|Additional related information may be found at:|
|Neuropsychopharmacology: The Fifth Generation of Progress|
Raymond R. Crowe, MD
Recombinant DNA technology has enabled us to explore the human genome in ways that would not have been thought possible twenty years ago. Widespread DNA polymorphism has been exploited through the use of restriction enzymes and, more recently, the polymerase chain reaction to create genetic markers and detect linkage to a growing list of human disease genes. These genetic markers are defining a human linkage map of increasingly greater resolution. Genes expressed in the brain are being cloned in large numbers and will ultimately be placed on the linkage map. This progress with the human genome is being paralleled by similar developments in rodents and other potential animal models. The end result of all of these efforts will be the creation of a set of tools that will enable us to identify disease producing mutations in increasingly complex genetic diseases.
These developments have obvious implications for Psychiatry. Collaboration efforts are underway to collect pedigrees, bank cell lines, and search for genes contributing to schizophrenia, bipolar illness, alcoholism, and Alzheimer's disease. Unlike the diseases that have been successfully mapped, behavioral disorders present unique challenges for molecular genetic strategies. These are illustrated by the number of linkage findings in Psychiatry that cannot be replicated. The special problems of applying molecular strategies to complex traits can be appreciated with an understanding of the genetic epidemiology of mental disorders and the theoretical underpinnings of the molecular genetic strategies. Understanding the interrelationship of these two disciplines is crucial to designing strategies that will avoid the pitfalls of the past and maximize the chances of success in the future.
Family, twin and adoption studies have supplied the raw material for a substantial foundation of genetic epidemiology. Family and twin studies can provide information on the penetrance of the genotype, as well as its full range of expression. Twin and adoption studies separate genetic from environmental transmission. Segregation analyses define the range of genetic hypotheses that can account for the observed transmission patterns. A growing number of molecular genetic studies aimed at identifying disease genes is now being added to this foundation of classical genetics.
Manic Depressive Illness
McGuffin and Katz (47) reviewed 12 family studies of bipolar illness and found the average morbidity risk among first-degree relatives to be 7.8% for bipolar and 11.4% for unipolar illness. This is a substantial increase over the respective population rates of approximately 1% and 3%, cited by the same authors. Probandwise monozygotic twin concordance rates for bipolar illness range from 62 to 72 percent, and an additional 18 to 25 percent have unipolar illness (9, 84). Comparative dizygotic concordance rates range from zero to 8 percent for bipolar illness with an additional unipolar range from zero to 11 percent. The role of genes in bipolar illness is further supported by observations on adoptees. Increased rates of both bipolar and unipolar illness are seen in biological parents but not in adoptive parents of bipolar adoptees, indicating that the family and twin data indeed reflect the action of genes (49). Collectively, the evidence points to a genetic predisposition with a relatively high penetrance and a range of expression including bipolar and unipolar affective disorder.
The most extensive observations on the bipolar phenotype and its transmission come from a comprehensive family study of bipolar I, bipolar II, schizoaffective, unipolar and control probands (26). The first four disorders, with the possible addition of cyclothymic personality disorder, aggregate in the families of schizoaffective and bipolar probands. Genetic analyses of these pedigrees indicate that a multifactorial-threshold model provides a plausible explanation of the transmission and that a single locus model can be excluded (26, 29). According to the model, schizoaffective disorder is the most severe form of the illness, followed by bipolar I, bipolar II and unipolar depressive disorder.
The discovery of large bipolar pedigrees among the Old Order Amish provides a rare opportunity to study manic-depressive illness in a population isolate with an apparently autosomal dominant form of the disease. Such genetically isolated populations are ideal for linkage studies because their gene pools are typically more homogeneous than those of the general population. Indeed, a gene appeared to have been found when linkage was reported between bipolar illness and two DNA markers (HRAS and INS) on the short arm of chromosome 11 (21).
In retrospect, failures to replicate the finding should have sounded a note of caution, but at the time, the discrepancy between the studies was felt to be evidence of genetic heterogeneity. This explanation seemed plausible because of the isolated nature of the Amish population; the linkage could represent a rare gene that would be difficult to replicate outside of that group. Therefore, the pedigree was extended to include two new branches and, at the same time, the diagnoses were updated (36). Linkage analyses based on these revisions resulted in a decisive erosion of the original lod score to the point that a disease locus for bipolar illness could now be excluded from the region in question. The lod scores at the two marker loci that had previously supported linkage, HRAS and INS, fell from 4.08 to -9.31 and from 2.63 to -7.75, respectively. The reversal of support for linkage was accounted for predominantly by two changes: two relatives who were unaffected in the original analyses subsequently became ill, and the two pedigree extensions introduced a number of new affected relatives. The net effect of both changes was to introduce a number of obligate recombinants; i.e., crossovers between affectation status and marker typings.
How can the steep fall in the lod scores be reconciled with the strong original evidence of linkage? One possibly is the inclusion of unipolar depression in the affected phenotype. Though unipolar depression is an expression of the bipolar genotype, the high population prevalence of depression could introduce cases into the pedigree that are unrelated to the bipolar gene. Indeed, it was unipolar depressives in the new pedigree extensions that were most responsible for the lod score changes. Alternatively, a second gene for bipolar illness could have entered the pedigree through marriage. The existence of positive assortative mating for affective disorder makes this a plausible possibility. Finally, the lod scores supporting linkage may have represented a false positive finding. The failure of other linkage reports in psychiatry to be replicated supports this explanation.
The X-linkage hypothesis of manic depressive illness has been an intriguing and elusive one since evidence for it first appeared over 25 years ago. Since that time, attempts to replicate X linkage have produced a confusing melange of supportive and contradictory results. A report of five Israeli pedigrees provided remarkably strong evidence for linkage to the G6PD and color blindness loci (located at Xq27-q28), supported by lod scores of 7.52 and greater (5). However, when three of the original pedigrees were reinterviewed and retyped with DNA markers the overall evidence for linkage declined dramatically. Lod scores in two of the pedigrees became negative, and in the third pedigree the score remained in the range of 2.00 (6).
Several reasons for the change in lod scores were identified (6). Females could not be typed at the G6PD and color blindness loci in the first study because DNA markers were not used. When these pedigree members were typed in the reanalysis a number of new recombinants were found. Diagnostic changes resulting from new interviews created additional recombinants. In addition, several key family members that had supported linkage were lost to reanalysis. Finally, a mother of two affected male offspring was found to be homozygous at the G6PD locus, and thus her sons, who had supported linkage in the first analysis, became uninformative.
The linkage studies just reviewed have tested the hypothesis that a major gene contributes to vulnerability to manic-depressive illness. A major gene may be thought of as necessary for the development of the disease, at least in the pedigree or pedigrees being studied. A locus of this nature would be expected to produce a large lod score and to be easily replicable in other pedigrees transmitting the disease at the same locus. Now that a number of genome scans of manic-depressive illness have been completed without identifying a major locus, it has become increasingly apparent that a different genetic mechanism must be anticipated. Most likely, the genetics will be explained by multiple genes of small effect acting in concert to cause the disease—oligogenic inheritance. Unlike major genes, each locus individually has a small effect on liability, no locus is necessary for the disease, and the statistical evidence of linkage at any locus is likely to be modest. An important consequence is that linkages will be difficult to replicate. The statistical power to detect any one locus is low; therefore, only a proportion of studies will detect it and nonreplications will be common (80). As forbidding as this may sound, it is an important mechanism of inheritance because it accounts for the common genetic traits that cause disease—hypertension, hyperlipidemia, diabetes, and mental disorder. Furthermore, finding genes responsible for Mendelian disease has become so straightforward that it is only a matter of time until all of these diseases have been mapped, and genetically complex traits are emerging as a new frontier of human genetics. It is also a frontier that is moving forward. Type 1 diabetes mellitus, a genetically complex phenotype, has been analyzed by linkage and found to be caused by at least five genes: a major locus in the major histocompatibility complex, the insulin locus, and at least three additional vulnerability loci (19).
The most promising candidate region for manic-depressive illness at this time is the pericentromeric region of chromosome 18, where two groups have found evidence of linkage (8, 77), although others have not replicated the finding (28). The regions identified by both groups were approximately 40 cM in length so that what was replicated was a region rather than a specific locus. Stine et al (77) noted that the evidence for linkage to chromosome 18 came almost entirely from pedigrees in which the disease was inherited from the father's side of the family. Gershon et al (27) have noted the same parental effect in the pedigrees that originally suggested linkage on chromosome 18. Thus, both groups found evidence of linkage on chromosome 18 in the pericentromeric region as well as a parental transmission effect.
Are we entitled to conclude at last that linkage to genes for manic-depressive illness has been found? Lander and Kruglyak (42) provide some guidance to the interpretation of lod scores and significance levels obtained in genome scans of complex traits. A nominal p-level of 0.00002 corresponds to a genomewise significance level of 0.05; a finding this significant would be expected to occur by chance in one out of 20 genome searches. However, replications need only reach conventional significance levels because they are testing a focused hypothesis. The significance levels obtained by the two groups for the broad regions of provisional linkage on chromosome 18 do not support confirmed linkage according to these criteria. The significance levels are more convincing in the paternally transmitted pedigrees, and it would be tempting to call this a replication were it not for the problem that the putatively linked regions overlap by only a few centimorgans (27, 77). Future developments on chromosome 18 in manic-depressive illness certainly bear watching, but it is premature to conclude at this time that a linkage has been found.
Another region that bears watching appears on the long arm of chromosome 21. Straub et al (78) obtained a lod score of 3.41 in one family out of 47 being typed in a genome scan, but they also attained a p-level of 0.0003 for the entire pedigree set using a nonparametric linkage analysis. Detera-Wadleigh et al (20) have reported statistically significant evidence of linkage at the locus with the greatest evidence of linkage in the first study. Finally, Pekkarinen et al (62) have revived the X-linkage hypothesis. They found evidence for a susceptibility locus at Xq24-27.1 in a large bipolar pedigree from Finland. A 20 cM region at that location cosegregated with affective disorder to a statistically significant degree (empirical p = 0.0006). This would not have been a remarkable finding had it not occurred in a region where so much prior evidence for a bipolar locus already existed, and it does sustain hope that a bipolar locus may be found on the X chromosome yet.
Recent family studies using current interview methods and diagnostic criteria have arrived at reasonably similar estimates of the recurrence risk for schizophrenia. The Roscommon Family Study was based on a complete ascertainment of all cases of schizophrenia in county Roscommon, Ireland, and it found a morbidity risk for schizophrenia of 6.5% (S.E. 1.6%) among interviewed relatives (40). This rate is 13 times the rate of 0.5% found among the relatives of control probands, and 7.6 times the usually quoted population rate of schizophrenia of 0.85%.
Twin studies provide further evidence of genetic vulnerability. Despite the wide diagnostic variation across studies, when MZ and DZ concordances are compared within studies the MZ/DZ ratios are remarkably consistent and indicate that genes account for a median of 65 percent of the variance in the transmission of schizophrenia (37). Adoption studies further support the role of genes by demonstrating a correlation in liability between biological relatives separated at birth (38).
The full expression of the genetic liability to schizophrenia is seen in MZ cotwins of schizophrenics. Therefore, the spectrum of illnesses that maximizes the MZ:DZ concordance ratio provides a genetic definition of the schizophrenia spectrum. In the Maudsley twin series, the range of DSM-III diagnoses that maximizes this ratio includes schizophrenia, schizotypal personality disorder, affective disorder with mood incongruent psychosis, atypical psychosis, and schizophreniform disorder (23). Although schizoaffective disorder was not placed in the spectrum in this series, a consensus exists that chronic schizoaffective disorder is related to schizophrenia. In the Roscommon Family Study increased rates of schizophrenia were found in the relatives of probands with schizoaffective disorder, schizotypal personality disorder, psychotic affective disorder, and other nonaffective psychoses (40).
When schizophrenia is considered as a unitary syndrome, the preponderance of the evidence supports polygenic, rather than mendelian inheritance (68). Mendelian diseases tend to be rare because of genetic selection against the gene, whereas the prevalence of schizophrenia is nearly one percent. MZ concordance rates of mendelian diseases are approximately twice the DZ rates, so the observed ratios of over four-to-one for schizophrenia are more consistent with polygenes. Familial morbidity risks of mendelian diseases decrease by one-half with each increasing degree of genetic distance from the proband (i.e., from first to second to third degree relatives). The more rapid regression seen in families of schizophrenics suggests polygenic inheritance. Finally, segregation analyses have not supported a mendelian model (4). If schizophrenia is genetically homogeneous the evidence argues convincingly against mendelian inheritance, but it is unlikely to be homogeneous. If, as is likely, the causes are multiple, it is possible that rare mendelian genes for schizophrenia exist and could be found in multiply affected pedigrees. This possibility has fueled interest in searching for vulnerability genes with linkage.
The search for a major gene effect received an impetus from the report of a pedigree segregating a chromosome 5 translocation with schizophrenia (7). As a result of the translocation, both the proband and a maternal uncle were trisomic for 5q11-q13, raising the possibility that excess activity of a gene in this region might have contributed to the schizophrenia. A linkage study with DNA markers in this region in seven Icelandic and English pedigrees resulted in what appeared to be convincing evidence of linkage: a lod score of 3.22 (74). Moreover, when the affected phenotype was broadened to include schizophrenia spectrum disorders the lod score increased to 4.33, and when all psychiatric disorders were considered as affected it reached 6.49. The last finding is puzzling because the inclusion of genetically unrelated disorders should result in a reduction in the lod score.
At first, when the chromosome 5 finding could not be replicated the nonreplications were attributed to genetic heterogeneity, but as failures to replicate continued to appear, heterogeneity became an increasingly unlikely explanation. Another argument against heterogeneity as an explanation is that it should be detectable within studies, but a meta-analysis found that the published studies were internally homogeneous but demonstrated heterogeneity between studies. The heterogeneity was accounted for by differences between the study reporting linkage and those attempting to replicate it (48). Since the genetic structure of the Icelandic population is not unique, there should be no reason for finding heterogeneity between these studies, and the most likely explanation is that the original lod scores were spuriously elevated.
As in manic-depressive illness, recent linkage studies of schizophrenia have made an increasingly plausible case for oligogenic inheritance. Several genomewide searches have now been completed without finding evidence of a major disease locus. Presently, the strongest candidate for a susceptibility locus lies on the short arm of chromosome 6. Straub et al (79) found evidence of linkage over approximately 40 cM in the region of chromosome 6p24-p22. That the maximum lod score obtained from 265 pedigrees was 3.51(adjusted for heterogeneity) indicates that the locus makes a relatively minor contribution to disease susceptibility. The Schizophrenia Linkage Collaboration Group (70) examined this region in 567 pedigrees contributed by 14 centers and obtained a maximum lod score of 2.19 in pedigrees exclusive of those used to make the original finding. The finding of linkage over regions as lengthy as those found in schizophrenia and bipolar illness presents a formidable obstacle to finding genes since one centimorgan equals approximately one million basepairs of DNA on average, and may contain as many as 30 genes.
Three additional loci showing initial support for linkage have now been followed up with large collaboration efforts to replicate them. These include a region on chromosome 22q12-q13.1 (65), which was supported by one collaboration (71), but not by another (66). Another promising region, located at chromosome 8p22-p21, was replicated (67, 70).
The linkage studies reviewed so far have searched for linkage to schizophrenia per se, but schizophrenia is characterized by a number of neurophysiological abnormalities that may provide better targets for linkage than the disease itself (Freedman et al, 1997). For example, patients with schizophrenia do not attenuate an evoked EEG potential (P50) as normals do, and the abnormality is found in their relatives in a distribution that is compatible with the trait being a vulnerability marker for schizophrenia. Nicotine can correct the abnormality, and this effect is mediated by the alpha7 nicotinic receptor. A polymorphism at this receptor locus is linked to the P50 deviation in schizophrenia pedigrees at a high level of statistical confidence (lod score 5.3). The ultimate question is whether the alpha7 receptor is linked to schizophrenia, but that question cannot be answered definitively at this time because the lod score for this analysis in the same pedigrees was 1.33, which is consistent with linkage but inconclusive (25).
Tourette's Syndrome (TS) is often familial and extensive pedigrees of the disease have been reported (41). Family members have an increased risk of Tourette's syndrome, chronic motor tics and obsessive-compulsive disorder. In one study of first-degree relatives of TS probands, the morbidity risk for Tourette's syndrome was 8.7%, for chronic tics 17.3%, and for obsessive-compulsive disorder 11.5%. Thus, the aggregate morbidity risk was 37.5%: over seven times the population rate of 5.2% found in the same study (61). The ratio of affected males to females was 4-to-1 for Tourette's syndrome and 1-to-2 for obsessive-compulsive disorder.
The genetic vulnerability is underscored by a MZ twin concordance rate of 53% for Tourette's syndrome among co-twins of TS probands and 77% for TS or chronic motor tics. The respective dizygotic concordances are 8% and 23% (64). Segregation analysis supports an autosomal dominant gene as the most likely mode of transmission (60).
The close fit to autosomal dominant transmission, the high penetrance, and the existence of large multiplex pedigrees make Tourette's syndrome an ideal choice for linkage studies, and the interim results of a collaboration genome search have been published (58). Ten large pedigrees typed with 228 markers have covered at least half, and perhaps as much as two-thirds, of the genome, but no convincing evidence of linkage has been found. Now that more informative and detailed human genome maps are available a complete genome search of Tourette's disease should be forthcoming.
The familial nature of panic disorder is supported by family history, with patients reporting secondary cases in 12 to 15 percent of their first-degree relatives (15). Interviews with family members found an age-corrected morbidity risk of 17% for panic disorder or agoraphobia with panic attacks, and the rate was 24% when subsyndromal panic attacks were included (15). The mean age of onset is 25 years and the sex ratio (F/M) is approximately 2-to-1. The range of disorders found in family members was limited to panic disorder, agoraphobia and subsyndromal panic attacks (54).
Limited twin data, based on 13 MZ and 16 DZ pairs, are consistent with the familial findings, both with respect to a genetic predisposition and its range of expression. Panic disorder, subsyndromal panic attacks, and agoraphobia appeared among monozygotic cotwins of panic disorder twins, and the aggregate concordance rate was 31%, compared with 0% for dizygotic cotwins (84). Segregation analyses of pedigrees and threshold models applied to aggregate recurrence rates are consistent with either a single-locus or a multifactorial mode of transmission (15, 59). Linkage studies of panic disorder are currently underway at a number of centers. In one study, approximately 30% of the genome has been screened with RFLP markers in 10-14 pedigrees, but no evidence of linkage was found (16).
Controlled family studies of alcoholism based on current interview methods and diagnostic criteria support a seven-fold increase in risk of alcoholism in first-degree relatives of alcoholics over controls, and a five-fold risk increase in male relatives over females (50). A representative study found the following rates of alcoholism among first-degree relatives of alcoholic probands: fathers 16.1%, brothers 12.4%, mothers 1.6%, and sisters 1.0% (63). The aggregate control rate was 1.2%.
Kendler's sample of female twins provides the best available estimate of the heritability of alcoholism because it was drawn from a population sample and all participating twins were interviewed (39). Probandwise MZ concordance rates ranged from 26 to 47 percent, depending on strictness of diagnosis, and the corresponding DZ rates ranged from 12 to 32 percent. These findings indicate that genes account for 50 to 60 percent of the variance in the transmission of alcoholism (39).
The role of genes in the etiology of alcoholism is further supported by adoption studies. Biological relatives separated by adoption show strong correlations in the diagnosis of alcoholism (31), and in temperance board registrations for alcohol abuse (14). The last study identified two patterns of inheritance: type 1 alcoholism was correlated with postnatal environment but not with criminality in biological fathers, while type 2 alcoholism followed the opposite pattern. In contrast to type 2 alcoholism, type 1 was more common, had a later onset, was less severe, and was more likely to include women.
A segregation analysis of 35 pedigrees found that neither a mendelian nor a polygenic model could account for the transmission (1). Since the disease is demonstrably genetic, it is puzzling that neither single nor multiple locus models could explain the transmission. The authors speculated that this result could reflect oligogenic inheritance (i.e., a small number of loci), phenocopies, sex effects, or genetic heterogeneity.
Ethanol is eliminated from the body by first being converted to acetaldehyde by the enzyme alcohol dehydrogenase (ADH), and acetaldehyde is then broken down to acetate and water by aldehyde dehydrogenase (ALDH). Genetic polymorphisms in both enzymes correlate with their activity. Three ADH alleles (ADH1-3) are responsible for marked variation in the elimination rate of ethanol. The breakdown of acetaldehyde is catalyzed primarily by the mitochondrial enzyme ALDH2. A dominant allele (ALDH2*2) confers an absence of detectable enzyme activity. The ALDH2*2 allele is prevalent in Asians and causes a flushing response to ethanol ingestion, similar that caused by the drug disulfiram, which inhibits ALDH activity. This mechanism appears to contribute to the low rate of alcoholism among Asians (75). Interestingly, compared to nonalcoholics, Chinese alcoholics may have a lower allele frequency of two ADH genes (ADH2*2 and ADH3*1) in addition to a lower frequency of ALDH2*2 (83). These three alleles have the net effect of decreasing blood acetaldehyde levels after ethanol ingestion by retarding production while accelerating elimination. These important findings demonstrate a mechanism by which genetic variation appears to influence drinking behavior in Asians. It is unfortunate that the ALDH2*2 allele is absent in Caucasians and, therefore, cannot account for their drinking behavior.
The role of polymorphism at the dopamine D2 receptor locus (DRD2) in the pathogenesis of alcoholism has been a source of continuing controversy (86). Mesolimbic and mesocortical dopaminergic systems appear to be involved in the mediation of reward, suggesting that dopamine could play a role in addictive behaviors. This hypothesis was supported by the report of an association between a RFLP allele (A2) at the DRD2 locus and alcoholism (11). A consensus of the literature supports a number of conclusions about the association. First, it has been replicated more often than would be expected by chance. Second, the association is stronger with more severe cases of alcoholism. Third, it is not limited to alcoholism but includes other substance abuse disorders and possibly an even broader spectrum of psychopathology. Finally, the DRD2 locus is not linked to alcoholism.
What do these results tell us about the genetics of substance abuse? If the association is real, the most likely explanation is that variation in receptor kinetics contributes to the liability to substance abuse and the variation is correlated with DRD2 alleles the way ADH and ALDH enzyme activity correlate with alleles at their respective loci. Even if D2 receptor kinetics contribute to alcoholism, a simple cause-and-effect relationship between DRD2 alleles and the disease is excluded by the lack of linkage. This implies that the allele would have to act epistatically; i.e., by modifying the effects of other genes that have a more direct causal role. This explanation is consistent with the association appearing stronger with more severe cases of alcoholism. It could also account for the D2 receptor influencing the pathogenesis of a broader range of diseases than alcoholism. This theoretical explanation for the DRD2 association is supported by observations that the A2 allele alters receptor kinetics (53).
GENETIC RESEARCH IN OTHER NEUROPSYCHIATRIC DISEASES
Examining how molecular genetic strategies have made inroads into several neuropsychiatric diseases that share many of the complexities of typical psychiatric disorders may provide clues to what can be anticipated in Psychiatry. Alzheimer's disease is reviewed in detail elsewhere in this volume (Lendon and Goate). The appearance of an Alzheimer-like dementia in persons with Down's syndrome, the location of the amyloid precursor protein gene (APP) on the long arm of chromosome 21, and the existence of apparently autosomal dominant forms of Alzheimer's disease led to an intense search for disease vulnerability genes on chromosome 21q. The search resulted in the discovery of mutations in APP which account for a rare, early-onset, chromosome 21-linked form of the disease. A and a systematic genome scan identified a second locus on chromosome 14 responsible for early-onset, familial cases, and a third locus on chromosome 19 causing a late-onset, familial form of the disease. Although the genetics of Alzheimer's disease are proving to be complex, they still follow known genetic mechanisms by which point mutations cause mendelian inheritance of a disorder. In the case of the fragile X syndrome and Huntington's disease an entirely new genetic mechanism was uncovered.
The Fragile X Syndrome
Families with male-limited mental retardation with dysmorphic features and X-linked recessive inheritance have been noted since the 1940s. Subsequently, the fragile-X syndrome, X-linked mental retardation associated with a fragile site on the long arm of X, was recognized as the most common cause of hereditary mental retardation, accounting for 1 in 1,250 retarded males and an additional 1 in 2,000 females with mild retardation (18). The fragile site created a biological marker for the fragile-X syndrome, and when family studies included an assessment of the fragile site a curious phenomenon was discovered (73). Twenty percent of males with the fragile site are unaffected, and conversely, 30% of female heterozygotes are mildly affected, so the gene is neither completely dominant nor fully recessive. When the gene passes from unaffected males through their daughters to their grandsons, the penetrance in grandsons is 40%. Yet, brothers of nonpenetrant males have a 9% penetrance. The low penetrance in brothers compared with the high penetrance in grandsons cannot be explained by mendelian inheritance and became known as the Sherman paradox.
The Sherman paradox was finally explained when the mutation responsible for the fragile X syndrome was discovered (24). It is a (CGG)n trinucleotide repeat located in the coding sequence of a gene of unknown function. The repeat is capable of expanding as it passes to successive generations. The number of repeats in normals ranges from 6 to 50. When the number expands into the range of 50-200 repeats it is termed a premutation because it is prone to further expansion into a range of from 200 to over 1,000 repeats. This last expansion causes the fragile X syndrome. The Sherman paradox was explained by the observation that premutations must pass through females to expand and cause the syndrome. The probability of expansion increases with the existing size of the repeat, and this explains another genetic paradox—anticipation: the tendency for some diseases to exhibit earlier onset, increased severity, or greater penetrance in successive generations. It should be reassuring to behavior geneticists that a complex pattern of inheritance at the phenotypic level was accounted for by a simple mechanism once the genetics were understood.
Huntington's disease is an autosomal dominant trait with complete penetrance, provided that persons at risk live long enough to develop it. Although the genetics are mendelian, it manifests some of the complexities of psychiatric diseases. Penetrance is age-dependent, with an average age of onset of approximately 45 years. Although juvenile-onset cases exist, their symptoms differ from those in the adult, with rigidity being more prominent than choreiform movements.
The gene has been localized to the short arm of chromosome 4 by linkage (33). The first step in moving from a linkage to the disease gene is to flank the locus with a second marker and narrow the flanked region by walking new markers toward the center. Once the region was flanked and narrowed to about 2 million basepairs the search for candidate genes began. Eventually, a candidate was found that contained a mutation in all patients tested but not in normals. The type of mutation proved to be as exciting as the discovery itself because it belongs to the recently identified class of triplet repeats (82). The (CAG)n repeat is present in all copies of the gene, but the repeat number varies from person to person. The number is capable of expanding during meiosis, and when it becomes large enough it inactivates the gene, causing the disease. Furthermore, the more severe, early-onset cases have larger numbers of repeats than the more typical cases. These findings illustrate how the phenotypic complexities of Huntington's disease can be explained by a simple genetic mechanism.
The next step will be to clarify the normal function of the gene and how it causes Huntington's disease. We now know that the protein produced by the Huntington gene, termed Huntington, is widely expressed throughout the body. If this is the case, why is Huntington's disease restricted to the brain? The answer may lie in the function of Huntington-associated protein (45). The CAG repeats in the Huntington gene code for a polyglutamine repeat in Huntington, and this repeat binds Huntington-associated protein, which is richly expressed in brain, possibly explaining the locus of the pathology. These exciting discoveries raise the possibility that the repeat expansion causes disease by altering the binding of Huntington-associated protein in a way that is toxic to the neuron.
The problem of why genetic strategies have delivered so many false starts in psychiatry has been the subject of intense analysis. As a result, the field has achieved a more critical understanding of the difficulties of applying linkage to diseases of ambiguous etiology, genetics, and phenotype (i.e., "complex" genetic traits). A number of factors may have contributed to the difficulties. Lod scores based on small numbers of pedigrees can be particularly labile in the face of new information, such as diagnostic changes and pedigree extensions. Since the lod score is a parametric test, ambiguity over the genetic parameters of the disease invalidates the conventional significance level of 3.00. When multiple analyses are done to cover a range of inheritance models and disease phenotypes they can inflate the lod score. Basing a linkage decision on the genetic and diagnostic model that maximizes the lod score increases the risk of false positives further by confusing exploratory with confirmatory analyses. Failure to appreciate the profound effect of the prior odds of linkage on the posterior odds, represented by the lod score, can prompt premature conclusions that linkage has been found.
The importance of the prior odds for any molecular genetic strategy is illustrated by the three examples of successful gene searches in neuropsychiatry. In each case a successful search began with a sound biological clue: in Huntington's disease, mendelian inheritance; in Alzheimer's disease mendelian inheritance and amyloid plaques; and in fragile X syndrome, the fragile site. These clues focused the research so that the prior probability of success was favorable. Biological findings of this nature do not exist currently for any psychiatric disorder. The absence of such clues to help dissect the genetic complexity of behavior disorders has been the Achilles' heel of psychiatric genetics.
MOLECULAR GENETIC STRATEGIES
The usual test for linkage is the lod score. This statistic is derived by calculating the binomial probability of the observed distribution of pedigree members who show recombination between the trait and the marker locus and those who do not. (Formally, the lod score at any recombination fraction is the logarithm to the base 10 of the ratio of the probability of the pedigree assuming linkage at the given recombination fraction divided by the probability assuming no linkage). The lod score is dependent on the correct specification of the mode of inheritance, disease allele frequency, penetrance, and the frequency of new mutations and nongenetic cases. Furthermore, the critical value of the lod score for linkage detection (3.00) is based on the assumption that a disease locus exists and will be detected if enough markers are tested; thus, the probability of false positive lod scores due to testing multiple markers is offset by the increasing probability of a true positive as the genome is systematically searched. When linkage analysis is limited to simple mendelian diseases these requirements are satisfied, and the analysis and interpretation of the results are straight forward. However, when it is applied to nonmendelian diseases the lod score and its interpretation can be seriously compromised by the need to specify information that is not known about the genetic transmission and phenotypic expression of the disease.
The mode of inheritance is not known for any of the major psychiatric disorders. Although it is hoped that major loci account for a large enough proportion of pedigrees to be detected by linkage, this may not be the case, and inheritance may be due to two or more additive loci. In polygenic inheritance, a large number of loci contribute small, equal, and additive effects to disease liability. No single locus produces an effect great enough to be detected by linkage. In practice, most polygenic traits have proven to be oligogenic when the loci were accounted for. In oligogenic inheritance a small number of loci contribute additively to the phenotype. Theoretically, oligogenic loci are detectable by linkage, but the smaller the effect of a locus the larger the number of pedigrees needed to detect it.
The allele frequency of the disease gene must also be known to calculate the lod score accurately. When penetrance is incomplete, unaffected carriers marrying into the pedigree can produce affected descendants who could appear as recombinants and adversely affect the lod score.
Etiologic heterogeneity may pose the most serious obstacle to finding genes through linkage. Mendelian diseases tend to be rare because selection against the gene keeps the allele frequency low. Therefore, common diseases are likely to be nonmendelian, and if mendelian forms exist, they are likely to account for a minority of the pedigrees. These arguments predict considerable etiologic heterogeneity of psychiatric disorders.
Heterogeneity can be either genetic or environmental. Environmental heterogeneity causes phenocopies, which are considered under the discussion of the phenotype. Genetic heterogeneity can occur within loci when multiple alleles cause the disease, or it can occur between loci as illustrated by Alzheimer's disease. Of the two, only interlocus heterogeneity complicates linkage. Multiple disease loci can confound linkage analysis by occurring either within or between pedigrees. Heterogeneity between pedigrees weakens positive lod scores by contributing negative scores to the total. Heterogeneity within pedigrees weakens lod scores by creating false recombinants. In both cases the net effect is loss of power to detect linkage.
Pedigree members must be dichotomized into affected and unaffected cases for linkage analysis. Misclassification due to either false positive or false negative diagnoses weakens the lod score and this can happen in a number of ways.
Incomplete penetrance causes genetically affected persons to appear unaffected. Penetrance can vary with age, sex, and birth cohort. Incomplete penetrance is typical of psychiatric disorders, as demonstrated by monozygotic twin concordances for psychiatric syndromes being consistently less than 100 percent.
One effect of incomplete penetrance on linkage analysis is illustrated by the Amish study of bipolar illness already discussed (36). Two pedigree members who had not developed symptoms at the time of the first analysis decreased the lod score by almost two lod units when they were counted as affected in the reanalysis. The usual effect of incomplete penetrance is to weaken the power to detect and exclude linkage. The lod score can be corrected for incomplete penetrance, but the lower the penetrance the less the unaffected relatives contribute to the lod score.
Variable expressivity also causes diagnostic misclassification. Expressivity refers to the range of clinical features the genotype can assume. The diagnostic boundaries of many psychiatric disorders are obscured by a spectrum of conditions that are genetically related to the core illness. For example, bipolar I illness may be genetically related to bipolar II, major depressive disorder, schizoaffective disorder, and cyclothymic disorder (26). The problem with including borderline and subclinical disorders in the affected phenotype definition is that some have high population prevalences (e.g., major depressive disorder), and are likely to introduce heterogeneity into the study. Expressivity also includes such phenotypes as biological markers but, at present, none of the biological markers in psychiatry can replace or extend clinical diagnosis in linkage studies.
Phenocopies are environmental copies of genetic traits. Diseases as common as depression and alcoholism are likely to include a large proportion of these. Obviously, phenocopies will also be created by inappropriately broad definitions of the disease. They create another source of false recombinants. The phenocopy rate can be modeled in linkage analyses, but the greater it is the less the affected relatives contribute to the lod score. Whether the phenocopy rate is modeled in the analysis or not, the net effect is to weaken the power to detect linkage.
PRACTICAL ISSUES OF LINKAGE
How can linkage studies be designed to minimize these limitations and maximize the chances of finding disease genes? Now that the availability of dense, highly informative genetic maps has ceased to be a limiting factor, linkage studies will succeed or fail on the grounds of pedigree ascertainment, clinical assessment, and statistical analysis.
The pedigree ascertainment strategy is the first line of defense against genetic heterogeneity. Not surprisingly, it has received considerable attention, and a number of solutions have been proposed (32, 88). One suggestion for minimizing heterogeneity is to study a single pedigree. This requires an extensive pedigree, such as the Venezuelan kindred used to map Huntington's disease (33). In psychiatry, this strategy might be suitable for Tourette's disease and perhaps for bipolar illness, but pedigrees of this size have not been reported for other psychiatric illnesses. This strategy has the added advantage that the genetic model can be estimated by a segregation analysis of the pedigree.
Extended pedigrees also have a number of disadvantages (32). By their multigenerational nature, they select against recessive genes. Furthermore, they are not immune to the effects of heterogeneity because of multiple persons marrying into the pedigree. In addition, lod scores in these pedigrees can be highly dependent on diagnostic changes in key pedigree members. This last phenomenon is illustrated by the reanalysis of the chromosome 11 linkage findings in bipolar illness already discussed (36). Finally, linkage in such pedigrees may be difficult to replicate because the pedigrees, by virtue of their unusual nature, may represent a rare form of the disease.
Genetic isolates provide another means of minimizing heterogeneity. Isolated populations that are descended from a small group of founders tend to be genetically more homogeneous than the general population. The genetic model can be estimated by segregation analysis with the assurance that it will be appropriate for linkage pedigrees selected from the same population. However, a major disadvantage of this approach is that findings might prove difficult to replicate in other populations.
Another strategy for overcoming the effects of heterogeneity is to exploit the sheer statistical power of large samples. The lod scores of individual pedigrees can be analyzed for heterogeneity and linkage detected, provided that the proportion of the pedigrees that are transmitting the same gene is large enough. Linkage to Alzheimer's disease was detected in this way, despite appreciable genetic heterogeneity. If less than 50 percent of the pedigrees are transmitting the linked form of the illness, the pedigree set would need to be quite large to detect linkage, and this consideration has prompted a number of collaboration linkage studies in psychiatry.
Ambiguous disease phenotypes present a different set of problems. These can be subsumed under the categories of incomplete penetrance and variable expressivity. With regard to penetrance, age, sex, and cohort effects can be modeled in the linkage analysis. Alternatively, the analysis can be restricted to affected persons, and analytical methods have been developed for this strategy (30, 87).
With regard to expressivity, multiple phenotypes can be analyzed or the analyses can be restricted to definite cases of the disease. An alternate approach is to include all phenotypic categories in a single analysis by weighting each one with the probability of its being an expression of the genotype (56). One strategy for coping with both types of phenotypic ambiguity is the affected sib-pair strategy (30). Pedigrees of two or more affected siblings are collected and the linkage analysis is based on those individuals only. Another advantage of affected pedigree member methods is that the results do not depend on an assumed genetic model because they are nonparametric. The tradeoff is that they are less powerful than the lod score method.
Clinical assessment is another critical issue in designing linkage studies, and broad agreement exists on a number of optimal methods to be used (50, 88 ). Standardized interview instruments, operationalized diagnostic criteria, and explicit diagnostic hierarchies facilitate replication of results and comparison of studies. Instruments that assess a broad range of psychopathology and accommodate multiple diagnostic systems are more likely to capture the full expression of the genotype. Periodically updating the diagnoses keeps the database current when penetrance is incomplete. Assessing the families of persons marrying into the pedigree helps to identify bilineal transmission. Keeping assessments blind to the marker typing protects against diagnostic contamination. The value of augmenting the clinical assessments with biological markers is demonstrated by juvenile myoclonic epilepsy, where EEGs greatly increased the power to detect linkage (32). However, none of the current biological markers in psychiatry have achieved the same level of validity that the EEG has achieved in diagnosing epilepsy.
The data analysis strategy needs to be as carefully chosen as the strategies for ascertainment and assessment. If a parametric analysis is used, genetic parameters must be estimated or else several models must be tested to cover the range of possibilities. On the other hand, the lod score method has several distinct advantages because it is parametric: it is a more powerful statistic, it can be used to map loci, and it can exclude a disease locus. In principle, a map of closely linked markers can be used to systematically exclude a disease locus from all of the genome except where linkage exists.
Unfortunately, the ambiguities of psychiatric data affect all three strengths of the method (57). Although misspecification of the genetic model does not inflate the lod score, it does weaken it and inflates the estimate of genetic distance. At the same time, the critical lod score levels for both linkage detection and exclusion are unknown. Therefore, in psychiatry, the differences between parametric and nonparametric methods may not be as critical as they are in mendelian genetics. Even with the more powerful lod score method, a large pedigree set may be needed to find linkage.
The genetic model can be estimated by segregation analysis, but only if the pedigrees are systematically ascertained and extended, and even then many of the same ambiguities that frustrate linkage analysis will complicate segregation analysis as well. If the genetic model cannot be estimated, a number of models may need to be analyzed, in addition to several diagnostic definitions. This creates the danger of lod score inflation. One solution is to adjust the score by subtracting the log10 of the number of models analyzed. Alternatively, the correct adjustment can be derived from simulations with the pedigrees used in the analysis.
The effect of analyzing multiple markers on the lod score needs to be considered as well. This is not a problem with mendelian diseases because as the probability of a false positive increases with each new marker tested, the probability of finding the gene increases correspondingly, and a lod score of 3.00 is adjusted for the prior probability to give a 5% posterior probability of a false positive result (57). However, in psychiatry the prior probability is unknown, and therefore, the lod score may not be self-correcting. For this reason, the lod score may need to be adjusted by as much as two lod units for a complete genome search. The net effect of all the lod score corrections may be that the lod score method has little advantage over nonparametric methods in psychiatry.
Given the limitations of the lod score method under these circumstances, nonparametric methods may provide a reasonable screening tool. They are based on the principle that if the disease and marker loci are linked, affected family members will share alleles at the marker locus more frequently than expected by chance. For example, the probability of a pair of siblings sharing 0, 1, or 2 alleles identical by descent is .25, .50, and .25 respectively. A significant departure from this expected distribution in the direction of increased allele sharing is evidence of linkage. With a sufficiently large sample, this method is more robust to intra-locus heterogeneity and oligogenic inheritance than is the lod score method. This consideration, and its short computation time, make sib-pair analyses an attractive screening tool for linkage.
These linkage strategies have been presented as competing alternatives when, in practice, studies typically incorporate features of several methods. Thus, ascertainment may be based on an affected sib-pair strategy but the pedigrees may be extended to include more relatives. Similarly, the primary analytical tool may be the lod score with a nonparametric analysis added to check its validity.
STRATEGIES NOT BASED ON LINKAGE
Candidate genes offer an alternative to an exhaustive genome search. A gene may be considered a candidate by virtue of its location near a linkage, its involvement in the disease pathogenesis, or both. The amyloid precursor protein exemplifies both types of candidate by its location on chromosome 21 and its role in the pathogenesis of amyloid plaques. Until recently, candidate gene approaches were felt to be too unlikely to succeed for serious consideration. A third to a half of the estimated 100,000 human genes are expressed in brain (81), whereas the genome can be screened with 300 markers at a 10 centimorgan level of resolution; thus, if 300 markers can exclude 100,000 genes it makes little sense to examine candidates individually. However, experience with Alzheimer's disease, where the amyloid precursor protein gene, which was thought to have been excluded by linkage, proved to be responsible for the disease in several families, demonstrates how easily exclusion maps can miss rare disease genes.
One appeal of candidate genes is their ability to circumvent the tedious task of positional cloning—the process of proceeding from a linkage to finding a disease gene. It consists of flanking the disease locus with additional markers and then progressively narrowing the inter-marker region until it is small enough to clone and begin the gene search. For the process to work, genetic distance must be measured by linkage and association, but the accuracy of these measurements depends on the accuracy of the genetic model and disease phenotype. Therefore, psychiatric disorders will complicate positional cloning for the same reasons that they complicate linkage analysis.
Candidate gene strategies fall into three categories: association (linkage disequilibrium), mutation screening (e.g., single strand conformational polymorphisms and denaturing gradient gel electrophoresis) (55, 72), and direct sequencing. Mutation screening procedures are based on electrophoresis methods that can detect single base-pair changes through altered mobility of DNA fragments. Sequencing detects mutations by reading the genetic code. Association studies depend on the detection of linkage disequilibrium.
Linkage disequilibrium results from a marker polymorphism occurring so close to a disease mutation that meiotic recombination has not had time to establish equilibrium among the allele frequencies at the two loci. Consequently, allele frequencies at the marker locus differ between affected and unaffected populations. This disequilibrium can be detected by a simple test of association between the disease and the marker allele. Since the test is nonparametric it is more robust than the lod score to the confounding effects of complex diseases. The effects of inter-locus heterogeneity, oligogenic inheritance, incomplete penetrance, variable expressivity, and phenocopies can, in principle, be overcome with large, narrowly diagnosed samples.
For linkage disequilibrium to occur, the loci must be less than 2.0 centimorgans apart (i.e., less than 2% recombination) and therefore, the approach is best suited to candidate genes. Linkage disequilibrium is one cause of disease-marker associations; others include sampling cases from either ethnic subgroups or inbred populations, interaction between the two loci, and evolutionary selection for alleles at both loci. Ethnic stratification is a potential source of false positive results. Allele frequencies can vary widely across ethnic groups, and if cases and controls differ on this critical variable, a spurious association will result. This bias can be avoided by sampling parents of cases and estimating population allele frequencies from the parental alleles that were not transmitted to the cases (the haplotype relative risk method) (22).
Mutation screening is more sensitive than association and linkage because it can detect mutation in a single patient, but its time intensiveness limits its general applicability. Still, it can play an important role in the overall search for disease genes (76). When a provisional linkage is found, candidate genes in the region can be screened for mutation in linked pedigrees, and if the disease gene is found the task of positional cloning will be circumvented.
If a proven linkage or some other biological mechanism existed for any psychiatric disorder a compelling case could be made for studying candidate genes. However, without a biological basis with which to prioritize candidates, any of the 30 to 50 thousand genes expressed in brain is a potential candidate. When the prior odds of a gene being pathogenic are so low, the posterior odds of a nominally significant result being correct can be discouragingly low (17). For example, if the 20 thousand genes nonconstitutively expressed in brain are all considered potential candidates, 98.5% of associations at the .05 significance level will be spurious. The same reasoning holds for sequence variants identified through mutation screening. Demonstrating that a sequence variant occurs in an evolutionarily conserved region of the gene or that it affects the protein structure of the gene product provides only circumstantial evidence connecting the variant with the disease. Sequence variation can be tied to a behavioral phenotype only by demonstrating association in the population or linkage in pedigrees. Ironically, the lack of biological clues to help focus molecular genetic strategies in psychiatry, the same problem that complicates linkage research and makes candidate gene approaches attractive, also complicates candidate gene approaches.
Trinucleotide repeats have now been identified as the cause of five neuropsychiatric diseases (Huntington's disease, fragile X syndrome, myotonic dystrophy, spinobulbar muscular atrophy, and spinocerebellar ataxia type 1) (46). Finding a mechanism that can explain genetic and phenotypic complexities in such diseases as the fragile X syndrome has generated enthusiasm for searching for triplet repeats in psychiatric disorders. Such repeats appear to be common in genes expressed in brain, being found in 35% of 40 cDNA clones tested in one study (45). Moreover, methods have been developed to detect triplet expansions in genomic DNA, opening the way to screening populations of psychiatric disorders (69). Indeed, this mechanism could be the biological clue needed to increase the prior odds of a candidate gene being etiologic.
Genomic Mismatch Scanning
The essence of linkage is to identify regions of the genome shared by affected family members through inheritance from a common ancestor. Genomic mismatch scanning is a promising strategy that could revolutionalize the search for genes underlying complex diseases by delivering a complete set of DNA clones from these regions (52). Genomic DNA from an affected pair of relatives is digested with a restriction enzyme to yield restriction fragments, and these are mixed under conditions that permit the formation of heteroduplex molecules (consisting of one strand from each individual). Enzymatic treatment is then used to degrade homoduplex molecules (those composed of both strands from the same individual). The remaining heteroduplexes are treated with an E. coli exonuclease that recognizes single basepair mismatches and nicks one strand of the DNA near the mismatch. Further enzymatic digestion strips off runs of bases from the nicked strand, and intact strands are separated from partially stripped ones by column separation. The final product is a collection of DNA molecules representing all regions of the genome where the two relatives match. These molecules can be labeled and used to probe a genomic library to identify matching genomic clones to search for candidate genes. Thus far, the method has only been applied to yeast, but if it can be adapted to affected relative pairs in humans it could become a powerful tool for finding candidate genes.
The Human Genome Project has created maps of the rat and mouse genomes in addition to the human map, and these maps can be used to find genes controlling behavioral traits in these organisms. Linkage analysis is far more controllable in laboratory animals than it is in humans. Animals are easily bred to provide the sample sizes needed. Inbred strains are isogenic: each strain is homozygous at all loci, and therefore, all members of any strain are genetically identical, eliminating the problem of genetic heterogeneity. Phenotypes can be carefully measured, and laboratory conditions held uniform to reduce phenotypic variation. The same strengths of the method will make it far easier to positionally clone genes once linkages are found. Alternatively, linkage in rodents may enable a search for linkage or candidate genes in homologous regions of the human genome.
Animal strains can be exploited in two ways (2). Strains bred to maximize and minimize a trait can be crossed and the offspring, which are heterozygous at all loci, backcrossed with a parental strain or intercrossed with each other. The trait then can be measured in the progeny and analyzed for linkage. Alternatively, the trait of interest can be measured in a number of inbred strains that are already typed with a genome map. Linkage can be detected by correlations between the trait and the genetic marker typings.
Detection of a locus for hypertension in the spontaneously hypertensive rat has sparked interest in using similar approaches in behavior disorders (35). Two examples of relevant animal models are the Maudsley Reactive and Nonreactive strains as a model of anxiety, and alcohol preferring and nonpreferring strains as a model of alcoholism (10, 44). Although animal models have many features of an attractive experimental system for linkage, their greatest weakness is uncertainty over whether the animal behavior accurately models the disease in question. If not, genes contributing to the animal behavior may be unrelated to the disease in humans. However, a counterbalancing strength is the immense importance of cloning any gene that determines a behavioral trait.
Genetics is a rapidly evolving field and considerable flexibility is required to adapt to the new technologies and insights. Our understanding of the complexities of applying linkage and other molecular methods in psychiatry has matured greatly. This maturation is the product of a productive interaction of theoretical, population, and laboratory genetics. Deploying new technologies effectively requires careful analysis to identify optimal research designs and analytical methods to complement the technologies, as well as to avoid hidden pitfalls inherent in them. Likewise, new disease mechanisms, such as triplet repeats, necessitate a reanalysis of epidemiological data to identify candidate diseases for the mechanism. Thus, psychiatric genetics rests on a foundation of molecular, clinical, and statistical science.
Finding disease genes in psychiatry is likely to require not only collaboration of these three sciences, but also a productive interaction of existing methods and strategies. If linkage searches do not deliver a confirmed linkage, they will certainly identify provisional linkages, both in the entire pedigree set and in subsets of it. These regions will then become target regions for candidate gene strategies. Association studies with candidate loci in large populations may have the power to replicate provisional linkages that are based on a small subsample of the pedigree set. Furthermore, the ability to study candidate genes in pedigrees where the candidate and disease are cosegregating will greatly enhance the chances of finding etiologic mutations.
Several thousand cDNA clones have now been partially sequenced and mapped as expressed sequence tags (ESTs), and the pace will certainly accelerate in the future. This effort will create a map of anonymous genes as candidates for diseases whenever a region is spotlighted by human or animal linkage studies. This change in candidate gene technology has prompted the speculation that the positional cloning strategy may become a positional candidate approach in the future (3). What is envisioned is that as more is known about these anonymous genes, their characteristics can be matched to those of the disease being studied. Genes expressed preferentially in limbic structures will be of considerable interest. Those containing triplet repeats will be candidates for diseases demonstrating anticipation. Membrane spanning domains suggest a receptor function. Developmentally expressed genes will be a priority for mental diseases thought to have a developmental etiology. The positional candidate strategy is certainly an attractive scenario in psychiatry, where the complexities of the disorders make it unlikely that linkage and association could narrow a region of interest to the point where a search for candidate sequences could be contemplated.
Two recent developments in psychiatric genetics will close the chapter on a note of optimism. The first is an association between familial thyroid hormone resistance and attention deficit-hyperactivity disorder (ADHD) (34). Symptoms of ADHD are often seen in persons with familial thyroid hormone resistance, an autosomal dominant disease caused by a variety of mutations in exons 9 and 10 of the receptor gene (hTRbeta). When ADHD was assessed in a number of these families strong co-segregation of DSM-III-R ADHD with hTRbeta mutations was found. Although the finding accounts for a small minority of ADHD cases, understanding the mechanism may put investigators on the trail of other causes of ADHD.The second development is new insight into the genetics of an unusual X-linked form of mental retardation (12). The disease is a nondysmorphic mental retardation characterized by aggressive and impulsive behavior that follows an X-linked recessive pattern of inheritance in a large Dutch pedigree. A gene for the disease was localized by linkage to Xp21-p11. Since monoamine oxidase A (MAOA) is located in this region, urinary monoamine metabolites were analyzed in several affected cases and found to be markedly abnormal. A point mutation in the eighth exon of the MAOA gene was identified in five of the affected males (13). The next step is to determine how frequently MAOA mutations account for mental retardation and aggressive behavior in the population.
Both of these developments are examples of the kind of biological insights that could reveal the etiology of two rare forms of behavior disorder. Insights like these have made molecular genetic strategies powerful tools for finding disease etiologies in other areas of medicine. More work will need to be done with both diseases before the genetic etiology of the disorder is considered to be found. If it is found, identifying the genetic code responsible for a behavior disorder will be a major advance in psychiatric genetics.