Clinical Chemistry 47:10 1782–1787 (2001)

Enzymes and Protein Markers

Comparison of HPLC and Small Column (CDTect) Methods for Disialotransferrin Ursula Turpeinen,1* Taina Methuen,2 Henrik Alfthan,1 Kalevi Laitinen,3 Mikko Salaspuro,2 and Ulf-Håkan Stenman1 Background: Current methods for determination of carbohydrate-deficient transferrin (CDT) are based on separation of the CDT fraction by ion-exchange chromatography on minicolumns and quantification by immunoassay. Alternatively, the transferrin isoforms can be separated by HPLC anion-exchange chromatography and quantified by absorbance. This method has been reported to improve the validity of CDT as a marker of chronic alcohol abuse. Methods: HPLC on either MonoQ or ResourceQ anionexchange columns was used to separate and quantify isoforms of transferrin with detection at 460 nm. The result was expressed as the percentage of the disialo form (pI 5.7) of total transferrin (DST). The commercial CDTectTM assay was used as a comparison method. Serum samples from nondrinkers (n ⴝ 57), moderate drinkers (n ⴝ 77), and heavy drinkers (n ⴝ 139) were analyzed. Results: In ROC analysis for differentiation between moderate and heavy drinkers, the area under the curve (AUC) for the HPLC method was 0.87 (95% confidence interval, 0.81– 0.93), whereas that for CDTect was 0.72 (95% confidence interval, 0.64 – 0.80). At 90% specificity, the sensitivity of DST was 63% (95% confidence interval, 53–73%) compared with 33% (22– 44%) for CDT. The reference interval of the HPLC method was 0.68 –1.7%. Conclusions: The HPLC anion-exchange method for quantification of CDT provides substantially better separation between moderate and heavy drinkers than the CDTect method. © 2001 American Association for Clinical Chemistry

Departments of 1 Clinical Chemistry and 3 Internal Medicine, Helsinki University Central Hospital, 00290 Helsinki, Finland. 2 Research Unit of Alcohol Diseases, University of Helsinki, 00290 Helsinki, Finland. *Address correspondence to this author at: Helsinki University Central Hospital, Department of Clinical Chemistry, Haartmaninkatu 2, 00290 Helsinki, Finland. Fax 358-9-471-4945; e-mail [email protected]. Received June 1, 2001; accepted July 23, 2001.

Carbohydrate-deficient transferrin (CDT)4 comprises transferrin isoforms lacking one or both of their carbohydrate chains (1 ). The proportion of serum CDT increases with increasing alcohol consumption and is widely used as a specific marker of excessive alcohol use (2, 3 ). Several analytical methods for CDT have been developed. Originally, transferrin isoforms were analyzed by isoelectric focusing and electrophoresis. Isoelectric focusing, because of its high selectivity, is regarded as a reference method (4 ). Rapid separation of a CDT fraction comprising isoforms with two, one, or no sialic acids and part of that with three sialic acids by anion-exchange chromatography on minicolumns followed by RIA (CDTectTM) has become widely used (5 ). The various transferrin isoforms can also be determined by anion-exchange HPLC followed by photometric detection of transferrin at 460 nm (6 ). By this method the ratios of the various serum CDT isoforms in relation to total transferrin can be determined, and it has been reported to provide better separation between alcohol abusers and controls (7 ). However, HPLC methods often are considered too expensive and complicated for routine use, but no exact comparisons have been reported. The aim of our study was to evaluate this HPLC method in comparison with the CDTect method in detecting heavy alcohol consumption and to estimate the costeffectiveness of the two methods.

Materials and Methods reagents All chemicals were of analytical reagent grade. CDTect reagent sets were purchased from Bio-Rad (Axis Biochemicals ASA).

apparatus For HPLC analyses, we used a chromatographic system consisting of a Varian 9010 HPLC pump (Varian Associ-

4 Nonstandard abbreviations: CDT, carbohydrate-deficient transferrin; DST, disialotransferrin; and AUC, area under the curve.

1782

1783

Clinical Chemistry 47, No. 10, 2001

ates), a Waters 717plus Autosampler (Millipore Corp.), and a Spectroflow 783 detector (ABI Analytical Kratos Division) combined with an C-R4A integrator (Shimadzu Corporation). We used a 50 ⫻ 5 mm MonoQ HR 5/5 anion-exchange column (Pharmacia Biotech). Alternatively, a Resource Q (1 mL) column was used. Disposable Millex-HV filter units (average pore size, 0.45 ␮m) were from Millipore S.A.

sample treatment for hplc One mL of fresh serum or serum that had been stored at ⫺20 °C was saturated with iron by the addition of 25 ␮L of 0.5 mol/L NaHCO3 and 18 ␮L of 10 mmol/L FeCl3. After the sample was mixed and incubated for 1 h at room temperature, the lipoproteins were precipitated by the addition of 10 ␮L of 100 g/L dextran sulfate (Sigma Chemical Co.) and 50 ␮L of 1 mol/L CaCl2. This mixture was kept at 4 °C for 30 – 60 min and then centrifuged at 10 000g for 10 min. The supernatant was diluted fivefold with water, filtered through a 0.45 ␮m filter unit.

hplc conditions

The pretreated sample (150 ␮L) was injected into the column. The transferrin isoforms were separated by a salt gradient. Buffer A was 20 mmol/L Bis-Tris buffer, pH 6.2; buffer B was buffer A plus 0.35 mol/L NaCl, pH 6.2. Solution C, consisting of 0.5 mol/L NaCl, was used for column regeneration. Before use, all solutions were filtered through a 0.45 ␮m filter. Separation was achieved by gradient elution (Table 1) at room temperature with a flow rate of 1 mL/min. The detection wavelength was 460 nm. Integration was performed in the horizontal baseline mode. The area for each individual isoform was reported as the percentage of the total area of transferrin. The chromatographic system is usually loaded with 70 – 80 samples and left unattended to run during the weekend.

participants Serum samples were obtained from Finnish men with well-documented drinking habits (8, 9 ). The participants were examined by an experienced physician and interviewed by trained staff. None of the participants reported or had clinical signs of liver damage. On the basis of data

Table 1. Solvent gradient used for HPLC separation of isotransferrins. Time, min

0a 1 2 24 25 28 29

%A

%B

%C

100 100 90 60 0 0 100

0 0 10 40 0 0 0

0 0 0 0 100 100 0

a The column was equilibrated for 10 min between each run with solutions containing the stated percentages at zero time.

elicited from a validated structured interview (10 ) and daily anonymous diaries (8 ), the participants were classified according to their alcohol intake during the last 30 days into the following groups: 57 nondrinkers; 77 moderate drinkers with an average alcohol consumption of ⬍210 g/week; and 139 heavy drinkers with an average alcohol consumption of ⬎210 g/week. Analyses were performed by operators who did not know the drinking habits. For determination of the reference interval, serum samples were obtained from 165 healthy blood donors (85 women and 80 men).

statistical analysis The reference interval was determined on the basis of the 2.5 and 97.5 percentiles. The validity of the test for separation of heavy drinkers from moderate drinkers and nondrinkers was estimated by ROC analysis (11 ).

costs We estimated the costs (in US$) of the tests based on 1000, 2000, and 3000 samples/year (Table 2). The instrument costs for the HPLC method were calculated on the basis of a leasing fee of 30% of the price of the automated HPLC instrument (US$38 000). The costs for HPLC columns was US$360 for the ResourceQ and US$780 for the MonoQ column. The cost for the CDTect reagent set (US$400) was calculated on the basis of optimal usage, i.e., 48 minicolumns were used for samples and 2 for controls. Labor costs were estimated according to our local expenses. Reagent cost per assay was calculated on the basis of single assays, with six calibrators (for CDTect) and two controls in each HPLC run.

Results hplc separation of transferrin isoforms Typical chromatograms of serum samples are shown in Fig. 1. Quantification relies on the selective absorbance of the iron-transferrin complex at 460 nm. Asialo-, disialo-, trisialo-, tetrasialo-, and pentasialotransferrin were separated as distinct peaks. Disialotransferrin (DST; pI 5.7) eluted as a separate peak at 11.7 min. The variation in retention time (CV) over 20 runs was 0.5%. The majority of the serum samples gave chromatograms similar to that shown in Fig. 1B. Virtually identical separation was

Table 2. Costs per sample for determination of CDT by HPLC and CDTect. HPLC Samples per year

1000

2000

3000

Reagents, US$ Labor, US$ Equipment, US$ Column, US$ Overhead, US$ Total, US$

0.57 4.3 11.3 0.4 3.1 19.7

0.57 4.3 5.7 0.4 3.1 14.1

0.57 4.3 3.8 0.4 3.1 12.2

CDTect 1000–3000

8.2 8.8 0.3 3.1 20.4

1784

Turpeinen et al.: HPLC Assay for DST

Fig. 1. HPLC profiles of iron-saturated transferrin isoforms obtained by ionexchange HPLC. Detection was at 460 nm. (A), sample with a high proportion of DST (eluting at 11.7 min) on a MonoQ column. (B), sample with normal proportion of DST on a MonoQ column. (C), same sample as in B, on a Resource Q column.

obtained with the ResourceQ column (Fig. 1C). For routine analysis we used the ResourceQ column because it is more economic and the separation of transferrin isoforms is similar to that obtained on the MonoQ. The interassay CVs for two samples (ResourceQ) were 6.5% and 4.3% (1.2% and 3.7% DST, respectively; 15 replicates each) and was 12% for CDTect (23.3 units/L; 15 replicates). The determinations were conducted on serum pools stored in frozen aliquots and thus reflect the entire process. The selection of HPLC instrumentation is not critical, but the detector needs to be sensitive. For example, the Spectroflow 783 (ABI Analytical) and the HP 1100 (Agilent Technologies) detectors provided reliable results, but some other detectors were less suitable. The gradient used in this study differs slightly from those used in other studies (6, 12 ), and it is likely that some adjustment must be made when switching to other HPLC systems.

moderate and heavy drinkers (Fig. 2). The correlation was: HPLC ⫽ 0.09CDTect ⫺ 0.26%; r ⫽ 0.72; Sy兩x ⫽ 1.3%. With the cutoff of 1.8% for DST and 20 units/L for CDT and by use of an alcohol consumption of 210 g/week as the cutoff for heavy drinking, 52% of the heavy drinkers had increased results by HPLC compared with 49% by the CDTect assay (Fig. 3). However, the specificity of the HPLC method (98%) was substantially higher than that for CDTect (84%). Among the moderate drinkers, 4% had increased results by CDTect and 2.6% by DST. Both methods showed a positive correlation between alcohol

reference interval

We analyzed serum samples (n ⫽ 165) from apparently healthy blood donors. After we eliminated eight obvious outliers with values from 2.2–3.1%, the proportion of CDT was 0.55–1.9%, the mean (⫾ SD) was 1.13% ⫾ 0.27%, and the median was 1.10%. The upper reference limit, based on the 97.5 percentile, was 1.7%, and the lower limit, based on the 2.5 percentile, was 0.68% for the whole group. The upper reference limit was 1.6% for women (n ⫽ 85) and 1.8% for men (n ⫽ 80). In women, the mean DST was 1.10% and the median was 1.10%, whereas the mean was 1.18% and the median was 1.10% in men (P ⫽ 0.066 between the groups).

correlation of methods The correlation of the HPLC (y) and the CDTect assay (x) was determined with 189 samples from nondrinkers and

Fig. 2. Correlation of CDT results obtained by HPLC (y axis) and CDTect (x axis) in 189 participants, including nondrinkers and moderate and heavy drinkers. The dashed lines indicate the upper reference limits. Corresponding regression equation is: y ⫽ 0.09x ⫺ 0.26; r ⫽ 0.72; Sy兩x ⫽ 1.3%.

1785

Clinical Chemistry 47, No. 10, 2001

Fig. 4. ROC plots showing the ability of the HPLC (thick solid line) and CDTect (thin solid line) methods to distinguish between 74 heavy drinkers (⬎210 g/week) and 78 moderate drinkers (⬍210 g/week). The dashed line indicates the line of identity (AUC ⫽ 0.50).

assays/year, were approximately one-half those of the CDTect assay (Table 2). The costs for the HPLC assay would be similar to those of the CDTect assay if we analyzed 1000 samples yearly. Because only 40 – 60% of the capacity of the HPLC instrumentation is used for the DST assay, the actual costs are ⬃10% lower. Because of reproducibility problems, we have in practice run duplicates in the CDTect assay; thus the cost of this assay has been approximately double that indicated in Table 2.

Discussion Fig. 3. Correlation of alcohol consumption (x axis) and CDT (y axis) in 214 sera from nondrinkers and moderate and heavy drinkers by HPLC (A) and CDTect (B). The dashed lines indicate the upper reference limits. Corresponding regression equations are: (A), y ⫽ 0.0003x ⫹ 1.12 (r ⫽ 0.54); (B), y ⫽ 0.002x ⫹ 16.1 (r ⫽ 0.40).

consumption and CDT values, but the correlation (r) was stronger with the DST method than with the CDTect, 0.54 vs 0.40 (Fig. 3). At 90% specificity, the sensitivity for DST was 63% (95% confidence interval, 22– 44%) compared with 33% (95% confidence interval, 53–73%) for CDT. ROC analysis showed that the area under curve (AUC) for differentiation between moderate drinkers and heavy drinkers (Fig. 4), was significantly larger (P ⫽ 0.0002) for DST (0.87; 95% confidence interval, 0.81– 0.93) than for CDT (0.72; 95% confidence interval, 0.64 – 0.80). We also determined the ratio of DST to trisialotransferrin by HPLC. The AUC for this ratio was lower (0.81) than that for DST. Estimated costs for the HPLC assay, assuming 3000

CDT has become the most important laboratory test for the detection of prolonged heavy alcohol consumption (7 ). In routine methods, the transferrin isoforms are separated into two fractions by anion-exchange chromatography on microcolumns and the CDT fraction quantified by RIA (CDTect) or immunoturbidimetry (%CDTTIA; Axis). The fraction referred to as CDT is not precisely defined and contains, in addition to asialo-, monosialo-, and disialotransferrin, variable amounts of trisialotransferrin, whereas part of DST is retained on the column (13 ). Isoelectric focusing provides good separation of the isotransferrins, but it requires immunostaining of the separated transferrins before quantification. It is therefore considered too complicated and time-consuming for routine use. The major transferrin isoforms can also be separated by anion-exchange chromatography. We evaluated a slight modification of the HPLC method originally developed by Jeppsson et al. (6 ), based on anionexchange chromatography after iron saturation of transferrin in the sample. Detection of the iron-transferrin complex at 460 nm is highly specific, and good separation of the transferrin isoforms is obtained. The chromatogram

1786

Turpeinen et al.: HPLC Assay for DST

is a visible document of successful isotransferrin determination. Common genetic variants of transferrin, which can cause erroneous results in interpretation of the CDTect method (13 ), do not pose a problem. Rare variants that cause problems in minicolumn chromatography have been reported to occur in Caucasians (7, 14 ). No such variants were detected in the present study, apparently because these variants are rare in the Finnish population. ROC analysis demonstrated that the HPLC method had a substantially better ability to distinguish between moderate and heavy drinkers than the CDTect. The probable explanation for this is the use of the proportion of DST in relation to total transferrin. This approach eliminates the confounding effect of variations in transferrin concentrations. This problem has been eliminated in a recently introduced method in which both of the fractions are measured by immunoassay (%CDT-TIA; Axis). This method has also been reported to have better diagnostic accuracy than the CDTect assay in some studies, but not in others (7, 15 ). The most likely explanation for the superior clinical performance of the DST assay is the good chromatographic separation. Nearly baseline separation between the important isoforms was achieved by HPLC. This is analogous to the improvement in performance when HPLC methods replaced batchwise methods for separation of glycohemoglobins (16 ). We also studied whether the ratio of disialo- to trisialotransferrin improved the discriminatory power of the HPLC method, but this was not the case, as evidenced by a smaller AUC. This is in agreement with previous results that indicated that the measurement of trisialotransferrin provides no diagnostic advantage (17, 18 ). The reference limits were determined based on samples from apparently healthy blood donors and reflect the general population in Finland. Some of these donors probably consume more alcohol than the moderate drinkers included in the study. We therefore eliminated outliers before calculating the reference interval. With these reference values, only 2.6% of the moderate drinkers had increased DST values, and the highest value was 3.1%. The upper reference limit for CDT established in this study, 1.7%, is higher than the previously reported value of 1% (12, 19 ). This difference can be explained by differences in the integration mode of the chromatographic profile. The horizontal baseline integration mode used in our study gives higher results for minor components than the valley-to-valley mode used in a previous study (6 ). We decided to use this method because it provides more realistic values and is less sensitive to variations in peak separation. Interestingly, we observed lower values in women than in men, although higher mean values (12 ) and spuriously increased CDT values are often observed in women (20 ). In spite of several hypotheses (7 ), the reason for this remains unclear. Because of the stigmatizing effect of an increased marker for alcohol abuse, these falsely increased results are of great concern. HPLC methods usually are considered more expensive

than immunoassays; in clinical chemistry, therefore, they are used mainly for determination of drugs and glycohemoglobin on dedicated instruments. Because higher cost is a common argument against the use of HPLC methods, we estimated the costs of the two methods. This showed that in our laboratory the HPLC method was actually much more cost-effective. This is because we analyze a fairly large number of samples. With 1000 samples/year, the costs of the HPLC and the CDTect assays would be similar. However, this calculation does not include expenses for establishing the assay. The costs of setting up an in-house HPLC method may be substantial unless experienced personnel are available. In conclusion, our results show that the ion-exchange HPLC method is more valid than the CDTect method for detection of excessive alcohol use. These results are in line with those of two previous studies (12, 19 ). In agreement with one of these studies (12 ), the diagnostic value of the CDTect method was so low that its clinical utility must be considered questionable and should probably be replaced by the newer %CDT-TIA assay from the same manufacturer or by HPLC. It remains to be determined which of these assays provides the most accurate clinical information.

This study was supported financially by the Helsinki University Central Hospital Research Funds and the Yrjo¨ Jahnsson Foundation. This report includes data and experiences obtained during the participation of M. Salaspuro and T. Methuen (Clinical Center, Finland) in the WHO/ ISBRA Project on Biomarkers of Alcohol Use, a project supported by the World Health Organization and the International Society for Biomedical Research on Alcoholism and funded by the participating field research centers. We are grateful to Anders Helander (WHO/ISBRA Project, Analytical Center, Sweden) for the CDT analyses. CDTect reagent sets were kindly supplied by Kabi Pharmacia (8 ), Pharmacia Diagnostics (Uppsala, Sweden), and Axis Biochemicals (Oslo, Norway).

References 1. Landberg E, Påhlsson P, Lundblad A, Arnetorp A, Jeppsson J-O. Carbohydrate composition of serum transferrin isoforms from patients with high alcohol consumption. Biochem Biophys Res Commun 1995;210:267–74. 2. Stibler H, Allgulander C, Borg S, Kjellin KG. Abnormal microheterogeneity of transferrin in serum and cerebrospinal fluid in alcoholism. Acta Med Scand 1978;204:49 –56. 3. Stibler H, Borg S. Evidence of a reduced sialic acid content in serum transferrin in male alcoholics. Alcohol Clin Exp Res 1981; 5:545–9. 4. Bean P, Husa A, Liegmann K, Sundrehagen E. Semi-automated carbohydrate-deficient transferrin in primary biliary cirrhosis: a pilot study. Alcohol Alcohol 1998;33:657– 60. 5. Stibler H, Borg S, Joustra M. Micro anion exchange chromatography of carbohydrate-deficient transferrin in serum in relation to alcohol consumption. Alcohol Clin Exp Res 1986;10:535– 44.

Clinical Chemistry 47, No. 10, 2001

6. Jeppsson J-O, Kristensson H, Fimiani C. Carbohydrate-deficient transferrin quantified by HPLC to determine heavy consumption of alcohol. Clin Chem 1993;39:2115–20. 7. Arndt T. Carbohydrate-deficient transferrin as a marker of chronic alcohol abuse: a critical review of preanalysis, analysis, and interpretation. Clin Chem 2001;47:13–27. 8. Elomaa V-V, Lo¨yttyniemi E, Ka¨rkka¨inen P, Salaspuro M, Laitinen K. Biological markers of alcohol consumption and effect of calcitonin in nonalcoholic men: a prospective, double-blind study. Alcohol Clin Exp Res 1996;20:830 –5. 9. Helander A, Tabakoff B, WHO/ISBRA Study Centres. Biochemical markers of alcohol use and abuse: experiences from the pilot study of the WHO/ISBRA collaborative project on state and trait markers of alcohol. Alcohol Alcohol 1997;32:133– 44. 10. Menninger JA, Baro´n AE, Conigrave KM, Whitfield JB, Saunders JB, Helander A, et al. Platelet adenylyl cyclase activity as a trait marker of alcohol dependence. Alcohol Clin Exp Res 2000;24: 810 –21. 11. Metz CE. Basic principles of ROC analysis. Semin Nucl Med 1978;8:283–98. 12. Werle E, Seitz GE, Kohl B, Fiehn W, Seitz HK. High-performance liquid chromatography improves diagnostic efficiency of carbohydrate-deficient transferrin. Alcohol Alcohol 1997;32:71–7. 13. Arndt T, Hackler R, Kleine TO, Gressner AM. Validation by isoelectric focusing of the anion-exchange isotransferrin fractionation

14. 15.

16.

17.

18.

19.

20.

1787

step involved in determination of carbohydrate-deficient transferrin by the CDTect assay. Clin Chem 1998;44:27–34. Kamboh MI, Ferrell RE. Human transferrin polymorphism. Hum Hered 1987;37:65– 81. Helander A. Absolute or relative measurement of carbohydratedeficient transferrin in serum? Experiences with three immunological assays. Clin Chem 1999;45:131–5. Turpeinen U, Stenman U-H. Analysis of HbA1C and some Hb variants by HPLC. In: Aboul-Enein HY, ed. Analytical and preparative separation methods of biomacromolecules. New York: Marcel Dekker, 1999:1–11. Renner F, Stratmann K, Kanitz RD, Wetterling T. Determination of carbohydrate-deficient transferrin and total transferrin by HPLC: diagnostic evaluation. Clin Lab 1997;43:955– 64. Dibbelt L. Does trisialo-transferrin provide valuable information for the laboratory diagnosis of chronically increased alcohol consumption by determination of carbohydrate-deficient transferrin? Clin Chem 2000;46:1203–5. Simonsson P, Lindberg S, Alling C. Carbohydrate-deficient transferrin measured by high-performance liquid chromatography and CDTectTM immunoassay. Alcohol Alcohol 1996;31:397– 402. Sarkola T, Eriksson CJP, Niemela¨ O, Sillanaukee P, Halmesma¨ki E. Mean cell volume and gamma-glutamyl transferase are superior to carbohydrate-deficient transferrin and hemoglobin-acetaldehyde adducts in the follow-up of pregnant women with alcohol abuse. Acta Obstet Gynecol Scand 2000;79:359 – 66.