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 P for trend

Quintile median2

Cases

Noncases

Study design model3

Multivariate model4

0.1 0.2 0.4 0.5 1.0

650 609 605 610 583

5843 5885 5890 5884 5911

1.00 0.92 (0.82, 1.03) 0.91 (0.81, 1.02) 0.90 (0.80, 1.02) 0.85 (0.76, 0.96) 0.01

1.00 0.97 (0.86, 1.09) 0.98 (0.87, 1.10) 0.99 (0.88, 1.12) 0.98 (0.86, 1.10) 0.87

0.2 0.3 0.4 0.5 0.9

706 617 635 564 535

5787 5877 5860 5930 5959

1.00 0.85 (0.75, 0.95) 0.85 (0.76, 0.95) 0.74 (0.66, 0.84) 0.69 (0.61, 0.78) 쏝0.001

1.00 0.90 (0.80, 1.01) 0.95 (0.84, 1.06) 0.85 (0.76, 0.97) 0.83 (0.74, 0.95) 0.004

0.08 0.2 0.3 0.4 0.8

681 636 614 597 529

5812 5858 5881 5897 5965

1.00 0.92 (0.82, 1.04) 0.89 (0.79, 1.00) 0.86 (0.76, 0.97) 0.74 (0.66, 0.84) 쏝0.001

1.00 0.96 (0.86, 1.08) 0.97 (0.86, 1.09) 0.97 (0.86, 1.09) 0.87 (0.77, 0.99) 0.07

0.4 0.5 0.7 0.9 1.2

648 593 613 644 559

5845 5901 5882 5850 5935

1.00 0.90 (0.80, 1.01) 0.93 (0.83, 1.05) 0.98 (0.88, 1.10) 0.84 (0.74, 0.94) 0.05

1.00 0.93 (0.83, 1.05) 0.98 (0.87, 1.10) 1.05 (0.93, 1.18) 0.89 (0.79, 1.01) 0.44

0.3 0.6 0.9 1.2 1.6

603 567 609 637 641

5890 5927 5886 5857 5853

1.00 0.90 (0.80, 1.02) 0.96 (0.86, 1.09) 1.01 (0.89, 1.14) 1.00 (0.89, 1.13) 0.38

1.00 0.90 (0.80, 1.02) 0.96 (0.85, 1.09) 1.01 (0.89, 1.14) 1.00 (0.88, 1.13) 0.46

0.05 0.09 0.1 0.2 0.4

653 630 619 591 564

5840 5864 5876 5903 5930

1.00 0.96 (0.85, 1.07) 0.93 (0.83, 1.04) 0.88 (0.78, 0.99) 0.85 (0.75, 0.96) 0.003

1.00 0.97 (0.87, 1.09) 0.95 (0.85, 1.07) 0.92 (0.82, 1.04) 0.92 (0.81, 1.03) 0.10

0.09 0.2 0.5 0.9 1.6

687 618 616 574 562

5806 5876 5879 5920 5932

1.00 0.90 (0.80, 1.01) 0.87 (0.77, 0.98) 0.79 (0.70, 0.88) 0.78 (0.69, 0.88) 쏝0.001

1.00 0.92 (0.82, 1.04) 0.94 (0.83, 1.05) 0.86 (0.76, 0.97) 0.87 (0.77, 0.99) 0.01

n ҃ 32 470. ORs and 95% CIs were estimated by using logistic regression. Quintiles are sex-specific. 3 Adjusted for study center, age at screening, (age at screening)2, and sex. 4 Adjusted for study center, age at screening, (age at screening)2, sex, race, education, family history of colon cancer, smoking, alcohol use, use of ibuprofen, use of aspirin, use of replacement hormones, physical activity, BMI, and BMI2. 5 P for trend across quintiles. 1 2

effect of fruit juice and the statistically significant OR comparing extreme quintiles for dark-green vegetables was removed. Further analyses were conducted to see whether the observed associations between risk of adenoma and intake of fruit and

vegetables could be explained by specific nutrients concentrated in fruit and vegetables, such as dietary folate and fiber (Table 3). Adjustment of the multivariate model for either nutrient could explain only part of the decreased risk associated with fruit con-

1760

MILLEN ET AL

TABLE 3 Odds ratios (ORs) and 95% CIs for prevalent colorectal adenoma among participants in quintile 5, relative to quintile 1, of fruit and vegetable intake in energy-adjusted pyramid servings/d: PLCO Cancer Screening Trial Food group Fruit Total fruit Fruit without juice Fruit juice Citrus, melon, berry Vegetables Total vegetables Total vegetables without potatoes Deep-yellow vegetables Dark-green vegetables Onions and garlic

MM2

MM23

MM34

MM45

0.75 (0.66, 0.86)6 0.72 (0.63, 0.82)6 0.87 (0.77, 0.98)6 0.86 (0.76, 0.97)6

0.80 (0.70, 0.91)6 0.76 (0.66, 0.86)6 0.89 (0.79, 1.01) 0.90 (0.79, 1.02)6

0.84 (0.73, 0.96)6 0.78 (0.68, 0.90)6 0.94 (0.83, 1.07) 0.96 (0.84, 1.10)

0.85 (0.74, 0.98)6 0.83 (0.71, 0.96) 0.88 (0.78, 0.99) 0.93 (0.82, 1.05)6

0.94 (0.83, 1.06) 0.90 (0.80, 1.02) 0.83 (0.74, 0.95)6 0.87 (0.77, 0.99) 0.87 (0.77, 0.99)6

0.95 (0.83, 1.07) 0.93 (0.82, 1.05) 0.86 (0.76, 0.98)6 0.92 (0.80, 1.04) 0.89 (0.79, 1.01)6

1.06 (0.93, 1.21) 1.04 (0.90, 1.19) 0.92 (0.80, 1.05) 0.98 (0.86, 1.13) 0.96 (0.84, 1.09)

1.21 (1.04, 1.40)6 1.17 (1.00, 1.37)6 0.99 (0.86, 1.15) 1.03 (0.89, 1.18) 0.89 (0.79, 1.01)6

1 n ҃ 32 470. ORs and 95% CIs were estimated by using logistic regression. Adjusted ORs and 95% CIs are presented for the multivariate model (MM), the multivariate model further adjusted for potential nutrient and dietary confounders (MM2), and the multivariate model adjusted for dietary factors that may explain the observed associations (MM3 and MM4). 2 Adjusted for study center, age at screening, (age at screening)2, sex, race, education, family history of colon cancer, smoking, alcohol use, use of ibuprofen, use of aspirin, use of replacement hormones, physical activity, BMI, and BMI2 (as also shown in Table 2). 3 Adjusted for all covariates in the multivariate model plus quintiles of current folate intake from supplements (energy-adjusted ␮g/d), quintiles of current calcium intake from foods and supplements combined (energy adjusted mg/d), quintiles of current vitamin D intake from foods and supplements combined (energy-adjusted ␮g/d), and quintiles of current red meat intake (energy-adjusted g/d). 4 Adjusted for all covariates in the multivariate model plus quintiles of current folate intake from foods (energy-adjusted ␮g/d). 5 Adjusted for all covariates in the multivariate model plus quintiles of current dietary fiber intake (energy-adjusted g/d). 6 P for trend 쏝 0.05.

sumption (total and without juice). Adjustment of the multivariate model for dietary folate explained practically all of the reduction in risk associated with the fruit juice; citrus, melon, and berry fruit; and all vegetable subgroups. Adjustment of the multivariate model for dietary fiber explained practically all the reduction in risk associated with deep-yellow vegetables and dark-green vegetables, but not fruit juice; citrus, melon, and berry fruit; and onions and garlic. After adjustment of the multivariate model for dietary fiber, the risk of adenoma among participants in quintile 5 versus 1 for intake of total vegetables was increased. After removal of potatoes from the total vegetable group, the risk of adenoma, when comparing extreme quintiles of intake, was no longer significant, although the P for trend remained 쏝0.05. Additional adjustment of the multivariate model independently for multivitamin use (yes, no), duration of use of multivitamin supplements, calcium supplements, or vitamin D supplements did not remove the statistically significant association between risk for adenoma and total fruit intake (data not shown). Similar patterns as the overall analyses were observed for nonadvanced and advanced colorectal adenomas and for single and multiple colorectal adenomas (Table 4). Increasing intake of total fruit was statistically significantly related to decreasing risk of nonadvanced, advanced, single, and multiple adenomas. No statistically significant relations were observed for total vegetable intake, although the P for trend for total vegetables without potatoes was statistically significant for advanced adenoma (P for trend ҃ 0.03) and multiple adenomas (P for trend ҃ 0.04). The observed decreased risk of adenoma with intake of fruit and vegetables was more apparent for adenomas in the colon than in the rectum. Total fruit intake was associated with a significant decrease in prevalence of colon adenomas and a nonsignificant decrease in prevalence of rectal adenomas, although the OR was 쏝1.0. Total vegetable intake, after potatoes were excluded, was

associated with a significant decrease in the prevalence of colon adenomas when comparing extreme quintiles (P for trend ҃ 0.10), but was unrelated to the prevalence of rectal adenomas.

DISCUSSION

Among the participants of the PLCO Cancer Screening Trial at baseline, we observed a statistically significant decreased risk of prevalent colorectal adenoma among participants who consumed high (앒5.7 ps/d) compared with low (앒1.2 ps/d) intakes of total fruit but not total vegetables. The statistically significant associations observed between adenoma and intake of total fruit were consistent regardless of adenoma type or number and were not explained by intake of dietary folate or fiber. However, intake of total fruit was statistically significantly protective only for adenomas in the colon and not the rectum. No statistically significant associations were observed between total vegetable consumption and advanced, nonadvanced, single, multiple, colon, or rectal adenoma. Our results for total fruit and total vegetable intake were similar whether we measured fruit and vegetable intake as frequencies or as pyramid servings. However, using pyramid servings may be beneficial from a public health perspective because they are easily translated into the Food Guide Pyramid’s (Internet: http://www.mypyramid.gov/) dietary guidance. Colorectal adenomas, precursors of most colorectal carcinomas, probably occur a decade or more before they evolve into colorectal cancer (59, 60). If diet is critical in the early stages of colorectal carcinogenesis, then assessing diet closer to the appearance of adenomas, which are often asymptomatic, may provide relevant information. To date, research on fruit and vegetable intake and risk of colorectal adenoma has been inconclusive.

1761

FRUIT AND VEGETABLE INTAKE AND COLORECTAL ADENOMA

TABLE 4 Adjusted odds ratios (ORs) and 95% CIs for prevalent adenoma by histopathology (nonadvanced or advanced), number (single or multiple), or location (colon or rectum) among participants in quintiles 2–5 compared with quintile 1 for total fruit and vegetable intake measured as pyramid servings/d: the PLCO Cancer Screening Trial1 OR (95% CI) Food groups

Cases

Noncases

Nonadvanced adenoma3 Total fruit Total vegetables Total vegetables without potatoes Advanced adenoma4 Total fruit Total vegetables Total vegetables without potatoes Single adenoma5 Total fruit Total vegetables Total vegetables without potatoes Multiple adenomas Total fruit Total vegetables Total vegetables without potatoes Colon adenoma6 Total fruit Total vegetables Total vegetables without potatoes Rectal adenoma Total fruit Total vegetables Total vegetables without potatoes

1708

29 413

1164

2322

735

2131

631

Quintile 2

Quintile 3

Quintile 4

Quintile 5

P2

0.89 (0.77, 1.03) 1.14 (0.97, 1.32) 1.02 (0.88, 1.19)

0.90 (0.77, 1.04) 1.04 (0.89, 1.22) 1.03 (0.88, 1.20)

0.78 (0.66, 0.92) 1.08 (0.92, 1.26) 1.05 (0.90, 1.23)

0.74 (0.63, 0.88) 0.96 (0.82, 1.13) 0.93 (0.79, 1.10)

쏝0.001 0.49 0.57

1.02 (0.86, 1.22) 0.97 (0.81, 1.16) 0.99 (0.83, 1.18)

0.85 (0.71, 1.03) 0.87 (0.72, 1.05) 0.91 (0.76, 1.10)

0.88 (0.72, 1.06) 0.93 (0.77, 1.12) 0.82 (0.68, 1.00)

0.76 (0.62, 0.93) 0.88 (0.73, 1.06) 0.86 (0.71, 1.04)

0.003 0.17 0.03

0.91 (0.80, 1.04) 1.01 (0.89, 1.16) 0.97 (0.85, 1.11)

0.84 (0.74, 0.96) 0.96 (0.84, 1.10) 1.01 (0.88, 1.15)

0.80 (0.70, 0.92) 1.02 (0.89, 1.17) 0.94 (0.82, 1.08)

0.76 (0.66, 0.88) 0.94 (0.82, 1.08) 0.95 (0.82, 1.09)

쏝0.001 0.46 0.39

1.07 (0.86, 1.33) 1.13 (0.91, 1.41) 1.03 (0.82, 1.28)

1.00 (0.80, 1.26) 1.00 (0.79, 1.26) 0.86 (0.68, 1.08)

0.97 (0.77, 1.23) 0.97 (0.76, 1.23) 0.96 (0.76, 1.21)

0.72 (0.55, 0.93) 0.90 (0.70, 1.15) 0.77 (0.60, 0.99)

0.02 0.18 0.04

0.91 (0.80, 1.05) 0.98 (0.85, 1.12) 0.90 (0.78, 1.03)

0.82 (0.71, 0.94) 0.98 (0.85, 1.13) 0.94 (0.82, 1.08)

0.83 (0.72, 0.96) 1.04 (0.91, 1.20) 0.94 (0.82, 1.08)

0.70 (0.60, 0.82) 0.89 (0.77, 1.03) 0.85 (0.73, 0.98)

쏝0.001 0.32 0.10

1.00 (0.78, 1.26) 1.16 (0.91, 1.48) 1.26 (0.99, 1.60)

1.01 (0.80, 1.30) 0.93 (0.72, 1.20) 1.00 (0.77, 1.29)

0.76 (0.58, 1.00) 0.86 (0.66, 1.12) 0.90 (0.69, 1.17)

0.89 (0.68, 1.16) 1.05 (0.82, 1.36) 1.15 (0.89, 1.49)

0.11 0.49 0.81

29 413

29 413

29 413

29 413

29 413

1 n ҃ 32 470. ORs were adjusted for study center, age at screening, age2 at screening, sex, race, education, family history of colon cancer, smoking, alcohol use, use of ibuprofen, use of aspirin, use of hormone replacement therapy, physical activity, BMI, and BMI2. ORs and 95% CIs were estimated by using logistic regression. 2 P for trend across quintiles. 3 Nonadvanced adenoma ҃ small (쏝1 cm), no high-grade dysplasia, and no villous elements. 4 Advanced adenoma ҃ large (욷1 cm), or with high-grade dysplasia, or with villous elements (including tubulovillous adenomas). 5 Single: one distal adenoma; multiple: 욷2 distal adenomas. 6 Colon adenoma ҃ adenoma of the sigmoid or descending colon; cases with adenomas in both rectum and colon were excluded.

Only one randomized dietary interventional trial evaluated the importance of vegetable and fruit intake in colorectal carcinogenesis. In this trial (n ҃ 1905), a high-fiber, low-fat diet enriched with fruit and vegetables (3.5 servings/1000 kcal) had no effect on adenoma recurrence among persons previously diagnosed with at least one colorectal adenoma (35). However, the 4-y intervention period may have been too short to affect tumor development, and the increase in vegetable and fruit intake in the intervention arm compared with the control arm was modest (1.1 servings/1000 kcal). Two cohort studies (32, 40), with prospectively collected dietary data and physician-confirmed self-reports of adenomas, have investigated the relation between risk of colorectal adenomas and intake of fruit and vegetables. In one of these cohorts, which was conducted in men (32), a statistically significant trend (P ҃ 0.03) for decreased risk of prevalent colorectal adenoma with intake of fruit (minus fruit juice), but not vegetables, was observed. In the other cohort study, which was conducted in women (40), a statistically significant decreased risk of prevalent adenoma in high versus low consumers of total fruit, but not vegetables, was observed. However, no statistically significant decreased risk for incident colorectal adenoma was observed for

high versus low intake of fruit or vegetables. However, those authors did observe a statistically significant decreased risk (12%) of incident colorectal adenoma with each additional serving of fruit per day. In our study. we observed a decreased risk of adenoma with intake of total fruit in men but not women. The sex difference could not be explained by a narrower distribution of pyramid servings of total fruit or vegetables in women, because the opposite was true. The difference could also not be explained by the smaller number of women than of men with adenoma, because we had 80% power to detect an OR of 0.75 among women and 0.80 among men. However, we may not have observed a statistically significant risk estimate in women because of the measurement error characteristics of the FFQ, which are shown to be greater in women (61). It is also possible that the difference is explained by a yet unknown sex-specific biological mechanism. We also observed a stronger protective effect for adenoma with intake of total fruit among persons who had ever smoked than among never smokers. This suggests that consumption of fruit may protect against adenoma development more so in persons who had ever smoked than in never smokers. This is the opposite of what Michels et al (40) found in the Nurses’ Health

1762

MILLEN ET AL

Study. More research on how smoking might modify the relation between fruit and vegetable intake and risk of colorectal adenoma is needed. We also observed a decreased risk of adenoma with high versus low intake of citrus, melon, and berry fruit; deep-yellow vegetables; dark-green vegetables; and onions and garlic. Other studies have analyzed associations between adenoma risk and intake of citrus fruit or fruit high in vitamin C (31, 38 – 40), green and yellow or high-carotenoid vegetables (24, 31, 32, 38 – 40), and onions or garlic (31, 38). Statistically significant decreased risks of adenoma were previously observed when comparing high with low consumers of citrus fruit (40), high–vitamin C fruit (31), folate-rich vegetables (32), green and yellow vegetables (24, 40), high-carotenoid vegetables (31), and garlic (31). However, unlike previous studies, we found no statistically significant decreased risks with consumption of cruciferous vegetables (23, 31) and legumes (24, 31, 40, 42, 43). Our results were strongest for intake of total fruit, and previous research on fiber and colorectal adenoma showed that intake of fruit fiber, in particular, may be protective for colorectal adenoma. A previous cohort study found no association between total fiber intake and colorectal cancer, but did observe a borderline statistically significant decreased risk of colorectal cancer with increasing intake of fruit fiber (62). A statistically significant effect of high versus low intake of fruit fiber on colorectal cancer risk was also observed in a different cohort study (63). In the PLCO trial (64), a statistically significant inverse association was observed between colorectal adenoma and fiber from fruit and fiber from grains and cereals but not fiber from legumes or vegetables. The association we observed between fruit intake and adenoma could not be fully explained by intake of dietary folate or fiber, both of which are thought to protect against colorectal cancer (65, 66). Perhaps, a combination of the nonfiber phytochemicals and soluble fiber like pectin found in fruit (1) may be protective against colorectal cancer by slowing glucose absorption and thus promoting better blood glucose control and reducing hyperinsulinemia (1, 67–70) or via production of short-chain fatty acids that have been show to reduce cancer cell growth in vitro (1, 71). Our study had limitations. Our results are not generalizable to all cases of colorectal adenoma because our cases and controls were defined on the basis of pathology in the distal colon and rectum; participants with adenomatous polyps in the proximal colon only would have been considered controls. Additionally, our cases with distal adenomas include subjects with advanced, nonadvanced, single, and multiple adenomas, but further analysis showed that our results did not vary greatly when they were stratified by histologic adenoma type, number, or location. Additionally, FFQs are subject to measurement error (61), which can attenuate the risk estimates (61, 72, 73), and dietary patterns change over time, and diet was assessed at only one time point. It is also possible, as in all observational studies, that residual confounding exists, and our multivariate models did not account for all confounding. The strengths of our study include its large number of cases (n ҃ 3707) of distal colorectal adenoma identified by a standardized flexible sigmoidoscopy screening protocol, characterized by clinical follow-up with histologic confirmation. Comparison controls for this study were also endoscopically determined to by polyp-free. Sigmoidoscopy also confirmed that the controls also

had no distal adenomatous or villous polyps. Participants were from diverse regions in the United States and reported a wide range of fruit and vegetable intake, which included both the national average of fruit and vegetable consumption (4.8 servings/d for persons ages 2 y and older in 2001; 47) and the amount recommended in the 2005 Dietary Guidelines for the average combined intake of fruit and vegetables (10 servings/d; 48). Overall, this study showed that diets high in fruit, dark-green vegetables, deep-yellow vegetables, and onions and garlic are associated with decreased risk of colorectal adenoma. In this population, the observed decreased risk of adenoma with increasing intake of fruit was not completely explained by the folate or fiber content of the fruit. We conclude, from the results of this screening trial, that to reduce the risk of development of colorectal adenoma, persons should continue to consume diets rich in fruit and vegetables, which contain many potential anticarcinogenic compounds. We thank Matthew Moore of Information Management Services Inc, Silver Spring, MD, for his support of analysis runs. The contributions of the authors were as follows—AEM: conducted the analyses, interpreted the work, and was the primary writer of this manuscript; AFS: was involved in the primary design, analyses, writing, and interpretation of this work; BIG: contributed to the statistical analysis, writing, and interpretation of this work; UP: contributed to the analysis, writing, and interpretation of this work; RBH: was involved in the conception, design, analysis, writing, and interpretation of this work; JLW: was involved in the conception, design, writing, and interpretation of this work; LAY: was involved in the conception, design, writing, and interpretation of this work; RGZ: contributed to the conception, design, analysis, writing, and interpretation of this work. None of the authors had a conflict of interest to report.

REFERENCES 1. Papas MA, Giovannucci E, Platz EA. Fiber from fruit and colorectal neoplasia. Cancer Epidemiol Biomarkers Prev 2004;13:1267–70. 2. Bostick RM. Nutrition and colon cancer prevention. Nestle Nutr Workshop Ser Clin Perform Programme 2000;4:67– 86. 3. Steinmetz KA, Potter JD. Vegetables, fruit, and cancer. II. Mechanisms Cancer Causes Control 1991;2:427– 42. 4. World Cancer Research Fund/American Institute for Cancer Research Food, nutrition and the prevention of cancer: a global perspective. Washington, DC: American Institute for Cancer Research, 1997. 5. Kato I, Akhmedkhanov A, Koenig K, Toniolo PG, Shore RE, Riboli E. Prospective study of diet and female colorectal cancer: the New York University Women’s Health Study. Nutr Cancer 1997;28:276 – 81. 6. Sellers TA, Bazyk AE, Bostick RM, et al. Diet and risk of colon cancer in a large prospective study of older women: an analysis stratified on family history (Iowa, United States). Cancer Causes Control 1998;9: 357– 67. 7. Singh PN, Fraser GE. Dietary risk factors for colon cancer in a low-risk population. Am J Epidemiol 1998;148:761–74. 8. Hsing AW, McLaughlin JK, Chow WH, et al. Risk factors for colorectal cancer in a prospective study among U.S. white men. Int J Cancer 1998;77:549 –53. 9. Pietinen P, Malila N, Virtanen M, et al. Diet and risk of colorectal cancer in a cohort of Finnish men. Cancer Causes Control 1999;10:387–96. 10. Voorrips LE, Goldbohm RA, van Poppel G, Sturmans F, Hermus RJ, van den Brandt PA. Vegetable and fruit consumption and risks of colon and rectal cancer in a prospective cohort study: The Netherlands Cohort Study on Diet and Cancer. Am J Epidemiol 2000;152:1081–92. 11. Michels KB, Edward G, Joshipura KJ, et al. Prospective study of fruit and vegetable consumption and incidence of colon and rectal cancers. J Natl Cancer Inst 2000;92:1740 –52. 12. Terry P, Giovannucci E, Michels KB, et al. Fruit, vegetables, dietary fiber, and risk of colorectal cancer. J Natl Cancer Inst 2001;93:525–33. 13. Flood A, Velie EM, Chaterjee N, et al. Fruit and vegetable intakes and the risk of colorectal cancer in the Breast Cancer Detection Demonstration Project follow-up cohort. Am J Clin Nutr 2002;75:936 – 43.

FRUIT AND VEGETABLE INTAKE AND COLORECTAL ADENOMA 14. McCullough ML, Robertson AS, Chao A, et al. A prospective study of whole grains, fruits, vegetables and colon cancer risk. Cancer Causes Control 2003;14:959 –70. 15. Sauvaget C, Nagano J, Allen N, Kodama K. Vegetable and fruit intake and stroke mortality in the Hiroshima/Nagasaki Life Span Study. Stroke 2003;34:2355– 60. 16. Khan MM, Goto R, Kobayashi K, et al. Dietary habits and cancer mortality among middle aged and older Japanese living in Hokkaido, Japan by cancer site and sex. Asian Pac J Cancer Prev 2004;5:58 – 65. 17. Sanjoaquin MA, Appleby PN, Thorogood M, Mann JI, Key TJ. Nutrition, lifestyle and colorectal cancer incidence: a prospective investigation of 10998 vegetarians and non-vegetarians in the United Kingdom. Br J Cancer 2004;90:118 –21. 18. Park Y, Subar AF, Kipnis V, et al. Fruit and vegetable intakes and risk of colorectal cancer in the NIH-AARP Diet and Health Study. Am J Epidemiol 2007;166:170 – 80. 19. Vainio H, Bianchini F. Fruit and vegetables. In: Vainio H, ed. IARC handbooks of cancer prevention. Lyon, France: IARC Press, 2003:72– 6. 20. Beresford SA, Johnson KC, Ritenbaugh C, et al. Low-fat dietary pattern and risk of colorectal cancer: the Women’s Health Initiative Randomized Controlled Dietary Modification Trial. JAMA 2006;295:643–54. 21. Morson B. President’s address. The polyp-cancer sequence in the large bowel. Proc R Soc Med 1974;67:451–7. 22. Macquart-Moulin G, Riboli E, Cornee J, Kaaks R, Berthezene P. Colorectal polyps and diet: a case-control study in Marseilles. Int J Cancer 1987;40:179 – 88. 23. Hoff G, Moen IE, Trygg K, et al. Colorectal adenomas and food. A prospective study of change in volume and total mass of adenomas in man. Scand J Gastroenterol 1988;23:1253– 8. 24. Kato I, Tominaga S, Matsuura A, Yoshii Y, Shirai M, Kobayashi S. A comparative case-control study of colorectal cancer and adenoma. Jpn J Cancer Res 1990;81:1101– 8. 25. Kune GA, Kune S, Read A, MacGowan K, Penfold C, Watson LF. Colorectal polyps, diet, alcohol, and family history of colorectal cancer: a case-control study. Nutr Cancer 1991;16:25–30. 26. Sandler RS, Lyles CM, Peipins LA, McAuliffe CA, Woosley JT, Kupper LL. Diet and risk of colorectal adenomas: macronutrients, cholesterol, and fiber. J Natl Cancer Inst 1993;85:884 –91. 27. Benito E, Cabeza E, Moreno V, Obrador A, Bosch FX. Diet and colorectal adenomas: a case-control study in Majorca. Int J Cancer 1993;55: 213–9. 28. Kono S, Imanishi K, Shinchi K, Yanai F. Relationship of diet to small and large adenomas of the sigmoid colon. Jpn J Cancer Res 1993;84: 13–9. 29. Little J, Logan RF, Hawtin PG, Hardcastle JD, Turner ID. Colorectal adenomas and diet: a case-control study of subjects participating in the Nottingham faecal occult blood screening programme. Br J Cancer 1993;67:177– 84. 30. Almendingen K, Trygg K, Larsen S, Hofstad B, Vatn MH. Dietary factors and colorectal polyps: a case-control study. Eur J Cancer Prev 1995;4:239 – 46. 31. Witte JS, Longnecker MP, Bird CL, Lee ER, Frankl HD, Haile RW. Relation of vegetable, fruit, and grain consumption to colorectal adenomatous polyps. Am J Epidemiol 1996;144:1015–25. 32. Platz EA, Giovannucci E, Rimm EB, et al. Dietary fiber and distal colorectal adenoma in men. Cancer Epidemiol Biomarkers Prev 1997; 6:661–70. 33. Lubin F, Rozen P, Arieli B, et al. Nutritional and lifestyle habits and water-fiber interaction in colorectal adenoma etiology. Cancer Epidemiol Biomarkers Prev 1997;6:79 – 85. 34. Faivre J, Boutron MC, Senesse P, Couillault C, Belighiti C, Meny B. Environmental and familial risk factors in relation to the colorectal adenoma– carcinoma sequence: results of a case-control study in Burgundy (France). Eur J Cancer Prev 1997;6:127–31. 35. Schatzkin A, Lanza E, Corle D, et al. Lack of effect of a low-fat, highfiber diet on the recurrence of colorectal adenomas. Polyp Prevention Trial Study Group. N Engl J Med 2000;342:1149 –55. 36. Almendingen K, Hofstad B, Trygg K, Hoff G, Hussain A, Vatn M. Current diet and colorectal adenomas: a case-control study including different sets of traditionally chosen control groups. Eur J Cancer Prev 2001;10:395– 406. 37. Breuer-Katschinski B, Nemes K, Marr A, Rump B, Leiendecker B, Breuer N, et al. Relation of serum antioxidant vitamins to the risk of colorectal adenoma. Digestion 2001;63:43– 8.

1763

38. Smith-Warner SA, Elmer PJ, Fosdick L, et al. Fruits, vegetables, and adenomatous polyps: the Minnesota Cancer Prevention Research Unit case-control study. Am J Epidemiol 2002;155:1104 –13. 39. Mathew A, Peters U, Chatterjee N, Kulldorff M, Sinha R. Fat, fiber, fruits, vegetables, and risk of colorectal adenomas. Int J Cancer 2004; 108:287–92. 40. Michels KB, Giovannucci E, Chan AT, Singhania R, Willett WC. Fruit and vegetable consumption and colorectal adenomas in the Nurses’ Health Study. Cancer Res 2006;66:3942–53. 41. Almendingen K, Hofstad B, Vatn MH. Dietary habits and growth and recurrence of colorectal adenomas: results from a three-year endoscopic follow-up study. Nutr Cancer 2004;49:131– 8. 42. Lanza E, Hartman TJ, Albert PS, et al. High dry bean intake and reduced risk of advanced colorectal adenoma recurrence among participants in the polyp prevention trial. J Nutr 2006;136:1896 –903. 43. Agurs-Collins T, Smoot D, Afful J, Makambi K, Adams-Campbell LL. Legume intake and reduced colorectal adenoma risk in AfricanAmericans. J Natl Black Nurses Assoc 2006;17:6 –12. 44. Austin GL, Adair LS, Galanko JA, Martin CF, Satia JA, Sandler RS. A diet high in fruits and low in meats reduces the risk of colorectal adenomas. J Nutr 2007;137:999 –1004. 45. Gohagan JK, Prorok PC, Hayes RB, Kramer BS. The Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial of the National Cancer Institute: history, organization, and status. Control Clin Trials 2000;21(suppl):251S–72S. 46. Prorok PC, Andriole GL, Bresalier RS, Buys SS, Chia D, Crawford ED, et al. Design of the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial. Control Clin Trials 2000;21(suppl):273S–309S. 47. Weissfeld JL, Schoen RE, Pinsky PF, Bresalier RS, Church T, Yurgalevitch S, et al. Flexible sigmoidoscopy in the PLCO cancer screening trial: results from the baseline screening examination of a randomized trial. J Natl Cancer Inst 2005;97:989 –97. 48. Subar AF, Thompson FE, Kipnis V, et al. Comparative validation of the Block, Willett, and National Cancer Institute food frequency questionnaires: the Eating at America’s Table Study Am J Epidemiol 2001;154: 1089 –99. 49. Block G, Woods M, Potosky A, Clifford C. Validation of a selfadministered diet history questionnaire using multiple diet records. J Clin Epidemiol 1990;43:1327–35. 50. Willett WC, Sampson L, Stampfer MJ, et al. Reproducibility and validity of a semiquantitative food frequency questionnaire. Am J Epidemiol 1985;122:51– 65. 51. Subar AF, Thompson FE, Smith AF, et al. Improving food frequency questionnaires: a qualitative approach using cognitive interviewing. J Am Diet Assoc 1995;95:781–90. 52. Thompson FE, Subar AF, Brown CC, et al. Cognitive research enhances accuracy of food frequency questionnaire reports: results of an experimental validation study. J Am Diet Assoc 2002;102:212–25. 53. Subar AF, Midthune D, Kulldorff M, et al. Evaluation of alternative approaches to assign nutrient values to food groups in food frequency questionnaires. Am J Epidemiol 2000;152:279 – 86. 54. Stolzenberg-Solomon RZ, Chang SC, Leitzmann MF, et al. Folate intake, alcohol use, and postmenopausal breast cancer risk in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial. Am J Clin Nutr 2006;83:895–904. 55. Cook AJ, Friday JE. 2004. Pyramid servings database for USDA survey food codes version 2.0. Internet: http://www.ba.ars.usda.gov/cnrg (accessed 18 October 2007). 56. Shaw A, Fulton L, Davis C, Hogbin M. Using the food guide pyramid: a resource for nutrition educators. Internet: http://www.nal.usda.gov/ fnic/Fpyr/guide.pdf (accessed 18 October 2007). 57. Millen AE, Midthune D, Thompson FE, Kipnis V, Subar AF. The National Cancer Institute diet history questionnaire: validation of pyramid food servings. Am J Epidemiol 2006;163:279 – 88. 58. Willett WC, Howe GR, Kushi LH. Adjustment for total energy intake in epidemiologic studies. Am J Clin Nutr 1997;65(suppl):1220S–31S. 59. Hill MJ, Morson BC, Bussey HJ. Aetiology of adenoma– carcinoma sequence in large bowel. Lancet 1978;1(8058):245–7. 60. Kim EC, Lance P. Colorectal polyps and their relationship to cancer. Gastroenterol Clin North Am 1997;26:1–17. 61. Subar AF, Kipnis V, Troiano RP, et al. Using intake biomarkers to evaluate the extent of dietary misreporting in a large sample of adults: the OPEN study. Am J Epidemiol 2003;158:1–13. 62. Michels KB, Fuchs CS, Giovannucci E, et al. Fiber intake and incidence

1764

63.

64.

65.

66. 67.

MILLEN ET AL

of colorectal cancer among 76,947 women and 47,279 men. Cancer Epidemiol Biomarkers Prev 2005;14:842–9. Bingham SA, Day NE, Luben R, et al. Dietary fibre in food and protection against colorectal cancer in the European Prospective Investigation into Cancer and Nutrition (EPIC): an observational study. Lancet 2003; 361(9368):1496 –501. Peters U, Sinha R, Chatterjee N, et al. Dietary fibre and colorectal adenoma in a colorectal cancer early detection programme. Lancet 2003; 361(9368):1491–5. Sanjoaquin MA, Allen N, Couto E, Roddam AW, Key TJ. Folate intake and colorectal cancer risk: a meta-analytical approach. Int J Cancer 2005;113:825– 8. Bingham S. The fibre-folate debate in colo-rectal cancer. Proc Nutr Soc 2006;65:216 –1. Saydah SH, Platz EA, Rifai N, Pollak MN, Brancati FL, Helzlsouer KJ. Association of markers of insulin and glucose control with subsequent colorectal cancer risk. Cancer Epidemiol Biomarkers Prev 2003;12:412– 8.

68. Giacco R, Parillo M, Rivellese AA, et al. Long-term dietary treatment with increased amounts of fiber-rich low-glycemic index natural foods improves blood glucose control and reduces the number of hypoglycemic events in type 1 diabetic patients. Diabetes Care 2000;23:1461– 6. 69. Giovannucci E. Insulin, insulin-like growth factors and colon cancer: a review of the evidence. J Nutr 2001;131(suppl):3109S–20S. 70. Ma J, Giovannucci E, Pollak M, et al. A prospective study of plasma C-peptide and colorectal cancer risk in men. J Natl Cancer Inst 2004; 96(7):546 –53. 71. Gamet L, Daviaud D, Denis-Pouxviel C, Remesy C, Murat JC. Effects of short-chain fatty acids on growth and differentiation of the human coloncancer cell line HT29. Int J Cancer 1992;52:286 –9. 72. Freudenheim JL, Marshall JR. The problem of profound mismeasurement and the power of epidemiological studies of diet and cancer. Nutr Cancer 1988;11:243–50. 73. Freedman LS, Schatzkin A, Wax Y. The impact of dietary measurement error on planning sample size required in a cohort study. Am J Epidemiol 1990;132:1185–95.

APPENDIX A Fruit and vegetables contained in the different food groups Food group Fruit Total fruit

Fruit, without juice Fruit juice Citrus, melon, berries

Vegetables Total vegetables

Total vegetables without white potatoes Total vegetables without beans, white potatoes, and starchy vegetables Cruciferous vegetables Deep-yellow vegetables Dark-green vegetables Tomatoes White potatoes Dry beans and peas Allium vegetables Total fruit, and vegetables Total fruit and vegetables

Fruit and vegetables Orange and grapefruit juice ѿ apple juice ѿ fresh apples in season ѿ fresh apples rest of the year ѿ applesauce ѿ fresh pears in season ѿ fresh pears rest of the year ѿ bananas ѿ peaches and nectarines in season ѿ peaches, canned ѿ fresh plums ѿ cantaloupe in season ѿ watermelon in season ѿ strawberries in season ѿ strawberries rest of the year ѿ fresh oranges in season ѿ fresh oranges rest of the year ѿ fresh grapefruit in season ѿ fresh grapefruit rest of the year ѿ grapes ѿ apricots ѿ raisins ѿ prunes ѿ pineapple, canned ѿ fruit cocktail Total fruits Ҁ fruit juice Orange and grapefruit juice ѿ apple juice, cider ѿ other juice Orange and grapefruit juice ѿ cantaloupe in season ѿ watermelon in season ѿ strawberries in season ѿ strawberries rest of the year ѿ fresh oranges in season ѿ fresh oranges rest of the year ѿ fresh grapefruit in season ѿ fresh grapefruit rest of the year Tomato and vegetable juice ѿ string and green beans ѿ peas ѿ corn in season ѿ corn rest of the year ѿ summer squash ѿ winter squash ѿ broccoli ѿ cauliflower ѿ Brussels sprouts ѿ spinach raw ѿ spinach cooked ѿ mustard greens, etc ѿ mixed vegetables ѿ coleslaw, cabbage, etc ѿ carrots, cooked ѿ carrots, raw ѿ head lettuce ѿ leaf lettuce ѿ green pepper ѿ cucumber ѿ celery ѿ beets ѿ fresh tomatoes in season ѿ fresh tomatoes rest of the year ѿ canned tomatoes ѿ tomato sauce ѿ ketchup, etc ѿ onions ѿ garlic ѿ French fries ѿ potatoes, other ѿ sweet potatoes, yam ѿ tofu, soybeans ѿ chili with beans ѿ beans, other Total vegetables Ҁ white potatoes Total vegetables Ҁ (beans ѿ white potatoes ѿ starchy vegetables) Broccoli ѿ cauliflower ѿ Brussels sprouts ѿ coleslaw, cabbage, etc Winter squash ѿ carrots, cooked ѿ carrots, raw ѿ sweet potatoes, yam Broccoli ѿ spinach raw ѿ spinach cooked ѿ mustard greens, etc ѿ leaf lettuce Tomato and vegetable juice ѿ fresh tomatoes in season ѿ fresh tomatoes rest of the year ѿ canned tomatoes ѿ tomato sauce ѿ ketchup, etc French fries ѿ potatoes, other Tofu, soybeans ѿ chili with beans ѿ beans, other Onions ѿ garlic Total fruit ѿ total vegetables

Early protein intake and later obesity risk: which protein sources at which time points throughout infancy and childhood are important for body mass index and body fat percentage at 7 y of age?1–3 Anke LB Gu¨nther, Thomas Remer, Anja Kroke, and Anette E Buyken ABSTRACT Background: A high early protein intake has been proposed to increase obesity risk. Objective: We examined whether a critical period of protein intake for later obesity may exist early in childhood and investigated the relation between protein intake from different sources and body mass index SD score and body fat percentage (BF%) at 7 y of age. Design: The study population included 203 participants of the Dortmund Nutritional and Longitudinally Designed Study with information on diet at 6 mo, 12 mo, 18-24 mo, 3-4 y, and 5-6 y. Life-course plots were constructed to assess when protein intake (% of energy) was associated with body mass index SD score and BF% at 7 y. Mean values were then compared among tertiles (T1-T3) of protein from different sources at the important time points. Results: The ages of 12 mo and 5-6 y were identified as critical ages at which higher total and animal, but not vegetable, protein intakes were positively related to later body fatness. In fully adjusted models, animal protein intake at 12 mo was associated with BF% at 7 y as follows [x៮ (95% CI) BF%]: T1, 16.20 (15.23, 17.25); T2, 17.21 (16.24, 18.23); T3, 18.21 (17.12, 19.15); P for trend ҃ 0.008. With respect to food groups, dairy, but not meat or cereal protein intake, at 12 mo was related to BF% at 7 y (P for trend ҃ 0.07). Animal protein at 5-6 y yielded similar results (P for trend ҃ 0.01), but food group associations were less consistent. Conclusion: A higher animal, especially dairy, protein intake at 12 mo may be associated with an unfavorable body composition at 7 y. The age of 5-6 y might represent another critical period of protein intake for later obesity risk. Am J Clin Nutr 2007;86:1765–72. KEY WORDS adiposity rebound

BMI, obesity, dairy protein, critical periods,

INTRODUCTION

Dietary factors during the sensitive period of infancy and early childhood are increasingly recognized as being potentially critical for adult disease and predisposition to obesity (1, 2). In this context, protein has received particular attention, because the “early protein” hypothesis postulates that high protein intakes in the first months of life increase the risk of subsequent obesity, possibly by inducing distinct hormonal responses, such as stimulating the secretion of insulin and insulin-like growth factor-I (IGF-1) (3). Although the evidence in favor of this hypothesis is still limited, there are 2 main reasons the first years of life may actually

represent a critical time window with regard to protein intake and later adiposity. First, infant formula is characterized by an 앒50 – 80% higher protein content than human milk (4), which has accordingly been discussed as one mechanism behind the commonly observed increased risk of later obesity in nonbreastfed children (5). Second, during the period of complementary feeding and the transition to the family diet, there is a rapid increase in protein intake. In various populations from 9 mo of age onward, it has been reported to exceed current recommendations (6), up to almost 5-fold in some children (7, 8). In fact, recent analyses of ours suggested that the persistence of a high protein intake at both 12 and 18 –24 mo is associated with a higher body mass index SD score (BMI SDS) and body fat percentage (BF%) at 7 y of age (9). Nevertheless, it cannot be excluded that dietary protein in later childhood, eg, during “adiposity rebound” [which itself represents a critical period for obesity development (10)], might play an ongoing role regarding later body fatness. Particularly in a critical period of intake, if present, it is conceivable that only certain protein qualities or protein intake from distinct food groups may be responsible for any relation with later obesity risk. Most studies in the context of the early protein hypothesis have so far concentrated on total protein intake (3, 9, 11–14); however, the main protein sources once complementary feeding has commenced [meat, dairy, and cereal (15)] are known to exert differential metabolic effects in children. In particular, several studies have suggested that cow milk but not meat intake or vegetable protein stimulates the secretion of insulin and IGF-1 in pediatric age groups (16 –18). Using data from the Dortmund Nutritional and Anthropometric Longitudinally Designed (DONALD) Study, the objectives of the present analysis were therefore 1) to evaluate whether certain time points or periods of protein intake in infancy, early 1 From the Research Institute of Child Nutrition, Affiliated Institute of the University of Bonn, Dortmund, Germany (ALBG, AEB, and TR), and the Department of Nutrition, Food and Consumer Sciences, Fulda University of Applied Sciences, Fulda, Germany (AK). 2 Supported by the Ministry of Science and Research of North Rhine Westphalia, Germany, and by a research grant from the International Foundation for the Promotion of Nutrition Research and Nutrition Education (to ALBG). 3 Reprints not available. Address correspondence to ALB Gu¨nther, Nutrition and Health Unit, Research Institute of Child Nutrition, Heinstueck 11, 44225 Dortmund, Germany. E-mail: [email protected]. Received May 8, 2007. Accepted for publication August 7, 2007.

Am J Clin Nutr 2007;86:1765–72. Printed in USA. © 2007 American Society for Nutrition

1765

1766

¨ NTHER ET AL GU

childhood, or the preschool years may be most important with regard to BMI and BF% at 7 y of age, and, based on these results, 2) to investigate whether only protein intake from distinct sources (animal, vegetable, dairy, meat, and cereal protein) could be responsible for potential associations between early total protein intake and later body fatness. SUBJECTS AND METHODS

Study population The DONALD Study is an ongoing, open cohort study that was started in 1985 in the area of Dortmund, Germany. Details of the study design have been published previously (19). In brief, an average of 40 –50 infants are newly recruited each year and first examined at the age of 3– 6 mo. From then on, detailed data on nutrition, growth, metabolism, and health status are collected at regular intervals between infancy and young adulthood, ie, up to 3 further visits in the first year of life, 2 in the second, and 1 per year thereafter. The study was approved by the Ethics Committee of the University of Bonn, and all assessments are performed with parental consent. For the purpose of this analysis, we only used data from term (gestational age 37– 42 wk) singletons with a minimum birth weight of 2500 g. Furthermore, several additional criteria had to be met: 1) complete anthropometric measurements (weight, height, 4 skinfold thicknesses) had to be available at ages 6 mo (baseline), 3– 4 y, and 7 y (endpoint) (n ҃ 340); 2) this number was reduced to those with plausible dietary records at the ages of 6 mo, 12 mo, 18 –24 mo (욷1 record out of possible 2), 3– 4 y (욷1 out of possible 2), and 5– 6 y (욷1 out of possible 2) (n ҃ 205); and finally, 3) information on potential confounders, eg, maternal overweight (BMI 욷 25; yes or no), maternal educational status (12 y schooling; yes or no), breastfeeding (full breastfeeding for 욷4 mo; yes or no), firstborn status (yes or no), and siblings in the dataset (yes or no), had to be available (n ҃ 203). Hence, the subcohort analyzed here consisted of 203 children (102 boys, 101 girls). This number was sufficient to detect a difference of 0.40 BMI SDS and 1.70% body fat (original scale) between 2 equal groups and a mean difference of 0.48 BMI SDS and 2.07% body fat between the highest and the lowest tertile with ␣ ҃ 0.05 and a power of 80% (two-tailed) (20). We chose 7 of age as our endpoint because at that age, BMI correlates well with BMI in adulthood (21, 22). Parental and birth characteristics On a child’s entry to the study, parents are asked to provide information about family characteristics and their educational status and employment, and their weight and height are measured by the same trained nurses who assess the anthropometrics of the participating children. Information on birth weight, birth length, and gestational age are abstracted from a standardized document given to all pregnant women in Germany. Anthropometric data At each visit, anthropometric measurements are performed by trained nurses according to standard procedures, with the children dressed in underwear only. The nurses undergo an annual quality-control check in which intra- and interobserver agreement is carefully monitored. Body weight is assessed to the nearest 100 g with a supine infant weighing scale (Mettler PS 15;

Mettler Toledo, Columbus, OH) or an electronic scale for subjects in the standing position (Seca 753 E; Seca GmbH & Co KG, Hamburg, Germany). Recumbent length in children up to 24 mo of age is measured to the nearest 0.1 cm by using a Harpenden stadiometer (Holtain Ltd, Crymych, United Kingdom). From 24 mo of age onward, standing height is measured to the nearest 0.1 cm with a digital telescopic wall-mounted stadiometer. Skinfold thicknesses are measured from the age of 6 mo onward on the right side of the body at the biceps, triceps, subscapular, and suprailiac sites to the nearest 0.1 mm by using a Holtain caliper (Holtain Ltd). For each child, age- and sex-independent SD scores of weight, height, and BMI were calculated by using the German reference curves (23). To correct for general deviations of our sample from the reference data, BMI SDS values were internally standardized (mean ҃ 0, SD ҃ 1; according to age and sex) for the multivariable analyses. Body density and BF% were calculated by using Deurenberg’s equations (24). Because BF% values were skewed, we applied log-transformations (lnBF%) in all analyses and present geometric means with 95% CIs throughout. Furthermore, we assessed the proportion of overweight children according to the definition of the International Obesity Task Force (25). In an analogous manner, the 85th percentile of the body fat reference curves published by McCarthy et al (26) formed the basis for the definition of overfat. Dietary data In the DONALD Study, dietary intake is assessed by use of 3-d weighed-diet records. Parents are asked to weigh all foods and beverages consumed by their children, including leftovers (eg, in milk bottles), to the nearest 1 g over 3 consecutive days with the use of regularly calibrated electronic food scales (initially Soehnle Digita 8000, Leifheit SG, Nassau, Germany; now WEDO digi 2000, Werner Dorsch GmbH, Muenster/Dieburg, Germany). With regard to breastfeeding, test weighing is performed, which means that the intake of breastmilk is assessed by weighing the infant before and after each meal to the nearest 10 g by using an infant-weighing scale (Soehnle multina 8300, Leifheit SG). In this analysis, 5% was added to the test weighing results to account for insensible water losses (27). Parents are instructed by trained dietitians. Semiquantitative recording with household measures (eg, number of spoons or scoops) is allowed when exact weighing is not possible (eg, foods eaten away from home); however, 91.5% of the meals reported in the present sample were eaten or prepared at home. Information on recipes or the types and brands of food items is also requested, and at the end of the 3-d record period, a dietitian visits the family and checks the record for completeness and accuracy. The dietary records are analyzed by using the continuously updated in-house nutrient database LEBTAB (28), which incorporates information from standard nutrient tables, product labels, or recipe simulation based on the labeled ingredients and nutrients (eg, commercial weaning foods). For the purpose of this study, total energy (kcal/d) and protein intakes (g/d) at all time points between 6 mo and 6 y were derived for each participant from the mean of the respective 3 dietary recording days. The reported energy intake was related to the basal metabolic rate (29), and age- and sex-specific cut-offs (30) were used to exclude potentially implausible records (2.6%). To represent diet at 18 –24 mo, 3– 4 y, and 5– 6 y, we took the mean

PROTEIN SOURCES AND LATER BODY FATNESS

of single standardized energy intakes (mean ҃ 0, SD ҃ 1) or the mean nutrient intakes at the respective time points. In addition to total protein intake, we also considered animal protein (excluding protein from human milk because its effect on obesity risk was expected to differ from that of other animal sources and might have confounded their associations) and vegetable protein intake. Animal and vegetable protein intakes were further divided into protein from the following food groups that are known to be main contributors once complementary feeding has commenced (15): 1) dairy protein (eg, from cow milk, custard and other milk desserts, yogurt, buttermilk, and cheese), 2) meat protein (eg, from beef, pork, poultry, and meat products like ham and sausages), and 3) cereal protein (eg, from bread, breakfast cereals, pasta, rice, and flour). To create these food groups, complementary and convenience foods were broken down into their components, eg, commercial baby meals were divided into vegetables and meat. Infant formula was included in the calculation of total and animal protein, but, as is common practice (15), not in the dairy food group. First, infant formula represents milk feeding, whereas dairy products are first introduced during the complementary feeding period. Second, it is unique to the first years of life and of only minor relevance once the family diet has been established. We also did not consider further food subgroups (eg, protein from fluid milk versus solid dairy foods, protein from full-fat versus reduced-fat dairy), because in the first 2 y of life, these foods were not regularly consumed. Statistical analysis To evaluate which time points or periods of protein intake might be most important with regard to BMI and BF% at 7 y of age (objective 1), life-course plots were constructed. This approach, which has been described in detail by Cole (31), deals with the question of how to present the relation between an explanatory variable that has been repeatedly measured and changes during childhood, eg, weight or diet, and an outcome. To use this method in the present analysis, total, animal, and vegetable protein intakes at the ages of 6 mo (ie, 0.5 y), 12 mo (1 y), 18 –24 mo (1.5–2 y), 3– 4, and 5– 6 y were corrected for total energy intake by expressing them as nutrient densities (% of energy) and were standardized (mean ҃ 0, SD ҃ 1) to facilitate comparability. They were then entered into multiple linear regression models as independent variables that were adjusted for each other and with BMI SDS or BF% at 7 y of age as the outcomes. To remove its known effect on BF% values, sex was additionally included in the models with BF% as the dependent variable; in addition, the mean standardized energy intake from 0.5– 6 y was entered into all models. The resulting regression coefficients were then plotted against age, and both their values (representing the strength of the relations at a distinct time point) and their changes (representing the associations between outcome and change in the explanatory variables over the corresponding time interval) were evaluated to try to identify sensitive time points or periods of protein intake with respect to later body fatness. To specifically investigate whether only protein intake from distinct sources could be responsible for potential associations between early total protein intake and later body fatness (objective 2), analyses subsequently concentrated on the critical ages identified by the life-course plot approach. Tertiles of animal, vegetable, dairy, meat, and cereal protein intake at those time

1767

points were created, and their association with BMI SDS and BF% at 7 y of age were investigated in depth. First, adjusted mean BMI SDS and BF% outcome levels were calculated for each tertile of intake; second, P values for linear trends were derived from multiple linear regression analyses, treating the nutrient intakes as continuous variables. Tests for interactions between diet and sex did not indicate a significant effect modification. Thus, all analyses were performed with the total sample of 203 children. Relevant confounders were evaluated on an individual basis and in full models or were considered because of a priori interest. These included sex, maternal overweight (BMI 욷 25; yes or no), maternal educational attainment (욷12 y schooling; yes or no), maternal age at birth of the child, gestational age, firstborn status (yes or no), smoking in the household (yes or no), and breastfeeding characteristics (full breastfeeding for 욷4 mo, yes or no; still 욷50 mL human milk/d at ages 6 or 12 mo, yes or no). Because a considerable number of siblings participate in the DONALD Study, we also took the presence of siblings in our subcohort (yes or no) into account. Furthermore, an indicator variable was created to consider whether children had consumed dairy or meat protein throughout the recording days, but was only retained in the models if it tended to be associated with the outcome (P 울 0.1) or affected the estimates for protein intake (dairy protein at age 12 mo in the model with BMI SDS, meat protein at age 12 mo in the models with BF% as the outcome). To adjust for total energy and to be consistent with the lifecourse plot analyses, we chose the nutrient density approach, ie, all protein variables were expressed as percentages of energy and total energy intake was additionally included (32). In addition, we included fat intake (% of energy) in all models, because it has been proposed that it is not (only) a high protein intake, but also the typical simultaneous decrease in fat intake that might predispose children to later obesity (10, 33). All models with vegetable or cereal protein intake as the independent variable further included fiber intake (g/kcal). Finally, the effect of adding the baseline values of BMI SDS and BF% at the preceding time point (eg, 6 mo for diet at 12 mo, 3– 4 y for diet at 5– 6 y) was evaluated, taking means of sex- and age-standardized lnBF% values if the preceding time point consisted of more than one measurement (eg, 3– 4 y). All statistical analyses were carried out by using SAS (version 8.2; SAS Institute Inc, Cary, NC), and a P value of 쏝 0.05 was considered statistically significant. RESULTS

A general description of the study sample with respect to birth, breastfeeding, and family characteristics is given in Table 1. Overall, the prevalence of overweight at 7 y of age was moderate, with 1 child in 7 having a BMI (in kg/m2) above the International Obesity Task Force cutoffs of 17.92 for boys and 17.75 for girls, respectively (24). The children’s energy and protein intakes at 6 mo, 12 mo, 18 –24 mo, 3– 4 y, and 5– 6 y of age are summarized in Table 2. In general, median protein intake at all ages exceeded the current recommendations [1.4 g 䡠 kgҀ1 䡠 dҀ1 at 6 mo, 1.2 g 䡠 kgҀ1 䡠 dҀ1 at 12 mo, 1.0 g 䡠 kgҀ1 䡠 dҀ1 at 18 –24 mo, and 0.9 g 䡠 kgҀ1 䡠 dҀ1 at 3– 6 y (6)]. With respect to food groups, dairy, meat, and cereal products were mainly introduced between 6 and 12 mo. The results of the life-course plot analyses interestingly yielded similar results for total and animal protein intake. At both

1768

¨ NTHER ET AL GU

12 mo (ie, 1 y) and 5– 6 y of age, intakes were consistently and, except for total protein intake at 5– 6 y and later BF% (P ҃ 0.06), statistically significantly associated with a higher BMI SDS and a higher BF% at 7 y of age (Figure 1, A and B). Total and animal protein consumption at 3– 4 y tended to be inversely related to the outcomes, but the CI of its regression coefficients included 0 throughout and hence the association was not significant. Because the same applied to the regression coefficients at the other time points (0.5 y, 1 y, and 1.5–2 y), there was no indication of a period during which a change in protein intake was of importance for later body fatness, because this would have made a clear switch of signs of the coefficients mandatory (34). By contrast with total and animal protein intakes, the lifecourse plots of vegetable protein and its association with BMI SDS and BF% at 7 y of age did not indicate significance at any time point or period (Figure 1C). Furthermore, additional adjustments for breastfeeding (full breastfeeding for 욷4 mo; yes or no) did not influence any of the conclusions (data not presented). To address the possibility that especially those children with a higher animal protein intake at both 1 y and 5– 6 y had a higher body fatness later on, we evaluated the interaction of these factors in an additional regression model. However, there was no suggestion of effect modification (P ҃ 0.3– 0.8). On the basis of the results of the life-course plots, we subsequently concentrated on the ages of 12 mo and 5– 6 y in separate TABLE 1 General characteristics of the DONALD Study sample Variable No. of subjects [n (% female)] Birth characteristics Firstborn [n (%)] Birth weight (g) Birth length (cm) Gestational age (wk) Breastfeeding characteristics [n (%)] Ever fully breastfed Fully breastfed for 욷 4 mo Still considerable breastfeeding at 6 mo2 Still considerable breastfeeding at 12 mo2 Family characteristics [n (%)] Siblings in the dataset Smoking in the home Maternal characteristics Overweight3 [n (%)] 욷 12 y of schooling [n (%)] Age at birth of the child (y) Anthropometric characteristics at age 7 y Weight (kg) Height (cm) BMI (kg/m2) Body fat percentage (%)4 Overweight5 Overfat6 1

203 (49.8) 127 (62.6) 3480 (3230, 3820)1 52 (51, 53) 40 (39, 41) 163 (80.3) 125 (61.6) 110 (54.2) 9 (4.4) 49 (24.1) 46 (22.7) 53 (26.1) 125 (61.6) 31 (29, 33) 24.7 (22.8, 27.9) 125.4 (122.2, 128.8) 15.5 (14.8, 17.1) 16.9 (16.3, 17.5) 30 (14.8) 30 (14.8)

Median; 25th, 75th percentiles in parentheses (all such values). Defined as 욷 50 mL human milk/d. 3 BMI 욷 25 kg/m2. 4 Geometric mean (95% CI). 5 Defined by using the International Obesity Task Force definition [BMI 쏜 17.92 for boys, 17.75 for girls (24)]. 6 Defined as 쏜85th percentile of the body fat reference curves by McCarthy et al (25). 2

TABLE 2 Total energy and protein intakes of the DONALD Study sample at different ages in infancy, early childhood, and midchildhood1 Variable Energy (kcal/d) 6 mo 12 mo 18–24 mo 3–4 y 5–6 y Total protein (g 䡠 kgҀ1 䡠 dҀ1) 6 mo 12 mo 18–24 mo 3–4 y 5–6 y Total protein (% of energy) 6 mo 12 mo 18–24 mo 3–4 y 5–6 y Animal protein (% of energy)2 6 mo 12 mo 18–24 mo 3–4 y 5–6 y Vegetable protein (% of energy) 6 mo 12 mo 18–24 mo 3–4 y 5–6 y Dairy protein (% of energy) 6 mo 12 mo 18–24 mo 3–4 y 5–6 y Meat protein (% of energy) 6 mo 12 mo 18–24 mo 3–4 y 5–6 y Cereal protein (% of energy) 6 mo 12 mo 18–24 mo 3–4 y 5–6 y 1 2

640.4 (575.4, 706.0) 832.0 (726.0, 906.1) 927.6 (854.9, 1006.0) 1153.0 (1060.4, 1267.1) 1349.8 (1240.8, 1490.8) 2.1 (1.5, 2.5) 2.7 (2.4, 3.2) 2.7 (2.4, 3.0) 2.3 (2.0, 2.6) 2.0 (1.8, 2.3) 9.8 (7.5, 11.8) 13.3 (11.7, 14.8) 13.8 (12.5, 15.1) 12.9 (11.5, 14.0) 12.4 (11.2, 13.7) 5.6 (1.2, 8.9) 8.4 (7.1, 9.8) 9.3 (8.1, 10.6) 8.5 (7.1, 9.7) 7.8 (6.7, 8.8) 2.1 (0.9, 3.3) 4.8 (4.0, 5.7) 4.4 (3.8, 5.0) 4.2 (3.7, 4.9) 4.5 (4.0, 5.1) 0.9 (0, 2.7) 4.4 (2.5, 6.3) 5.6 (4.0, 7.2) 4.1 (3.0, 5.3) 3.5 (2.6, 4.5) 0.8 (0, 1.7) 2.0 (1.1, 2.7) 2.4 (1.6, 3.4) 2.7 (1.6, 3.7) 2.5 (1.8, 3.5) 0 (0, 0.5) 2.0 (1.2, 2.7) 2.2 (1.7, 2.8) 2.4 (1.9, 3.0) 2.7 (2.3, 3.4)

Values are medians; 25th and 75th percentiles in parentheses. Excluding human milk protein.

models because of age-specific confounders (eg, baseline anthropometrics) to investigate the role of different protein qualities and protein from distinct food groups on BMI SDS and BF% at 7 y of age. As shown in Table 3, a higher animal protein intake (% of energy) at 12 mo was related to a higher BMI SDS at 7 y, which was not explained by family and maternal characteristics or the inclusion of BMI SDS at baseline (P for continuous trend ҃ 0.03). A similar, albeit weaker, tendency was obtained by a full

1769

PROTEIN SOURCES AND LATER BODY FATNESS A

Regression coefficient

0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4

P=0.001 P=0.7

P=0.02 P=0.7 P=0.1

0.5

1

1.5-2.0

3.0-4.0

Regression coefficient

lnBF%

BMI SDS

0.12 0.1 0.08 0.06 0.04 0.02 0 -0.02 -0.04 -0.06 -0.08

5.0-6.0

P=0.06

P=0.03 P=0.6 P=0.99 P=0.1

0.5

1

Age (y)

1.5-2.0

3.0-4.0

5.0-6.0

Age (y)

B

Regression coefficient

0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4

P=0.006

P=0.008 P=0.6

P=0.6 P=0.2

0.5

1

1.5-2.0

3.0-4.0

Regression coefficient

lnBF%

BMI SDS

0.12 0.1 0.08 0.06 0.04 0.02 0 -0.02 -0.04 -0.06 -0.08

5.0-6.0

P=0.02

P=0.03 P=0.7

P=0.95 P=0.4

0.5

1

Age (y)

1.5-2.0

3.0-4.0

5.0-6.0

Age (y)

C

Regression coefficient

0.6 0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4

P=0.7

P=0.8

P=0.7

P=0.4

0.5

P=0.3

1

1.5-2.0

3.0-4.0

5.0-6.0

Regression coefficient

lnBF%

BMI SDS

0.12 0.1 0.08 0.06 0.04 0.02 0

P=0.9 P=0.6

P=0.6

P=0.7 P=0.2

-0.02 -0.04 -0.06 -0.08 0.5

1

1.5-2.0

3.0-4.0

5.0-6.0

Age (y)

Age (y)

FIGURE 1. Life-course plots of multiple linear regression analyses with BMI SD score (SDS) and the natural log of body fat percentage (lnBF%) at 7 y of age as the outcome and standardized A) total protein, B) animal protein, and C) vegetable protein intakes (% of energy) at different time points throughout infancy and childhood as the explanatory variables, adjusted for each other, mean standardized energy intake from 0.5– 6 y, and for sex in the case of lnBF%. n ҃ 203 participants of the Dortmund Nutritional and Longitudinally Designed (DONALD) Study. Values are regression coefficients (95% CI).

model with animal protein intake at 5– 6 y (P ҃ 0.07). By contrast, higher protein consumption from vegetable sources at either 12 mo or 5– 6 y was not associated with the outcome (P ҃ 0.8 and P ҃ 0.2, respectively). With respect to food groups, a higher protein intake from dairy foods (% of energy) at 12 mo was associated with a higher BMI SDS at 7 y, similar to animal protein intake (P ҃ 0.02). However, dairy protein intake at 5– 6 y did not yield comparable results (P ҃ 0.7). For both 12 mo and 5– 6 y, there was no relation of meat or cereal protein intake with later BMI SDS. Regarding BF% at 7 y of age, the results were comparable overall with those for BMI SDS (Table 4). Again, dairy but not meat or cereal protein intake at the age of 12 mo showed similar associations as did animal protein consumption and tended to be related to a higher later BF% (P ҃ 0.07). A higher animal protein

intake at 5– 6 y was also associated with BF% at 7 y; however, a similar tendency existed for dairy protein also (P ҃ 0.1). In contrast with BMI SDS, a higher vegetable protein intake at 5– 6 y (but not 12 mo) showed an inverse relation to later BF% of borderline significance (P ҃ 0.05). Additional adjustments for other potential confounders, such as breastfeeding or maternal educational status, did not change the results notably, nor did the additional inclusion of the protein intakes at the 3 other time points initially examined (6 mo, 18 –24 mo, and 3– 4 y; data not shown).

DISCUSSION

This is the first study to compare the relevance for later body fatness of protein intake from different sources at various time

¨ NTHER ET AL GU

1770

TABLE 3 Adjusted mean BMI standard deviation scores (BMI SDS) at 7 y of age according to tertiles of protein intake (% of energy) at ages 12 mo and 5– 6 y for participants of the DONALD Study (n ҃ 203)1 Tertiles of protein intake

Animal protein (% of energy) 12 mo 5–6 y Vegetable protein (% of energy)3 12 mo 5–6 y Dairy protein (% of energy) 12 mo 5–6 y Meat protein (% of energy) 12 mo 5–6 y Cereal protein (% of energy)3 12 mo 5–6 y

3

P for trend2

1

2

Ҁ0.04 (Ҁ0.29, 0.20) 0.07 (Ҁ0.08, 0.22)

0.01 (Ҁ0.21, 0.24) 0.07 (Ҁ0.08, 0.22)

0.39 (0.16, 0.61) 0.19 (0.05, 0.34)

0.03 0.07

0.18 (Ҁ0.10, 0.46) 0.16 (Ҁ0.02, 0.34)

0.01 (Ҁ0.23, 0.25) 0.16 (0.01, 0.30)

0.23 (Ҁ0.04, 0.50) 0.03 (Ҁ0.15, 0.20)

0.8 0.2

0.03 (Ҁ0.21, 0.27) 0.03 (Ҁ0.12, 0.19)

0.02 (Ҁ0.22, 0.26) 0.17 (0.03, 0.32)

0.35 (0.13, 0.57) 0.13 (Ҁ0.02, 0.27)

0.02 0.7

0.13 (Ҁ0.10, 0.36) 0.05 (Ҁ0.10, 0.20)

0.25 (0.02, 0.48) 0.14 (Ҁ0.01, 0.29)

0.05 (Ҁ0.19, 0.29) 0.15 (0.00, 0.30)

0.3 0.4

0.12 (Ҁ0.12, 0.36) 0.15 (Ҁ0.01, 0.32)

0.07 (Ҁ0.16, 0.30) 0.09 (Ҁ0.06, 0.24)

0.26 (0.01, 0.51) 0.11 (Ҁ0.04, 0.27)

0.5 0.9

1

Values are adjusted means (95% CI). Based on multiple linear regression analyses (dietary intakes as continuous variables) with adjustment for total energy and fat (% of energy) intake, siblings in the dataset (yes or no), maternal overweight (BMI 욷 25; yes or no), and BMI SDS at baseline (6 mo and 3– 4 y, respectively). 3 Models with vegetable or cereal protein intake as explanatory variables also included fiber intake (g/kcal). 2

points in infancy and childhood. Our results support the hypothesis that the end of the first year of life, when children undergo the transition from breast milk or formula feeding to a diet based on family foods, may represent a critical phase with respect to protein intake and subsequent obesity risk. Our results indicate that animal, in particular dairy, protein intake might actually be responsible for these associations. To identify critical phases of protein intake, we evaluated closely spaced, repeatedly collected dietary data and used a lifecourse plot analysis as proposed by Cole (31). To the best of our

knowledge, this approach has not been used in nutritional epidemiology so far, although taking a life-course perspective has been explicitly called for when investigating determinants of later disease (35). Interestingly, the results of the life-course plot analyses suggested that 5– 6 y of age may represent a second sensitive period of total as well as animal protein intake for subsequent body fatness. Most pediatric studies describing associations between protein intake and subsequent obesity measures focused on diet in the first 2 y of life and did not include protein consumption later

TABLE 4 Adjusted mean body fat percentages (BF%) at 7 y of age according to tertiles of protein intake (% of energy) at ages 12 mo and 5– 6 y for participants of the DONALD Study (n ҃ 203)1 Tertiles of protein intake

Animal protein (% of energy) 12 mo 5–6 y Vegetable protein (% of energy)3 12 mo 5–6 y Dairy protein (% of energy) 12 mo 5–6 y Meat protein (% of energy) 12 mo 5–6 y Cereal protein (% of energy)3 12 mo 5–6 y 1

1

2

3

P for trend2

16.20 (15.23, 17.25) 16.83 (16.14, 17.55)

17.21 (16.24, 18.23) 16.95 (16.25, 17.68)

18.21 (17.12, 19.15) 18.02 (17.29, 18.79)

0.008 0.01

17.64 (16.47, 18.89) 17.42 (16.55, 18.34)

16.66 (15.70, 17.67) 17.48 (16.78, 18.21)

17.28 (16.12, 18.52) 16.89 (16.06, 17.76)

0.3 0.05

16.88 (15.90, 17.92) 16.80 (16.08, 17.55)

16.75 (15.75, 17.81) 17.21 (16.53, 17.96)

17.99 (17.01, 19.02) 17.60 (17.00, 18.48)

0.07 0.1

17.77 (16.74, 18.85) 17.07 (16.35, 17.82)

17.56 (16.57, 18.61) 17.33 (16.59, 18.09)

16.41 (15.46, 17.42) 17.44 (16.71, 18.21)

0.3 0.4

16.92 (15.96, 17.95) 17.36 (16.58, 18.19)

17.06 (16.12, 18.07) 17.40 (16.68, 18.14)

17.70 (16.65, 18.82) 17.06 (16.34, 17.80)

0.3 0.2

Values are adjusted geometric means (95% CI). Based on multiple linear regression analyses (dietary intakes as continuous variables) with adjustment for sex, total energy and fat (% of energy) intake, siblings in the dataset (yes or no), firstborn status (yes or no), maternal overweight (BMI 욷 25; yes or no), and lnBF% at baseline (6 mo and 3– 4 y, respectively). 3 Models with vegetable or cereal protein intake as explanatory variables also included fiber intake (g/kcal). 2

PROTEIN SOURCES AND LATER BODY FATNESS

on (3, 12–14, 36). However, Skinner et al (37) reported associations between mean protein intake in mid-childhood (2– 8 y) and BMI at 8 y of age in 70 children. By contrast, another prospective study (n ҃ 112) did not see a relation between protein intake at 8 y of age and BMI change over a period of 4 y (38). In this context, our results underline that it is crucial to discuss nutrients and sustained dietary consequences within distinct developmental or metabolic stages. At the age of 5– 6 y, the children in our cohort typically experience adiposity rebound (11), a potentially critical period for obesity development when weight gain starts to regain importance compared with height gain (10). It may well be that, in addition to complementary feeding, a high animal protein intake at this developmental stage also results in adverse effects on body composition and triggers gain in fat mass, as suggested by our findings. It can further be speculated that at time points other than these 2 possibly critical windows, the potentially favorable effects of higher dietary protein intakes, such as increased satiety, which have predominantly been suggested in adults (39, 40), are of greater relevance, and that different, age- and development-dependent protein effects on body composition might exist. The apparently differential associations of vegetable and dairy protein at 12 mo versus 5– 6 y with body fatness at 7 y of age may also be interpreted in this way. In the fully adjusted models, however, the results for protein intake at 5– 6 y of age were overall less convincing than those for 12 mo, eg, because of different results depending on whether BMI SDS or BF% was the outcome. Furthermore, the conclusions that can be drawn from our findings may be limited because of the methodologic limitation of a small time gap between the influencing and the dependent variable (diet at 5– 6 y, outcome at 7 y), which may result in difficulties in separating our prospective analyses from a cross-sectional approach. It can therefore not be excluded that the short-term effect seen for protein intake at 5– 6 y is lost if an endpoint further away is chosen, as opposed to the effect of intake at 12 mo of age. Here, the sustained effect on body fatness suggests a true programming effect of early diet. The fact that animal, in particular dairy, protein at age 12 mo might be responsible for the associations reported for total protein consumption agrees with both clinical and observational studies that have suggested a specific effect of these protein sources on IGF-1 and insulin secretion. The discussed mechanism behind the early protein hypothesis is that particularly these hormonal responses adversely affect preadipocyte differentiation and multiplication (3, 5). In a 7-d intervention study in 8-y-old boys (n ҃ 24), a higher protein intake from skimmed milk but not meat increased the serum concentrations of IGF-1 and the molar ratio of IGF-1 to its binding protein IGFBP-3 (17). Interestingly, the children in the milk group simultaneously displayed a significant increase in BMI, but it must be mentioned that the intervention also resulted in a difference in energy intake between the milk and the meat group. A stimulatory effect of animal protein in general, and dairy protein in particular, on the secretion of IGF-1 has further been suggested by cross-sectional, observational studies in children aged 2.5 y (n ҃ 90) (18) and 7– 8 y (n ҃ 521–538), respectively. In addition to stimulating IGF-1 secretion, it has further been shown in adults that milk seems to have potent insulinotropic properties (41). Correspondingly, in the 7-d intervention in 8-y-old boys cited above, fasting serum insulin and insulin resistance significantly increased in the milk but not in the meat group (16). With respect to this insulinotropic effect, the whey fraction of milk protein might be responsible

1771

(42). However, it is still a matter of debate which milk compound actually stimulates IGF-1 secretion (43). Likewise, it is possible that the adverse effects of dairy protein suggested by the present study stem from other, or at least additional, mechanisms. Only a few studies have prospectively examined the role of cow milk or dairy foods in the first 2 y of life and obesity measures so far. In a large observational study in infants that was nested within the cluster-randomized “Promotion of Breastfeeding Intervention Trial” PROBIT (n ҃ 17 046), a higher consumption of milk other than breast milk or formula at age 9 mo was associated with a larger weight-for-length z score gain between 9 and 12. This effect was greater than that of formula only, which is characterized by a lower protein content than that of whole-fat cow milk. Comparable with our results, cereal intake was not associated with weight-for-length z score in this time period (44). In general, however, the long-term effects of cow milk in complementary feeding are still under discussion, and recommendations about when it should first be introduced vary considerably among industrialized countries (45). Should the present results be confirmed, the role of cow milk and dairy products during complementary feeding needs further consideration. A potential adverse role with respect to obesity development would contradict recent findings of a beneficial effect of calcium or dairy for adiposity (46), but the evidence in children is controversial (47–50). Thus, as already discussed for protein intake in general, results obtained from studies in adults should probably not be simply adopted for early childhood. For example, patterns of dairy consumption in the first years of life can be expected to differ considerably from later ones, eg, because of an even higher intake of fluid milk instead of solid milk products. Several limitations of our study should be mentioned. First, in previous analyses of a subcohort of the DONALD Study, a high total protein intake at 12 mo of age was similarly associated with higher later body fatness but was statistically significant only when the intake had stayed high throughout the second year of life (9). Whether such an interaction with subsequent diet also existed for protein intake from different sources could not be answered by the life-course plot, but is not to be excluded given the fact that 93.6% of the children in the present sample were included in the previous study also. Second, the inclusion of blood samples with IGF-1 measurements would have been advantageous to confirm hypotheses on potential mechanisms but are unfortunately not available in the DONALD Study before young adulthood. Third, we also could not extend our analyses to puberty or young adulthood because the ongoing design of the DONALD Study means that many participants have not yet reached that age. This would have been interesting, especially for the purpose of evaluating the longer-term effect of diet at adiposity rebound. However, the closely spaced, detailed data represent a major strength of our study, as do the large number of family and dietary characteristics that we could consider in our analyses. In summary, our study suggests that a higher intake of animal protein at age 12 mo, in particular from cow milk and dairy products, might be associated with an unfavorable body composition at the age of 7 y. Furthermore, adiposity rebound (age 5– 6 y) might represent a second critical period of protein intake for subsequent body fatness. We thank the staff of the Research Institute of Child Nutrition for carrying out the anthropometric measurements and for collecting and coding the dietary records and all participants of the DONALD Study.

1772

¨ NTHER ET AL GU

The contributions of the authors were as follows—ALBG: conducted the statistical analyses and wrote the manuscript; ALBG and AEB: conceived the research project; all authors: made substantial contributions to the study design or the interpretation of the results. None of the authors had any financial or personal conflicts of interest.

REFERENCES 1. Lucas A. Programming by early nutrition: an experimental approach. J Nutr 1998;128:401S– 6S. 2. Martorell R, Stein AD, Schroeder DG. Early nutrition and later adiposity. J Nutr 2001;131:S874 – 80. 3. Rolland-Cachera MF, Deheeger M, Akrout M, Bellisle F. Influence of macronutrients on adiposity development: a follow up study of nutrition and growth from 10 months to 8 years of age. Int J Obes Relat Metab Disord 1995;19:573– 8. 4. Koletzko B, von Kries R. Are there long term protective effects of breast feeding against later obesity? Nutr Health 2001;15:225–36. 5. Koletzko B. Long-term consequences of early feeding on later obesity risk. In: Rigo J, Ziegler EE, eds. Protein and energy requirements in infancy and childhood. Nestlé Nutr Workshop Ser Pediatr Programm. 2006;58:1–18. 6. Garlick PJ. Protein requirements of infants and children. In: Rigo J, Ziegler EE, eds. Protein and energy requirements in infancy and childhood. Nestlé Nutr Workshop Ser Pediatr Programm. 2006;58:39 –50. 7. Rolland-Cachera MF, Deheeger M, Bellisle F. Increasing prevalence of obesity among 18-year-old males in Sweden: evidence for early determinants. Acta Paediatr 1999;88:365–7. 8. Michaelsen KF, Hoppe C, Schack-Nielsen L, Molgaard C. Does an excessive protein intake early in life cause health problems such as obesity later in life? In: Black RE, Fleischer Michaelsen K, eds. Public health issues in infant and child nutrition. Nestlé Nutr Workshop Ser Pediatr Program. 2002;48:279 –93. 9. Gu¨nther ALB, Buyken AE, Kroke A. Protein intake levels during the period of complementary feeding and early childhood and their association with BMI and body fat percentage at age 7. Am J Clin Nutr 2007;85:1626 –33. 10. Rolland-Cachera MF, Deheeger M, Maillot M, Bellisle F. Early adiposity rebound: causes and consequences for obesity in children and adults. Int J Obes (Lond) 2006;30(suppl 4):S11–7. 11. Gu¨nther ALB, Buyken AE, Kroke A. The influence of habitual protein intake in early childhood on BMI and age at adiposity rebound: results from the DONALD Study. Int J Obes (Lond) 2006;30:1072–9. 12. Gunnarsdottir I, Thorsdottir I. Relationship between growth and feeding in infancy and body mass index at the age of 6 years. Int J Obes Relat Metab Disord 2003;27:1523–7. 13. Hoppe C, Molgaard C, Thomsen BL, Juul A, Michaelsen KF. Protein intake at 9 mo of age is associated with body size but not with body fat in 10-y-old Danish children. Am J Clin Nutr 2004;79:494 –501. 14. Scaglioni S, Agostoni C, Notaris RD, et al. Early macronutrient intake and overweight at five years of age. Int J Obes Relat Metab Disord 2000;24:777– 81. 15. Fox MK, Reidy K, Novak T, Ziegler P. Sources of energy and nutrients in the diets of infants and toddlers. J Am Diet Assoc 2006;106:S28 – 42. 16. Hoppe C, Molgaard C, Vaag A, Barkholt V, Michaelsen KF. High intakes of milk, but not meat, increase s-insulin and insulin resistance in 8-year-old boys. Eur J Clin Nutr 2005;59:393– 8. 17. Hoppe C, Molgaard C, Juul A, Michaelsen KF. High intakes of skimmed milk, but not meat, increase serum IGF-I and IGFBP-3 in eight-year-old boys. Eur J Clin Nutr 2004;58:1211– 6. 18. Hoppe C, Udam TR, Lauritzen L, Molgaard C, Juul A, Michaelsen KF. Animal protein intake, serum insulin-like growth factor I, and growth in healthy 2.5-y-old Danish children. Am J Clin Nutr 2004;80:447–52. 19. Kroke A, Manz F, Kersting M, et al. The DONALD Study. History, current status and future perspectives. Eur J Nutr 2004;43:45–54. 20. Erdfelder E, Faul F, Buchner A. GPOWER: a general power analysis program. Behav Res Methods Instrum Comput 1996;28:1–11. 21. Williams S, Davie G, Lam F. Predicting BMI in young adults from childhood data using two approaches to modelling adiposity rebound. Int J Obes Relat Metab Disord 1999;23:348 –54. 22. Freedman DS, Kettel Khan L, Serdula MK, Srinivasan SR, Berenson GS. BMI rebound, childhood height and obesity among adults: the Bogalusa Heart Study. Int J Obes Relat Metab Disord 2001;25:543–9. 23. Kromeyer-Hauschild K, Wabitsch M, Kunze D, et al. Perzentile fu¨r den

24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37. 38. 39. 40. 41. 42.

43. 44. 45. 46. 47. 48. 49. 50.

Body-mass-Index fu¨r das Kindes- und Jugendalter unter Heranziehung verschiedener deutscher Stichproben. [Body mass index percentiles for children and adolescents using various German samples.] Monatsschr Kinderheilk 2001;149:807–18 (in German). Deurenberg P, Pieters JJ, Hautvast JG. The assessment of the body fat percentage by skinfold thickness measurements in childhood and young adolescence. Br J Nutr 1990;63:293–303. Cole TJ, Bellizzi MC, Flegal KM, Dietz WH. Establishing a standard definition for child overweight and obesity worldwide: international survey. BMJ 2000;320:1240 –3. McCarthy HD, Cole TJ, Fry T, Jebb SA, Prentice AM. Body fat reference curves for children. Int J Obes (Lond) 2006;30:598 – 602. Reilly JJ, Wells JC. Duration of exclusive breast-feeding: introduction of complementary feeding may be necessary before 6 months of age. Br J Nutr 2005;94:869 –72. Sichert-Hellert W, Kersting M, Chahda C, Schafer R, Kroke A. German food composition data base for dietary evaluations in children and adolescents. J Food Compos Anal 2007;20:63–70. Schofield WN. Predicting basal metabolic rate, new standards and review of previous work. Hum Nutr Clin Nutr 1985;39(suppl 1):5– 41. Sichert-Hellert W, Kersting M, Schoch G. Underreporting of energy intake in 1 to 18 year old German children and adolescents. Z Ernahrungswiss 1998;37:242–51. Cole TJ. Modeling postnatal exposures and their interactions with birth size. J Nutr 2004;134:201– 4. Willett W, Stampfer M. Implications of total energy intake for epidemiologic analyses. In: Willett W, ed. Nutritional epidemiology. 2nd ed. New York, NY: Oxford University Press, 1998:273–301. Rolland-Cachera MF, Deheeger M, Bellisle F. Nutrient balance and body composition. Reprod Nutr Dev 1997;37:727–34. De Stavola BL, Nitsch D, dos Santos Silva I, et al. Statistical issues in life course epidemiology. Am J Epidemiol 2006;163:84 –96. Ben-Shlomo Y, Kuh D. A life course approach to chronic disease epidemiology: conceptual models, empirical challenges and interdisciplinary perspectives. Int J Epidemiol 2002;31:285–93. Dorosty AR, Emmett PM, Cowin S, Reilly JJ. Factors associated with early adiposity rebound. ALSPAC Study Team. Pediatrics 2000;105:1115– 8. Skinner JD, Bounds W, Carruth BR, Morris M, Ziegler P. Predictors of children’s body mass index: a longitudinal study of diet and growth in children aged 2– 8 y. Int J Obes Relat Metab Disord 2004;28:476 – 82. Maffeis C, Talamini G, Tato L. Influence of diet, physical activity and parents’ obesity on children’s adiposity: a four-year longitudinal study. Int J Obes Relat Metab Disord 1998;22:758 – 64. Astrup A. The satiating power of protein–a key to obesity prevention? Am J Clin Nutr 2005;82:1–2. Westerterp-Plantenga MS, Luscombe-Marsh N, Lejeune MP, et al. Dietary protein, metabolism, and body-weight regulation: dose-response effects. Int J Obes (Lond) 2006;30(suppl 3):S16 –23. Ostman EM, Liljeberg Elmstahl HG, Bjorck IM. Inconsistency between glycemic and insulinemic responses to regular and fermented milk products. Am J Clin Nutr 2001;74:96 –100. Nilsson M, Stenberg M, Frid AH, Holst JJ, Bjorck IM. Glycemia and insulinemia in healthy subjects after lactose-equivalent meals of milk and other food proteins: the role of plasma amino acids and incretins. Am J Clin Nutr 2004;80:1246 –53. Hoppe C, Molgaard C, Michaelsen KF. Cow’s milk and linear growth in industrialized and developing countries. Annu Rev Nutr 2006;26:131–73. Kramer MS, Guo T, Platt RW, et al. Feeding effects on growth during infancy. J Pediatr 2004;145:600 –5. Michaelsen KF. Cows’ milk in complementary feeding. Pediatrics 2000; 106:1302–3. Zemel MB, Miller SL. Dietary calcium and dairy modulation of adiposity and obesity risk. Nutr Rev 2004;62:125–31. Skinner JD, Bounds W, Carruth BR, Ziegler P. Longitudinal calcium intake is negatively related to children’s body fat indexes. J Am Diet Assoc 2003;103:1626 –31. Moreira P, Padez C, Mourao I, Rosado V. Dietary calcium and body mass index in Portuguese children. Eur J Clin Nutr 2005;59:861–7. Berkey CS, Rockett HR, Willett WC, Colditz GA. Milk, dairy fat, dietary calcium, and weight gain: a longitudinal study of adolescents. Arch Pediatr Adolesc Med 2005;159:543–50. Venti CA, Tataranni PA, Salbe AD. Lack of relationship between calcium intake and body size in an obesity-prone population. J Am Diet Assoc 2005;105:1401–7.

Reassessing folic acid consumption patterns in the United States (1999 –2004): potential effect on neural tube defects and overexposure to folate1–3 Eoin P Quinlivan and Jesse F Gregory III ABSTRACT Background: In the United States, folic acid fortification of cerealgrain foods has significantly increased folate status. However, blood folate concentrations have decreased from their postfortification high as a result, in part, of decreasing food fortification concentrations and the popularity of low-carbohydrate weight-loss diets. Objectives: The objectives of the study were to quantify changes in folate intake after folic acid fortification and to estimate the effect on neural tube defect (NTD) occurrence. Design: Expanding on an earlier model, we used data from 11 intervention studies to determine the relation between chronic folate intervention and changes in steady state serum folate concentrations. With serum folate data from the National Health and Nutrition Examination Survey (NHANES), we used reverse prediction to calculate postfortification changes in daily folate equivalents (DFEs). With the use of NHANES red blood cell folate data and a published equation that related NTD risk to maternal red cell folate concentrations, we calculated NTD risk. Results: Folate intake decreased by 앒130 ␮g DFE/d from its postfortification high, primarily as a result of changes seen in women with the highest folate status. This decrease in folate intake was predicted to increase the incidence of NTD by 4 –7%, relative to a predicted 43% postfortification decrease. In addition, the number of women consuming 쏜1 mg bioavailable folate/d decreased. Conclusions: Folate consumption by women of childbearing age in the United States has decreased. However, the decrease in those women with the lowest folate status was disproportionately small. Consequently, the effect on NTD risk should be less than would be seen if a uniform decrease in folate concentrations had occurred. These results reinforce the need to maintain monitoring of the way fortification is implemented. Am J Clin Nutr 2007;86: 1773–9. KEY WORDS defects, nutrition

Food fortification, folic acid, folate, neural tube

from fortification was almost twice that originally envisioned (2, 3). This larger-than-expected increase in folic acid consumption resulted in part from wide-scale overfortification of enrichedgrain products; initial studies suggested that fortified foods typically contained 160% (4) to 175% (5) of the mandated amount of folic acid. Recent anecdotal and empirical evidence suggests that folic acid fortification levels have decreased in recent years and that they are coming more in line with mandated levels (6). This apparent reduction in fortification levels, coupled with the popularity of low-carbohydrate diets (7), has been cited as an explanation for the decreases in serum and red blood cell (RBC) folate concentrations observed in the annual National Health and Nutrition Examination Survey (NHANES) from 1999 through 2004 (8). We set out to quantify the change in folate consumption since the instigation of folic acid fortification and to quantify the effect this change may have had on the incidence of NTDs and the extent of folate overconsumption. METHODS

Relation between changes in folate consumption and changes in serum folate concentration Previously, we (2) identified 4 studies (9 –12) in which serum or plasma folate concentrations were measured before and after oral folic acid intervention (Table 1). Another 7 studies (13–19), which were published after our original report, were also identified. In all studies, intervention periods were sufficient to achieve plateau serum folate concentrations (9, 10). Daily folate intervention from each of these studies was expressed as daily folate equivalents (DFEs); DFEs are calculated as folic acid ҂ 1.7 on the basis that folic acid (20) and, by extension, folate monoglutamates (13, 21), are 1.7 times more bioavailable than are food folates. From this calculation, we plotted the change in 1

INTRODUCTION

In January 1998, in an attempt to reduce the incidence of neural tube defects (NTDs), the addition of folic acid to enriched-grain products became mandatory in the United States (1). The folic acid fortification levels were originally set so as to maximize folate consumption by women of childbearing age and to minimize the number of persons consuming 쏜1 mg folic acid/d. However, it soon became evident that folic acid intake derived

From the Biomedical Mass Spectrometry Laboratory, General Clinical Research Center (EPQ), and the Food Science and Human Nutrition Department (JFG), University of Florida, Gainesville, FL. 2 Supported by NIH grants no. DK37481 and no. DK56274 and by the Florida Agricultural Experiment Station. 3 Reprints not available. Address correspondence to JF Gregory, Food Science and Human Nutrition Department, PO Box 110370, University of Florida, Gainesville, FL 32611-0370. E-mail: [email protected]. Received July 30, 2007. Accepted for publication August 10, 2007.

Am J Clin Nutr 2007;86:1773–9. Printed in USA. © 2007 American Society for Nutrition

1773

1774

QUINLIVAN AND GREGORY

TABLE 1 Change in serum or plasma folate concentrations observed in intervention studies on the effect of oral folic acid consumption1

Study group

Male subjects

Duration of intervention

Intervention type2

Folate species

% van Oort et al (19) van Oort et al (19) Venn et al (14) Venn et al (14) Ward et al (9) Melse-Boonstra et al (17) van Oort et al (19) Lamers et al (13) Schorah et al (11) Schorah et al (11) Wald et al (12) PACIFIC Study (16) Ward et al (9) Riddell et al (10) Carrero et al (18) Ashfield-Watt et al (15) Ward et al (9) Lamers et al (13) Lamers et al (13) Wald et al (12) van Oort et al (19) Riddell et al (10) Wald et al (12) van Oort et al (19) Wald et al (12) van Oort et al (19) Wald et al (12)

NS NS 57 62 100 NS NS 0 52 58 83 82 100 62 100 42 100 0 0 83 NR 62 83 NS 83 NS 83

12 wk 12 wk 16 wk 16 wk 6 wk 12 wk 12 wk 16 wk 24 wk 24 wk 3 mo 6 mo 6 wk 12 wk 12 wk 4 mo 14 wk 16 wk 16 wk 3 mo 12 wk 12 wk 3 mo 12 wk 3 mo 12 wk 3 mo

Supplement Supplement Supplement Supplement Supplement Supplement Supplement Supplement Fortification Fortification Supplement Supplement Supplement Fortification Supplement Supplement Supplement Supplement Supplement Supplement Supplement Supplement Supplement Supplement Supplement Supplement Supplement

Folic acid Folic acid 5-CH3-THF Folic acid Folic acid Folic acid Folic acid 5-CH3-THF Folic acid Folic acid Folic acid Folic acid Folic acid Folic acid Folic acid Folic acid Folic acid Folic acid 5-CH3-THF Folic acid Folic acid Folic acid Folic acid Folic acid Folic acid Folic acid Folic acid

Dose3

DFE dose4

Subjects

Change in folate concentrations5

␮g/d

␮g/d

n

␮g/L

49 99 100 100 100 145 198 200 200 200 200 200 200 298 340 352 400 400 400 400 408 437 600 633 800 872 1000

83 168 170 170 170 247 337 340 340 340 340 340 340 507 578 598 680 680 680 680 694 743 1020 1076 1360 1482 1700

42 41 53 52 30 54 43 32 33 31 25 68 30 16 30 108 30 34 35 25 43 16 25 43 25 43 25

1.9 3.2 2.4 2.3 2.1 4.9 5.4 5.8 5.1 6.0 4.5 4.9 4.6 4.9 6.0 6.5 11.0 10.0 9.8 11.5 13.0 11.9 13.9 18.8 20.3 27.3 24.4

1 Studies listed more than once reported the results of multiple interventions. DFE, daily folate equivalent; NS, not stated; 5-CH3-THF, 5-methyltetrahydrofolate. 2 Folic acid was administered either in tablet form (supplement) or through consumption of fortified breakfast cereal (fortification). 3 Amount of additional folic acid consumed daily by subjects. 4 Folate and 5-CH3-THF (adjusted for differences in molecular mass) dose multiplied by 1.7, to adjust for their greater bioavailability than that of dietary folate. 5 Change in median or mean serum or plasma folate concentrations after intervention.

serum or plasma folate concentration due to the intervention versus daily folate dose and calculated the linear regression equation describing that relation. The regression line defined by the data of van Oort et al (19) appeared to be different from the regression defined by the other data points. We used a coincidence test, with the equation defined by Kleinbaum et al (22), calculated by using an Excel spreadsheet (Microsoft Corp, Redmond, WA) to determine whether the difference was significant. The regression linearity was determined by using SIGMASTAT for WINDOWS statistical software (version 3.00; SPSS Inc, Chicago, IL). Red blood cell and serum folate concentrations Prefortification (23) and postfortification (8) serum and RBC folate concentrations for women of childbearing age were taken from published NHANES studies. Changes in folate consumption By subtracting the 1989 –1994 NHANES III (8) serum folate data from the data of each of the NHANES postfortification

surveys (20), we calculated the change in serum folate concentration over this period. With the use of reverse prediction and comparing the changes in serum folate concentrations with the linear regression equation derived above, we calculated the apparent change in daily folate consumption since fortification. Total daily folate consumption On the grounds that we were able to determine the change in daily folate consumption by the change in serum folate concentration, we then decided to test whether we could determine total daily folate consumption on the basis of total serum folate concentrations. With the use of reverse prediction and comparing total serum folate concentrations from the NHANES surveys (8, 23) with the linear regression equation derived above, we calculated the apparent total DFEs. To validate this hypothesis, we used serum folate concentrations from the Framingham Offspring Cohort (24) to calculate apparent total DFEs for that group. We compared these calculated values to published DFE values from the same cohort that were derived by using food-frequency questionnaires (3). (The

1775

MODELING FOLATE INTAKE AND ITS CONSEQUENCES

surveys. NTD risk was expressed relative to the median prefortification group.

Change in serum or plasma folate concentration (µg/L)

30

25

20

RESULTS

15

Relation between changes in folate consumption and changes in serum folate concentration

10

5

0 0

200

400

600

800

1000

1200

1400

1600

Folate intake due to intervention (µg DFE/d) FIGURE 1. Relation between controlled folate intake and the resulting change in median or mean serum or plasma folate concentration. Data were derived from intervention studies of the effect of longitudinal folate supplementation or fortification with known daily amounts of folate on median or mean serum or plasma folate concentrations. Intervention was with folic acid or 5-CH3-THF. Daily folic acid intervention was converted to dietary folate equivalents (DFE) by multiplying by 1.7 (20). Likewise, daily 5-CH3-THF intervention was converted to DFE after adjustment for differences in mass between 5-CH3-THF and folic acid. •••, the regression line derived in our original study (2); - - - (with 〫), the regression line derived from van Oort et al (19)— data that were excluded from the final analysis because the slope was determined by a coincidence test (22) to be significantly different (P 쏝 0.01) from the slope defined by the other data points; — (with f), the regression line derived from all intervention studies except van Oort et al (y ҃ 0.0145x ѿ 0.132; r ҃ 0.979, P 쏝 0.001).

Changes in serum or plasma folate concentrations were plotted against DFEs (Figure 1). Because the slope in the study of van Oort et al (19) differed significantly [P 쏝 0.01, as determined by a coincidence test (22)] from the slope defined by the data points from the other studies, it was excluded from the final analysis. Because van Oort et al used a competitive binding assay (Immulite 2000; Diagnostic Products Co, Los Angeles, CA) to calculate the serum folate concentrations, it is possible that the elevated folate concentrations observed were due to systematic error within the assay. Such folate-binding assays can show a high degree of variability (26), depending on the folate species present, the calibrants, and the assay conditions. For instance, the Immulite 2000 assay has been reported to give higher RBC folate concentrations than does the microbiological assay (27). The slope of the data from the remaining studies was linear (r ҃ 0.979, P 쏝 0.001). Comparing the data points from our original study (2) with those from the current, expanded study, we found no significant (P 쏜 0.6) difference in slope between the 2 sets of data. The age of the subjects in the intervention studies had no significant effect on serum folate response (data not shown). Red blood cell and serum folate concentrations

food composition tables used to analyze the food-frequency questionnaires had been adjusted to reflect actual folic acid fortification.) Neural tube defect risk Daly et al (25) derived an equation that defines the relation between RBC folate concentrations and NTD risk. We used this equation and RBC folate concentrations from the NHANES surveys (8, 20) to calculate the NTD risk for each of the NHANES

Both RBC (Table 2) and serum (Table 3) folate concentrations increased between 1988 –1994 and 1999 –2000, and then they decreased each year from 1999 –2000 to 2003–2004. Between 1988 –1994 and 1999 –2000, the percentage increase in serum and RBC folate concentration was smallest in the women with the highest folate status. In contrast, the percentage decline in serum and RBC folate concentrations between 1999 –2000 and 2003–2004 was greatest in the women with the highest folate status.

TABLE 2 Change in red blood cell (RBC) folate concentrations between the third National Health and Nutrition Examination Survey (NHANES III) and the annual NHANES surveys from 1999 through 2004 Percentile of red blood cell folate concentration 10

25

50

75

90

92 164 163 155

119 200 208 188

160 255 260 235

222 329 318 298

296 409 395 367

1002 180

1002 171

1002 164

1002 155

1002 146

1002 99 95

1002 104 94

1002 102 92

1002 97 91

1002 97 90

1

RBC folate (ng/mL) 1988–1994 1999–2000 2001–2002 2003–2004 Change in RBC folate concentration from 1994 (%) 1988–1994 1999–2000 Change in RBC folate concentration from 1999 (%) 1999–2000 2001–2002 2003–2004 1 2

Values are from references 23 (1988 –1994) and 8 (1999 –2000, 2001–2002, and 2003–2004). Reference group.

1776

QUINLIVAN AND GREGORY

TABLE 3 Change in serum folate concentrations between the third National Health and Nutrition Examination Survey (NHANES III) and the annual NHANES surveys from 1999 through 2004 Percentile of serum folate concentration 10

25

50

75

90

2.3 6.3 6.4 6.0

3.1 8.9 8.5 7.8

4.8 12.6 11.4 10.6

7.8 17.3 15.2 14.1

11.7 24.7 19.7 18.5

4.0 4.1 3.7

5.8 5.4 4.7

7.8 6.6 5.8

9.5 7.4 6.3

13.0 8.0 6.8

1

Serum folate (ng/mL) 1988–1994 1999–2000 2001–2002 2003–2004 Change in serum folate from 1988–1994 (ng/mL)2 1999–2000 2001–2002 2003–2004 Change in serum folate concentration from 1994 (%) 1988–1994 1999–2000 Change in serum folate concentration from 1999 (%) 1999–2000 2001–2002 2003–2004

1003 274

1003 287

1003 263

1003 222

1003 211

1003 102 95

1003 96 88

1003 90 84

1003 88 82

1003 80 75

1

Values are from references 23 (1988 –1994) and 8 (1999 –2000, 2001–2002, and 2003–2004). Calculated by subtracting the NHANES III concentrations from the concentration in each of the annual NHANES surveys. 3 Reference group. 2

Median folate consumption (Table 4) increased by 529 ␮g DFE/d between 1988 –1994 (before fortification) and 1999 –2000 (after fortification); it then decreased by 135 ␮g DFE/d between 1999 –2000 and 2003–2004. This overall decrease in folate consumption was primarily the result of changes in subjects with the highest folate status: eg, folic acid consumption decreased by 20 and 74 ␮g DFE/d in women in the 10th and 25th percentile of serum folate, respectively, whereas the decrease was 215 and 417 ␮g DFE/d for women in the 75th and 90th percentile, respectively.

centrations. As shown in Figure 2, we found a strong correlation (r ҃ 0.9761, P 쏝 0.001) between total daily folate intakes, estimated in this manner, and published intake values (3), estimated by using food-frequency questionnaires (updated for actual folic acid concentrations in food). As expected, total folate consumption increased in the year after mandatory fortification (1999 –2000), so that we estimate that subjects in the 90th percentile consumed a total of 1666 ␮g DFE/d. However, by 2003–2004, total folate consumed by subjects in the 90th percentile had decreased to 1249 ␮g DFE/d.

Total folate consumption

Neural tube defect risk

The following findings supported our hypothesis that total daily folate intake can be estimated by using serum folate con-

Our analysis predicted a 43% decrease in NTD risk between 1988 –1994 and 1999 –2000. However, it also predicted that

Changes in folate consumption

TABLE 4 Change in daily folate intake between the third National Health and Nutrition Examination Survey (NHANES III) and the annual NHANES surveys from 1999 through 2004 and total daily folate intake in each study year, stratified by percentile of serum folate concentration1 Percentile of serum folate concentration

Change in folate intake from 1988–1994 (␮g DFE/d) 1999–2000 2001–2002 2003–2004 Total folate consumed (␮g DFE/d)3 1988–1994 1999–2000 2001–2002 2003–2004

10

25

50

75

90

273 280 253

394 368 320

529 448 394

643 502 428

879 542 462

159 428 435 408

213 603 576 529

327 852 771 717

529 1168 1027 953

791 1666 1329 1249

2

1

DFE, daily folate equivalents. Calculated from the change in serum folate concentration between NHANES III (23) and the annual NHANES surveys (8) from 1999 through 2004 by using a regression equation relating changes in serum folate concentration to changes in daily folate intake (derived from Figure 1). 3 Calculated from the total serum folate concentrations in NHANES III (23) and the annual NHANES surveys (8) from 1999 through 2004 by using a regression equation relating changes in serum folate concentration to changes in daily folate intake (derived from Figure 1). 2

1777

MODELING FOLATE INTAKE AND ITS CONSEQUENCES

over this period from 25% in 1995 to 29% in 1998 (28) and to 31% in 2003 (29). We estimate that supplement use increased average folate consumption by 164 ␮g DFE/d more than did fortification alone [ie, average folate consumption increased by 529 ␮g DFE/d (Table 4)] in all women in the NHANES cohort, but by only 365 ␮g DFE/d [215 ␮g folic acid/d (2)] in supplement nonusers in the Framingham Offspring Cohort (529 ␮g DFE/d Ҁ 365 ␮g DFE/d ҃ 164 ␮g DFE/d). Risk of folate overconsumption

FIGURE 2. Relation between total daily folate intake calculated by regression (see Figure 1) versus intakes calculated by using corrected foodfrequency questionnaires (y ҃ 1.106x ѿ 10.387; r ҃ 0.9761, P 쏝 0.001).

NTD risk increased by 4 –7% between 1999 –2000 and 2003– 2004 (calculated by subtracting the relative NTD risk in 1999 – 2000 from that in 2003–2004) (Table 5). DISCUSSION

Changes in serum and red blood cell folate concentration Women with the lowest percentiles of folate status had the largest percentage increase in RBC (Table 2) and serum (Table 3) folate concentrations between 1988 –1994 and 1999 –2000. In contrast, women in the same percentiles had the smallest percentage decrease in folate concentrations between 1999 –2000 and 2003–2004. This disproportionate change in folate status cannot be attributed solely to changes in a single factor. For instance, if folic acid fortification causes a relatively large increase in folate concentrations, withdrawal of fortification cannot then cause a relatively small decrease in folate concentration in the same women. Several factors must have combined to give this disproportionate response. Other factors that may have affected folate status include the changing use of folic acid supplements and the increasing popularity of diets low in enriched-grain products and breakfast cereal (eg, fortified products). For instance, the popularity of low-carbohydrate diets (low in grain products) has increased in recent years (7). Daily folic acid supplement use has increased

The Food and Drug Administration’s upper safe limit for total folate (folic acid plus natural folate) is 1 mg/d. This value is somewhat arbitrary and makes no allowances for the differences in bioavailability between folic acid and natural folates. Natural food folates, because of their polyglutamyl tails, may be less bioavailable than are folate monoglutamates (21), and folic acid is 1.7 times as bioavailable as are natural food folates. Therefore, when setting upper tolerable limits for total folate consumption, a better measure may be total bioavailable folate ҃ DFE 쐦 1.7. Using this definition and the data from Table 4, we estimated that, soon after fortification (1999 –2000), 10% of women (ie, those above the 90th percentile) were consuming 쏜980 ␮g bioavailable folate/d (1666 ␮g DFE/d 쐦 1.7). However, by 2003–2004, the amount of bioavailable folate consumed by women in the 90th percentile had decreased by almost 25%, to 687 ␮g/d (1168 ␮g DFE/d 쐦 1.7), which suggested that the number of women consuming 쏜1 mg bioavailable folate/d had decreased significantly. Neural tube defect risk There is some confusion concerning the extent to which folic acid fortification reduced the incidence of NTDs in the United States. One widely cited study (30) suggested that NTD occurrence decreased by 19%, whereas the Centers for Disease Control and Prevention (CDC) estimated a slightly higher value of 26% (31). However, both of these studies used post-partum medical records in their analysis and may have underestimated NTD frequency by excluding cases of spontaneous or medical abortion. More thorough estimates that included prenatal diagnosis data suggested that NTD incidences decreased by 40% (32). Such a value would be in line with the 43% decrease predicted in Table 5. Thus, assuming the validity of our estimates in Table 5, we predict that the NTD risk would have increased by 4 –7% between 1999 and 2004 (relative to the 43% decrease between 1989 –1995 and 1999).

TABLE 5 Relative risk of having a child with a neural tube defect (NTD) by percentile of red blood cell folate concentration during the third National Health and Nutrition Examination Survey (NHANES III) and each of the annual NHANES surveys from 1999 through 20041 Percentile of red blood cell folate concentration

Relative NTD risk versus prefortification median (%) 1988–1994 1999–2000 2001–2002 2003–2004

25

50

75

90

196 97 98 104

143 76 73 82

1002 57 55 63

67 42 43 47

47 32 33 36

RCF, red (blood) cell folate. The risk was estimated by fitting red blood cell concentrations into an equation from reference 23: NTD risk ҃ . 2 Reference group. 1

(0.6489Ҁ1.2193 ҂ ln[RCF (mg/L)]

exp

10

1778

QUINLIVAN AND GREGORY

Moore et al (33) conducted an NTD risk assessment similar to that of Daly et al (23) but using daily folate intake, calculated from food-frequency questionnaires, rather than RBC folate concentration. An estimate of the change in NTD risk, based on our estimates for daily folate intakes, and the regression equation of Moore et al are included elsewhere (See Supplemental Data, including Table S1, under “Supplemental data” in the current online issue at www.ajcn.org).

women with low folate status and, at the same time, limit folate overconsumption. It should be noted that the present study did not address changes in folate consumption by men. Such an analysis should be conducted before any further changes are made in fortification programs. The authors’ responsibilities were as follows—EPQ: developed the original concept and conducted the statistical analysis; JFG (principal investigator): provided input on the execution of the concept; and EPQ and JFG: wrote the report. Neither author had a personal or financial conflict of interest.

Conclusions The recent decrease in serum and RBC folate concentrations in the United States has resulted primarily from changes in women with the highest folate concentrations. Consequently, we estimate that the effect on NTD occurrence would be less than that seen if a uniform decrease in folate concentrations had occurred. In addition, the large decrease in folate consumption in women with the highest folate status may limit the potential danger from folate overconsumption. This change in folate consumption patterns is in accordance with the FDA’s aim of maximizing folate intake in women with low folate status (and thus reducing the incidence of NTDs) at the same time that the incidence of folate overconsumption is minimized. However, the manner in which this change in folate consumption occurred is entirely fortuitous, and the mechanism remains unregulated. It is quite conceivable that the events leading to this trend, particularly the decrease in folate overconsumption, could easily be reversed or could even become exacerbated. Of concern is the effect that the decrease in serum and RBC folate concentrations may have on the monitoring of the potential risks of folic acid fortification. Some models of cancer development predict that, whereas folic acid fortification may prevent precancerous cells from turning cancerous, it may also promote the proliferation of neoplastic cells—ie, cells that already are cancerous (34). Thus, folic acid fortification may decrease cancer rates by preventing noncancerous cells from turning neoplastic but also may increase cancer rates by increasing neoplastic cell proliferation. An extension of this model is the possibility that an increase in cancer rates due to fortification may be transitory, because fortification also reduces the transformation of normal cells to neoplastic cells. Such a model may explain the recent report by Mason et al (35) showing that rates of colorectal cancer in the United States increased between 1995 and 1998, coincident with the introduction of folic acid fortification. However, there are 욷2 possible explanations for the subsequent decrease in colorectal cancer rates observed by Mason et al between 1998 and 2002. First, as predicted by the model, folic acid fortification may have prevented noncancerous cells from becoming neoplastic. Second, serum and RBC folate concentrations also decreased over this period, however, and thus it is possible that colorectal cancer rates decreased because folate concentrations were no longer sufficient to sustain the elevated cancer proliferation rates. The possibility of this second scenario, although unlikely, is of concern, because it presents the possibility that colorectal cancer rates may increase again if folate concentrations revert to 1999 levels. We, therefore, call for the continued monitoring of the way in which fortification is implemented—particularly with regard to the monitoring of food consumption patterns. Furthermore, a better understanding of this phenomenon may provide information on ways in which we may increase folate consumption in

REFERENCES 1. Food and Drug Administration. Food labeling: health claims and label statements; folate and neural tube defects. Fed Regist 1993;58:53254 – 95. 2. Quinlivan EP, Gregory JF. Effect of food fortification on folic acid intake in the United States. Am J Clin Nutr 2003;77:221–5. 3. Choumenkovitch SF, Selhhub J, Wilson PWF, Rader JI, Rosenberg IH, Jacques PF. Folic acid intake from fortification in United States exceeds predictions. J Nutr 2002;132:2792– 8. 4. Rader JI, Weaver CM, Angyal G. Total folate in enriched cereal-grain products in the United States following fortification. Food Chem 2000; 70:275– 89. 5. Whittaker P, Tufaro PR, Radar JI. Iron and folate in fortified cereals. J Am Coll Nutr 2001;20:247–54. 6. Po´o-Prieto R, Haytowitz DB, Holden JM, Rogers G, Choumenkovitch SF, Jacques PF, Selhub J. Use of the Affinity/HPLC method for quantitative estimation of folic acid in enriched cereal-grain products. J Nutr 2006;136:3079 – 83. 7. Blanck HM, Gillespie C, Serdula MK, Kettel Khan L, Galuska DA, Ainsworth BE. Use of low-carbohydrate, high-protein diets among Americans: correlates, duration, and weight loss. Med Genet Med 2006; 8:5. 8. Centers for Disease Control and Prevention (CDC). Folate status in women of childbearing age, by race/ethnicity—United States, 1999 – 2000, 2001–2002, and 2003–2004. MMWR Morb Mortal Wkly Rep 2007;55:1377– 80. 9. Ward M, McNulty H, McPartlin J, Straine JJ, Weir DG, Scott JM. Plasma homocysteine, a risk factor for cardiovascular disease, is lowered by physiological doses of folic acid. QJM 1997;90:519 –24. 10. Riddell LJ, Chisholm A, Williams S, Mann JI. Dietary strategies for lowering homocysteine concentrations. Am J Clin Nutr 2000;71:1448 – 54. 11. Schorah CJ, Devitt H, Lucock M, Dowell AC. The responsiveness of plasma homocysteine to small increase in dietary folic acid: a primary care study. Eur J Clin Nutr 1998;52:407–11. 12. Wald DS, Bishop L, Wald NJ, et al. Randomized trial of folic acid supplementation and serum homocysteine levels. Arch Intern Med 2001; 16:695–700. 13. Lamers Y, Prinz-Langenohl R, Moser R, Pietrzik K. Supplementation with [6S]-5-methyltetrahydrofolate or folic acid equally reduces plasma total homocysteine concentrations in healthy women. Am J Clin Nutr 2004;79:473– 8. 14. Venn BJ, Green TJ, Moser R, Mann JI. Comparison of the effect of low-dose supplementation with L-5-methyltetrahydrofolate or folic acid on plasma homocysteine: a randomized placebo-controlled study. Am J Clin Nutr 2003;77:658 – 62. 15. Fohr IP, Prinz-Langenohl R, Brönstrup A, et al. 5,10-Methylenetetrahydrofolate reductase genotype determines the plasma homocysteine-lowering effect of supplementation with 5-methyltetrahydrofolate or folic acid in healthy young women. Am J Clin Nutr 2002;75:275– 82. 16. PACIFIC Study Group. Dose-dependent effects of folic acid on plasma homocysteine in a randomized trial conducted among 723 individuals with coronary heart disease. Eur Heart J 2002;23:1509 –15. 17. Melse-Boonstra A, Lievers KJA, Blom HK, Verhoef P. Bioavailability of polyglutamyl folic acid relative to that of monoglutamyl folic acid in subjects with different genotypes of the glutamate carboxypeptidase II gene. Am J Clin Nutr 2004;80:700 – 4. 18. Carrero JJ, Lopez-Huertas E, Salmeron LM, Baro L, Ros E. Daily supplementation with (n–3) PUFAs, oleic acid, folic acid, and vitamins B-6 and E increases pain-free walking distance and improves risk factors in men with peripheral vascular disease. J Nutr 2005;135:1393–9.

MODELING FOLATE INTAKE AND ITS CONSEQUENCES 19. van Oort FVA, Melse-Boonstra A, Brouwer IA, et al. Folic acid and reduction of plasma homocysteine concentrations in older adults: a doseresponse study. Am J Clin Nutr 2003;77:1318 –23. 20. Yang TL, Hung J, Caudill MA, et al. A long-term controlled folate feeding study in young women supports the validity of the 1.7 multiplier in the dietary folate equivalency equation. J Nutr 2005;135:1139 – 45. 21. Gregory JF, Quinlivan EP, Davis SR. Integrating the issues of folate bioavailability, intake and metabolism in the era of fortification. Trends Food Sci Technol 2005;16:229 – 40. 22. Kleinbaum DG, Kupper LL, Muller KE, Nizam A. Applied regression analysis and multivariable methods. 3rd ed. Pacific Grove, CA: Duxbury Press, 1998:322– 4. 23. Centers for Disease Control and Prevention (CDC). Folate status in women of childbearing age, by race/ethnicity—United States, 1999 – 2000. MMWR Morb Mortal Wkly Rep 2002;51:808 –10. 24. Jacques PF, Selhub J, Bostom AG, Wilson PWF, Rosenberg IH. The effect of folic acid fortification on plasma folate and total homocysteine concentrations. N Engl J Med 2000;340:1449 –54. 25. Daly LE, Kirke PN, Molloy A, Weir DG, Scott JM. Folate levels and neural tube defects. Implications for prevention. JAMA 1995;274: 1698 –702. 26. Quinlivan EP, Hanson AD, Gregory JF. The analysis of folate and its metabolic precursors in biological samples. Anal Biochem 2006;348: 163– 84. 27. Icke GC, Dennis M, Sjollema S, Nicol DJ, Eikelboom JW. Red cell

28. 29. 30. 31. 32. 33. 34. 35.

1779

N5-methyltetrahydrofolate concentrations and C677T methylenetetrahydrofolate reductase genotype in patients with stroke. J Clin Pathol 2004;57:54 –7. Centers for Disease Control and Prevention (CDC). Knowledge and use of folic acid by women of childbearing age—United States, 1995 and 1998. MMWR Morb Mortal Wkly Rep 1999;48:325–7. Centers for Disease Control and Prevention (CDC). Use of vitamins containing folic acid among women of childbearing age—United States, 2004. MMWR Morb Mortal Wkly Rep 2004;53:847–50. Honein MA, Paulozzi LJ, Mathews TJ, Erickson JD, Wong L-YC. Impact of folic acid fortification of the US food supply on the occurrence of neural tube defects. JAMA 2001;285:2981– 6. Centers for Disease Control and Prevention (CDC). Spina bifida and anencephaly before and after folic acid mandate–United States, 19951996 and 1999-2000. MMWR Morb Mortal Wkly Rep2004;53:362-5. Olney RS, Mulinare J. Trends in neural tube defect prevalence, folic acid fortification and vitamin supplement use. Semin Perinatol 2002;26:277– 85. Moore LL, Bradlee ML, Singer MR, Rothman KJ, Milunsky A. Folate intake and the risk of neural tube defects: an estimate of dose response. Epidemiology 2003;14:200 –5. Ulrich CM, Potter JD. Folate Supplementation: too much of a good thing? Cancer Epidemiol Biomarkers Prev 2006;15:189 –93. Mason JB, Aaron Dickstein A, Jacques PF, et al. A temporal association between folic acid fortification and an increase in colorectal cancer rates may be illuminating important biological principles: a hypothesis. Cancer Epidemiol Biomarkers Prev 2007;16:1325–9.

See corresponding editorial on page 1579.

Calcium intake and hip fracture risk in men and women: a metaanalysis of prospective cohort studies and randomized controlled trials1–3 Heike A Bischoff-Ferrari, Bess Dawson-Hughes, John A Baron, Peter Burckhardt, Ruifeng Li, Donna Spiegelman, Bonny Specker, John E Orav, John B Wong, Hannes B Staehelin, Eilis O’Reilly, Douglas P Kiel, and Walter C Willett ABSTRACT Background: The role of total calcium intake in the prevention of hip fracture risk has not been well established. Objective: The objective of the study was to assess the relation of calcium intake to the risk of hip fracture on the basis of meta-analyses of cohort studies and clinical trials. Results: In women (7 prospective cohort studies, 170 991 women, 2954 hip fractures), there was no association between total calcium intake and hip fracture risk [pooled risk ratio (RR) per 300 mg total Ca/d ҃ 1.01; 95% CI: 0.97, 1.05]. In men (5 prospective cohort studies, 68 606 men, 214 hip fractures), the pooled RR per 300 mg total Ca/d was 0.92 (95% CI: 0.82, 1.03). On the basis of 5 clinical trials (n ҃ 5666 women, primarily postmenopausal, plus 1074 men) with 814 nonvertebral fractures, the pooled RR for nonvertebral fractures between calcium supplementation (800 –1600 mg/d) and placebo was 0.92 (95% CI: 0.81, 1.05). On the basis of 4 clinical trials with separate results for hip fracture (6504 subjects with 139 hip fractures), the pooled RR between calcium and placebo was 1.64 (95% CI:1.02, 2.64). Sensitivity analyses including 2 additional small trials with 쏝100 participants or per-protocol results did not substantially alter results. Conclusions: Pooled results from prospective cohort studies suggest that calcium intake is not significantly associated with hip fracture risk in women or men. Pooled results from randomized controlled trials show no reduction in hip fracture risk with calcium supplementation, and an increased risk is possible. For any nonvertebral fractures, there was a neutral effect in the randomized trials. Am J Clin Nutr 2007;86:1780 –90. KEY WORDS Meta-analysis, hip fracture, nonvertebral fracture, calcium intake, calcium supplementation, cohort studies, randomized controlled trials KEY WORDS

INTRODUCTION

Calcium supplementation or the consumption of calcium-rich foods such as milk is commonly recommended for the prevention of osteoporosis and fractures. These recommendations are primarily based on evidence from randomized controlled trials (RCTs) with bone density as the outcome. However, in a 2004 meta-analysis of RCTs, supplementation with 500 –2000 mg Ca/d in postmenopausal women provided only a modest benefit

1780

for bone density: 2.05% difference in total-body bone density, 1.66% difference in lumbar spine bone density, and 1.64% difference in hip bone density (1, 2). The implications of such differences with respect to fracture risk prevention are unclear. In the same meta-analysis, limited evidence from RCTs (222 subjects in 2 trials) suggested only a modest and nonsignificant benefit of calcium supplementation for the risk of nonvertebral fractures [pooled risk ratio (RR) ҃ 0.86; 95% CI: 0.43, 1.72]. Furthermore, an earlier meta-analysis published in 1997 that summarized observational studies in postmenopausal women found no clear benefit of a 300-mg increment in daily calcium intake for hip fracture risk [pooled RR among 28 511 women from 5 cohorts was 0.96 (95% CI: 0.91, 1.02)] (3). Consequently, considerable uncertainty exists regarding optimal intakes of calcium, and this uncertainty is reflected in markedly different recommended daily intakes among countries. For example, for adults 쏜50 y old, the recommended daily intake is 700 mg Ca/d in the United Kingdom and 1200 mg Ca/d in the United States (4). 1

From the Departments of Nutrition (HAB-F, EO, and WCW), Epidemiology (RL and DS), and Biostatistics (RL, DS, and JEO), Harvard School of Public Health, Boston, MA; the Department of Rheumatology and the Institute for Physical Medicine and Rehabilitation, University Hospital Zurich, Zurich, Switzerland (HAB-F); the Jean Mayer USDA Human Nutrition Research Center on Aging, Tufts University, Boston, MA (BD-H); the Departments of Medicine and of Community and Family Medicine, Dartmouth Medical School, Hanover, NH (JAB); Clinique Bois Cerf/Hirslanden, Lausanne, Switzerland (PB); the EA Martin Program in Human Nutrition, South Dakota State University, Brookings, SD (BS); the Division of Clinical Decision Making, Tufts-New England Medical Center, Boston, MA (JBW); the Department of Geriatrics, University of Basel, Basel, Switzerland (HBS); and the Institute for Aging Research, Hebrew SeniorLife, Harvard Medical School, Boston, MA (DPK). 2 Supported by grants from the Medical Foundation (Charles H Farnsworth Trust; US Trust Company; Trustee and the Charles A King Trust; Fleet National Bank) and the International Foundation for the Promotion of Nutrition Research and Nutrition Education (ISFE); the Swiss Foundation for Nutrition Research (SFEFS), and the Swiss National Foundation (SNF Professorship grant). 3 Reprints not available. Address correspondence to HA Bischoff-Ferrari, Department of Rheumatology and Institute of Physical Medicine, University Hospital Zurich, Gloriastrasse 25, 8091 Zurich, Switzerland. E-mail: [email protected]. Received June 22, 2007. Accepted for publication June 28, 2007.

Am J Clin Nutr 2007;86:1780 –90. Printed in USA. © 2007 American Society for Nutrition

CALCIUM INTAKE AND HIP FRACTURE RISK: META-ANALYSIS

Several RCTs tested the combined effect of calcium plus vitamin D, and this evidence was summarized in 2 recent metaanalyses that suggested, irrespective of trial quality, a small but significant reduction in hip fracture risk [pooled RR ҃ 0.81; 95% CI: 0.68, 0.96 (5); pooled RR ҃ 0.82; 95% CI: 0.71, 0.94 (6)]. However, the benefit of calcium supplementation alone was not addressed in these analyses. Because increased calcium intake alone is still commonly recommended as a fracture-prevention strategy (6, 7), the assessment of calcium intake and its effect on hip fracture risk reduction is of clinical importance. Thus, we conducted a systematic review and meta-analysis of prospective cohort studies to address these relations with respect to hip fracture prevention. Hip fractures are the most severe and the most frequent fractures in older persons (8, 9). We focused on prospective cohort studies because they are less susceptible to selection and recall biases than are case-control studies. Because several RCTs of calcium supplementation without vitamin D have been conducted since the mid-1990s, and all have had samples sizes that were somewhat small for the assessment of hip fractures (10 –14), we also summarized findings regarding the effect of calcium supplementation on all nonvertebral fractures from randomized trials. METHODS

Search strategy and data extraction For both prospective cohort studies and RCTs, we conducted a systematic search for relevant English and non-English publications by using MEDLINE (Ovid and PubMed) for the period from January 1960 to December 2006 and by using EMBASE for January 1991 to December 2006. We also contacted experts in the field and searched reference lists and abstracts presented at the meetings of the American Society for Bone and Mineral Research from 1995 through 2006. For prospective cohort studies, we used numerous medical subject headings (MeSH terms)—“cohort studies” or “prospective studies,” or “retrospective studies,” and “fracture,” or “hip fracture,” and “calcium,” or “calcium analogs or derivates,” or “calcium carbonate,” or “calcium citrate,” or “calcium gluconate,” or “calcium phosphate,” or “milk,” or “dairy products.” To update the most recent meta-analysis of RCTs on calcium and fracture risk, we also searched for RCTs and fracture risk by using the MeSH terms above plus “randomized-controlled trial” or “controlled-clinical trial” or “random allocation” or “doubleblind method” or “uncontrolled trial.” In addition, we searched for fracture data in trials that had bone density as the primary outcome. We received unpublished data from one large published trial for calcium supplementation, the Randomised Evaluation of Calcium Or vitamin D (RECORD) trial (12)]. Eligibility and exclusion criteria were specified in advance. Data extraction was conducted independently by 2 investigators (HAB-F and EO). Eligible studies For prospective cohort studies, we included only studies in which calcium intake had been assessed before the fracture events. Our primary outcome was the first incident hip fracture in middle-aged or older men and women. For trials that addressed fracture incidence, we included only double-blind RCTs that studied any dose of calcium supplementation compared with

1781

placebo. Because of limited data on hip fractures, in separate analyses, we also included all nonvertebral fractures. We included only double-blind RCTs that studied calcium supplementation with a minimum follow-up of 1 y and that required 쏜100 study participants. Trials with 쏝100 participants were added in a sensitivity analysis. Ineligible studies We excluded uncontrolled trials, cross-sectional and casecontrol studies, and animal investigations. Of prospective cohort studies, we excluded studies that did not provide separate data for men and women (15) or for hip fracture (16). Of RCTs, we excluded studies in which calcium was combined with other agents, such as vitamin D (17–19), because the effects of the two agents could not be separated. The combination of calcium plus vitamin D has been addressed in 2 recent meta-analyses of RCTs (5, 6). Studies identified Prospective cohort studies A total of 8 separate studies were identified—7 that included women (3, 20 –25) and 5 that included men (20 –23, 26) (Table 1 and Figure 1A). Total calcium intake included dietary and supplement sources in 4 studies (3, 21, 25, 26) and only dietary calcium intake in 4 studies (20, 22–24). This omission may not have been important in the older studies (20, 23), because calcium intake from supplements became widespread only in the late 1980s (21). Randomized controlled trials We identified 5 RCTs that met our criteria for the primary analysis (10, 12–14, 27) and 2 additional smaller trials that were included in the sensitivity analysis (11, 28). RCTs primarily included postmenopausal women (Table 2). Five of these 7 RCTs provided separate data on hip fractures (10 –14). Our analyses followed the intention-to-treat principle. Only 3 trials provided a per-protocol analysis(10, 12, 13). Despite limited data for men, results by sex were pooled for the primary intention-to-treat analysis examining both hip and any nonvertebral fractures. Statistical methods Sex-specific cohort study analyses were conducted because men and women differ in fracture risk (29) and calcium intake (30). The primary outcome of the pooled analysis was the RR of hip fracture for a 300-mg increment in daily calcium intake, the amount of calcium in an 8-ounce (237-mL) glass of milk or one slice (50 g) of hard cheese. For the highest and the lowest open-ended calcium intake categories, we chose a previously defined value for a corresponding median—30% lower than the lowest cutoff and 30% higher than the upper cutoff. RRs adjusted for multiple covariates were used whenever available. The study of Holbrook et al (20) provided, for each sex, the RR, the overall sample size, and the overall number of hip fractures. From these numbers, we calculated the number of hip fracture cases in each exposure group and re-ran the analyses of the 2 ҂ 2 tables to retrieve the corresponding 95% CIs for the RRs. To compare studies on the same scale in the pooled analysis, we calculated the RR for a 300-mg increment in total daily

National Health Screening Norway

Swedish Mammography Screening Study

Nurses’ Health Study

13

18

Study of Osteoporotic Fractures

6.6

13.8

Health Professionals Follow-up Study

NHANES I follow-up study

California retirement community

8

16

6.8

72 337

60 689

20 035 19 752

9704

43 063

2116 2226

2966 5752

426 531

34–59

53.6 (40–74)

47.1 (40–53)

71

54 (40–75)7

50–74

735

y 50–794

White community, California

Age

y

Sex

14

Population

5608

719/763 558/558

634/787 583/620

mg

Calcium intake (Fx/no fx)2

603

1535

49 154

306

56

44 122

50 216

15 18

n

Total cases

FFQ (yes)

FFQ (no)

FFQ (no)

FFQ (yes)

FFQ (yes)

24-h recall (no)

FFQ (yes)

24-h recall (no)

Calcium assessment (calcium supplement included)

Protein intake, retinol intake, total vitamin D intake, age, BMI, postmenopausal hormone use, physical activity, smoking, calcium supplement use, multivitamin use, vitamin K, vitamin A intake, total energy intake, alcohol use, caffeine intake

Age, BMI, energy intake, protein intake, retinol intake, meat consumption, marital status, nulliparity, education level

Age, height, BMI, physical activity, DM, disability pension, marital status, smoking

Age, clinic, weight, history of fracture since age 50 y, fall in past 12 mo, protein intake, caffeine intake, recreational physical activity, walking for exercise, use of vitamin D supplements and Tums antacids

Age, BMI, smoking, physical activity, total energy, alcohol, vitamin D intake

Alcohol, smoking, physical activity, BMI, HRT use

Age

None

Covariates adjusted for

2

Fx, fracture; NHANES, National Health and Nutrition Examination Survey; HRT, hormone replacement therapy; FFQ, food-frequency questionnaire; DM, diabetes mellitus. If total mean calcium was given per 1000 kcal, the value was multiplied by 2 in women and by 2.5 in men. 3 Subjects with prior hip fracture were excluded. 4 Range (all such values). 5 Median. 6 Subjects with high-trauma fractures were excluded. 7 x៮ ; range in parentheses (all such values). 8 Mean across quintiles.

1

Women

Women Meyer et al, 1997 (23)6 Men Women Michaelsson et al, 2003 (24)3 Women Feskanich et al, 2003 (25)3,6

Holbrook et al, 1988 (20)3 Men Women Paganini-Hill et al, 1991 (21)3 Men Women Looker et al, 1993 (22)3 Men Women Owusu et al, 1997 (26)3,6 Men Cumming et al, 1997 (3)3,6

Reference

Mean duration of follow-up

TABLE 1 Prospective cohort studies that assessed total calcium intake or milk intake and hip fracture risk1

1782 BISCHOFF-FERRARI ET AL

CALCIUM INTAKE AND HIP FRACTURE RISK: META-ANALYSIS

A 34 Potentially relevant cohort studies identified and screened for retrieval

21 Excluded with reason

13 Retrieved for more detailed evaluation

4 Excluded: 1 no calcium intake assessment 1 milk intake alone 1 no separate hip fracture assessment 1 review

9 Potentially appropriate for inclusion

1783

intake. Results from all studies were then pooled by using random-effects models (32). RCT outcomes were pooled on an intention-to-treat basis with random-effects models, which control for both within-trial and between trial variance. Heterogeneity among RCTs and among cohort studies was evaluated by using the Q statistic, which is considered significant for P 쏝 0.10 (33, 34). To address the effect of poor adherence in some larger trials, a sensitivity analysis was performed on the basis of the per-protocol results of those trials. To assess potential publication bias, we used the Begg and Egger tests and Begg’s funnel plot (35, 36); no evidence of bias was seen in the prospective cohort studies or the RCTs. Statistical analysis was performed by using STATA software (version 7.0; Stata Corp, College Station, TX).

RESULTS

Prospective cohort studies 1 Excluded for no separate fracture assessment by sex

8 Cohort studies included in primary analyses - 7 for women - 5 for men

96 Potentially relevant RCTs identified and screened for retrieval

52 Excluded with reason

44 Retrieved for more detailed evaluation

36 Excluded: 10 no fracture outcome 4 only vertebral fractures assessed 12 calcium plus vitamin D 10 vitamin D alone

8 Potentially appropriate for inclusion

1 Excluded for cohort too young (young military) 2 Excluded for sample size < 100 5 RCTs included in primary analyses - 4 published as full manuscripts - 1 published in abstract form - 4 RCTs with separate results for hip fractures

FIGURE 1. A: A Quality of Reporting of Meta-analyses (QUOROM) flow diagram for the prospective cohort studies. B: A QUOROM flow diagram for the randomized controlled trials (RCTs).

calcium intake, which assumes a log-linear association of intake with risk. Because the relative risks within each cohort study depend on a common reference group, they are correlated. Thus, we used a method developed by Greenland et al (31), which yields an efficient point estimator and a consistent variance estimate under these circumstances, to calculate for each study the RR of hip fracture per 300-mg increase in total daily calcium

Characteristics of the 8 prospective cohort studies that met our inclusion criteria are shown in Table 1. Of these 8 studies, 7 included 170 991 women who sustained 2954 hip fractures, and 5 included 68 606 men who sustained 214 hip fractures. The median age at baseline ranged from 41 to 72 y. Mean follow-up varied between 3 and 18 y. Six studies were from the United States (3, 20 –22, 25, 26), one was from Norway (23), and one was from Sweden (24). Primary analysis The Forest plots for the RR of hip fracture for a 300-mg increase in daily calcium intake are shown in Figure 2. In women, there was no association between total calcium intake and hip fracture risk (pooled RR for additional 300 mg Ca/d intake ҃ 1.01; 95% CI: 0.97, 1.05). In men, the pooled RR was 0.92 (95% CI: 0.82, 1.03). Subgroup analyses for hip fracture risk and total calcium intake In the 4 studies of women that reported calcium intake from dietary sources alone, ie, ignoring supplemental calcium (20, 22–24), we found no association between dietary calcium intake and risk of hip fracture (pooled RR per additional 300 mg Ca/d intake from dietary calcium ҃ 1.01; 95% CI: 0.96, 1.06). In the 2 studies that measured dietary calcium intakes by using 24-h recall (20, 22), the association between calcium intake and hip fracture risk (pooled RR per additional 300 mg Ca/d intake ҃ 0.90; 95% CI: 0.76, 1.07) was somewhat stronger than that in the 5 studies that used food-frequency questionnaires (pooled RR per additional 300 mg dietary Ca/d calcium ҃ 1.01; 95% CI: 0.98, 1.05). Because the efficacy of calcium may be enhanced by additional vitamin D, as found in RCTs involving both institutionalized (17) and ambulatory (19) women, we sorted studies by latitude (south to north), taking the mean of the state capitals of the multistate US studies (Figure 2A). This analysis showed no stronger protective effect of calcium intake on hip fracture risk in women living in southern latitudes, who possibly had higher vitamin D status due to increased sunshine exposure, than in women living in northern latitudes. The limited data for men did not allow useful subgroup analyses.

1600 mg Calcium citrate/d

1200 mg Calcium carbonate/d

1000 mg Calcium citrate/d

1000 mg Calcium carbonate/d

1200 mg Calcium carbonate/d

58 앐 4

66.3 앐 2.6

75.1 앐 2.7

74 앐 4

77 앐 6

61 앐 9

86 postmenopausal women (86 of 122 women agreed to the 2-y extension of the original 2-y study) Fx ascertainment for nonvertebral fxs is not described in the manuscript 236 postmenopausal women Fx ascertainment for nonvertebral fx is not described 1460 postmenopausal women Fxs were reported in a diary and confirmed by radiographic reports

1471 postmenopausal women Fxs were assessed at every 6mo visit and confirmed by a radiograph or a report 2643 (2241 women and 402 men)

930 (258 women and 672 men) Fxs were assessed by yearly questionnaires throughout the trial phase and followup. Fxs were confirmed by a radiology report or a physician report of a fracture

Reid et al, 1995 (11)2

Reid et al, 2006 (10)

BischoffFerrari et al, 2006 (14) (abstract)

2

y

4 y of treatment (mean followup of 10.8 y)

2–5

5

5

4

4

1.5

Duration

25(OH)D, 25-hydroxyvitamin D. Small trials including 쏝100 subjects. 3 x៮ 앐 SD (all such values). 4 x៮ ; interquartile range in parentheses. 5 This report was from the Randomized Evaluation of Calcium or Vitamin D (RECORD) trial.

1

Grant et al, 2006 (12)5

Prince et al, 2006 (13)

Riggs et al, 1998 (27)

800 mg Ca from calcium carbonate or osseinomineral complex versus placebo (300 000 IU vitamin D3 orally in a single dose at study entry) Calcium gluconate: 1000 mg/d or placebo

Intervention

72.1 앐 0.613

y

Age

93 (82 ambulatory women, 11 ambulatory men) Fx ascertainment for nonvertebral fx is not described in the manuscript

n

Total subjects, fx ascertainment

Chevalley et al, 1994 (28)2

Author

TABLE 2 Randomized controlled trials of calcium supplementation and fracture (fx) risk1

865 앐 423 (placebo group) 889 앐 451 (calcium group)

861 앐 390 (calcium group) 853 앐 381 (placebo group) 829 앐 353

897 (704–1146)4

714 앐 286

745 앐 298

Nonvertebral fxs

Nonvertebral fxs

Nonvertebral fxs

Nonvertebral fx

Nonvertebral fx No information on hip fx

Symptomatic nonvertebral fxs

73.0 in the calcium group, 72.8 in the placebo group

At 24 mo, 30% were not returning questionnaires (lost to follow-up) Compliance: 42% at 24 mo (among those in the calcium group returning the questionnaire) 10.5% during 4 y of treatment Average compliance: 75– 77%

Hazard ratio of withdrawal or death: 0.86 (calcium group) and 0.76 (placebo group) Average compliance: 57% 15% Average compliance: 55– 58%

67.5 앐 35 (winter) 87.5 앐 30 (summer)

51.5 앐 19 (calcium group) 52 앐 19.5 (placebo group) 38 앐 16.5 (subgroup of 60 subjects)

25%

10.3%

14%

Loss to follow-up and compliance

76 앐 25.8

92.5 앐 5

nmol/L 62.5 앐 15

Vertebral fx in subjects with no hip fx No new hip fxs (according to information from the TC)

Baseline 25(OH)D

mg/d

Fx outcome and hip fx

619 앐 33

Baseline calcium intake

1784 BISCHOFF-FERRARI ET AL

CALCIUM INTAKE AND HIP FRACTURE RISK: META-ANALYSIS

1785

there is no apparent association between calcium intake and hip fracture risk over a wide range of calcium intakes in women. The limited data for men did not allow useful categorical doseresponse analyses.

A Holbrook et al (20) Paganini-Hill et al (21)

Randomized controlled trials

Cummings et al (3) Looker et al (22) Michaelsson et al (24) Meyer et al (23)

Combined 0.1

0.5

1

5

RR

B Holbrook et al (20)

Looker et al (22) Meyer et al (23)

Paganini-Hill et al (21)

Combined 0.1

0.5

1

5

10

RR

FIGURE 2. Relative risk (RR) of hip fracture for a 300-mg increase in total calcium intake per day from the prospective cohort studies. The size of the squares is proportional to the inverse of the variance. Error bars represent the 95% CIs. The confidence limits for the pooled RR are indicated by the diamond-shaped figure. There was no significant heterogeneity among studies in men or women (P ҃ 0.34 and 0.55 for women and men, respectively; Q test). A: In the women, the pooled RR per 300 mg total Ca/d was 1.01 (95% CI: 0.97, 1.05). B: In the men, the pooled RR per 300 mg total Ca/d was 0.92 (95% CI: 0.82, 1.03).

To examine the relation between calcium intake and hip fracture risk in more detail, we pooled RRs for categories of total calcium intake and hip fracture risk among women from each cohort study (Figure 3), using the lowest category as a reference and corresponding RRs for higher intake categories. Figure 3 confirms the findings suggested by the analysis in Figure 2A—ie,

Characteristics of the 5 RCTs that met our criteria for the primary analysis(10,12–14, 27) and 2 smaller trials that were included in the sensitivity analysis (11, 28) are shown in Table 2. The mean age ranged from 58 to 77 y. The mean duration of follow-up varied between 1.5 and 10.8 y. One study was from Switzerland (28), 1 study was from the United Kingdom (12), 2 studies were from New Zealand (10, 11), 2 were from the United States (14, 27), and 1 was from Australia (13). The primary analysis for nonvertebral fracture risk included 5 RCTs (Table 3; Figure 4) in which calcium supplementation between 800 and 1600 mg/d was compared with placebo. Among 6740 subjects (5666 women, primarily postmenopausal, plus 1074 men) who had a total of 814 nonvertebral fractures, the pooled RR was 0.92 (95% CI: 0.81, 1.05). This result did not change substantially in the sensitivity analyses: including the 2 additional small trials, the pooled RR was 0.91 (95% CI: 0.80, 1.03), and including only adherent subjects, the pooled RR was 0.83 (95% CI: 0.64, 1.09). Intention-to-treat results comparing calcium with placebo in women alone (pooled RR ҃ 0.92; 95% CI: 0.81, 1.06) were similar to the results including men. The pooled RR for men from 2 trials was 0.94 (95% CI: 0.64, 1.37) (12, 14). The primary analysis for hip fracture risk included 4 RCTs (Table 4; Figure 5) in which calcium supplementation between 800 and 1200 mg/d was compared with placebo. Among 6504 subjects (5430 women, primarily postmenopausal, plus 1074 men) who had a total of 139 hip fractures, the pooled RR was 1.64 (95% CI: 1.02, 2.64). This result did not change significantly in sensitivity analyses: including the 2 additional small trials, the pooled RR was 1.57 (95% CI: 0.96, 2.55), and including only adherent subjects, the pooled RR was 1.42 (95% CI: 0.81, 2.49). Intention-to-treat results comparing calcium with placebo in women alone (pooled RR ҃ 1.66; 95% CI: 0.97, 2.86) were similar to the results including men. The pooled RR for men from 2 trials was 1.55 (95% CI: 0.62, 3.96) (12, 14).

FIGURE 3. Pooled analysis for categories of calcium intake and hip fracture risk in the women from the prospective cohort studies. The reference intake categories in the various studies ranged from 280 to 554 mg total Ca/d.

1786

BISCHOFF-FERRARI ET AL

TABLE 3 Evidence table for randomized controlled trials comparing calcium supplementation with placebo for risk of all nonvertebral fractures (fxs)1

Chevalley et al (28)2 Reid et al (11)2 Riggs et al (27) Prince et al (13) Reid et al (10) Grant et al (12)4 Bischoff-Ferrari et al (14) Pooled results Primary analysis (intention-to-treat)5 Sensitivity analysis (intention-to-treat)6 Per-protocol analysis7

Outcome

Year

Treated subjects

Control subjects

n

n

Nonvertebral fx Nonvertebral fx Nonvertebral fx Nonvertebral fx Nonvertebral fx Nonvertebral fx Nonvertebral fx

1994 1995 1998 2006 2006 2006 2006

2/543 2/42 11/119 83/730 92/732 156/1311 46/464

2/253 7/44 12/117 94/730 94/739 172/1332 54/466

0.46 (0.07, 3.02) 0.30 (0.07, 1.22) 0.91 (0.41, 1.96) 0.88 (0.67, 1.16) 0.99 (0.76, 1.29) 0.92 (0.75, 1.13) 0.87 (0.60, 1.27)

79 86 236 1460 1471 2643 930

388/3356

426/3384

0.92 (0.81, 1.05)

6740

392/3452

435/3453

0.91 (0.80, 1.03)

6905

152/1314

202/1464

0.83 (0.64, 1.09)

2778

RR (95% CI)

Total n

RR, risk ratio. Intention-to-treat results for women alone based on the primary analysis comparing calcium with placebo (pooled RR ҃ 0.92; 95% CI: 0.81, 1.06; Q ҃ 0.50; P ҃ 0.97, Q test) were similar to the results including men. The few data available for men alone from the trials of Grant et al (RR ҃ 0.97) and Bischoff-Ferrari et al (RR ҃ 0.92) (pooled RR ҃ 0.94; 95% CI: 0.64, 1.37; P ҃ 0.90, Q test) were similar to the pooled risk among women. 2 Trial was excluded for the primary analysis. According to the intention-to-treat results, Table 3 shows the pooled results for the RR of having any nonvertebral fx with calcium supplementation compared to placebo. 3 n with fracture/total n of the group (all such values). 4 This report was from the Randomized Evaluation of Calcium or Vitamin D (RECORD) trial. 5 In the primary pooled analysis, there was an 8% (NS) lower risk with calcium than with placebo (95% CI: 0.81, 1.05; homogeneity: Q ҃ 3.3; P ҃ 0.77, Q test). 6 In the sensitivity analysis, including 2 additional small trials, nonvertebral fx risk was 9% lower with the 95% CI including 1 (95% CI: 0.80, 1.03; homogeneity: Q ҃ 0.48; P ҃ 0.98, Q test). 7 Per-protocol results were available for references 11–13, which included a total of 2778 subjects (all women). The pooled RR excluding subjects with 쏝80% compliance in the trials of Prince et al and Grant et al and excluding subjects with at least one 6-mo episode with 쏝60% compliance in the trial of Reid et al was 0.83 (95% CI: 0.64, 1.09; Q ҃ 3.64; P ҃ 0.16, Q test). 1

DISCUSSION

In our meta-analysis of prospective cohort studies, calcium intake was not significantly associated with hip fracture risk in men or women. Similarly, our meta-analysis of RCTs, which included data largely from postmenopausal women, yielded a neutral effect of

calcium supplementation as compared with placebo for any nonvertebral fracture and suggested a significantly (64%) greater risk of hip fractures with calcium supplementation. There are several possible explanations for the lack of overall association between total calcium intake and hip fracture risk in

Riggs et al (27)

Prince et al (13)

Reid et al (10)

Bischoff-Ferrar i et al (14)

Grant et al (12)

Combined

0.1

0.5

1 RR

5

10

FIGURE 4. Forest plot comparing intention-to-treat data from 5 randomized controlled trials for the risk of nonvertebral fractures between calcium-treated and placebo groups. The squares represent the relative risk (RR) of fracture between subjects who took calcium in any dose and those who took placebo. In a total of 6740 subjects, the pooled RR was 0.92 (95% CI: 0.81, 1.05). There was no heterogeneity between studies (P ҃ 0.77, Q test).

1787

CALCIUM INTAKE AND HIP FRACTURE RISK: META-ANALYSIS TABLE 4 Evidence table for randomized controlled trials comparing calcium supplementation with placebo for risk of hip fractures (fxs)1

Reid et al (11)2 Prince et al (13) Reid et al (10) Grant et al (12)4 Bischoff-Ferrari et al (14) Pooled results Primary analysis (intention-to-treat)5 Sensitivity analysis (intention-to-treat)6 Per-protocol analysis7

Outcome

Year

Treated subjects

Control subjects

n

n

Hip fracture Hip fracture Hip fracture Hip fracture Hip fracture

1995 2006 2006 2006 2006

0/423 11/730 17/732 49/1311 6/464

2/443 6/730 5/739 41/1332 4/466

0.33 (0.03, 4.23) 1.83 (0.69, 4.86) 3.43 (1.35, 8.70) 1.21 (0.81, 1.82) 1.51 (0.43, 5.26)

86 1460 1471 2643 930

83/3237 83/3279 27/1314

56/3267 58/3311 22/1464

1.64 (1.02, 2.64) 1.57 (0.96, 2.55) 1.42 (0.81, 2.49)

6504 6590 2778

RR (95% CI)

Total n

RR, risk ratio. Intention-to-treat results for women alone based on the primary analysis comparing calcium with placebo (pooled RR ҃ 1.66; 95% CI: 0.97, 2.86; Q ҃ 4.41; P ҃ 0.22, Q test) were similar to the results including men. The few data available for men alone from the trials of Grant et al (RR ҃ 1.55) and Bischoff-Ferrari et al (RR ҃ 1.58) (pooled RR ҃ 1.56; 95% CI: 0.62, 3.96; P ҃ 0.98, Q test) were similar to the pooled risk among women. 2 Trial excluded for the primary analysis. Based on the intention-to-treat results, Table 4 shows the pooled results for the RR of having a hip fracture with calcium supplementation compared with placebo, including data from men and women. 3 n with fracture/total n of the group (all such values). 4 This report was from the Randomized Evaluation of Calcium or Vitamin D (RECORD) trial. 5 In the primary pooled analysis, there was a significant 1.64-fold risk with calcium than with placebo (95% CI: 1.02, 2.64; P ҃ 0.04; homogeneity: Q ҃ 4.3; P ҃ 0.24, Q test). 6 In the sensitivity analysis, including one additional small trial, hip fracture risk was still 1.57-fold, although the 95% CI included 1 (95% CI: 0.96, 2.55; homogeneity: Q ҃ 5.64; P ҃ 0.23, Q test). 7 Per-protocol results were available for the trials of Prince et al, Reid et al, and Grant et al including a total of 2778 subjects (all women). The pooled RR excluding subjects with 쏝80% compliance in the trials of Prince et al and Grant et al and excluding subjects with at least one 6-mo episode with 쏝60% compliance in the trial of Reid et al was 1.42 (95% CI: 0.81, 2.49; Q ҃ 1.65; P ҃ 0.44, Q test). 1

prospective cohort studies. Calcium intake is imperfectly measured in observational studies, and this measurement error—if nondifferential—would lead to an underestimation of a true calcium effect. However, in several of the studies, the validity of the calcium intake estimates was assessed by comparison with more detailed methods (37, 38), and the correlations between food-frequency questionnaires and 1-wk diet records were high, 앒0.75 (37, 39). This modest degree of

measurement error would tend to cause a conservative bias, but, with the large number of cases in women than in men, an important association should not have been missed. Moreover, in some of the same studies that reported fracture data, dietary calcium intake has been inversely associated with the risk of kidney stones (38, 40, 41) and colon cancer (42), which showed that the calcium intake measures used were at least accurate enough to detect those relations.

Prince et al (13)

Reid et al (10)

Bischoff-Ferrari et al (14)

Grant et al (12)

Combined

0.1

0.5

1 RR

5

10

FIGURE 5. Forest plot comparing intention-to-treat data from 4 randomized controlled trials for the risk of hip fracture between calcium-treated and placebo groups. The squares represent the relative risk (RR) of fracture between subjects who took calcium in any dose and those who took placebo. In a total of 6504 subjects, the pooled RR was 1.64 (95% CI: 1.02, 2.64). There was no heterogeneity between studies (P ҃ 0.24, Q test).

1788

BISCHOFF-FERRARI ET AL

Another possible explanation for the lack of association is that patients with recognized osteoporosis are generally advised to increase calcium intake, which could mask an inverse association between calcium intake and fracture risk (29). However, in the Nurses’ Health Study, the exclusion of women with a history of diagnosed osteoporosis did not appreciably affect its negative findings (25). Alternatively, calcium alone may not prevent hip fractures in women. In fact, our meta-analysis of RCTs suggests an increased risk with calcium supplementation among men and women. It is possible that, among the frail subjects at risk of hip fracture, other deficiencies, such as vitamin D deficiency and phosphate deficiency due to low protein intake, should be corrected along with ensuring adequate calcium intake (17). Calcium carbonate or calcium citrate supplements can reduce phosphate absorption (43), which may be detrimental, because a balanced ratio of calcium to phosphate is needed for bone mineralization (44). Phosphate deficiency [defined as an intake 쏝70% of adult recommended dietary allowance (700 mg/d)] is found in 10% of US women 쏜60 y old and in 15% of US women 쏜80 y old (45). Each increase in calcium intake by 500 mg/d decreases phosphorus absorption by 166 mg (43), so a calcium supplement of 1000 mg may shift an elderly person with a relatively low phosphorus intake into phosphate deficiency (43, 46). This change could augment bone resorption (43, 47, 48) and thus increase fracture risk. Conversely, in the trial by Chapuy et al (17), the beneficial effect of vitamin D plus calcium on hip fracture risk in frail elderly women may have been enhanced by the use of tricalcium phosphate, which may have avoided a calcium-related phosphate deficiency. Furthermore, vitamin D stimulates phosphate absorption (49), which may enhance phosphate uptake from nutritional sources in calcium supplement users. Such a benefit is supported by a recent meta-analyses showing that hip fracture risk is significantly reduced in trials that combined any calcium supplement with vitamin D (5, 6). Similarly, the main dietary sources of calcium also contain phosphorus, which could explain the lack of a positive association between calcium intake and fracture risk in the cohort studies. An alternative explanation for the pooled RCT data on hip fractures is poor statistical power. The 4 studies that were pooled for the hip fracture outcome were limited by a relatively small number of cases (139 fractures), and one trial had poor adherence and a focus on secondary fracture prevention (12). Thus the apparent elevation in risk may be due to chance. However, if we considered only adherent subjects from the 3 largest trials, the elevated risk was maintained. Furthermore, each of the 4 trials included in the primary analysis indicated an elevated risk of hip fracture with RRs between 1.21 and 3.43, although only 1 of these trials reached significance (10). Thus the consistency of the evidence argues against a chance finding. Moreover, as discussed above, commonly used calcium supplements with a carbonate or citrate component have been shown to shift older persons into phosphate deficiency (43, 46) and to induce bone loss (43, 47, 48). To build calcium into bone, a calcium-phosphate product is needed, which may be disturbed by the described calcium supplement–induced phosphate malabsorption. This may be especially critical in frail older persons ndividuals at risk of both hip fracture and phosphate deficiency because of low protein intake. Thus, our findings are mechanistically plausible. It is most important, however, that, with a CI excluding 1 for the primary analysis, an important reduction in hip fracture risk

with calcium supplementation seems unlikely. Furthermore, despite the limited data on men, a differential effect of calcium on hip fracture risk by sex appears unlikely: the pooled RRs were 1.66 among women and 1.56 among men. Still, more data on fracture risk in men are needed. The strengths of our meta-analysis of prospective cohort studies are the large number of cases, the long duration of follow-up, and the inclusion of both men and women. Prospective cohort studies have less potential for bias than do other observationally designed studies, because the data on calcium intake are assessed before occurrence of fractures. In the presence of limited data from RCTs, as confirmed in our meta-analysis, summarization of these studies is likely to be the most informative approach. However, our analysis has limitations. Prospective cohort studies may still be susceptible to bias, including loss to follow-up and residual confounding. Another limitation of our study is that the calcium intake from supplements was not assessed in all of the studies. Nonetheless, we did not detect heterogeneity between studies that assessed calcium intake from both dietary and supplement sources and those that studied only dietary intake. Our study was also limited by the lack of information on baseline 25-hydroxyvitamin D concentrations, phosphate intake, and physical activity. These factors could potentially modify the associations between calcium intake and fracture risk (17, 19, 50). In the RCTs included in our metaanalysis, participants’ mean baseline calcium intakes exceeded the estimated mean intake of persons aged 욷50 y in the general US population—ie, 763 mg/d in men and 558 mg/d in women (22). Thus, we cannot exclude the possibility that persons with very low baseline calcium intakes may benefit more from calcium supplementation than those with higher calcium intakes; another reason for the negative findings could be that the subjects in the RCTs pooled for this meta-analysis already had “enough” calcium. In summary, our results do not support an overall beneficial effect of greater calcium intake on hip fracture risk. Among women, the cohort data suggest a neutral effect of calcium intake on hip fractures, but data from RCT’s of calcium supplementation suggest an adverse effect, even among adherent women. In addition, RCT data for any nonvertebral fractures indicate a neutral effect of calcium with respect to fracture reduction. Thus, future studies of the prevention of hip fracture or any nonvertebral fracture in women should not consider calcium supplementation alone but, rather, should focus on the optimal combination of calcium plus vitamin D and possibly on the correction of phosphate deficiency by using calcium-phosphate supplements. RCT and prospective cohort study findings for men did not support a beneficial effect of calcium intake on hip fracture risk in men, but further studies in men are needed as data are limited. The authors thank the investigators of the RECORD trial for providing unpublished data for this meta-analysis. The authors’ responsibilities were as follows—HAB-F had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis; HAB-F, WCW, JAB, and BD-H: study concept and design; HAB-F and EO: acquisition of data; all authors: analysis and interpretation of data and critical review of the manuscript; HAB-F, WCW, JAB, BD-H, JEO: writing the manuscript draft; HAB-F, WCW, JEO, DS, and RL: statistical analysis; HAB-F: obtained funding; and WCW: administrative, technical, or material support. JBW receives funding from federal agencies, Schering Plough, and Centocor for work unrelated to studies of calcium or falls and fractures; JAB receives study agents (medications)

CALCIUM INTAKE AND HIP FRACTURE RISK: META-ANALYSIS from Wyeth, and, with Dartmouth College, holds a use patent for calcium as a cancer chemopreventive agent. No other authors had any personal or financial conflict of interest.

REFERENCES 1. Shea B, Wells G, Cranney A, et al. Meta-analyses of therapies for postmenopausal osteoporosis. VII. Meta-analysis of calcium supplementation for the prevention of postmenopausal osteoporosis Endocr Rev 2002;23(4):552–9. 2. Shea B, Wells G, Cranney A, et al. Calcium supplementation on bone loss in postmenopausal women. Cochrane Database Syst Rev 2004;1(1): CD004526. 3. Cumming RG, Cummings SR, Nevitt MC, et al. Calcium intake and fracture risk: results from the study of osteoporotic fractures. Am J Epidemiol 1997;145(10):926 –34. 4. Feskanich D, Willett WC, Stampfer MJ, Colditz GA. Milk, dietary calcium, and bone fractures in women: a 12-year prospective study. Am J Public Health 1997;87(6):992–7. 5. Avenell A, Gillespie WJ, Gillespie LD, O’Connell DL. Vitamin D and vitamin D analogues for preventing fractures associated with involutional and post-menopausal osteoporosis. Cochrane Database Syst Rev 2005;20(3):CD000227. 6. Boonen S, Lips P, Bouillon R, Bischoff-Ferrari HA, Vanderschueren D, Haentjens P. Need for additional calcium to reduce the risk of hip fracture with vitamin D supplementation: evidence from a comparative metaanalysis of randomized controlled trials. J Clin Endocrinol Metab 2007; 92:1415–23. 7. Bischoff-Ferrari HA, Willett WC, Wong JB, Giovannucci E, Dietrich T, Dawson-Hughes B. Fracture prevention with vitamin D supplementation: a meta-analysis of randomized controlled trials. JAMA 2005; 293(18):2257– 64. 8. Barrett JA, Baron JA, Karagas MR, Beach ML. Fracture risk in the U.S. Medicare population. J Clin Epidemiol 1999;52(3):243–9. 9. Gullberg B, Johnell O, Kanis JA. World-wide projections for hip fracture. Osteoporos Int 1997;7(5):407–13. 10. Reid IR, Mason B, Horne A, et al. Randomized controlled trial of calcium in healthy older women. Am J Med 2006;119(9):777– 85. 11. Reid IR, Ames RW, Evans MC, Gamble GD, Sharpe SJ. Long-term effects of calcium supplementation on bone loss and fractures in postmenopausal women: a randomized controlled trial. Am J Med 1995; 98(4):331–5. 12. Grant AM, Avenell A, Campbell MK, et al. Oral vitamin D3 and calcium for secondary prevention of low-trauma fractures in elderly people (Randomised Evaluation of Calcium Or vitamin D, RECORD): a randomised placebo-controlled trial Lancet 2005;365(9471):1621– 8. 13. Prince RL, Devine A, Dhaliwal SS, Dick IM. Effects of calcium supplementation on clinical fracture and bone structure: results of a 5-year, double-blind, placebo-controlled trial in elderly women. Arch Intern Med 2006;166(8):869 –75. 14. Bischoff-Ferrari HA, Rees JR, Grau MV, Barry EL, Baron JA. Effect of calcium supplementation on fracture risk: a double-blind randomized controlled trial. J Bone Miner Res 2006;21(suppl):S60 (abstr). 15. Wickham CA, Walsh K, Cooper C, et al. Dietary calcium, physical activity, and risk of hip fracture: a prospective study. BMJ 1989;299(6704):889 –92. 16. Huopio J, Kroger H, Honkanen R, Saarikoski S, Alhava E. Risk factors for perimenopausal fractures: a prospective study. Osteoporos Int 2000; 11(3):219 –27. 17. Chapuy MC, Arlot ME, Duboeuf F, et al. Vitamin D3 and calcium to prevent hip fractures in the elderly women. N Engl J Med 1992;327(23): 1637– 42. 18. Chapuy MC, Pamphile R, Paris E, et al. Combined calcium and vitamin D3 supplementation in elderly women: confirmation of reversal of secondary hyperparathyroidism and hip fracture risk: the Decalyos II Study. Osteoporos Int 2002;13(3):257– 64. 19. Dawson-Hughes B, Harris SS, Krall EA, Dallal GE. Effect of calcium and vitamin D supplementation on bone density in men and women 65 years of age or older. N Engl J Med 1997;337(10):670 – 6. 20. Holbrook TL, Barrett-Connor E, Wingard DL. Dietary calcium and risk of hip fracture: 14-year prospective population study. Lancet 1988; 2(8619):1046 –9. 21. Paganini-Hill A, Chao A, Ross RK, Henderson BE. Exercise and other

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

32. 33.

34.

35. 36. 37.

38.

39.

40.

41.

42.

43.

44.

1789

factors in the prevention of hip fracture: the Leisure World Study. Epidemiology 1991;2(1):16 –25. Looker AC, Harris TB, Madans JH, Sempos CT. Dietary calcium and hip fracture risk: the NHANES I Epidemiologic Follow-Up Study. Osteoporos Int 1993;3(4):177– 84. Meyer HE, Pedersen JI, Loken EB, Tverdal A. Dietary factors and the incidence of hip fracture in middle-aged Norwegians. A prospective study. Am J Epidemiol 1997;145(2):117–23. Michaelsson K, Melhus H, Bellocco R, Wolk A. Dietary calcium and vitamin D intake in relation to osteoporotic fracture risk. Bone 2003; 32(6):694 –703. Feskanich D, Willett WC, Colditz GA. Calcium, vitamin D, milk consumption, and hip fractures: a prospective study among postmenopausal women. Am J Clin Nutr 2003;77(2):504 –11. Owusu W, Willett WC, Feskanich D, Ascherio A, Spiegelman D, Colditz GA. Calcium intake and the incidence of forearm and hip fractures among men. J Nutr 1997;127(9):1782–7. Riggs BL, O’Fallon WM, Muhs J, O’Connor MK, Kumar R, Melton LJ III. Long-term effects of calcium supplementation on serum parathyroid hormone level, bone turnover, and bone loss in elderly women. J Bone Miner Res 1998;13(2):168 –74. Chevalley T, Rizzoli R, Nydegger V, et al. Effects of calcium supplements on femoral bone mineral density and vertebral fracture rate in vitamin-D-replete elderly patients. Osteoporos Int 1994;4(5):245–52. Cummings SR, Kelsey JL, Nevitt MC, O’Dowd KJ. Epidemiology of osteoporosis and osteoporotic fractures. Epidemiol Rev 1985;7:178 – 208. Bischoff HA, Dietrich T, Orav JE. Dawson-Hughes B. 2002 Positive association between 25-hydroxy vitamin D levels and bone mineral density: a population-based study of younger and older adults. Am J Med 2004;116:634 –9. Greenland S, Longnecker MP. Methods for trend estimation from summarized dose-response data, with applications to meta-analysis. Am J Epidemiol 1992;135(11):1301–9. Berkey CS, Hoaglin DC, Mosteller F, Colditz GA. A random-effects regression model for meta-analysis. Stat Med 1995;14(4):395– 411. Feit F, Brooks MM, Sopko G, et al. Long-term clinical outcome in the Bypass Angioplasty Revascularization Investigation Registry: comparison with the randomized trial. BARI Investigators. Circulation 2000; 101(24):2795– 802. Egger M, Juni P, Bartlett C, Holenstein F, Sterne J. How important are comprehensive literature searches and the assessment of trial quality in systematic reviews? Empirical study. Health Technol Assess 2003;7(1): 1–76. Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997;315(7109):629 –34. Egger M, Smith GD, Altman DG, eds. Systematic reviews in health care. 2nd ed. London, United Kingdom: BMJ Books, 2001:211–7. Martinez ME, Willett WC. Calcium, vitamin D, and colorectal cancer: a review of the epidemiologic evidence. Cancer Epidemiol Biomarkers Prev 1998;7(2):163– 8. Curhan GC, Willett WC, Knight EL, et al. Dietary factors and the risk of incident kidney stones in younger women: Nurses’ Health Study II. Arch Intern Med 2004;164(8):885–91. Cummings SR, Block G, McHenry K, Baron RB. Evaluation of two food frequency methods of measuring dietary calcium intake. Am J Epidemiol 1987;126(5):796 – 802. Curhan GC, Willett WC, Rimm EB, Stampfer MJ. A prospective study of dietary calcium and other nutrients and the risk of symptomatic kidney stones. N Engl J Med 1993;328(12):833– 8. Curhan GC, Willett WC, Speizer FE, Spiegelman D, Stampfer MJ. Comparison of dietary calcium with supplemental calcium and other nutrients as factors affecting the risk for kidney stones in women. Ann Intern Med 1997;126(7):497–504. Brandt KD, Heilman DK, Slemenda C, et al. Quadriceps strength in women with radiographically progressive osteoarthritis of the knee and those with stable radiographic changes. J Rheumatol 1999;26(11): 2431–7. Heaney RP, Nordin BE. Calcium effects on phosphorus absorption: implications for the prevention and co-therapy of osteoporosis. J Am Coll Nutr 2002;21(3):239 – 44. Chen TC, Persons K, Liu WW, Chen ML, Holick MF. The antiproliferative and differentiative activities of 1,25-dihydroxyvitamin D3 are

1790

BISCHOFF-FERRARI ET AL

potentiated by epidermal growth factor and attenuated by insulin in cultured human keratinocytes. J Invest Dermatol 1995;104(1):113–7. 45. Alaimo K, McDowell MA, Briefel RR, et al. Dietary intake of vitamins, minerals, and fiber of persons ages 2 months and over in the United States: third National Health and Nutrition Examination Survey, Phase 1, 1988 –91. Adv Data 1994;(258):1–28. 46. Heaney RP. Phosphorus nutrition and the treatment of osteoporosis. Mayo Clin Proc 2004;79(1):91–7. 47. Raisz LG, Niemann I. Effect of phosphate, calcium and magnesium on

bone resorption and hormonal responses in tissue culture. Endocrinology 1969;85(3):446 –52. 48. Lotz M. The diagnostic importance of hypophosphatemia. Med Times 1968;96(12):1166 – 8. 49. DeLuca HF. Overview of general physiologic features and functions of vitamin D. Am J Clin Nutr 2004;80(suppl):1689S–96S. 50. Specker BL. Evidence for an interaction between calcium intake and physical activity on changes in bone mineral density. J Bone Miner Res 1996;11(10):1539 – 44.

A prospective investigation of the relations among cognitive dietary restraint, subclinical ovulatory disturbances, physical activity, and bone mass in healthy young women1–3 Esther J Waugh, Janet Polivy, Rowena Ridout, and Gillian A Hawker ABSTRACT Background: Cognitive dietary restraint (CDR) may mediate subclinical ovulatory disturbances, which may result in loss of bone mineral density (BMD). CDR is associated with greater physical activity, which may modify the effect of CDR and ovulatory disturbances on bone mass. Objective: We aimed to investigate the relations among CDR, ovulatory disturbances, and physical activity and their effect on BMD in healthy premenopausal women over a 2-y period. Design: In this prospective cohort study, key explanatory factors, important covariates, and BMD were measured at baseline and at 12 and 24 mo; 225 women completed the baseline assessment, and 189 completed the study. CDR was measured with the Three-Factor Eating Questionnaire, and physical activity was measured with the Baecke scale. An average of 9.8 menstrual cycles in 2 y were monitored by using salivary progesterone measurements and urinary ovulation detection kits. Ovulatory disturbances included anovulatory cycles or short luteal phase lengths of 쏝10 d. BMD at the lumbar spine, femoral neck, and total body was measured by using dualenergy X-ray absorptiometry. General linear mixed modeling was used to determine predictors of change in BMD over time. Results: CDR was not associated with ovulatory disturbances or changes in BMD. The average annual rate of change in lumbar spine BMD was decreased by 0.01 g/cm2 in women who had experienced 욷3 monitored cycles with ovulatory disturbances (P ҃ 0.02). Conclusions: CDR did not predict bone loss, and there was no relation between CDR and ovulatory disturbances. Ovulatory disturbances had a negative effect on the rate of change at the lumbar spine. The cause of these disturbances is unknown. Am J Clin Nutr 2007;86:1791– 801. KEY WORDS Dietary restraint, bone mineral density, BMD, ovulatory disturbances, premenopausal women, physical activity

INTRODUCTION

Cognitive dietary restraint (CDR) is defined as the conscious attempt to limit and monitor food intake to achieve or maintain a desired weight. CDR is associated with eating behavior that is governed by cognitive processes rather than by physiologic mechanisms such as hunger and satiety (1). This CDR behavior may lead to impairment of the psychophysiologic regulation of food intake (2), which could result in alternating periods of restricted eating and overeating and in weight fluctuations (3).

Other characteristics associated with CDR include heightened anxiety (4) and greater physical activity levels (5, 6). CDR may have important biological consequences in healthy premenopausal women, particularly with respect to menstrual function and bone health. Three previous studies reported that CDR is associated with ovulatory disturbances, including a greater proportion of anovulatory cycles and short luteal phase lengths (7–9). These disturbances may have a negative effect on bone mass. Prior et al (10) found that anovulatory cycles and short luteal phase lengths were related to spinal bone loss over a 1-y period; 2 studies by others did not confirm these findings (11, 12). Four studies directly investigated the relation between CDR and bone mineral density (BMD) in healthy premenopausal women, and they provide modest evidence for a negative association (8, 13–15). One study found that total-body (TB) bone mineral content (BMC) but not BMD was lower in women with high restraint scores and body weight 쏝 71 kg (15). A second study reported that the restraint score was a negative predictor of TB BMD and BMC in normal-weight university students (14), and a third study found a negative correlation between femoral BMC and cognitive restraint in obese premenopausal women with a history of chronic dieting (13). The fourth study did not find an association between restraint and bone density, but it was limited by a small sample size (8). Further investigation of the relation among CDR, ovulatory disturbances, and BMD in healthy young women is required. Four of the prior studies (7–9, 14) were conducted in normalweight or underweight women, who may represent a unique subgroup of successful restrained eaters who may experience more pronounced physiologic and metabolic effects. Evidence of 1 From the Osteoporosis Research Program, Women’s College Hospital, Toronto, Canada (EJW and GAH), and the Departments of Health Policy, Management and Evaluation (EJW and GAH), Psychology (JP), and Medicine (RR and GAH), University of Toronto, Toronto, Canada. 2 Supported by grant no. MT15645 from the Canadian Institutes of Health Research and by a doctoral fellowship from the Canadian Institutes of Health Research (to EJW). GAH was a Canadian Institutes of Health Research Scientist and is the FM Hill Chair in Academic Women’s Medicine at the University of Toronto. 3 Reprints not available. Address correspondence to EJ Waugh, Osteoporosis Research Program, Women’s College Hospital, 76 Grenville Street, 8th Floor E, Room 812C, Toronto, ON M5S 1B2, Canada. E-mail: [email protected]. Received January 15, 2007. Accepted for publication August 9, 2007.

Am J Clin Nutr 2007;86:1791– 801. Printed in USA. © 2007 American Society for Nutrition

1791

1792

WAUGH ET AL

a negative relation between subclinical ovulatory disturbances and BMD remains inconsistent. Moreover, the effect of physical activity, which is known to have a protective effect on BMD (16), but which may be associated with both CDR (5, 6) and ovulatory disturbances (17, 18), has been addressed in only one study (14), which reported that the positive effect of exercise was moderated in women with high restraint. Finally, none of the studies concurrently evaluated the relations among CDR, ovulatory disturbances, physical activity, and BMD in one sample. The purpose of the present study was to build on these prior findings by investigating these relations over a 2-y period in a large sample of healthy premenopausal women who represent a broad spectrum of body weights. Specifically, we addressed 4 questions. 1) Is CDR associated with subclinical ovulatory disturbances? 2) Does CDR or subclinical ovulatory disturbances (or both) predict loss of BMD? 3) Is the relation between CDR and changes in BMD moderated by ovulatory disturbances, physical activity, or body weight? 4) What are the key independent predictors of change in BMD over 2 y? SUBJECTS AND METHODS

Subjects Participants were healthy, white, premenopausal women aged 21– 40 y who were recruited from the community. Between June

Previous cohort: n = 662 white women

1999 and August 2001, participants were recruited from 2 groups: from an existing cohort that was described previously (19) and from women who responded to advertisements in local newspapers or to posted flyers seeking volunteers for a longitudinal study on determinants of peak bone mass. Of the 842 women who initially agreed to participate, 510 were ineligible, and 107 completed the baseline assessment but subsequently refused to monitor their menstrual cycles; thus, the baseline sample was 225 women. Twelve participants did not return for BMD testing at year 1, and an additional 24 did not complete year 2. A total of 189 participants completed the study (Figure 1). Exclusion criteria for the study included prior diagnosis of osteoporosis or specific comorbid conditions known to be associated with bone loss (ie, Crohn disease, symptomatic hyperthyroidism, rheumatoid arthritis, hysterectomy or bilateral oophorectomy, or anorexia nervosa); medications known to affect bone metabolism (ie, oral contraceptives or other exogenous hormones in the previous 3 mo, systematic glucocorticoids, hydrochlorothiazide and bisphosphonates in the previous 6 mo, treatment for infertility or endometriosis, use at any time of anticonvulsants for 욷3 mo, or daily use of inhaled corticosteroids); 쏝8 menstrual cycles in the previous year; or pregnancy or active breastfeeding at the time of the study or within the previous 12 mo. Written informed consent was obtained from all subjects. Ethical approval was obtained from the ethics review committees of

New recruits: 733 calls received

Ineligible: 250 • pregnant or breastfeeding: 58 • medications or illness: 10 • oral contraceptives: 148 • hysterectomy or irregular periods: 12 • trying to conceive: 20 • other: 2 Refused: 98 Unable to contact or out of town: 212

Ineligible: 260 • nonwhite: 165 • medications or illness: 68 • oral contraceptives: 27 Refused: 243

n = 102

n = 230 332 completed baseline

107 subsequently refused to monitor cycles

n = 225 completed baseline and at least 1 monitored cycle

n = 213 completed year 1

n = 189 completed year 2

36 dropped out • pregnant: 8 • oral contraceptives: 17 • refused final bone scan: 1 • moved: 3 • cycle monitoring too difficult: 7

FIGURE 1. Flow chart depicting recruitment of participants from a previous cohort and new community sample to obtain a sample of 225 participants who completed baseline assessment, 213 who completed the year 1 assessment, and 189 who completed the study.

DIETARY RESTRAINT AND BONE MASS IN YOUNG WOMEN

the Sunnybrook and Women’s College Health Sciences Centre and the University of Toronto. Study design This study was a 2-y prospective cohort study. Key explanatory factors, important covariates, and BMD were measured at baseline and at 12 and 24 mo. Menstrual function was measured over the 2-y study period. Physical measurements Height (cm) was measured with a stadiometer and weight (kg) was measured with a balance-beam scale. These measurements were used to calculate body mass index (BMI; in kg/m2). Body composition (lean mass and fat mass) was measured by using dual-energy X-ray absorptiometry (Lunar DPX-L bone densitometer; Lunar Corporation, Madison, WI) (20). Questionnaires Participants completed a detailed osteoporosis risk factor questionnaire to elicit information regarding lifestyle factors (ie, cigarette smoking and alcohol and caffeine consumption), family history of osteoporosis, calcium and vitamin D supplementation, medication use, and prior surgeries and illnesses. CDR was measured by using the restraint subscale of the Three-Factor Eating Questionnaire (TFEQ-R) (21). The TFEQ-R is widely used to measure the construct of CDR (21). Higher scores on this measure are predictive of reduced caloric intake, avoidance of dietary fat, consumption of greater amounts of artificial sweeteners and diet products, and an intense concern about one’s physical appearance in the absence of typical features of a psychiatric illness or eating disorder (22–24). The TFEQ-R consists of 21 items, each scored on a 2-point scale (0 or 1); a higher score is indicative of greater cognitive restraint. Participants also completed the State Trait Anxiety Inventory (STAI), a widely used measure of trait anxiety consisting of 20 items (25); a higher score indicates greater levels of general psychological stress. Physical activity level was determined by using the Baecke questionnaire, a valid and reliable self-administered questionnaire developed for the measurement of habitual physical activity in epidemiologic studies (26, 27). The Baecke questionnaire consists of work, leisuretime, sports, and total activity indexes. Finally, an intervieweradministered food-frequency questionnaire was completed to measure daily dietary intake of calcium. Food models and standardized portion sizes were used to aid participant recall and accuracy. Assessment of ovulatory disturbances and hormone analyses Three consecutive menstrual cycles were evaluated twice during each of the 2 y of the study (total of 12 cycles) with the use of salivary progesterone measurements and urinary ovulation detection kits. Each block of 3 menstrual cycles was monitored during a different season so that all 4 seasons were included in the 2-y study. Salivary progesterone measurement is an established method by which to study luteal function in epidemiologic studies (28, 29). For each of the monitored cycles, participants collected saliva samples each morning before breakfast or tooth brushing. Samples were stored at Ҁ20 °C until they were analyzed. Salivary samples underwent radioimmunoassay (Diagnostic Systems Laboratories, Webster, TX) for progesterone

1793

with the use of a modified version of the method of Walker et al (30), described previously (31), in the laboratory of M Steiner (Women’s Health Concerns Clinic, St Joseph’s Healthcare, (Hamilton, Canada). The interassay and intraassay CVs were 15.7% and 13.9%, respectively. Luteal phase samples were pooled to derive time-integrated mean luteal and mean midluteal progesterone values for each cycle. Specifically, mean luteal progesterone values were determined by pooling aliquots from the day when a urinary luteinizing hormone (LH) surge was detected until the day before the next menstrual cycle commenced. The days used to determine mean midluteal progesterone values were dependent on the day of the LH surge. For example, in a cycle in which the LH surge was detected 14 d before the start of the next menstrual cycle, mean midluteal progesterone was determined by pooling 5 samples from the span of time 9 –5 d before commencement of the next cycle. To detect an LH surge, participants were provided with urinary ovulation detection kits (ClearPlan Easy; Unipath Ltd, Bedford, United Kingdom). These kits provide a valid and reliable method of identifying the LH surge that indicates the onset of ovulation, and thus they allow determination of the luteal phase length (32). ClearPlan Easy has been found to have an 82– 88% positive predictive value for ovulation within 1 d (33). Participants followed the instructions provided with the kit, including testing at the same time of the day and not urinating for 4 h before testing. On the basis of information on menstrual cycle patterns that was provided during the baseline visit, participants were asked to begin urinary testing 14 d before the anticipated end of the cycle and to continue testing for 5 consecutive days or until the test became positive. A positive test indicated that the LH surge had occurred. The duration of the luteal phase was calculated from the day after the LH surge until the day before the onset of menstrual flow. The follicular phase was the the period from first day of menstrual flow to the day before the onset of the luteal phase. The criterion for a short luteal phase length based on urinary LH peak data was 쏝10 d (17, 34 –36). When no urinary LH surge was detected, salivary progesterone concentrations were assayed consecutively from samples collected beginning 14 d before day 1 of the following cycle. If there was no rise in salivary progesterone to 욷2 SDs above the mean progesterone concentrations of the previous days after 7 d of testing, and if these concentrations did not remain elevated for 욷2 d, the cycle was considered to be anovulatory (31, 37– 40). Participants kept a menstrual calendar for the duration of the study, in which they recorded testing days, the day on which the test became positive, and the onset of each menstrual period. To measure serum concentrations of estradiol and free testosterone, we obtained serum samples once between 0800 and 1000 on days 3–5 of 1 of the 12 monitored cycles, chosen at random. All samples were frozen at Ҁ70 °C and analyzed by using commercial radioimmunoassays in the Endocrinology and Metabolism Laboratory at the University of Toronto (Toronto, Canada). Measurement of bone density Lumbar spine (L1–L4) (LSP), hip (femoral neck) (FN), and TB BMD measurements were obtained with the use of dualenergy X-ray absorptiometry. Measurements were performed by a single, certified densitometry technician. The CVs, determined by test-retest with repositioning, were 1.18% at LSP, 1.56% at FN, and 0.72% at TB.

1794

WAUGH ET AL

Statistical analysis All analyses were performed by using SAS statistical software (version 9.1; SAS Institute Inc, Cary, NC). The level of significance was established at P 울 0.05 (2-tailed). BMD at the LSP was considered the primary outcome, because it consists primarily of trabecular bone. It was anticipated that the variables being investigated would have the most demonstrable effect on this site. To ensure completeness and to enable comparisons with other studies, BMD at the FN and TB was also included in the analysis as secondary outcomes. All variables were assessed for normality. Paired t tests were used to evaluate changes in mean BMD values and in predictor variables between year 2 and baseline. Because there was minimal change in predictor variables over the 2-y study, evaluation of potential associations between and among variables was conducted on baseline values. Participants were categorized into 2 groups according to whether they had experienced 쏝3 or 욷3 monitored menstrual cycles with subclinical ovulatory disturbances (anovulatory cycles or short luteal phases). Participants were also categorized into tertiles of CDR on the basis of their baseline TFEQ-R scores. Comparisons of physical and lifestyle characteristics and hormone values between ovulatory disturbance groups and CDR groups were done by using unpaired t tests or Wilcoxon rank-sum tests as appropriate for continuous variables and chi-square tests for categorical variables. The lowest and middle tertiles of CDR were compared with the highest tertile. A general linear mixed method was used to model the effects of predictor variables on interindividual and intraindividual changes in BMD. The mixed model extends the general linear model by the addition of random effects, which allows each subject’s initial BMD status (intercept) and rate of change (slope) to vary from the population average. This method also enables the estimation of individual-specific and population-specific fixed effects through the inclusion of time-varying and timeinvariant variables, respectively (41– 43). An unstructured covariance structure was specified, and maximum likelihood estimation method was used for parameter estimation (42). Random effects for both intercept and slope were included. The change variable was named “time” and was centered at time 0 to facilitate interpretation of the intercept (42). SAS PROC MIXED was used for this analysis. Our initial models evaluated the effects of CDR and subclinical ovulatory disturbances on changes in BMD after control for BMI and physical activity (total Baecke score). Interactions between CDR and ovulatory disturbances, between CDR and BMI, and between CDR and physical activity were evaluated to determine whether the effect of CDR on BMD was moderated by these variables. Extended models to determine all key predictors of change in BMD were then generated. Time-varying predictors that were included in the complex model were BMI, percentage lean mass, CDR (TFEQ-R score), calcium intake (does not meet recommended daily amount or meets recommended daily amount), and smoking status (nonsmoker or current smoker); time-invariant predictors included were subclinical ovulatory disturbances (쏝3 or 욷3 cycles with disturbances), hormone values (ie, estradiol, testosterone, and mean luteal progesterone), age at menarche, Baecke total score, physically active as an adolescent (no or yes), alcohol consumption (no. of drinks/mo), age, and family history of osteoporosis (쏝3 or 욷3 relatives).

Variables were removed one-by-one if P 울 0.05, starting with the variable with the highest P value, to produce the final model (backward stepwise selection). RESULTS

Two hundred twenty-five women consented to the study and completed baseline assessments and 욷1 monitored cycle. Two hundred thirteen women completed year 1 assessments, and 189 women completed the 2-y study (16% were lost to follow-up). Of those who did not complete the study, 25 became ineligible because of pregnancy or commencing use of oral contraceptives, 7 found menstrual cycle monitoring too difficult, 3 moved, and 1 refused a third BMD test because of radiation concerns. Baseline demographic characteristics of the sample are summarized in Table 1. Physical and lifestyle characteristics and BMD values at each of the 3 time points are summarized in Table 2. There was a small but significant (P 쏝 0.0001) increase of 1.4% in mean BMD at LSP over the 2 y in the 189 participants who completed the study. There were nonsignificant increases of 0.5% at the FN (P ҃ 0.09) and 0.09% at the TB (P ҃ 0.45). Of the predictor variables, only calcium intake changed significantly, increasing by 12.7% (P ҃ 0.001). Subclinical ovulatory disturbances A total of 2213 menstrual cycles were monitored over the 2-y study period. Sixty percent of participants monitored 12 cycles, and 14% monitored 쏝6 cycles (average: 9.8 앐 3.4 cycles/participant). Blood samples were obtained from 205 women. The mean menstrual cycle length was 28.9 앐 3.9 d (range: 23.6 –55.2 d). Eight women had long mean cycle lengths (쏜36 d), and none had short mean cycle lengths (쏝21 d). The menstrual characteristics, hormone values, and subclinical ovulatory disturbances experienced by these women are summarized in Table 3. One hundred forty-one cycles (6.4%) showed ovulatory disturbances, as indicated by short luteal phase lengths (72 cycles) or anovulation (69 cycles). Forty-one women had 욷1 short luteal phase (쏝10 d), and 45 women had 욷1 anovulatory cycle. In total, 75 women (33.3%) experienced 욷1 subclinical ovulatory disturbance, and 16 women experienced 욷3 cycles with disturbances. Eight women experienced ovulatory disturbances TABLE 1 Demographic characteristics of the sample of 225 white premenopausal women Characteristic Age (y) Education (%) Elementary school High school Postsecondary school Marital status (%) Single Married or common-law spouse Separated or divorced Reproductive status (%) Nulliparous 1 birth 쏜1 birth History of 욷3 relatives with osteoporosis (%) 1

x៮ 앐 SD.

Value 32.4 앐 4.61 0.9 12.0 87.1 57.3 39.1 3.6 74.2 5.3 20.5 8.5

1795

DIETARY RESTRAINT AND BONE MASS IN YOUNG WOMEN TABLE 2 Summary of physical and lifestyle characteristics and bone mineral density values at baseline and at 12 and 24 mo1 Baseline (n ҃ 225) BMI (kg/m2) Lean mass (%) Baecke total score (/15) Active as an adolescent (%) TFEQ-R (/21) STAI (/80) Calcium intake (mg/d) Met RDA of 1000 mg/d (%) Alcohol consumption (drinks/mo) Smoking status (%) Current smoker Past smoker Bone mineral density (g/cm2) Lumbar spine (L1–L4) Femoral neck Total body

24.3 앐 4.62 66.1 앐 9.5 8.3 앐 1.3 82.2 7.6 앐 4.0 37.0 앐 9.0

Year 1 (12 mo) (n ҃ 213) 24.6 앐 4.7 65.6 앐 9.6 8.3 앐 1.3 81.7 7.3 앐 3.9 36.8 앐 8.7

Year 2 (24 mo) (n ҃ 189) 24.7 앐 4.9 65.6 앐 9.7 8.4 앐 1.4 81 7.5 앐 4.2 36.1 앐 8.6

876.5 앐 440.7 32.4 10.4 앐 12.8

811.5 앐 440.43 27.2 11.2 앐 12.6

965.2 앐 447.24 42.8 12.5 앐 14.7

6.2 16.9

7.5 16.0

7.4 14.8

1.200 앐 0.138 1.013 앐 0.139 1.181 앐 0.140

1.218 앐 0.1433 1.016 앐 0.144 1.183 앐 0.081

1.223 앐 0.1483 1.023 앐 0.1444 1.183 앐 0.080

1

TFEQ-R, Three Factor Eating Questionnaire–Restraint subscale; STAI, State Trait Anxiety Inventory; RDA, recommended daily amount. x៮ 앐 SD (all such values). 3 Significantly different from baseline in the 189 participants who completed the study. P 울 0.05 (paired t test). 4 Significantly different from Year 1 in the 189 participants who completed the study, P 울 0.05 (paired t test). 2

in 쏜40% of their monitored cycles. Baseline characteristics of women who experienced ovulatory disturbances over the 2-y study are shown in Table 4. There were no significant differences in physical and lifestyle characteristics or hormone values between women with 쏝3 cycles with ovulatory disturbances and women with 욷3 such cycles. Having a greater percentage of cycles with ovulatory disturbances was associated with significantly lower mean luteal and mean midluteal progesterone (P ҃ 0.0009 and P 쏝 0.001, respectively) but not with significantly lower estradiol (P ҃ 0.77) or testosterone (P ҃ 0.11). Cognitive dietary restraint and subclinical ovulatory disturbances TFEQ-R scores ranged from 0 to 18 out of a possible score of 21 with a weighted mean over the 2 y of 7.5 앐 4.0. Analysis of baseline scores showed that TFEQ-R scores were correlated with BMI (r ҃ 0.264, P 쏝 0.0001), percentage lean mass (r ҃ Ҁ0.167, P ҃ 0.01), Baecke physical activity score (r ҃ 0.163, P ҃ 0.01), Baecke sport index score (r ҃ 0.22, P ҃ 0.0005), and STAI score (r ҃ 0.193, P ҃ 0.007). The physical characteristics, menstrual characteristics, and hormone values for women in each tertile of CDR at baseline are shown in Table 5. Women who were in the highest tertile had significantly higher BMIs (P ҃ 0.0002), lower percentage lean mass (P ҃ 0.008), and higher Baecke sport index scores (P ҃ 0.006) but not significantly higher total Baecke scores (P ҃ 0.08) than did those who were in the lowest tertile. Women in the highest tertile also had significantly (P ҃ 0.03) higher BMIs than did those in the middle tertile. The mean menstrual cycle length and mean luteal phase length did not differ significantly, and the percentage of women with 욷3 cycles with ovulatory disturbances did not differ significantly across the restraint groups. There were no significant differences in estradiol, testosterone, or mean luteal progesterone values between the lowest, middle, and highest tertiles.

Cognitive dietary restraint, subclinical ovulatory disturbances, and 2-y change in BMD The results of the mixed-model analysis to evaluate the effect of CDR and subclinical ovulatory disturbances on changes in BMD, after adjustment for BMI and Baecke activity scores, are shown in Table 6. CDR had no effect on the average initial values or change in BMD at the LSP. Women who had 욷3 cycles with ovulatory disturbances had an average annual rate of change in LSP BMD (Ҁ0.01g/cm2) significantly (P ҃ 0.03) different from that in women with 쏝3 such cycles. There was no difference between these 2 groups in initial BMD values. There were no significant interactions between CDR and ovulatory disturbances, BMIs, or Baecke score at the LSP. Neither CDR nor ovulatory disturbances had significant effects at the FN or TB, and there were no significant interactions between CDR and the other 3 variables. Expanded predictive models of 2-y change in bone mineral density Final mixed models are summarized in Table 7. The results for the LSP show that BMI was positively associated with BMD and a family history of osteoporosis and not being physically active as an adolescent had a negative affect on average initial BMD values, whereas each additional alcoholic drink/mo had a positive effect on initial BMD values. Having 욷3 cycles with ovulatory disturbances had a negative effect on rate of change. The model indicates that average initial LSP values (when BMI is zero) in women who had no family history of osteoporosis, who were physically active as an adolescent, and who consumed no alcohol was 1.0780 g/cm2. Women with a family history of osteoporosis had a difference of Ҁ0.0784 g/cm2 in average initial BMD (P ҃ 0.01), women who did not exercise as an adolescent had a difference of Ҁ0.0569 g/cm2 in average initial BMD (P ҃ 0.01), and each additional alcoholic drink/mo was associated

1796

WAUGH ET AL

TABLE 3 Summary of menstrual characteristics, hormone values, and subclinical ovulatory disturbances based on assessment of 9.8 앐 3.4 menstrual cycles over 2 y in a sample of 225 premenopausal women1 Value Menstrual characteristics Age at menarche (y) Gynecologic age (y) Mean cycle length (d) Mean luteal phase length (d) Hormone values2 Estradiol (pg/mL) Free testosterone (pg/mL) Mean luteal progesterone (pmol/L) Mean midluteal progesterone (pmol/L) Subclinical disturbances Ever experienced short luteal phase length [n (%)] 1 short luteal phase (n) 2 short luteal phases (n) 3 short luteal phases (n) 4 short luteal phases (n) 5 short luteal phases (n) Ever experienced anovulatory cycle [n (%)] 1 anovulatory cycle (n) 2 anovulatory cycles (n) 3 anovulatory cycles (n) 4 anovulatory cycles (n) 6 anovulatory cycles (n) Ever experienced 욷1 short luteal phase length or anovulatory cycle [n (%)] 1 cycle with ovulatory disturbance (n) 2 cycles with ovulatory disturbance (n) 3 cycles with ovulatory disturbance (n) 4 cycles with ovulatory disturbance (n) 5 cycles with ovulatory disturbance (n) 6 cycles with ovulatory disturbance (n) Monitored cycles with ovulatory disturbances per participant [n (%)] 0% 1–10% 11–20% 21–30% 31–40% 41–50% 51–60% 쏜60% 1 2

12.7 앐 1.51 19.7 앐 4.9 28.9 앐 3.9 13.6 앐 1.7 37.9 앐 36.2 1.3 앐 0.6 373.4 앐 155.6 492.7 앐 222.5 41 (18.2) 23 8 8 1 1 45 (20.0) 32 7 3 2 1 75 (33.3) 39 20 9 2 3 2

150 (66.7) 30 (13.3) 26 (11.6) 9 (4.0) 2 (0.9) 7 (3.1) 1 (0.4) 0

x៮ 앐 SD (all such values). Estradiol and testosterone were collected once in 205 subjects.

with an increase of 0.0016 g/cm2 in average initial BMD (P ҃ 0.02) Each 1-point increase in BMI at each time point was associated with a BMD increase of 0.0052 g/cm2 (P 쏝 0.0001). These variables did not affect the average annual rate of change, which was 0.0096 g/cm2 (P 쏝 0.0001), after control for BMI. This rate of change was decreased by 0.0109 g/cm2 in women with 욷3 cycles with ovulatory disturbances, so that they experienced an annual loss of 0.0013 g/cm2 (P ҃ 0.02) (Figure 2). Having ovulatory disturbances did not affect initial BMD values. The FN model indicates that there was no significant annual change in BMD (P ҃ 0.08). BMI was positively associated with BMD, having a family history of osteoporosis had a negative effect on initial values and each 1-point increase in the Baecke

total score had a positive effect on initial values. The model indicates that initial FN BMD (when BMI is zero) was 0.7972 g/cm2 for women who did not have a family history of osteoporosis and who were not physically active (Baecke score of zero). Women with a family history of osteoporosis had a difference of Ҁ0.0732 g/cm2 in average initial BMD (P ҃ 0.02), and every 1-point increase in the Baecke total score was associated with an increase of 0.0157 g/cm2 in average initial BMD (P ҃ 0.02). Each 1-point increase in BMI at each time point was associated with a BMD increase of 0.0035 g/cm2 (P ҃ 0.0005). The TB model indicates that there was no significant annual change in BMD (P ҃ 0.77). BMI and percentage lean mass were positively associated with BMD, and not being physically active as an adolescent had a negative effect on initial values. The model indicates that average initial TB BMD (when BMI and percentage lean mass are zero) was 1.0212 g/cm2 in women who were physically active as adolescents. Women who did not exercise as adolescents had a difference of Ҁ0.0319 g/cm2 in average initial BMI (P ҃ 0.008). Each 1-point increase in BMI at each time point was associated with a BMD increase of 0.0048 g/cm2 (P 쏝 0.0001), and each 1% increase in lean mass at each time point was associated with a BMD increase of 0.00074 g/cm2 (P ҃ 0.02). DISCUSSION

The primary purpose of this study was to evaluate the relations among CDR, subclinical ovulatory disturbances, and physical activity and their effect on change in BMD over a 2-y period in healthy, premenopausal women. In this large community cohort of young women, we did not confirm prior reports of either a TABLE 4 Baseline characteristics of women with subclinical ovulatory disturbances categorized as having 쏝3 or 욷3 cycles with short luteal phase lengths or anovulation1

Physical and lifestyle characteristics Age (y) Gynecologic age (y) Age at menarche (y) BMI (kg/m2) Lean mass (%) Baecke total score (/15) TFEQ-R score (/21) STAI score (/80) Menstrual cycle length (d) Hormone values Estradiol (pg/mL) Free testosterone (pg/mL) Mean luteal progesterone (pmol/L) Mean midluteal progesterone (pmol/L)

Had 쏝 3 cycles with ovulatory disturbances (n ҃ 209)

Had 욷 3 cycles with ovulatory disturbances (n ҃ 16)

31.52 18.5 12.6 앐 1.03 24.7 64.6 앐 9.6 7.9 앐 1.2 8.4 앐 4.2 35.7 앐 8.9 27.5

33 20 12.7 앐 1.5 23.5 66.2 앐 9.5 8.3 앐 1.3 7.6 앐 4.0 37.2 앐 9.0 28.0

30.2 1.1 320.3 412.1

30.7 1.2 353.6 468.3

1 TFEQ-R, Three Factor Eating Questionnaire–Restraint subscale; STAI, State Trait Anxiety Inventory. Characteristics were compared between the 2 groups by t test for normally distributed variables and by Wilcoxon rank-sum test for skewed variables; level of significance was P 울 0.05. 2 Median (all such values). 3 x៮ 앐 SD (all such values).

1797

DIETARY RESTRAINT AND BONE MASS IN YOUNG WOMEN TABLE 5 Baseline characteristics of women categorized by tertile of baseline dietary restraint scores1

Physical or lifestyle characteristics Age (y) Age at menarche (y) BMI (kg/m2) Lean mass (%) Baecke total score (/15) Sport score (/5) TFEQ-R score (/21) STAI score (/80) Alcohol intake (drinks/mo) Calcium intake (mg/d) Current smoker (%) Menstrual characteristics Mean cycle length (d) Mean luteal phase length (d) Had 욷3 cycles with ovulatory disturbances (%) Hormone values Estradiol (pg/mL) Free testosterone (pg/mL) Mean luteal progesterone (pmol/L)

Lowest tertile of restraint (score 쏝 5) (n ҃ 76)

Middle tertile of restraint (n ҃ 70)

Highest tertile of restraint (score 쏜 9.4) (n ҃ 79)

342 12.9 앐 1.53 21.24 67.9 앐 9.66

33 12.8 앐 1.5 23.55 66.4 앐 1.5

34 12.4 앐 1.4 25.2 64.1 앐 8.3

8.0 앐 1.5 2.7 앐 0.86 3.5 앐 1.26 35.2 앐 8.7 5 827.2 6.6

8.4 앐 1.3 2.9 앐 0.7 7.1 앐 1.37 38.0 앐 9.4 6.3 844.5 5.7

8.4 앐 1.2 3.1 앐 0.8 12.1 앐 2.4 38.0 앐 8.7 7 806.9 6.3

28.0 13.8 5.3

28.0 13.9 10

27.6 13.7 6.3

30.2 1.3 345.9

29.2 1.1 342.0

32.1 1.2 364.2

1 TFEQ-R, Three Factor Eating Questionnaire–Restraint subscale; STAI, State Trait Anxiety Inventory. Characteristics of the lowest and middle tertiles are each compared with those of the highest tertile by t test for normally distributed variables, Wilcoxon rank-sum test for skewed variables, and chi-square for categorical variables. 2 Median (all such values). 3 x៮ 앐 SD (all such values). 4 Significantly different from highest tertile, P 울 0.05 (Wilcoxon rank-sum test). 5 Significantly different from highest tertile, P 울 0.05 (Wilcoxon rank-sum test). 6 Significantly different from highest tertile, P 울 0.05 (t test). 7 Significantly different from highest tertile, P 울 0.05 (t test).

negative relation between CDR and BMD or an association of CDR with subclinical ovulatory disturbances. Although unrelated to CDR or physical activity, having subclinical ovulatory disturbances negatively affected the rate of change in LSP BMD over 2 y, and it resulted in slight bone loss, whereas women with 쏝3 cycles with disturbances had a slight increase in bone density (Figure 2). The inconsistency of our findings with those of previous studies examining the effects of CDR on ovulatory function and BMD may be the result of important differences in the populations studied—namely, the range in BMI values and the mean age of the participants in our cohort—and of our method of determining ovulatory disturbances. First, in contrast to prior studies that evaluated CDR in normal-weight persons (BMI 18 –25) (7–9, 14), our sample was composed of women with a broad range of BMI values (16.2–39.3), and it included obese, normalweight, and low-weight persons. In our study, higher CDR scores were associated with higher BMI, lower mean percentage lean mass, and higher Baecke sport scores, which suggests that these scores were indicative of overweight women who were attempting to diet and exercise to lose weight. We hypothesized that normal-weight or underweight persons who have high CDR scores may be a unique subgroup who are “successful” dieters and are consistently restricting caloric intake. This caloric restriction, if severe enough, may be contributing to the previously

documented ovulatory disturbances. Our participants may represent “typical” restrained eaters who are “unsuccessful” dieters (3, 44). Our results suggest that, in this population, CDR does not induce alterations in the reproductive axis. We examined the interaction between CDR and BMI to determine whether the effect of CDR on BMD was different in women with lower weight, but we did not confirm this hypothesis. Further investigation is required to clarify the relation found in prior studies between CDR and ovulatory function in normal-weight and underweight women. A second important difference between the population of the present study and the populations of previous studies was the age of our participants: the subjects in the present study had a greater mean age than did subjects in prior studies that found significant associations between CDR and BMD (14) or between CDR and ovulatory disturbances (7, 8). Our sample represents women who have reached gynecologic maturity, but who are not yet in the perimenopausal period. Gynecologic maturity has been found to be associated with reduced variability in cycle length (45, 46). In addition, durability of the reproductive axis in response to a moderate endurance training program was shown in a group of gynecologically mature, eumenorrheic women (gynecologic age: 17.8 앐 0.9 y) (47). This would suggest that the women in our study may have more robust reproductive systems and therefore

1798

WAUGH ET AL

TABLE 6 Summary of results of general linear mixed-model analysis to evaluate the effect of cognitive dietary restraint (CDR) and subclinical ovulatory disturbances on 2-y changes in bone mineral density (BMD) after adjustment for BMI and physical activity (Baecke total score)1 Parameter estimate Spinal BMD (L1–L4) Intercept (initial status) Time (annual rate of change) CDR Lowest vs highest tertile Middle vs highest tertile Subclinical ovulatory disturbances BMI (kg/m2) Baecke total score (/15) CDR ҂ time Lowest vs highest tertile Middle vs highest tertile Subclinical ovulatory disturbances ҂ time Femoral neck BMD Intercept (initial status) Time (annual rate of change) CDR Lowest vs highest tertile Middle vs highest tertile Subclinical ovulatory disturbances BMI (kg/m2) Baecke total score (/15) CDR ҂ time Lowest vs highest tertile Middle vs highest tertile Subclinical ovulatory disturbances ҂ time Total-body BMD Intercept (initial status) Time (annual rate of change) CDR Lowest vs highest tertile Middle vs highest tertile Subclinical ovulatory disturbances BMI (kg/m2) Baecke total score (/15) CDR ҂ time Lowest vs highest tertile Middle vs highest tertile Subclinical ovulatory disturbances ҂ time

P

0.9899 0.0123

쏝0.0001 쏝0.0001

0.0047 0.0068 Ҁ0.0292 0.0053 0.0099

0.36 0.15 0.39 쏝0.0001 0.13

Ҁ0.0046 Ҁ0.0042 Ҁ0.0103

0.15 0.22 0.03

0.7952 0.0048

쏝0.0001 0.10

Ҁ0.0058 Ҁ0.0035 Ҁ0.0037 0.0040 0.0146

0.36 0.54 0.91 0.0001 0.03

Ҁ0.0033 Ҁ0.0025 0.0021

0.40 0.56 0.74

1.0324 Ҁ0.0001

쏝0.0001 0.96

Ҁ0.0016 0.0017 Ҁ0.0105 0.0034 0.0077

0.58 0.51 0.56 쏝0.0001 0.03

0.0001 Ҁ0.0011 0.0002

0.95 0.57 0.95

1

Interactions between CDR and each of BMI, ovulatory disturbances, and Baecke score were analyzed; all were not statistically significant (P 욷 0.05); only main effects are included in this table for simplicity. CDR was categorized by tertile on the basis of scores on the Three Factor Eating Questionnaire–Restraint subscale. Subclinical ovulatory disturbances were defined as having 욷 3 cycles with anovulation or short luteal phase lengths (쏝 10 d). Time was set at 0, 1, and 2, which represented baseline and 12- and 24-mo BMD assessments, respectively.

may be less likely to experience ovulatory disturbances in response to minor physiologic stressors, such as CDR or moderate exercise, than would women in the earlier studies who were of a younger age. Our method of determining ovulatory function differed from that used in prior studies documenting a relation between ovulatory disturbances and CDR (7, 8) and ovulatory disturbances and BMD (10), and that difference may have contributed to our

different results. Those other studies used the basal body temperature method, which has been shown to be an unreliable method of determining ovulation and which may have led to misclassification (48 –50). Indeed, those investigators reported a much higher prevalence of ovulatory disturbances (67– 80% of participants had disturbances) (7, 10) than we and others (12, 51) who used progesterone measurements have reported. Both salivary and urinary progesterone are highly correlated with serum progesterone, and measuring them is a reliable method of assessing ovarian function (28, 30, 52). We also used urinary ovulation detection kits as our primary method of determining ovulation; these kits have been shown to have very good psychometric properties (33). Despite the different method of ovulatory assessment, and although we did not find a relation between CDR and ovulatory disturbances, we confirmed the results of Prior et al (10, 53) that ovulatory disturbances are predictive of spinal bone loss. No significant change in BMD at FN or TB was observed over the 2-y study, and that lack of such an observation was likely due to slower bone turnover rates at these sites. Ovulatory disturbances also may have a negative effect on these sites if observed over a longer period. Debate continues over the underlying cause of reduced bone mass due to ovulatory disturbances. Prior et al (10) proposed that the observed bone loss was due to reduced luteal progesterone production in the presence of normal concentrations of estrogen, but this possibility has not been supported by other investigators. Sowers et al (51) found lower concentrations of both luteal phase estrogen and progesterone urinary metabolites in women with BMD in the bottom 10th percentile than in women with BMD in the 50 –75th percentile. DeSouza et al (11) observed lower luteal phase progesterone in women with short luteal phases (but found no difference in BMD) than in women with normal luteal phases. DeSouza et al concluded that, if estradiol status is maintained, BMD is not affected by a disturbance of progesterone production associated with luteal phase abnormalities. In the present study, although decreased luteal progesterone was associated with ovulatory disturbances, we found no relation between progesterone and bone loss. The extended models of all key predictors of BMD indicated that BMI was positively associated with BMD at each skeletal site. The strong association between body weight and bone density has been documented in many prior studies of premenopausal women (54). Physical activity also was consistently predictive of higher BMD. Because activity scores remained constant throughout the study, the relation between activity and change in BMD could not be evaluated to determine the extent of the benefit of activity or the amount of activity required to increase BMD in premenopausal women. Although current activity was the best predictor at the FN, being physically active as an adolescent had a stronger effect at the LSP and TB. This finding provides support for the findings of others of the existence of a critical window of opportunity during puberty, when the skeleton is particularly responsive to mechanical loading that results in the optimization of peak bone mass (55). Additional significant predictors included a family history of osteoporosis (negative effect) and the subject’s alcohol consumption (positive effect), which were previously reported (19, 56 –58). Finally, lean mass was found to have a positive effect on BMD at TB after control for BMI and physical activity. Although questions remain about the relative benefit of lean mass and fat mass for bone density, recent studies have shown that lean mass has a greater effect than fat

1799

DIETARY RESTRAINT AND BONE MASS IN YOUNG WOMEN TABLE 7 Key predictors of 2-y change in bone mineral density (BMD) at the lumbar spine (L1–L4), femoral neck, and total body determined by general linear mixed-model analysis1

Spinal BMD (L1–L4) Intercept (initial status) Time (annual rate of change) BMI (kg/m2) Family history of osteoporosis (욷3 relatives) Not physically active as an adolescent Alcohol consumption (drinks/mo) Subclinical ovulatory disturbances Subclinical ovulatory disturbances ҂ time Femoral neck BMD Intercept (initial status) Time (annual rate of change) BMI (kg/m2) Family history of osteoporosis (욷3 relatives) Baecke total score (/15) Total-body BMD Intercept (initial status) Time (annual rate of change) BMI (kg/m2) Lean mass (%) Not physically active as an adolescent

Parameter estimate

95% CI

P

1.0779 0.0096 0.0052 Ҁ0.0755 Ҁ0.0524 0.0016 Ҁ0.0351 Ҁ0.0109

1.0286, 1.1272 0.0071, 0.0121 0.0034, 0.0070 Ҁ0.1350, Ҁ0.0161 Ҁ0.0962, Ҁ0.0085 0.0003, 0.0030 Ҁ0.0995, 0.0292 Ҁ0.0200, Ҁ0.0018

쏝0.0001 쏝0.0001 쏝0.0001 0.01 0.02 0.02 0.28 0.02

0.7972 0.0028 0.0037 Ҁ0.0732 0.0157

0.6735, 0.9209 Ҁ0.0004, 0.0060 0.0016, 0.0057 Ҁ0.1353, Ҁ0.0112 0.0028, 0.0287

쏝0.0001 0.08 0.0005 0.02 0.02

1.0212 Ҁ0.0002 0.0048 0.0007 Ҁ0.0319

0.9493, 1.0932 Ҁ0.0016, 0.0012 0.0033, 0.0063 0.0001, 0.0013 Ҁ0.0555, Ҁ0.0082

쏝0.0001 0.77 쏝0.0001 0.02 0.008

1

Final reduced model was determined through backward stepwise selection from complex model that included the following variables: time (varying)— BMI, percentage lean mass, cognitive dietary restraint (Three Factor Eating Questionnaire–Restraint scale score), smoking status (nonsmoker or current smoker), calcium intake (does not meet or meets recommended daily amount); time (invariant)—subclinical ovulatory disturbances (쏝3 or 욷3 cycles with disturbances), hormone values (ie, estradiol, testosterone, and mean luteal progesterone), age at menarche, Baecke total score, physically active as an adolescent (no or yes), alcohol consumption (no. of drinks/mo), age, and family history of osteoporosis (쏝3 or 욷3 relatives). Time was set at 0, 1, and 2, which represented baseline and 12- and 24-mo BMD assessments, respectively. Subclinical ovulatory disturbances were defined as having 욷3 cycles with anovulation or short luteal phase lengths (쏝10 d).

Lumbar spine bone mineral density (g/cm2)

mass in younger women (59 – 61). This finding suggests that the dynamic load generated by muscle contraction stimulates a greater osteogenic response than that generated by the static load associated with body weight (62). This relation may not hold true

1.4 1.35 1.3 1.25