Genetic analysis of manic-depressive Illness

AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 62:51-59 (1983) z zyxwvuts zyxwv zyxw Genet ic Analysis of Manic-Depressive IIIness D.H. O’ROURKE, P. McGUFFIN, AND T. REICH Department ofAnthropology, University of Utah, Salt Lake City, Utah 84112 (D.H.O.); Institute of Psychiatry, Maudsley Hospital, London (P.M.1; Department of Psychiatry (P.M., T.R.),and Department of Genetics, Washington University School of Medicine and the Jewish Hospital of St. Louis (T.R.), St. Louis, Missouri 63110 KEY WORDS Bipolar affective disorder, Segregation analysis, Combined model ABSTRACT Two threshold models, a single locus model, and two combined models are fitted to data on familial incidence of bipolar affective disorder in 194 nuclear families ascertained through a bipolar proband. The relative fit of alternative tramsmission models is tested by a likelihood ratio chi-square with the degrees of freedom defined by the difference in the number of parameters estimated by each model. All parameters are estimated by the method of maximum likelihood. The simplest threshold model, permitting only a single background familial correlatiron, is found to provide a statistically poorer fit than any of the alternative models, and may be rejected as a model for the etiology of bipolar affective disorder. The four remaining models are statistically indistinguishable. It is suggested, however, that the involvement of a major locus in the etiology of this disorder deserves further scrutiny since any of the models incorporating a major locus, with or without a multifactorial background, are consistently associated with greater likelihoods than the complex threshold model. It is also noted that diagnostic criteria are critical in the analysis. In the present study, relatives of probands are considered affected if a diagnosis of bipolar or unipolar affective disorder is present. When only bipolar relatives are considered affected, none of the transmission models may be rejected. Finally, the results of these analyses are found to be independent of the ascertainment parameter. The interaction of biology and behavior is central to the investigations of biological anthropology. Of particular interest are those behavioral phenotypes that relate to fitness and have a demonstrable biological basis. Manicdepressive illness is just such a phenotype. So called “bipolar affective disorder” (BP) is characterized by extreme mood swings between elation (mania) and depression, whereas unipolar (UP) disorder is characterized by depressive episodes alone. Mania comprises (1) elated, unstable, and fluctuating moods; (2) pressure of speech, occasionally associated with rhyming or punning and circumlocution; (3) flight of ideas, grandiosity, and increased distractability; (4) increased motor activity and energy output; ( 5 )exercise of poor judgement; and (6) increased aggressive or sexual behaviors. In contrast, a depressive episode is marked by (1)depressed mood; (2) psychomotor retar- dation; (3) deterioration of thought processes, which may include paranoid ideation, accusatory hallucinations, and delusions; (4) suicidal ideation; ( 5 ) decreased sexual interests; and (6) abnormalities of physiological function such as loss of appetite, weight loss, and sleep disturbance (Balis et al., 1978). Although females exhibit UP disorder nearly twice as often as males, recent surveys indicate that the population prevalence of BP disorder is roughly comparable between the sexes (see Reich et al., 1983 and McGuffin and Reich, 1983 for reviews). Population prevalence (Kp) estimates for BP disorder range from a low of 0.0036 to a high of 0.022 (Slater and Roth, 1969). Family, twin, and adoption studies have demonstrated the importance of genetic factors zyxwvutsr (9 1983 ALAN R. LISS. INC Received June 5, 1982, accepted Marc’, 16, 1983 52 D.H. O’ROURKE. P. M ~ G U F F I N ,AND r r . REICH in the etiology of bipolar affective disorder. However, the mode of inheritance of this nonMendelian trait has remained obscure. One advance for genetic studies of affective disorders was the diagnostic distinction proposed by Leonhard (1959)between bipolar and unipolar affective disorders. Subsequent studies by Angst (1966) and Perris (1966)illustrated that family members of UP probands showed increased frequencies of U P disorder but no increase of BP disorder. In contrast, relatives of BP probands showed elevated rates of affectation for BP disorder and, in most studies (Angst 1966; Gershon, et al. 1975a,b; Gershon and Leibowitz, 1975; Helzer and Winokur 1974; Smeraldi, et al. 1977; Winokur and Clayton 1967) increased frequencies of primary depression without episodes of mania, suggesting that depressed relatives of BP probands should be considered affected whether or not a manic episode has been observed. This dichotomy and diagnostic procedure is supported by data suggesting that conversion to BP disorder after several depressive episodes is reasonably high (perhaps approaching 18%, e.g., Akiskal et al. 1977). Consequently, in the present study, relatives of BP probands diagnosed as suffering from primary depression are considered affected. ance where specific combinations of alleles at multiple loci, together with environmental contributions, additively determine liability to the disease (Falconer, 1965, 1967; Crittenden, 1961; Reich et al., 1972, 1975).Liability to the disease is considered normally distributed (or transformable to a normal distribution) within the population (Fig. 11, and only those individuals above the liability threshold are affected. It may be emphasized that even where evidence exists for unequal effects (e.g., a few loci of major effect), the assumption of equal effects of all factors underlying the liability distribution is reasonable since it leads to accurate estimates of risks to the disease as well as response to selection (Falconer, 1960, 1965). Moreover, the assumption of normality of the liability distribution is independent of the equality of effects of underlying factors since it depends only on the distribution of phenotypic values. This assumption, then, merely reflects the choice in scale of measurement. The Central Limit Theorem assures rapid approach to normality as long as many underlying factors are relevant (Cloninger et al., 1983; Reich et al., 1972). Under this mode of inheritance, relatives of affected probands will show a distribution on the liability scale with a higher mean than the general population and, hence, a greater proportion of affected individuals. Since the distribution of liability is considered normal, we may take it as a standardized distribution with the unit of measurement the standard deviation, the threshold a t zero, and zyxwvutsrq MATERIALS AND METHODS The data come from three family studies of BP disorder conducted by members of the Psychiatry Department, Washington University School of Medicine, St. Louis (Winokur et al., 1969, Helzer and Winokur, 1974, and unpublished data on 103 black families ascertained through a BP proband) between 1965 and 1975. In all cases the diagnostic criteria used were either those of Feighner et al. (1972),or closely comparable criteria developed in the Psychiatry Department, which were the forerunners of the Feighner criteria. These three studies were first analyzed separately and remarkably similar results obtained. Therefore, in subsequent analyses, the results of which are presented here, the studies of the St. Louis group have been pooled to include data on 194 BP probands, their sibs, and parents. Various forms of three genetic transmission models have been fitted to these data: two multifactorial threshold models, a single major locus model, and two combined models; that is, a major locus with multifactorial background contributing to liability to the disease. z zyx zyxwv zyxw Threshold 2. r) 2 w 3 X zyxwvu zyxwvuts The models Multifuctoriul (MF). Discrete, presence/absence traits, such as BP disorder may be the result of polygenic or multifactorial inherit- Xr LIABILl7Y Fig. 1. Multifactorial model with single threshold. Probands are ascertained from pool of affected in general population. Relatives of probands show a n increase in mean of the distribution on the liability scale as well as a n increase in the proportion of affected individuals. 53 GENETICS OF BIPOLAR DISORDER the total variance in liability for the population equal to one. Taking xp and XR as the normal deviates of the thresholds for probands and relatives of probands for their respective distributions on the liability scale, the correlation between relatives on the liability scale is (Reich et al., 1972): r = xp THRESHOLD zyxwvuts - XR I V‘[I - (x; - x;) (1 - (xp/a))l a + x&(a - xp) where a is the deviation of the proband mean from the population mean. This correlation and the threshold values (or the prevalences in the general population and relatives of probands) determine the MF model under the following additional assumptions (Falconer, 1965; Reich et al., 1972): (1)all genetic and environmental causes of the disease may be combined into a singie continuous variable termed the liability; (2) the population is divided into a t least two recognizable classes by one or more thresholds; (3) genes that are relevant to the etiology of the disease are separately of small effect relative t o the total variation; (4)although not required (see above), genes are assumed to act additively in order t o allow data on sibs to be combined with those from parents and offspring; ( 5 ) there are multiple environmental contributions t o etiology that act additively; and (6) for purposes of genetic parameter estimation, common environmental effects among relatives are negligible. If the assumptions of the model are met, r is independent of the population mean and is invariant over changes in definition of the threshold. Further, this model “applies only to those diseases whose genetic component is multifactorial, or if there are few genes, where these have effects that are small in relation to the non-genetic variation” (Falconer, 196553). In those cases where the genetic component is a single locus of major effect, an alternative model must be employed. Single major locus (SML). Such traits may be the result of a single diallelic locus (A and a) where each of the three genotypes may be incompletely penetrant (Fig. 2). This model is defined by the a gene frequency (q = 1 - p) and the penetrance values (fl, fz, f3) for each of the three genotypes (AA, Aa, and aa, respectively), where the penetrances are the probability of each genotype exhibiting the phenotype under study. Collectively, the penetrance values are referred to as the penetrance vector. James (1971) showed that morbid risk data permit estimation of only three parameters; the population prevalence (Kp), the additive genetic variance (VA),and the dominance var- LIABI L I M Fig. 2. General single major locus model. The “threshold,” analagous to that in multifactorial model, is defined by the penetrance vector; the probability of each genotype exhibitng the phenotype under study. zyxwvu zyxwvutsr zyxw zyxwvu iance (V,) irrespective of the number of classes of relatives of probands studied. Thus, a parameter problem is encountered. Determination of these three values is insufficient to estimate the four underlying parameters uniquely (Kempthorne, 1957): Kp VA VD + $f3 = 2pq [q(f3 - f2) + p(f2 - fAI2 = pZq2[f1 - 2f2 + fJ2. = p2fi + 2pqf2 As noted by James (1971)incidence of a disease in relatives of a proband is a function of Kp and covariance (COVR)between relatives, where the covariance between relatives is the sum of weighted proportions of the two genetic variances (i.e., COVR = uVA + uVD). The weights (u and u ) are simply the probabilities that relatives share one particular allele and both alleles at a locus identical by descent, respectively. For the class of relatives considered here these weights are 0.5 and 0 for parents and offspring and 0.5 and 0.25 for sibs. Fortunately, the parameter problem just discussed may be resolved by maximizing the information contained in nuclear family units as opposed to population prevalences and correlations between relatives. Fishman et al. (1978) have shown that examination of the joint prevalence probability for sibs (e.g., both sibs in a sibship of size two are affected) in conjunction with the observed parental phenotypes permits unique estimation of the underlying parameters of the single major locus model. Here, the phenotype is dichotomized as affectational sta- zyx zyxwv 54 D.H. O’ROURKE, P. McGUFFIN, AND T. REICH zyxwvuts zyxwv zyxwvut zyxwvut zyxwvutsrq zyxwv tus where an affected individual may be coded as 1 and an unaffected individual as 0. Fishman et al. (1978) prove the theorem that the joint probability of affectation in sibs, conditional on parental phenotypes for a presence-absence trait, e.g., zero and separate, estimated variances such that their sum is the variance of e. Similarly, c, the polygenic effect may be subdivided into the midparent breeding value as well as the individual deviation from this value with each normally distributed around mean zero. The major locus is defined by its mean value P, (1,llij) = VAi2 vD/4 across genotypes, the degree of dominance, fre+ K? + (y1 + ~ J K P V A quency, and its displacement (that is, the effect + yiyj K P V A ~ +D (yi + yj) 5114 + ~ 1 J2~ 1 of substituting one allele for the other, see Morton and MacLean, 1974 for complete discussion with of the model). It should be noted that since the means of c and e are taken t o be zero, the phenJ1 = (p - 4) Ct (VA + VD) + 3VAuD + VDUD, otypic mean is the same as the mean of the major locus (Fig. 3). This is intuitively obvious 5 2 = [VA + VDi2 + (p - q)(YUD/2l2- vA/2 since everyone in the population has one of the and three genotypes defined by the diallelic major locus. Ct = p(f2 - fi) + q(f3 - f2) It has been suggested that such a model may provide the final resolution to the mode of inyo = -(1 - Kp1-l heritance of BP disorder (Reich et al., 1983; y1 = Kp-I McGuffin and Reich, 1983). The present study is the first to assess the fit of a combined model uniquely specifies the four underlying paramto the familial distribution of manic-depressive eters if Kp, Po (111)and one each of Po Cllij) illness. and P, (1,llij) are known with VA # 0. The Table 1summarizes the models used and paconditional probabilities Po (111)and Po Cl)ij) rameters estimated in the present study. All refer to the prevalence of affected offspring of parameters are estimated by the method of affected individuals and prevalence of affected maximum likelihood. The two multifactorial offspring given parental mating type, respec- models differ only in estimation of correlations tively. between relatives. The first model (MF1) asIn fact, it is further shown that P, (1,llij) sumes a single familial background correlaneed only be known for one mating type as is tion, whereas MF2 estimates six possible corusually the case (especially for rare recessive relations. By definition the familial correlations traits). Thus, the present study utilizes comare assumed to be zero for the single major plex segregation analysis to examine the patlocus model, although the three genotypic petern of distribution of the trait within nuclear netrances are allowed to vary between the sexes. families and to maximize the log, likelihood For the two combined models the six specific over families, rather than just correlations between pairs of family members. Combined model (CM). Finally, the underlying genetic mechanism may represent a comCOMBINED MODEL bination of the two preceding models; a major locus and an additive multifactorial background that augments correlations between relatives and contributes to liability (Morton and Maclean, 1974). Under this model, the presence of a dichotomous trait (x) is the result of an effect due to a major locus (g), background correlations between relatives (c) due to multiple additive genetic factors, and an environmental component. x = g + c + e. % 92 93 The environmental effect may be subdivided Fig. 3. The combined model. Phenotype is determined into common familial and random environ- by a major locus in conjunction with multifactorial backments, each with a normal distribution of mean ground. See text for discussion. + zyxwvu zy zyxwv zyxwvuts zyxwvut zyxwvu zyxwvu 55 GENETICS OF BIPOLAR DISORDER T A B LE 1 , Models used and parameters estimated Model Multifactorial (MF1) Multifactorial (MF2) Single major locus (SML) Combined (CM1) Combined (CM2) ~ ~~~ Parameters estimated Number of parameters d threshold, 9 threshold, 1 familial correlation 6 threshold, P threshold, 6 familial correlations' gene frequency (q), 6 penetrances' gene frequency (q), 6 penetrances,' 1 familial correlation gene frequency (q), 6 penetrances,' 6 familial correlations' 3 'Parentioffspring = male-male, male-female, male-female. 2Males = fim,fz,, f3,,,; females = fir, fir, f3r female-female; penetrance values are allowed to vary but only CM2 estimates multiple familial correlations. In all these preliminary analyses an ascertainment probability (1~)of 0.0001 is assumed (i.e., the probability of an individual being affected and a proband is small ( 1 + ~ 01, so that we have effectively single ascertainment with one proband per family). One advantage of the present approach is that a single likelihood value defines the fit of each model used. It is thus possible to test statistically the relative fit of nested models to a single data set. The appropriate test follows a chi-square distribution. x2 = 8 7 8 13 Sibisib = male-male, male-female, fe- TABLE 2. Parameter estimates under multifactorial models Estimates MF1 Correlations' 0.43 K, (MIF) 0.02810.042 - 700.1 25.8 Likelihood XZ zy MF2 0.21 0.57 0.58 0.33 0.30 0.37 0.0310.04 687.2 (p < 0.01) ~ zyxwvu zy - 2 (L, - Lz), where L, and Lz are the observed log, likelihoods for the two models being compared, and the degrees of freedom are defined by the difference in number of parameters estimated by each model. The possibility exists, then, to select the most adequate model to describe the familial distribution of BP disorder, and test the significance of its superiority over competing models. RESULTS AND DISCUSSION Table 2 summarizes the results of the fitting of the two multifactorial models to the data. It is worth noting that while both MF models predict population prevalences of approximately 0.03 and 0.04 for males and females, respectively, the observed Kp for the St. Louis data was 0.034 for both sexes. The small difference between the estimated values, and their symmetry around the observed prevalence, testifies to the roughly equal prevalence rates in both sexes for this disorder. The moderately high familial correlations estimated by both models is not surprising; it reflects the well known elevation in morbid risk to BP disorder 'Parent/offspring = MM, MF, FF; sib/sib = MM, MF, FF in first degree relatives of probands. Somewhat surprising are the high mother-offspring correlations in MF2, suggesting some form of maternal effect. Indeed, such a pattern led earlier workers to postulate X-linkage for BP disorder (e.g., Winokur and Tanna, 1969; Mendlewicz and Fleiss, 1974; Mendlewicz et al., 1972). However, X-linkage for BP disorder has not been unequivocally demonstrated in general, and several cases of father-son transmission are now known. Thus, undue importance should not be ascribed t o the high mother-offspring correlations reported here. Perhaps of greater interest is the fact that the MF2 model, with the full complement of familial correlations, provides a significantly better fit than MF1. A summary of the results from the single major locus model are presented in Table 3. With an estimated gene frequency of 0.031, the male and female population prevalences of 0.029 and 0.038 are virtually identical to those estimated under the assumptions of multifactorial inheritance (see Table 2). Given the relatively low gene frequency, all of the homozygotes are predicted to be affected. However, considerably larger proportions of females than males are affected as heterozygotes, while a slight excess of male sporadics is predicted under this model. 56 zyxwv zyxwv zyxwvutsrqpon zyxwv zyx zy zyxwvu zyxwvut zyxwvutsrqp D.H. O’ROURKE, P. McGUFFIN, AND 1:REICH T A B L E 3 . Parameter estimates under single major locus model Estimates Males Females Penetrances fi f2 f3 KP q Likelihood 0.006 0.010 0.306 0.999 0.029 0.530 0.999 0.038 0.031 - 683.9 T A B L E 4 . Percentage contrrbutron to pool of affected hy genotype and sex for S M L Genotype Male Female “Sporadics” (pLfl/K,I Heterozygotes ( 2 p q t K,I Homozygotes (qLfiK,I 32 8 14 I 63 8 82 8 33 25 T A B L E 5 . Parameter estimates under wnihined models In Table 4 these values, taken together with the predicted Kp values, reveal more clearly the subtle sex differences. The majority of affected individuals of both sexes are predicted to be heterozygous under this model, although nearly one-third are found to be sporadics among males. Finally, two separate combined models were fitted to these data. Table 5 summarizes these results. A somewhat surprising result is obtained. For both CM1 and CM2 the familial correlation estimates reduce to essentially zero, approximating a simple, single major locus model. In fact, all remaining parameter estimates are found to be exactly the same as those noted earlier for the SML. Given the lack of evidence for appreciable background familial correlations contributing to liability in the combined models, the proportions affected by sex and genotype under these models are the same as those seen earlier for the SML (see Table 4). Having reviewed the analyses of these data sets through the implementation of multifactorial and complex segregation models, it is now possible to test statistically the relative efficacy of each model. Table 6 presents the results of these tests. MF1 provides a statistically poorer fit than either of the combined models. Since CM1 is, in effect, a single major locus model, the latter would be viewed as providing a statistically better fit than MF1 as well. This is not a trivial point since the SML is not, technically, a model nested within the parameterization of the MF models and, hence, a direct statistical comparison is precluded. Here the clear cut distinctions end. ALthough it is not possible to compare MF2 and CM1 directly (they estimate equal numbers of parameters and, hence, have no degrees of freedom), MF2 is not significantly different from CM2. Since the two combined models and the SML model are statistically indistinguishable, all four of the remaining models must be considered to fit the data equally well. It is not surprising, then, that numerous authors, using transmis- Estimates CMl CM2 Correlations’ 00 Penetrancesl 0 010 0 304 0 999 0 006 0 530 0 999 0 02910 038 0 031 683 9 000000 000000 0 010 0 304 0 999 0 006 0 530 0 999 0 029/0 038 0 031 683 9 K , iM Fi 9 1~kelihood T A B L E 6. Direct coinparison of likelihood i.alues for models fitted to hipolar afrec.tir,e disiircier f a m i l y studies Models a n d likelihoods’ Chi-square ~ M F l I700 11-CM1 M F l 1700 11-CM2 MF2 I687 2I-CM2 1638 91 1683 9 1 16839 1 ‘Absolute values of Ilkellhim!. 32 4 32 4 66 df Probability 5 10 5 LOO1 ~~ < 0 01 ,005 in o a r e n t h e w s sion models similar to those described here, have argued for single major locus inheritance (Winokur et al., 1969; Crowe and Smouse, 1977; Gershon, 1975a,b,c), including X-linkage (Mendlewicz and Fleiss, 1974; Mendlewicz et al., 1972; Winokur and Tanna, 1969; Baron, 1977), as well as multifactorial transmission (Slater and Tsuang, 1968; Bucher and Elston, 1981; Bucher et al., 1981; Baron, 1980). The present analyses suggest that a combined model may be added to the list, bearing in mind that the combined models employed provided essentially a single major locus with little evidence for the importance of a multifactorial background. Although only MF1 may be statistically rejected using these data, it is worth noting that the presence of a major locus, with or without a polygenic background, has the highest likelihood based on the combined St. Louis data. This at least suggests that it is a reasonable working hypothesis. Indeed, such a working hypothesis is not inconsistent with recent reports of linkage of a disease susceptibility locus for BP disorder to the major histocornpatability GENETICS OF BIPOLAR DISORDER complex on the short arm of chromosome 6, although these reports need to be verified (e.g., Smeraldi and Bellodi, 1981; Weitkamp et al., 1981).For example it has been shown recently (Suarez and Van Eerdewegh, 1981) that misspecification of the mode of inheritance of a disease susceptibility locus may give spuriously high lod scores for linkage to a known marker locus. Since it is by no means clear whether the genetic etiology of BP disorder is multifactorial, the result of a single major locus, or some combination, much less whether the putative single locus is dominantly or recessively transmitted, such reports must await further confirmation and testing. The results presented here are concordant with some previous studies but at variance with others. Gershon et al. (1975b, 1976; Gershon and Leibowitz, 1975)fitted MF and SML models to incidence data on UP and BP disorders in the Jewish population of Jerusalem. In these studies, both MF and SML models were found to predict incidence of affectation adequately in relatives of probands. Although these results are similar to those presented here, methodological differences preclude a more rigorous comparison of the studies. Gershon et al. (1975b, 1976; Gershon and Leibowitz, 1975) were in effect testing the validity of the Leonhard (1959) dichotomy of polarity for affective disorders. Thus, they used a two-threshold model for both the MF and SML analyses with BP and UP disorders representing narrow and broad forms of a disease on a single liability scale. Given this major distinction between the underlying assumptions and diagnostic criteria used by Gershon and colleagues and the present study, it is interesting that the results and inferences are similar. It should be noted, however, that in the present case the less complex form of the MF model may be statistically rejected. Moreover, the greater likelihood associated with those models containing a major locus suggests that further examination of the role of a major locus in this disorder is warranted. In contradistinction to this position, Bucher and Elston (1981)and Bucher et al. (1981)have explicitly rejected the notion of involvement of a major locus in the etiology of BP disorder. Once again, methodological differences make direct comparisons of analyses and inferences difficult. The methodology used by Bucher and Elston (1981) and Bucher et al. (1981) is Elston’s approach to segregation analysis (e.g., Elston 1980; Elston and Stewart 1971).This is a different parameterization of the SML than that used in the present study, and may ac- 57 count for some of the differences in result. Additionally, rather than testing the relative fit of competing models directly, Bucher and colleagues test a Mendelian model against a general unrestricted model. That is, one in which the transmission probabilities of each phenotype are not constrained by Mendelian values. By not being able to reject this general unrestricted model, Bucher and coworkers suggest that a major locus hypothesis for BP disorder is untenable. However, no model other than the SML was tested. Moreover, the nature of the test suggests that deviations of the individual data sets from Hardy-Weinberg equilibrium conditions could give rise to the results obtained. Further work in this area is imperative since Bucher and coworkers used some of the same data utilized in the present report (Helzer and Winokur, 1974;Winokur et al., 19691, but came to dramatically different conclusions. In addition, the effects of small sample size, and the pooling of separate studies that utilize different diagnostic criteria on the results of genetic model fitting to incidence data require further investigation (see O’Rourkeet al., 1982 for brief review). The approximation of both combined models to a major locus model in the present analysis not only suggests the importance of a major gene in this behavioral disorder, but raises once again the question of whether this putative disease susceptibility locus is autosomal or Xlinked. Evaluation of this question is outside the scope of the present paper. However, Van Eerdewegh et al. (1980) have recently examined the fit of three separate X-linked threshold models to familial data on bipolar affective disorder. Two of these models were single threshold models and differed only in regard to whether individuals suffering unipolar depression were considered affected or unaffected. All families were ascertained through a BP proband. Using three separate data sets, Van Eerdewegh et al. (1980) found that Xchromosome single locus models did not consistently describe the distribution of affected relatives of BP probands. They note that their results may suggest etiological heterogeneity. We are currently reexamining the St. Louis data in light of the present reports’ suggestion of major locus involvement in order to evaluate the relative efficacy of autosomal versus Xchromosome transmission models. The result of this research will appear elsewhere. Finally, bipolar affective disorder is known t o have a variable age of onset. Unfortunately, adjustment for age-dependent penetrance is not 58 D.H. O’ROURKE, P. McGUFFIN, AND T. REICH possible with the analytical techniques used in this analysis. We are, however, modifying the routines employed here to permit adjustment for the age of onset distribution in order to evaluate what effect this may have on the reported results. The age distribution of the probands and family members is such that we do not believe age adjustment will radically alter the results. SUMMARY AND CONCLUSIONS of multifactorial inheritance or single major locus involvement through complex segregation analysis. Different programs use different algorithms and different parameterizations of the models. It is not yet clear that they all produce concordant results. Until some degree of standardization is achieved, results may be considered tenative. (6) Finally, results and hypotheses generated by analysis of data using one analytic technique should be continually retested using different analytical methods based on different sets of assumptions. We are currently analyzing these data using several different methodologies in order t o more fully elucidate the genetic mechanisms involved in bipolar affective disorder. zyxwvuts zyxwv zyxwv While these preliminary analyses suggest that a major locus may be important in the etiology of bipolar affective disorder, several cautionary notes need be appended. (1)We have not tested the “fit”of a specific model to a set of data. Rather, we have attempted t o evaluate the relative value of several transmission models to account for the distribution of bipolar affective disorder in families. (2) Although we have carried out these analyses assuming an ascertainment probability ( 7 ~ )of 0.0001, we cannot know that this is, in fact, the correct value. To test our assumption, we have reanalyzed the St. Louis data under a range of different values of 7~ (range = 0.0001-0.999), and the results are relatively invariant. (3) Although BP disorder has a variable age of onset, we were unable to use these genetic models on age-corrected data. However, since computational constraints placed a limit on sibship size to five, no family member under the minimum age a t risk (15 years) was included in the analysis. Moreover, in those sibships where more than five members had entered the risk period, the youngest were deleted so as to maximize subjects who had passed through the maximum period of risk. Since no birth order effect has been noted for BP disorder, we do not believe this introduces any bias. (4)Diagnostic criteria are clearly important. In the present study, we have considered relatives affected if they were diagnosed as BP or primary depression only. When only BP relatives are considered affected, none of the five models used are statistically distinguishable. Given the possibility of genetic heterogeneity in BP disorder, further work in this area is imperative. (5) These analyses were carried out using computer programs developed a t Washington University School of Medicine in St. Louis under the supervision of the third author. These programs are but one set out of several that allow analysis of data under the assumptions ACKNOWLEDGMENTS We gratefully acknowledge the assistance of Dr. John Helzer, and Mr. Joe Mullaney for his able programming skills. This work was supported in part by USPHS grants MH31302, MH14677, MH25430, and GM28067, and an MRC (U.K.) Fellowship (Dr. P. McGuffin). LITERATURE CITED Akiskal, HS, Djenderedjan, AH, Rosenthal, RH, and Khani, MK (1977) Validating criteria for inclusion in the bipolar affective group. Am. J. Psychiat. 134:1227-1233. Angst, J (19661 Zur Atiologie and Nosologie endogener depressiver Psychosen. In: Monographien aus der Neurologie und Psychiatrie, No. 122. Berlin: Springer Verlag, pp. 1-118. Balis, GU, Wurmer, L. McDaniel, E (1978) Clinical Psychopathology. Boston: Butterworth. Baron, M (1977) Linkage between a n X-chromosome marker (deutan color blindness) and bipolar affective illness. Arch. Gen. Psychiat. 34:721-725. Baron, M (1980) Genetic models of affective disorder: application to twin data. Acta Genet. Med. Gemellol. 29:289-294. Bucher, KD, and Elston, RC (19811 The transmission of manic-depressive illness. I. Theory, description of the model and summary of results. J. Psychiat. Res. 16:53-63. Bucher, KD, Elston, RC, Green, R, Whybrow, P, Helzer, J, Reich, T, Clayton, P, and Winokur, G (1981) The transmission of manic-depressive illness. 11. Segregation analysis of three sets of family data. J. Psychiat. Res. 16:6578. Clonginger, CR, Reich, T, Suarez, BK, Rice, J, Gottesman, I1 (1983) The principles of psychiatric genetics. In M. Shepherd (edl: Handbook of Psychiatry, vol. 5: The Scientific Foundation of Psychiatry. Cambridge: Cambridge University Press (in press). Crittenden, LB (1961) An interpretation of familial aggregation based on multiple genetic and environmental factors. Ann. N.Y. Acad. Sci 91:769-780. Crowe, RR, and Smouse, PE (1977) The genetic implications of age-dependent penetrance in manic-depressive illness. J. Psychiat. Res. 13:273-285. Elston, RC (1980) Segregation analysis. In J H Mielke and MH Crawford (eds): Current Developments in Anthropological Genetics. New York: Plenum, pp. 327-354. zyxwvut GENETICS OF BIPOLAR DISORDER Elston, RC, and Stewart, J (1971) A general model for the genetic analysis of pedigree data. Hum. Hered. 21t523-542. Falconer, DS (1960) Introduction to Quantitative Genetics. London: Oliver and Boyd. Falconer, DS (1965) The inheritance of liability to certain diseases estimated from incidence among relatives. Ann. Hum. Genet. 29t51-76. Falconer, DS (1967) The inheritance of liability to diseases with variable age of onset with particular reference to diabetes mellitus. Ann. Hum. Genet. 31:l-20. Feighner, J P , Robins, E, Guze, SB, Woodruff, RA, Winokur, G, and Munoz, R (1972) Diagnostic criteria for use in psychiatric research. Arch. Gen. Psychiat. 26t57-63. Fishman, PM, Reich, T, Suarez, BK, and James, J W (1978) A note on the essential parameters of the two-allele autosomal locus model. Am. J. Hum. Genet. 30:283-292. Gershon, E, Baron, M, and Leckman, J (1975a) Genetic models of the transmission of affective disorders. J. Psychiat. Res. 12t301-317. Gershon, E, and Leibowitz, T (1975) Sociocultural and demographic correlates of affective disorders in Jerusalem. J. Psychiat. Res. 12~37-50. Gershon, E, Mark, A, Cohen, N, Belizon, N, Baron, M, and Knobe, K (1975b) Transmitted factors in the morbid risk of affective disorders: a controlled study. J. Psychiat. Res. 12t283-299. Gershon, E, Bunney, WE, Leckman, JF, Van Eerdewegh, M, and DeBauche BA (1976) The inheritance of affective disorders: a review of data and hypotheses. Behav. Genet. 6227-261. Helzer, J, and Winokur, G (1974) A family interview study of male manic-depressives. Arch. Gen. Psychiat. 31t73-77. James, J W (1971) Frequency in relatives for an all-or-none trait. Ann. Hum. Genet. 35r47-49. Kempthorne, 0 (1975) An Introduction to Genetic Statistics. New York: John Wiley and Sons, Inc. Leonhard, K (1959) Aufteilung der endogenen Psychosen. Akademie Verlag Berlin. McGuffin, P, and Reich, T (1983) Psychopathology and Genetics. In: HE Adams and PB Sutker (eds): Comprehensive Handbook of Psychopathology. New York: Plenum Press (in press). Mendlewicz, J , and Fleiss, J (1974) Linkage studies with Xchromosome markers in bipolar (manic-depressive) and unipolar (depressive) illness. Biol. Psychiat. 9%-294. Mendlewicz, J , Fleiss, J, and Fieve, RR (1972) Evidence for X-linkage in the transmission of manic-depressive illness. J. A. M. A. 222:1624-1627. Morton, NE, and MacLean, CJ (1974) Analysis of family resemblance. 111. Complex segregation analysis of quan- zyxw 59 titative traits. Amer. J. Hum Genet. 26t489-503. ORourke, DH, Gottesman, 11, Suarez, BK, Rice, J, and Reich, T (1982) Refutation of the general single-locus model for the etiology of schizophrenia. Am. J. Hum. Genet. 34:630-649. Perris, C (1966) A study of bipolar (manic-depressive) and unipolar recurrent depressive psychoses. Acta Psychiat. Scand. Suppl. No. 194. Reich, T, James, J W, and Morris, CA (1972) The use of multiple thresholds in determining the mode of transmission of semicontinuous traits. Ann. Hum. Gen. 36:163-184. Reich, T, Cloninger, CR, and Guze, SB (1975) The multifactorial model of disease transmission: I. Description of the model and its use in psychiatry. Br. J. Psychiat. 127:l-10. Reich, T, Cloninger, CR, Suarez, B, and Rice, J (1983) Genetics of the affective psychoses. In: J K Wing (ed): Handbook of Psychiatry. London: Cambridge University Press, in press. Slater, E, and Roth, M (1969) Clinical Psychiatry. New York: Williams and Wilkins. Slater, E, and Tsuang, MT (1968) Abnormality on paternal and maternal sides: observations in schizophrenia and manic depression. J. Med. Genet. 6t197-199. Smeraldi, E, and Bellodi, L (1981) Possible linkage between primary affective disorder susceptibility locus and HLA haplotypes. Am. J. Psychiat. 138t1232-1234. Smeraldi, E, Negri, F, and Melic, AM (1977) A genetic study of affective disorders. Acta Psychiat. Scand. 56:382-398. Suarez, BK, and Van Eerdewegh, P (1981) Type I (insulin dependent) diabetes mellitus. Is there strong evidence for a non-HLA linked gene? Diabetologia 20:524-529. Van Eerdewegh, MM, Gershon, ES, and Van Eerdewegh, PM (1980) X-chromosome threshold models of bipolar manic-depressive illness. J. Psychiat. Res. 15215-238. Weitkamp, LR, Stancer, HC, Persad, E, Flood, C, and Guttormsen, S (1981) Depressive disorders and HLA: a gene on chromosome 6 thay can affect behavior. N. Engl. J. Med. 305:1301-1306. Winokur, G, and Clayton, P (1967) Family history studies. I. Two types of affective disorders separated according to genetic and clinical factors. Recent Adv. Biol. Psychiat. 9:35-50. Winokur, G, Clayton, P, and Reich, T (1969) Manic-Depressive Illness. St. Louis: Mosby. Winokur, G, and Tanna, J (1969) Possible role of X-linked dominant factors in manic-depressive disease. Dis. Nerv. Sys. 3039. zyxwvutsrqp zy