Genetic Strategies for the Analysis of Childhood Behavioral Traits

VOL. 8, NO. 2, 1982 by David L. Pauls and Kenneth K. Kidd Genetic Strategies for the Analysis of Childhood Behavioral Traits Abstract Genetic strat...
Author: Francis Gaines
2 downloads 0 Views 1MB Size
VOL. 8, NO. 2, 1982

by David L. Pauls and Kenneth K. Kidd

Genetic Strategies for the Analysis of Childhood Behavioral Traits

Abstract Genetic strategies appropriate for the analysis of childhood behavioral traits are reviewed. The problem of etiologic heterogeneity is discussed, and suggestions for minimizing its effect are offered. Two traits, stuttering and Tourette syndrome, are given as examples to illustrate methodologies useful for the genetic analysis of childhood disorders. In both examples, it is shown that the disorder clusters in families, that it is vertically transmitted, and that the transmission is sex-modified. Several genetic models can explain the pattern of transmission, although cultural hypotheses also should be considered. Hence, additional analyses need to be attempted to help clarify the modes of transmission and the factors involved. The genetic element of a disease is obvious if an enzyme or structural protein is altered or absent. Certain other biochemical aberrations are presumptive evidence of genetic defects even in the absence of family data. However, for most childhood behavioral disorders, biochemical studies have not yet provided evidence for any genetic defect. Four other types of evidence can suggest that genetic factors are responsible for a disease of unknown etiology: (1) a higher concordance among monozygotic (MZ) twins than among dizygotic (DZ) twins; (2) significant aggregation of the illness within families; (3) genetic linkage of the illness with an identifiable allele at a marker locus; (4) a higher incidence of the trait among biological offspring of affected individuals that among biological offspring of


unaffected individuals, even when those children have been raised in adoptive homes. Twin studies and studies of biological families may yield data suggesting genetic involvement, but such data are incapable of proving the existence of genetic factors for any traits that might have a major environmental component in their etiology. Genes and familial environment are confounded in determining similarities among relatives and cannot be separately quantified. For this reason geneticists today stress the distinction between familial and nonfamilial environmental factors as well as the possibility of confounding familial environmental similarity with genetic similarity among relatives (Cavalli-Sforza and Feldman 1973; Rao, Morton, and Yee 1976; Kidd and Matthysse 1978; Rice, Cloninger, and Reich 1978; Cloninger, Rice, and Reich 1979). The familial environmental factors that come most readily to mind include those that tend to be shared by all members of a family, such as religion, language, and diet. Actually, most environmental factors show some familial aggregation, but the degree of their concentration in families varies considerably. Among the environmental factors showing a familial concentration, some, especially the cultural ones, can actually be transmitted, like genes, from parent to child. In practice, it may be very difficult to distinguish cultural from genetic transmission. One of the strongest forms of evidence for a genetic factor is the Reprint requests should be sent to Dr. D.L. Pauls at Department of Human Genetics, Yale University School of Medicine, New Haven, CT 06510.



demonstration of genetic linkage. Genetic linkage is detectable, at least in theory, if a known genetic marker locus is sufficiently close to a locus segregating for alleles affecting the trait that nonrandom segregation of alleles at the two loci results in an association of phenotypes within a family. Thus, if it is possible to demonstrate genetic linkage between a hypothesized locus for a disorder and a well-known marker locus, that demonstration provides convincing evidence that there is a major gene contributing to the disorder since it is highly unlikely that any other explanation could mimic linkage with a marker locus. Some problems in detecting linkage in human data include fairly small family sizes, inability to control matings, and the small prior probability that two loci are linked. In addition, linkage studies are often not efficient since only a small proportion of families studied will actually show segregation for any particular marker locus. Other problems with linkage studies have to do with the disorder in question. The presence of reduced penetrance and a variable age of onset will reduce the power of the statistical methods that have been developed to test for linkage. Most of these methods assume high penetrance for the trait being studied. The program LIPED version 3 developed by Ott (1976) allows for incomplete penetrance and, in addition, has been modified to allow for sex- and age-dependent penetrance values (Hodge et al. 1979). Adoption or separation studies also can give evidence for a genetic factor. If a trait is genetic, then children should more closely resemble their biological relatives than they do their adoptive rela-

tives. Separation studies also provide the ability to see to what extent environment influences the trait in question. If the children resemble their adoptive parents and siblings to any extent, then that resemblance is, at least in theory, due to environmental similarities. It may then be possible to partial out some of the environmental and genetic influences of the trait being studied. There are some difficulties with adoption studies. Very often the adoptive children are put in homes that closely resemble the homes of the biological parents. Hence, when comparisons are made between child and biologic parent, the similarity may result from selective placement and not genetic contribution (Munsinger 1975). It may be that for serious psychiatric problems this may be less of a problem than for traits like IQ. However, it is a problem that needs to be noted. Another problem with separation studies is that it is often difficult to get adequate data on the biologic parents. Since these studies tend to be retrospective, information necessary for diagnoses needs to be derived from records obtained when the mother was pregnant. Often the information is not adequate. Also, to do a good genetic analysis, information from both parents is necessary. Unfortunately, much of the time information on the biologic father is missing completely. Still another difficulty with adoption studies is that the information obtained will not allow any statements about possible mode of transmission. Unless information is available about other biologic siblings of the adoptive children, nothing can be said about how the trait might be transmitted. Hence,

adoption studies as well as genetic linkage studies can demonstrate that genetic factors must be important, but such studies are difficult to conduct and are not commonly done. Even when done, they can still give ambiguous results. In the absence of a wellunderstood etiology, demonstration of the presence of genetic factors does not usually resolve questions about the nature of the genes involved and how they interact with environmental factors. For example, the adoption studies of adult schizophrenia (Heston 1966; Rosenthal et al. 1971; Kety et al. 1975) have provided strong evidence for a genetic component to schizophrenia, but many questions still remain. Is a particular gene necessary for the disease to develop? Will a certain genotype always lead to illness? Are there ameliorating environments that prevent illness in persons otherwise susceptible? Answers to questions such as these usually require data of many different types. Adoption studies and comparisons of MZ and DZ concordances provide little, if any, information on these questions. Family segregation patterns (how many and which specific relatives are affected) with or without genetic linkage do provide relevant information and may provide partial answers. However, the full scope of geneenvironment interactions can only be understood after the full disease process is understood. Until such a time, family data— pedigrees—and data on genetic linkage can be the bases for an initial understanding.

Possible Complications Etiologic Heterogeneity. In the design of both psychological and bio-

VOL. 8, NO. 2, 1982

logical studies, it is essential to consider that the disorder being studied may be heterogeneous. Heterogeneity can be present at several different levels. When a syndrome subdivides into distinct classes by onset, course, or response to treatment, a corresponding heterogeneity in etiology is a possibility that must be considered. For example, it is likely that infantile autism is a behavioral syndrome with multiple etiologies (Rutter 1974). It is known that the syndrome can develop in association with conditions as pathologically diverse as congenital rubella (Chess, Korn, and Fernandez 1971) and infantile spasms (Taft and Cohen 1971). These environmental agents may be causal in these cases, but infantile autism can also develop in the absence of any environmental agents such as these. Hence, the search for possible genetic factors for autism must take into account the likely etiologic heterogeneity. For example, recurrence risks in these families of autistic children should be calculated separately according to environmental factors present and compared to risks in families with no known causes for the disorder. An apparently homogeneous syndrome may also be heterogeneous, with quite different etiologies producing nearly indistinguishable symptoms. For example, the mucopolysaccharide storage diseases encompass defects of seven different enzymes. Initially, all affected children were considered to have "gargoylism." Now the symptoms for most of the defects are recognized to be different, largely as a result of biochemical/genetic differences having been demonstrated. However, two dif-


ferent enzyme deficiencies give rise to clinical disorders that are still indistinguishable except by enzyme assays (Kolodny 1976). Often, the multiple causes of a "single disease" become easily identifiable as new phenotypic levels are defined: What is apparently homogeneous at the level of gross symptomatology becomes obviously heterogeneous as physiological and/or biochemical aberrations are used to redefine the phenotype. There are many other examples of genetic heterogeneity, the same phenotype being caused by defects at different gene loci, e.g., hemophilia, congenital deafness, and anemia (McKusick 1975). Various types of heterogeneity may be confounded. For example, several different genetic types may exist, and each may have its own separate array of environmental precipitants. Diabetes may be such a disorder, with different modes of inheritance for susceptibilities to juvenile-onset diabetes, to the maturity-onset type of diabetes of the young, and to maturity-onset diabetes. Different environmental agents are thought to be involved for each type (Ganda and Soelder 1977), and genetic heterogeneity may also be present within each type. The anemias are another example of multiple etiologies. There are several different genetic types. Some, such as /3-thalassemia and hereditary spherocytosis, have pathologic processes largely unaffected by environmental variation. Others, such as favism, are quite susceptible to environmental variation. There are also nongenetic anemias caused by environmental factors, such as iron deficiency. Minimizing Heterogeneity. Separating out different genetic types

solely on the basis of the inheritance pattern is possible for some disorders such as retinitis pigmentosa (Spence, Elston, and Cedarbaum 1974). In this case, most large pedigrees can be placed reasonably well into one of three separate classes: autosomal recessive, autosomal dominant, and X-linked. However, some pedigrees are compatible with both autosomal dominant and X-linked inheritance (Gladstien and Spence 1979). For many disorders there is no obvious heterogeneity among the patterns in different families. For the psychiatric disorders and many other disorders with a reasonably large nongenetic component, no exact Mendelian frequency is expected; some families might show apparent recessive inheritance, while others might show apparent dominant inheritance, simply as a function of chance. To select out pedigrees that appear to represent a particular mode of inheritance, autosomal dominant, for example, is likely to lead to false conclusions. Even in the absence of incomplete penetrance, a mixture of rare recessive and rare dominant pedigrees will be difficult to untangle by pattern alone if the family size is small. Some other techniques for resolving heterogeneity are needed. Diagnostic Precision. Genetic studies of childhood behavioral disorders—whether twin, family, linkage, or adoption studies— have been hampered by the imprecision and variability of diagnostic criteria. In fact, much variation in results among different studies might be explained by the differences in diagnostic methods. There are several sources of



unreliability in diagnosis; we shall consider two. The first is the method of collecting information necessary for making diagnostic distinctions; here childhood disorders present special problems. In genetic studies, it is desirable to obtain comparable information from all family members regardless of age. This means interviewing adults about their behavior some 20 to 40 years ago. Such retrospective elicitation is necessary because of the transitory nature of symptoms in many childhood behavioral disorders. Often the disorder is outgrown completely; in other cases the symptomatology changes over time. It is difficult to determine if an adult being studied actually had the appropriate symptoms during childhood, but accurate diagnoses are a necessity in genetic studies. Clearly, work needs to be done in this area. A second source of unreliability in diagnosis is in the rules for using the data to classify patients. The systematically collected information on symptoms needs to be used in a precisely defined way to assign valid diagnoses. The problem of validity is one that can be finally resolved only with a complete understanding of the etiology of the disorder, but must still be considered in early research on the disorder. New diagnostic methodologies that take into account these sources of variation and improve reliability considerably, such as DSM-III (American Psychiatric Association 1980), are now being applied to genetic studies. Improved definitions of diagnostic categories and the greater accuracy with which patients can be assigned valid diagnoses will undoubtedly improve future genetic studies.

But much work remains to be done. Followup studies are needed to determine whether any relationships exist between childhood and adult behavioral disorders. For example, is the depression seen in children an early manifestation of depression in adults (Orvaschel, Weissman, and Kidd 1980)? Diagnostic criteria should then become available to facilitate more accurate diagnoses. When the use of these diagnostic criteria becomes widespread, the resulting diagnostic uniformity will allow meaningful comparisons between studies. Sample Definition. After diagnostic criteria have been used to define the phenotype as rigorously as possible, there are additional ways to minimize possible heterogeneity within the data set before any genetic analysis. These are (1) controlling in the sample for possibly relevant nongenetic variables and (2) collecting data from a population as genetically homogeneous as possible. Genetic homogeneity is more likely in a sample from the same ethnic group than in a completely random sample; it is even more likely to exist in an inbred isolate; genetic homogeneity is more likely to be found in a single pedigree. Even in situations that are relatively homogeneous, genotypic heterogeneity may exist among affected, i.e., some are heterozygous and some are homozygous. The frequency of homozygosity may be increased by special ascertainment criteria for families. Matthysse and Kidd (1976) have shown that ascertainment through two affected children changes the expected genotype distribution of the family markedly because the probability

of affected individuals being homozygous is increased considerably. Selecting for additional affected individuals among remote relatives in multigenerational data should enhance etiologic homogeneity and greatly simplify interpretation of biochemical data. However, if those same pedigrees are to be used for a statistical analysis to estimate genetic parameters, the bias introduced by the high-density ascertainment must be estimated and removed. Morton and Kidd (1979) have developed one method for removing the bias from a likelihood procedure for parameter estimation. Their method is impractical for general use because it requires extensive computer simulation, but their results do show the necessity of considering the high-density ascertainment bias. If homogeneity is achieved by collecting data only from a population with a restricted gene pool (e.g., an isolate or a single pedigree), the results obtained may not extrapolate to the same phenotype in other populations. That should not be a deterrent for special ascertainment strategies, however. If it is possible to identify a specific etiology for a homogeneous group selected with a unique ascertainment technique, it may be possible to establish clinical differences which could then be used to ascertain other groups for additional study. Another strategy for enhancing possible homogeneity has been suggested by Buchsbaum and Rieder (1979). If a biochemical abnormality has been suggested to be associated with a particular o trait, then selecting subjects on the basis of that biochemical pheno-


VOL. 8, NO. 2, 1982

type rather than a psychiatric phenotype might enhance homogeneity. After that sample has been selected, psychiatric evaluation may then be done to see if a higher than expected number of these subjects have a psychiatric disorder. As the authors point out, this allows the investigator to determine if a variety of symptomatologies (i.e., a spectrum) is associated with the biochemical abnormality. In childhood disorders the investigator could determine whether the biochemical abnormality remains observable into adult years and what, if any, symptoms are associated then. Genetic Analyses

Methods. Once homogeneity has been obtained to the best of our ability and the data have been collected, the analysis of the data can begin. The best of the current approaches to genetic analyses of childhood psychiatric disorders and other complex traits is based upon testing of specific genetic hypotheses and attempting to discriminate among the many different genetic models proposed over the past few years (cf. Kidd and Pauls, in press). It is to be hoped that some reasonable set of parameter values for one single model will provide an adequate fit to the observed data while no parameter values for other models will allow acceptable fits. The acceptable model can then be taken, with reasonable confidence, as the true explanation for the genetics of the trait. There are, however, difficulties with this approach. For example, a familial aggregation that does not appear to follow standard Mendelian inheritance patterns is com-

mon for psychiatric disorders (Rosenthal 1970), other behavioral disorders such as stuttering (Kidd and Records 1979), and a variety of birth defects, such as cleft lip and cleft palate (Chung, Ching, and Morton 1974). The simplest explanations of these inheritance patterns incorporate the notion that not all individuals who have a genetic susceptibility actually manifest the disease, i.e., there is incomplete penetrance of the susceptible genotype(s). When twin studies demonstrate that MZ concordance is less than 100 percent, incomplete penetrance can be accepted as a demonstrated phenomenon. Several approaches to genetic analysis of these disorders have been based on incomplete penetrance (Crittenden 1961; Falconer 1965; Morton, Yee, and Lew 1971; Elston and Campbell 1971; Elston and Yelverton 1975; Kidd and Cavalli-Sforza 1973; Morton and MacLean 1974). These models differ in whether the genetic basis is a single major locus or is polygenic (many independent loci, each of small effect), but in every model the susceptible genotypes do not always result in illness. Moreover, most models consider the possibility of phenocopies—individuals in whom the disorder is produced by purely environmental factors in the absence of the susceptible genotypes. These models give rise to different levels of heterogeneity —the same disorder is produced by either a genetic aberration (interactively with the environment) or by purely environmental factors such that patients represent all possible susceptibility genotypes. The incomplete penetrance also results in all genotypes being represented among the unaffected.

Hence, for both theoretical and pragmatic reasons, the model fitting approach has been less than satisfactory. From a statistical perspective the greatest difficulty in the genetic models is the large number of inherent assumptions and simplifications (Kidd 1981). Because of these, the possibility exists that a statistical test may often be testing the validity of the assumptions for a particular trait rather than the applicability of a model's basic genetic mechanism to that trait. From a purely pragmatic perspective, it is now clear that conclusions on a specific mode of inheritance are difficult to reach. Many analyses of real (as opposed to simulated) data have been unable to discriminate between quite different genetic models. Simulation studies have shown that most available types of data on complex disorders do not allow clear discrimination among different genetic hypotheses— most reasonable models are able to explain data on presence/absence of the disorder approximately equally well. Discrimination has so far been reasonably successful only when genetic linkage data or data on obviously related biochemical traits are also incorporated in the analysis (e.g., Kravitz et al. 1979). To counter this pessimistic conclusion, there are steps that can be taken to improve the outlook for understanding complex disorders such as childhood behavioral traits. One of the most important steps is to choose models that are appropriate to the known biology of the disorder. For example, if sex effects are known to exist, then sex-specific parameters should be included in the analysis. As obvious as this seems, it has not al-



ways been done in the past, in part because of limitations in computer methodology. Those limitations no longer exist. Another important step is the preliminary estimation of some parameters using data not included in the genetic analyses. This approach improves the accuracy of estimation for the genetic parameters since, as more parameters are included to make the genetic models more realistic, the precision of simultaneous estimates will be lessened. Of course, some parameters cannot be estimated separately from the genetic model, but others can be. For example, age-of-onset distributions can be estimated using data on unrelated individuals (Heimbuch, Matthysse, and Kidd 1980). Using a "known" distribution as a first approximation in a pedigree analysis should allow a clearer estimation of the remaining parameters and a greater ability to discriminate among alternative hypotheses. If we momentarily forego the ultimate goal of elucidating the precise roles and interrelationships of all factors relevant to a disorder, and instead consider only more immediate goals, the prospects seem brighter. No one expects a complete understanding of etiology to emerge from one study alone; each study serves to provide insight into the design of further experiments aimed at elucidating etiology. If one merely identifies one significant factor or eliminates one possible hypothesis, the study can be judged successful. In this context, even though our ultimate goal is an understanding of the etiology of the trait, it seems appropriate at the moment to take a "backward" step from precise genetic models and

relax the rigor of hypothesis testing. Instead, let us approach family data in the spirit of exploratory data analysis—use of statistics to expose the data to "visual" examination. Many existing analytic techniques might be considered in the context of exploration rather than testing of specific genetic models. One is Slater's (1966) examination of the distribution of uniparental and biparental occurrences of secondary cases in the ancestry of probands. He suggested that biparental occurrences would represent approximately one third of the cases for polygenic inheritance but be exceedingly uncommon for single-locus inheritance. Little has been done with this measure, but it might be worth more exploration as a quick and simple part of an initial exploratory analysis of a set of data. Another method is the use by Reich et al. (1979) of correlation .coefficients and complex segregation analysis to determine whether apparent subtypes of a disorder represent etiologic heterogeneity, part of a genetic continuum, or random variation in the phenotype. Another approach is the use of a logistic analysis (Cox 1970) for study of multiway contingency tables. The logistic model is related to the log linear model but is considered more appropriate for qualitative data like diagnostic categories (well or affected) among relatives. The model can be used to test whether chance alone can explain the differences in frequencies of affected relatives observed among the subsets defined by classifications used. The objective is the elucidation of patterns of risk among relatives. The method has proved useful for preliminary

analysis of family data on human behavioral traits. It has been applied to affective disorders (Smeraldi et al. 1979), stuttering (Kidd and Records 1979; Kidd, Heimbuch, and Records 1980), and Tourette syndrome (Kidd, Prusoff, and Cohen 1980). Examples of Genetic Analyses. The analyses of data on stuttering illustrate many aspects of our approach to the genetics of childhood disorders and its value. The data discussed here were collected in a 5-year study of 600 stutterers and their relatives. The analyses done so far are restricted to the data from probands of European descent. The details of ascertainment and data collection have been presented elsewhere (Kidd and Records 1979). The study accepted only probands with a clinical diagnosis of stuttering and no evidence of any other neurologic problem such as epilepsy and cerebral palsy. Data were collected by interview and self-report questionnaire on presence of stuttering at any time in each first-degree relative. Several preliminary analyses were done before more complete analysis. Kidd, Kidd, and Records (1978) examined the possible causes for the excess of males among individuals who stutter. The possible factors examined included ascertainment bias, incomplete reporting, social role differences, X-linked inheritance, and a combination of environmental and genetic elements. Only the geneenvironment interaction hypothesis was consistent with the data. The sex difference was shown to be real and to be related to transmission of the trait. It was con-


VOL. 8, NO. 2, 1982

eluded that the sex difference observed should be included in any subsequent analysis aimed at understanding etiology of this trait. The data were also examined to determine whether some biases in the modes of ascertainment had yielded distinct subsets of the data with sufficient differences that their existence might affect subsequent analyses. Three major sources of data were defined: (1) child probands were obtained primarily through speech pathologists in schools and private clinics; (2) most adult probands were obtained through enrollment in an intensive therapy program; (3) other adult probands were obtained through clinics and the Councils of Adult Stutterers in several large cities. A comparison of the familial incidences of stuttering for each of these three groups, with incidences calculated separately according to sex of relative, sex of proband, and relationship, showed the families of the children to be significantly different from those of the two adult groups. There was no difference between the adult sample from intensive therapy clinics and the other adult sample. However, the families of child stutterers were different from the families of the adult stutterers. Simple tabulation of the frequencies of affected relatives showed that for child probands, parents more frequently had a history of stuttering than they had for adult probands. This is contrary to an intuitive expectation that, because most children who stutter recover before adulthood, chiidhood stuttering is usually a milder, less severe form than adult stuttering. Two possible biases could account for the elevated (according

to that intuitive expectation) frequency of stuttering among parents of child probands: better reporting of stuttering in the parents because the parents themselves are providing the information and/or a higher frequency of participation by parents who themselves have a history of stuttering. A comparison of the frequencies of stuttering among siblings, conditional on whether or not either parent ever stuttered, showed no differences between the adult and child proband groups. Thus, the difference appears to be in the frequency with which families in which a parent of the proband also stutters agreed to participate in our study. This apparent bias could shift analyses toward accepting the hypothesis that stuttering is transmitted. Therefore, only adult probands have so far

been used for subsequent analyses. Among the adult probands, there were three times as many males as females. While there were more than 100 female probands, total, some of the first degree relative categories contained such small numbers as to be comparatively meaningless. Thus, female probands were eliminated in the next stage of the analyses. The data from adult male probands were divided into four categories based on whether any parent had ever stuttered. The data for siblings of these male probands are given in table 1. Analyses were done to test for transmission of stuttering using the underlying principle that if each affected individual was an independent event, then selecting for families with multiple affected would not in-

Table 1. Number of relatives (percent who ever stuttered) by parental type for male probands1 Parental type2 Number of families:





267 families

65 families

19 families

5 families

58 (17.21) 279 337

24 (32.43) 50 74

6 (35.29) 11 17

2 (33.33) 4 6

7 (2.56) 266 273

8 (10.39) 69 77

1 (5.56) 17 18

3 (37.5) 5 8


S NS Total Sisters3 S NS Total

1 Reprinted with permission from Kidd, K.K., and Records, M.A. Genetic methodologies for the study of speech. In: Breakefield, X.O., ed. Neurogenetics: Genetic Approaches to the Nervous System, 1979. Copyright © American Elsevier Publishing Company.

'Parental types: N—neither parent ever stuttered; F—only father ever stuttered; M—only mother ever stuttered; B—both parents stuttered. Classifications of siblings: S—ever stuttered; NS—never stuttered.



crease the frequency of stuttering among the remaining relatives. From these data it is possible to examine whether parental stuttering affected the frequency of stuttering among the proband's siblings. For statistical analyses only the first two categories (where neither parent stuttered or where only the father stuttered) were used. The other two categories of parental type were excluded because their numbers were too few to allow meaningful interpretations. The effect of having a father who stuttered as opposed to having no stuttering parent is quite marked and statistically significant: the frequencies of affected siblings of a proband are increased if that proband has a father who stuttered. Though the statistical analysis has no genetic hypotheses incorporated in it, this significant difference provides strong support for the hypothesis that stuttering is transmitted. What cannot be resolved by this analysis is whether the transmission is solely genetic, solely cultural, or a combination of the two. Following these preliminary analyses, a more elaborate statistical analysis has been done on the families of all adult probands. The effects of all the variables shown to be significant were tested simultaneously using a logistic model (Cox 1970). As stated earlier, this model has some similarities to the log-linear model but uses the logistic function to estimate the frequency (9) of affected individuals in a category as

e = —e-y

where 1/ is a linear function of the parameters. Specifically, in these analyses t/ is the sum of an overall mean, n, and the values associated with the specific classification criteria for the category. The classification criteria used were sex-ofproband (male or female), sex-of-relative (male or female), and parental type (neither or atleast-one parent affected). The full model includes parameters for all main classifications, three twoway interactions, and the threeway interaction. Statistical significance of a parameter estimate can be tested in two ways: by setting that parameter to zero and doing a likelihood ratio test and by testing whether the estimated value is significantly different from zero. This model is directly analogous to analysis of variance, which can also be formulated as a general linear problem. Unlike a chi-square test for association, the linear logistic model allows some inference concerning the relative magnitude and direction of the effects within a classification criterion. The overall null hypothesis for this analysis is that any observed differences in frequencies of affected individuals among subsets of relatives are due to chance alone. This null hypothesis assumes that the subsets defined by the classification criteria are all drawn from a statistically homogeneous population. Based on the classifications used, we can define null hypotheses, rejection of which would support general hypotheses about the nature of the familial pattern present. Specific null hypotheses that would lead to a hypothesis of vertical transmission of susceptibility include: (1) Regardless of other classifi-

cation criteria applied to the relatives, parental type has no significant effect upon frequency of affected relatives of probands. (2) Regardless of other criteria, the overall frequency of affected relatives is not different from the general population prevalence. The results of this more comprehensive analysis for stuttering have helped define the pattern with which stuttering occurs in families (Kidd, Heimbuch, and Records 1981). A total of 294 adult male probands and 103 adult female probands and their families were studied. The samples obtained by the two methods of ascertainment were not significantly different and were pooled for this analysis. Likelihood ratio tests indicated that no interaction terms were significant. The main factors found to be significant were sex of proband, sex of relative, paternal stuttering, maternal stuttering, and type of relative. Maximum likelihood parameter estimates for the main effects are given in table 2 for the logistic model. When these parameters are used, the agreement between observed and expected frequencies of stuttering is quite good. Next the data were examined to determine if these factors and other variables were predictors of severity of stuttering in the proband (Kidd et al. 1980). This possibility was examined using data on a subset of 184 adult stutterers and their families. Frequency of stuttering during a pretreatment oral reading task was available as the severity measure for each of these index cases and information on whether a relative ever stuttered was available on all first degree relatives. The family


VOL. 8, NO. 2, 1982

Table 2. Maximum likelihood parameter estimates for main effects Maximum likelihood estimates Parameter







Sex of proband Female Male Sex of relative Male Female Paternal stuttering Present Absent Maternal stuttering Present Absent Type of relative Sibling Child

data variables, including sex and exact relationship, combined with birth date and sex of index case were used in three types of analyses: multiple regressions, AID regressions, and stepwise regressions. None of the variables tested, including stuttering among first degree relatives, was a predictor of severity of stuttering in the index case. Apparently this measure of severity is not related to the factors which predispose to stuttering. The next questions consider how to explain the transmission. Genetic models might be used to explain the transmission. Although stuttering does not follow a clear single gene pattern, it may very well be explained by some threshold model in which a major locus segregates three genotypes, but

Standard error



.25 -.25


.63 -.63


.32 -.32


.35 -.35


-.22 .22


nongenetic variation modifies the risk of stuttering for each genotype. A threshold then determines who does and who does not stutter. We could have heterozygotes

who stutter if they have a sufficiently stressful environment, and we could have homozygotes who do not stutter if they have a sufficiently benign or ameliorating environment. The position of the threshold and the gene frequencies can be varied to find the best explanation of the data. For stuttering the results are given in table 3. This solution explains the family data quite well, with the transmission having a purely genetic basis. The gene frequency in the solution is 4 percent, not a rare gene, but not a very common one either. Most individuals in the total population lack the gene. About 8 percent of the population is heterozygous, and less than 2 in 1,000 are actually homozygous. The penetrances—the probability that an individual with a particular genotype will ever stutter, as a function of sex—are virtually zero for the normal genotype; they are one (or 100 percent) for homozygotes for the stuttering allele. A striking sex difference is found for heterozygotes: penetrance is about 38 percent for a heterozygous male but only 11 percent for a

Table 3. Best fit of the single major locus model to the family incidence data on stuttering1 Stuttering allele (S) frequency = .040 ± .007 Predicted general incidences: .035 for males; .010 for females Penetrances for each genotype and sex: Genotypes

Male Female

NN 0. 005 ± 0.003 0. 0002 ± 0.0002

NS 0 .378 ± 0.025 0 .107 ± 0.019




The goodness of fit xl = 465, p = 0.22 1 Reprinted with permission from Kidd, K.K., and Records, M.A. Genetic methodologies for the study of speech. In Breakefield, X.O., ed. Neurogenetics: Genetic Approaches to the Nervous System, 1979 © American Publishing Company.



heterozygous female. These parameter values predict lifetime prevalances of about 4 percent of the population of males and about 1 percent of the population of females, roughly the observed values for the lifetime prevalance of ever stuttering at some time in childhood. Finding an acceptable solution does not constitute proof that a single locus is responsible for stuttering, but it is highly suggestive. Especially noteworthy is that the model does not require that the interaction of genetic and nongenetic factors be confined to heterozygotes and yet biological principles would suggest that genotype as the one in which that interaction is most likely to be significant. Morton and Kidd (1979) corrected for selection bias in high density pedigrees (those with five or more affected individuals) and obtained additional support for parameter estimates which closely correspond to the estimates in table 3 except that the allele frequency estimate is lower. This difference gives population frequency estimates lower than those above. The most likely gene frequency was .01 which gives population frequencies between 1.3 for males and .25 for females. Arguing against a single major locus (SML) explanation is the finding that a multifactorial polygenic threshold model (MFP) can also explain the data (Kidd 1977). Genetic factors also account for the transmission of stuttering in the MFP model and the model also accounts for the sex differences observed. The solution gives sexspecific thresholds, with the female threshold being farther from the population mean than the threshold for males. The predicted

general population incidence for males is about 5.5 percent and 1.5 percent for females with the correlation among relatives estimated at 0.38. These incidence figures correspond quite well with the observed values reported earlier, and the correlation coefficient indicates that stuttering clusters within families. In summary, the overall pattern of occurrences of stuttering within families is definitely nonrandom. Genetic analyses of the data reveal that this pattern of stuttering can be explained by genetic transmission of susceptibility and sexmodified interaction with environment. Although the mode of transmission is not yet clearly defined, two genetic models with sex-specific thresholds closely fit the data and may help to identify the genetic nature of susceptibility to stuttering. However, nongenetic hypotheses must also be considered before any firm conclusion can be reached. Certain models of purely cultural inheritance can be excluded by the data. The simplest cultural model—imitating the speech of a family member who stutters—could not explain more than a small fraction of all stuttering in children (Kidd, Kidd, and Records 1978). This finding is consistent with the high familial concentration of stuttering because most stutterers recover before adulthood. For example, in 20 percent of these families the father had at some time stuttered, but in one half of those cases he had recovered before the birth of the child who later stuttered. Some more complex cultural hypotheses have involved a general "nervousness" being culturally transmitted since, for example, anxiety is

known to exacerbate symptoms in a stutterer. Such hypotheses predict that the severity of stuttering would be associated with the frequency of relatives who stutter. One reliable measure of the severity of stuttering is the frequency of words on which a stutterer has some difficulty. This measure of severity is not associated with the frequency or distribution of stuttering among relatives, and we therefore conclude that severity is not related to the transmitted factors that predispose to stuttering. Hence, hypotheses of general anxiety being the culturally transmitted cause of stuttering appear to be excluded by these data. At this time definite proof is elusive, yet all available evidence suggests that susceptibility to stuttering is genetically transmitted. The analysis of data on Tourette syndrome (Kidd, Prusoff, and Cohen 1980) is another illustration of the use of the logistic model. Several questions were addressed: Are multiple tics (MT) a milder form of Tourette syndrome (TS)? Is there a familial concentration? Is the sex difference real and not some artifact of ascertainment? If the sex difference is real, is it important in transmission? Collection of data for this analysis has been described elsewhere (Kidd, Prusoff, and Cohen 1980). Briefly, 200 questionnaires were mailed out by the Tourette Syndrome Association (TSA) of which only 75 were returned. Followup contact with some individuals who had not returned completed questionnaires indicated a variety of reasons: questionnaire not received, individual did not have TS, etc. Of the 75 returned questionnaires, 66 included information on symptomatology on a total

VOL. 8, NO. 2, 1982


of 231 first degree relatives in these families. Only these relatives were included in this analysis. The frequencies of TS and/or multiple tics among these relatives are shown in table 4. Table 5 shows the data when divided into a three-way contingency table by three classifications: sex of proband, sex of sibling, and affected status of parents. The logistic model was used to test simultaneously whether each of these classification criteria resulted in statistically significant differences in the frequencies of affected siblings and whether there were any statistically meaningful interactions among these main classifications. The analyses showed that each main classification was statistically significant, but none of the interaction terms were. Therefore the values from tables 5 and 6 indicate that: (1) multiple tics appear to be a mild form of TS; (2) there is vertical transmission since the parental effect is significant; (3) the sex difference in the prevalence of TS is real; and (4) this sex difference is related to the transmission of susceptibility for either multiple tics and/or TS. Hence, a

Table 5. Frequency of TS and/or MT among siblings of TS patients Probands Type of family Neither parent affected


.074 ± .043 ± .400 ± .200 ±

Brothers Sisters

One or more parents Brothers affected Sisters

sex-modified form of transmission was demonstrated without having to assume a specific etiologic model—either genetic or cultural. These results support previous findings which indicate that TS shows a familial concentration, especially if multiple tics are considered a minor manifestation of the trait (Eldridge et al. 1977; Shapiro et al. 1978; Nee et al. 1980). Wilson, Garron, and Klawans (1978), however, pointed out that most data purporting to show> familial aggregation of multiple tics do not provide support for the hypothesis that the condition is

Table 4. Frequency of TS and/or MT among 230 first-degree relatives Sex of proband Type of relative Fathers Mothers Brothers Sisters



.229 .125 .162 .091

± ± ± ±

.061 .048 .061 .050

.157 ± .029


.333 .111 .462 .188

± ± ± ±

.111 .074 .138 .098

.262 ± .055

.050 (2/27) .042 (1/23) .155(4/10) .126 (2/10)


.400 ± .200 ± .667 ± .167 ±

.155(4/10) .126(2/10) .272 (2/3) .152(1/6)

transmitted. Most previous studies have reported only the proportion of patients with a positive family history of TS or multiple tics. Wilson, Garron, and Klawans (1978) suggest that, because the frequency of tics is approximately 10 percent in the general population, about 30 percent of patients could be expected to have a positive family history of tics (i.e., at least one affected relative) by chance alone. The conclusion reached by Kidd, Prusoff, and Cohen (1980) is that each relative has an elevated risk, thereby confirming the familial aggregation. These data suggest that, because affected status of parent (with either MT or TS) significantly affects the risk to relatives, multiple tics and TS are transmitted vertically and may have a similar etiologic base. These data also suggest that, because sex of proband (with TS) affects the risk of MT and/or TS in relatives and because opposite sex relatives are at greater risk, the transmission is sex-modified. What these data do not do, however, is explain the cause of that transmission. While genetic models may adequately fit the data, other cultural/environmental mod-



Table 6. The values of the parameters of the logistic model applied to family data on Tourette syndrome

References American Psychiatric Association.

Parameter fi, overall "mean"

Estimated value -1.25 ± .29

Sex of proband Male' Female

- . 6 6 ± .29 .66 ± .29

Sex of sibling Male Female

.58 ± .30 - . 5 8 ± .30

Parental type Neither affected At least one affected

- . 6 6 ± .29 .66 ± .29

DSM-IH: Diagnostic and Statistical Manual of Mental Disorders. 3rd ed.

Washington, DC: APA, 1980. Buchsbaum, M.S., and Rieder, R.O. Biologic heterogeneity and psychiatric research: Platelet MAO activity as a case study. Archives of General Psychiatry, 36:1163-1169, 1979. Cavalli-Sforza, L.L., and Feldman, M.W. Cultural versus biological inheritance: Phenotypic transmission from parent to children (a theory of the effect of parental phenotypes on children's phenotype). American Journal of Human

els may as well. The results reviewed here have been confirmed by additional data (Pauls et al., 1981), and therefore the patterns observed need to be incorporated into subsequent attempts to find appropriate explanations of how the trait is transmitted. For example, the sex differences need to be incorporated into both a possible genetic or environmental model. Evaluation of these findings will require systematic investigation. Homogeneous groups identified by symptomatology, family history, response to treatment, and long-term prognosis need to be identified and studied to help in our understanding of this disorder. Conclusion Childhood behavioral traits are complex disorders and hence do not lend themselves to simple,

straightforward genetic analyses. Methodologies are available for these kinds of traits, but the assumptions inherent in their formation often make interpretation of results, difficult. There are several approaches appropriate for preliminary analyses which can help diminish the number of assumptions necessary for more complex analyses. Data on stuttering and Tourette syndrome are examples of an approach to genetic analysis that helps identify parameters to be included in subsequent analyses. These data suggest that the two disorders are transmitted and that in both cases the transmission is sex-modified. Therefore, it is important to incorporate sex differences in subsequent analyses for these two disorders. It is also apparent that in subsequent analyses of TS, multiple tics should be considered as a milder manifestation of the trait.

Genetics, 25:618-637, 1973. Chess, S.; Korn, S.J.; and Fernandez, P.B. Psychiatric Disorders of Children With Congenital Rubella. New York: Brunner/Mazel, 1971. Chung, C.S.; Ching, G.H.S.; and Morton, N.E. A genetic study of cleft lip and palate in Hawaii: II. Complex segregation analysis and genetic risks. American Journal of Human Genetics, 26:177-188, 1974. Cloninger, C.R.; Rice, ].; and Reich, T. Multifactorial inheritance with cultural transmission and assortative mating. II. A general model of combined polygenic and cultural inheritance. American Journal of Human Genetics, 31:176-198, 1979. Cox, D.R. Analysis of Binary Data.

New York: Chapman and Hall, 1970. Crittenden, L.B. An interpretation of familial aggregation based on multiple genetic and environmental factors. Annals of the New York Academy of Sciences, 91:769-780, 1961.


VOL. 8, NO. 2, 1982

Eldridge, R.; Sweet, R.; Lake, C.R.; Ziegler, M.; and Shapiro, A. Gilles de la Tourette's syndrome: Clinical, genetic, psychologic and biochemical aspects of 21 selected families. Neurology, 27:115-124, 1977. Elston, R.C., and Campbell, M.A. Schizophrenia: Evidence for the major gene hypothesis. Behavior Genetics, 1:3-10, 1971. Elston, R.C., and Yelverton, K.C. General models for segregation analysis. American Journal of Human Genetics, 27:31-45, 1975. Falconer, D.S. The inheritance of liability to certain diseases, estimated from the incidence among the relatives. Annals of Human Genetics, 29:51-76, 1965. Ganda, O.P., and Soeldner, S.S. Genetic, acquired and related factors in the etiology of diabetes mellitus. Archives of Internal Medicine, 137:461-469, 1977. Gladstien, K., and Spence, M.A. Counseling for dominant traits: A correction for the ascertainment bias due to referral for analysis. Clinical Genetics, 15:171-175, 1979. Heimbuch, R.C.; Matthysse, S.W.; and Kidd, K.K. Estimating age-ofonset distributions for disorders with variable onset. American Journal of Human Genetics, 31:564-574, 1980. Heston, L.L. Psychiatric disorders in foster home reared children of schizophrenic mothers. British Journal of Psychiatry, 112:819-825, 1966. Hodge, S.E.; Morton, L.A.; Tideman, S.; Kidd, K.K.; and Spence, M.A. Age of onset correction available for linkage analysis (LIPED). American Journal of Human Genetics, 31:761-762, 1979.

Kety, S.S.; Rosenthal, D.; Wender, P.H.; Schulsinger, F.; and Jacobsen, B. Mental illness in the biological and adoptive families of adoptive individuals who have become schizophrenic: A preliminary report based on psychiatric interviews. In: Fieve, R.; Rosenthal, D.; and Brill, H., eds. Genetic Research in Psychiatry. Baltimore: Johns Hopkins University Press, 1975. pp. 147-165. Kidd, K.K. A genetic perspective on stuttering. Journal of Fluency Disorders, 2:259-269, 1977. Kidd, K.K. Genetic models for psychiatric disorders. In: Gershon, E.; Matthysse, S.; Ciaranello, R.D.; and Breakefield, X.O., eds. Genetic Strategies in Psychobiology and Psychiatry. Pacific Grove, CA: Boxwood Press, 1981. pp. 369-382. Kidd, K.K., and Cavalli-Sforza, L.L. An analysis of the genetics of schizophrenia. Social Biology, 20:254-265, 1973. Kidd, K.K.; Heimbuch, R.C.; and Records, M.A. Vertical transmission of susceptibility to stuttering with sex-modified expression. Proceedings of the National Academy of Sciences, 78:606-610, 1981. Kidd, K.K.; Kidd, J.R.; and Records, M.A. The possible causes of the sex ratio in stuttering and its implications. Journal of Fluency Disorders, 3:13-23, 1978. Kidd, K.K., and Matthysse, S.W.; with Carmelli, D.; Cavalli-Sforza, L.L.; Elston, R.; Lange, K.; Motulsky, A.; Rao, D.C.; Reich, T.; and Wyatt, R.J. Research designs for the study of gene-environment interactions in psychiatric disorders: Report of an FFRP panel. Archives of General Psychiatry, 35:925-932, 1978.

Kidd, K.K.; Oehlert, G.; Heimbuch, R.C.; Records, M.A.; and Webster, R.L. Familial stuttering patterns are not related to one measure of severity. Journal of Speech and Hearing Disorders, 23:539-545, 1980. Kidd, K.K., and Pauls, D.L. Considerations in research designs and data analysis for psychiatric genetics. In: Tsuang, M.T., ed. Psychosocial Epidemiology Monograph Series (Genetics). Washington, DC: Superintendent of Documents, Government Printing Office, in press. Kidd, K.K.; Prusoff, B.A.; and Cohen, D.J. Familial pattern of Gilles de la Tourette syndrome. Archives of General Psychiatry, 37:1336-1339, 1980. Kidd, K.K., and Records, M.A. Genetic methodologies for the study of speech. In: Breakefield, X.O., ed. Neurogenetics: Genetic Approaches to the Nervous System. New York: Elsevier-North Holland, 1979. pp. 311-314. Kolodny, E.H. Current concepts in genetics: Lysosomal storage diseases. New England Journal of Medicine, 294:1217-1220, 1976. Kravitz, K.; Skolnick, M.; Cannings, C ; Carmelli, D.; Baty, B.; Amos, B.; Johnson, A.; Mendell, N.; Edwards, C ; and Cartwright, G. Genetic linkage between hereditary hemochromatosis and HLA. American Journal of Human Genetics, 31:601-619, 1979. Matthysse, S.W., and Kidd, K.K. Estimating the genetic contribution to schizophrenia. American Journal of Psychiatry, 133:185-191, 1976. McKusick, V.A. Mendelian Inheritance in Man: Catalogs of Autosomal



Dominant, Autosomal Recessive, and X-Linked Phenotypes. Baltimore: Johns Hopkins University Press, 1975. Morton, L.A., and Kidd, K.K. Likelihood analysis of highdensity pedigrees of stuttering. Behavior Genetics, 9:470A (Abstract), 1979. Morton, N.E., and MacLean, C.J. Analysis of family resemblance: III. Complex segregation of quantitative traits. American Journal of Human Genetics, 26:489-503, 1974. Morton, N.E.; Yee, S.; and Lew, R. Complex segregation analysis. American Journal of Human Genetics, 23:602-611, 1971. Munsinger, H. The adopted child's IQ: A critical review. Psychological Review, 82:623-659, 1975. Nee, L.E.; Caine, E.D.; Polinsky, R.J.; Eldridge, R.; and Ebert, M.H. Gilles de la Tourette Syndrome— Clinical and family study of 50 cases. Annals of Neurology, 17:41-49, 1980. Orvaschel, H.; Weissman, M.J.; and Kidd, K.K. Children and depression: The children of depressed parents; the childhood of depressed patients; depression in children. Journal of Affective Disorders, 2:1-16, 1980. Ott, J. A" computer program for linkage analysis of general human pedigrees. American Journal of Human Genetics, 28:528-529, 1976. Pauls, D.L.; Cohen, D.J.; Heimbuch, R.; Detlor, J.; and Kidd, K.K. The familial pattern and transmission of Tourette syndrome and multiple tics. Archives of Gen-

eral Psychiatry, 38:1091-1093, 1981. Rao, D.C.; Morton, N.E.; and Yee, S. Resolution of cultural and biological inheritance by path analysis. American Journal of Human Genetics, 28:228-242, 1976. Reich, T.; Rice, J.; Cloninger, C.R.; Wette, R.; and James, J. The use of multiple thresholds and segregation analysis in analyzing the phenotypic heterogeneity of multifactorial traits. Annals of Human Genetics, 42:371-389, 1979. Rice, J.; Cloninger, C.R.; and Reich, T. Multifactorial inheritance with cultural transmission and assortative mating. I. Description and basic properties of the unitary models. American Journal of Human Genetics, 30:618-643, 1978. Rosenthal, D. Genetic Theory and Abnortnal Behavior. New York: McGraw-Hill Book Co. Inc., 1970. Rosenthal, D.; Wender, P.H.; Kety, S.S.; Welner, J.; and Schulsinger, F. The adopted-away offspring of schizophrenics. American Journal of Psychiatry, 128:307-311, 1971.

A.M. Genetic studies of affective disorders. In: Obiols, J.; Ballu's, C ; Gonzalez Moncliis, E.; and Pujol, J., eds. Biological Psychiatry Today. Elsevier-North Holland Biomedical Press, 1979. pp. 60-65. Spence, M.A.; Elston, R.C.; and Cedarbaum, S.D. Pedigree analysis to determine the mode of inheritance in a family with retinitis pigmentosa. Clinical Genetics, 5:338-343, 1974. Taft, L.T., and Cohen H.J. Hypsarrhythmia and infantile autism: A clinical report. Journal of Autism and Childhood Schizophrenia, 1:327-336, 1971. Wilson, R.S.; Garron, D.C.; and Klawans, H.L. Significance of genetic factors in Gilles de la Tourette syndrome: A review. Behavior Genetics, 8:503-510, 1978. Acknowledgment

Rutter, M. The development of infantile autism. Psychological Medicine, 4:147-163, 1974.

The preparation of this article was supported, in part, by grants MH-14235, MH-30929, NS-11786, and NS-16648 from the U.S. Public Health Service and a grant from the Stuttering Center, Department of Neurology, Baylor College of Medicine, Houston, TX.

Shapiro, A.; Shapiro, E.; Brunn, R.; and Sweet, R. Gilles de la Tourette Syndrome. Raven Press: New York, 1978.

The Authors

Slater, E. Expectation of abnormality on paternal and maternal sides: A computational model. Journal of Medical Genetics, 3:159-161, 1966. Smeraldi, E.; Kidd, K.K.; Negri, F.; Heimbuch, R.; and Melica,

David L. Pauls, Ph.D., is a Research Associate and Kenneth K. Kidd, Ph.D., is Associate Professor, Department of Human Genetics, Yale University School of Medicine, New Haven, CT.

Suggest Documents