The search for linkage is the attempt to locate predisposing genes by identifying sites (polymorphisms) in the genome at which sequence variations (alleles) travel with illness within families that have more than one affected member. It is thus a method of locating a disease-inducing gene at a specific site on a particular chromosome.

It has been nearly two decades since Sherrington et al. (1) reported in Nature that they had identified a susceptibility locus for schizophrenia on the long arm of chromosome 5. In seven families that included 39 individuals with DSM-III diagnoses of schizophrenia or schizophreniform disorder, these authors recorded lod scores ranging from 3.22 to 6.49 with different analyses and inclusion or exclusion of cases. They concluded that their findings provided the “first strong evidence for the involvement of a single gene in the causation of schizophrenia.” Within 2 years, however, a number of tests of linkage in the same region (e.g., references 2–5) had been conducted without yielding comparable findings.

Egeland et al. (6), studying affective disorders, reported that “an analysis of the segregation of restriction fragment length polymorphisms in an Old Order Amish pedigree has made it possible to localize a dominant gene conferring a strong predisposition to manic depressive disease to the tip of the short arm of chromosome 11”—but then reported 2 years later (7) that a reanalysis that included several new individuals and two changes in clinical status “markedly reduces the probability of linkage” and that, indeed, linkage could be excluded from that region. Shortly afterward, the same journal published a paper (8) linking bipolar affective illness to color blindness and glucose-6-phosphate dehydrogenase deficiency as markers on the long arm of the X chromosome with lod scores between 7.52 and 9.17—“that is, the odds in favor of linkage range between 3×10⁷ and 10⁹ to 1.” No magnitude of lod score comparable to those reported in these early studies is now seriously entertained. Nevertheless, the case for a genetic contribution to schizophrenia (9) and manic-depressive illness is strong (10).

The question in the light of the subsequent literature is how such apparently convincing evidence could have been obtained. Given this history and the accumulation of much data from studies with larger sample sizes, it is surprising that a claim for linkage with a lod score of 6.5 of schizophrenia to a locus on chromosome 1q21-q22 in a study of 22 families appeared in Science in April 2000 without any clear explanation of the discrepancy with the more modest findings that had by then accumulated (11). The subsequent failure to replicate the finding in a large collaborative study (12) could perhaps have been predicted.

Meta-Analyses of Genome Scans

Scientific progress requires replication. What has made replication an attainable target in linkage studies has been the pursuit of systematic genome scans in which spaced markers throughout the genome—a hypothesis-free approach—are evaluated against the transmission of disease within families, generally including affected pairs of siblings. In principle, such studies have the potential to reveal linkage to multiple genes, and with increasing sample size, they can be expected to show increasing uniformity of findings across studies. There have now been two meta-analyses of the genome scans in relation to schizophrenia and two in relation to bipolar disorder (Table 1).

Taken together, the findings are instructive:

1. One meta-analysis of schizophrenia draws attention to three and the other to nine possible loci of significance. Badner and Gershon (13) pointed to 8p, 13q, and 22q (p and q indicating the short and long arms, respectively, of the relevant chromosomes) as the regions of greatest significance, whereas Lewis et al. (14), by the method they adopted (16), concluded that their strongest finding was on 2p, with significant evidence for linkage on 5q, 3p, 11q, 6p telomere, chromosome 1 at the centromere, and 22 at the short arm telomere all yielding stronger evidence than the strongest candidate, 8p, in the first study. Thus the two meta-analyses applied to a material with an overlap of approximately 500 pedigrees out of a total of 681 (13) or 1,208 (14) agree on only two out of 10 chromosome arms as candidates to harbor a locus for schizophrenia. The strongest finding in one study (14) (linkage to 2p) was not detected in the other analysis; this locus emerged only from a single large study (17) (382 sib pairs) that was included in the second but not the first meta-analysis.

2. While the second meta-analyses of schizophrenia (14) and of bipolar disorder (15) were conducted with an identical methodology, the conclusions that the authors drew are quite different. Whereas the schizophrenia meta-analysis concluded that there were eight bins that yielded genomewide evidence of linkage, the bipolar meta-analysis revealed no such evidence (the locations listed in Table 1 for this study are the nominal 5% by-bin significances), and the locations that yielded the largest rank sums showed no overlap with the bins identified for schizophrenia. This is surprising in light of evidence that there is some genetic commonality between these major phenotypes (18–21), and all the more surprising given that the core phenotype in meta-analysis included diagnoses of schizoaffective disorder.

3. The two findings in the first analysis of bipolar disorder (13) of 353 families were not replicated in the second meta-analysis (15), which included 347 families from essentially the same published literature but identified four different locations of interest, although none reached genomewide significance by strict criteria.

Comparison of Four Large Linkage Studies

There have now been claims for linkage of schizophrenia to locations on 21 of the 23 pairs of chromosomes, but for any one of these claims the majority of genome scans has failed to detect linkage. Figure 1 presents a comparison of the findings of DeLisi et al. (17), Williams et al. (22), and Suarez et al. (23), the three largest linkage studies published thus far that include schizophrenic and schizoaffective psychosis, together with findings from a study of bipolar illness of comparable size and design (Lambert et al. [24]). The first two studies (17, 22) each draw attention to two loci—the first to loci on chromosomes 2 and 10, and the second to loci on chromosomes 10 and 17, with the loci on chromosome 10 being located at the short arm and long arm telomeres, respectively—and the third (23) points to loci on chromosomes 5, 8, 10, and 11. The findings in each study are weak (lod scores of less than 3 are generally taken as lacking significance), and there is no agreement on a single locus between the three studies of schizophrenia and schizoaffective psychosis and no commonality with the findings in bipolar disorder (modest peaks on chromosomes 4 and 6). Moreover, none of the studies gives support to a number of candidate genes that have recently been suggested (25), such as RGS4 and DISC1 (on chromosome 1), dysbindin (on chromosome 6), neuregulin (on chromosome 8), DAAO (on chromosome 12), G72 (on chromosome 13), and COMT (on chromosome 22). If any of these candidates (identified on the basis of other linkage studies) were really associated with disease, why would this association not be apparent in these four largest studies?

The inconsistency of the findings in these comparisons deserves emphasis. First, there is no agreement between these three large linkage studies, each conducted with broadly similar techniques on samples of sibling pairs identified by comparable diagnostic criteria. Second, there is no consistency in the findings of two meta-analyses conducted on a body of literature that included a core of the same studies. Third, there is no relationship between the locations identified in either of the meta-analyses and the modest peaks that emerged from the three sibling pair studies summarized here. The eye is drawn to chromosome 8 as a possible exception, but the peak in the study of Suarez et al. (23) has no counterpart in the other two linkage studies. Fourth, the candidate genes that have attracted widespread interest in the recent literature are not in any systematic way related either to the locations identified in the meta-analyses or to any conjunction of peaks from the three largest sibling pair studies. Again in relation to chromosome 8, Suarez et al. (23) conclude that the broad peak that emerged from their studies was clearly telomeric to neuregulin, and the same authors (Duan et al. [26]) have conducted a systematic association study of this region with negative outcome.

Nor do studies of genetic isolates clarify the situation. There have now been nine claims for linkage in such populations (see Table 3 in DeLisi et al. [27]), but in general there is no agreement between isolates and no obvious agreement between any one of these claims and the consensus of either of the meta-analyses.

Where does this leave the search for genes for the spectrum or continuum of psychosis? At the very least, it is clear that no strong signal is emerging, and more important, there is no obvious criterion for distinguishing signal from noise. One possible conclusion is that many genes of small effect are relevant, that there is significant heterogeneity across populations, and that samples much larger than those currently available will be required to detect reliable linkage. While this view (“multiple genes of small effect”) is popular among psychiatric geneticists, there is no indication at all in the current literature that it has a chance of success. The field of psychosis genetics, as based on linkage studies, is thus at an impasse—there is no clear way ahead.

A quite different answer is conceivable: that the genetic variation is relatively homogeneous in form but of a nature that is in principle invisible to the linkage approach. It is possible that the relevant variation is epigenetic, that is, that it lies in modifications, such as methylation of the sequence or in variations in methylation, phosphorylation, and acetylation of the associated histones that are components of the chromosome, rather than in alterations of the DNA sequence itself, and for this reason is not detected by the linkage strategy.

There are already indications of this in the data from monozygotic twins, the cornerstone of the evidence for the genetic theory. Whereas concordance (40%–50%) is greater than that seen in dizygotic twins (12%–15%), consistent with a genetic contribution, it falls well short of 100%. The discrepancy is often interpreted as evidence for an environmental interaction, but no consistent differences in exposure to putative risk factors between affected and unaffected members of discordant pairs have been identified (28). The alternative is that discordance reflects stochastic differences in gene expression in the course of development. The steep decline from around 10% in first-degree relatives to between 1% and 2% in second-degree relatives, close to the risk in the general population, is often argued to be consistent with “multiple genes of small effect” (29) but is also compatible with a genetic imprint imposed or reimposed at meiosis with a short half-life between generations.

The Case for a Single Gene

Contrary to recent interpretations of linkage findings, there are arguments for a single gene underlying psychotic disorder (e.g., reference 30):

1. The structural changes in the brain in psychotic illness are relatively homogeneous. There is a degree of ventricular enlargement (31), and the range of values in the patient group reflects an upward shift of the distribution relative to controls, without an increase in variance (32). Ventricular enlargement is thus a consistent correlate of psychosis. It has been demonstrated in a number of ethnically separated groups. It is unclear how this could be the result of a number of unrelated genes varying independently in different populations.

2. Incidences are relatively uniform across populations, particularly when the core (nuclear) symptoms are assessed (33). This is true also when bipolar illness is the case definition (34). The fact that incidence is apparently unrelated to geographic or other environmental variation (see references 30 and 35 for a discussion of the absence of evidence for environmental factors) is the key observation that has to be explained by theories of etiology. If psychosis is due to multiple genes, why should these not vary independently in separated populations?

3. Although it has been claimed in the past that certain forms of psychosis (e.g., periodic catatonia and cycloid psychosis) “breed true” within families, this claim is not defended in the contemporary literature. What is observed is that the form of psychosis varies within families encompassing individuals with schizoaffective and affective illness as well as more characteristically nuclear forms of schizophrenia. Each of the larger collaborative studies of both schizophrenic and affective psychoses includes schizoaffective illnesses in the core phenotype (see also Table 1). This suggests some sort of a continuum. If different genes were responsible for different psychotic illnesses in different families, a homogeneity of the clinical picture within families would be expected, and this is not observed.

4. In a graphic analysis of the relationship between penetrance and concordance in monozygotic and dizygotic twins, Pritchard (36) has shown that only a single-gene model is consistent with the data. Polygenic models predict greater discrepancies between monozygotic and dizygotic concordance than are observed.

5. It is necessary to postulate a significant genetic advantage to account for the persistence of psychosis in association with a fecundity disadvantage. If two genes were relevant, it is unlikely that the balance of advantage and disadvantage associated with each would be similar (37). One gene would be selected out.

Any argument for a single gene must explain why such a gene has not long since been detected by the linkage approach. The answer proposed here is that the variation relates to the characteristic that separates Homo sapiens as a species from other primate and hominid species, and that interindividual variation is epigenetic in form. The symptoms of psychosis can readily be seen as disturbances of the H. sapiens-specific capacity for language. Hallucinations (voices), disturbances of thought processes (thoughts experienced as alien, loss of direction), and even delusions (distortions of meaning) can all be conceived as deviations in the transition from thought to speech production or from perceived speech to meaning (38, 39). Thus the phenomena of psychosis are associated with the core characteristic that defines the species. The function to be accounted for is language, and this faculty has long posed difficulties for evolutionary theory. It represents an apparent discontinuity or saltation that is not easily assimilated by the conventional gradualist view that species separate in response to differential environmental selective pressures. That the variation associated with psychosis relates precisely to those changes that distinguish H. sapiens from prior hominid species raises the question of the identity of the species transition.

What Is Characteristic of the Human Brain?

According to the Out of Africa hypothesis (40), all modern human populations have an origin in Africa sometime between 100,000 and 150,000 years ago, at which time a species arose that replaced prior hominid species such as H. Heidelbergensis, H. erectus, and the Neanderthals (Figure 2). This species had the capacity to represent the environment in symbols, a capacity manifest in language that can be traced back in the archaeological record through artifacts to 90,000 years ago, but not much earlier (41, 42). It is parsimonious to assume that the transition to language and the origin of the species reflect the same genetic event. To judge by the relative stability of the pattern of artifacts associated with preceding hominid species, the transition was relatively abrupt.

What can this genetic event have been? Broca (43), following Dax (44), reported that language production in the frontal lobes was located on the left side. In 1877 he wrote, “Man is, of all the animals, the one whose brain in the normal state is the most asymmetrical. He is also the one who possesses the most acquired faculties. Among these faculties . . . the faculty of language holds pride of place. It is this that distinguishes us most clearly from the animals” (45).

That Broca was right in supposing that directional asymmetry is the critical factor is now supported by studies of handedness in chimpanzees (46) and other primates (47) and studies of minicolumn separation in the planum temporale (48). Directional asymmetries are absent in other primates, including the great apes, although there is some evidence that tool use in H. erectus was preferentially right-handed (49).

Thus, to understand the human brain, it is necessary to understand the genetic basis of asymmetry (and its anatomical correlate—the cerebral “torque” from right frontal to left occipital). Two theories of handedness (50, 51) postulate a single gene acting together with a random factor. Although both originally predicted an autosomal location, there is an association between handedness and sex within families (52) that is consistent with the presence of the gene on both X and Y chromosomes. Earlier, a gene in a region of X-Y homology was suggested on the basis of observations of sex chromosome aneuploidies (53), and this is supported by intergenerational evidence from family studies (54, 55). The question remains: How could an apparently new feature—in this case cerebral asymmetry—arise from the existing genome complement? One hypothesis, championed, for example, by Goldschmidt (56), an advocate of saltational change in evolution, is that chromosomal rearrangements play a role. Have there been chromosomal changes that might be relevant? One such change is the X to Y translocation followed by a paracentric inversion on the Y short arm that created a region of homology between Xq21.3 and two blocks on Yp after the separation of the chimpanzee from the hominid lineages (57–59). These rearrangements ensured that gene sequences that are present on the X in all mammals are present in man on the Y chromosome also. The direction of the sequence on the Y was reversed in the original translocation (estimated to have taken place approximately 6 million years ago—that is, close to the time of, and maybe the cause of, separation of the chimpanzee and hominid lineages) and in the paracentric inversion was reversed again in the distal block in Yp (Figure 3). Because the arrangement of the homologous regions on the Y is universal in modern human males, these changes must have been selected (i.e., the gene content of the block duplicated on the Y is relevant), and the chromosomal changes constitute speciation events in hominid evolution. The first rearrangement is a possible correlate of the transition from the chimpanzee-hominid precursor to Australopithecus, and the second (the paracentric inversion which cannot yet be dated) is a possible correlate of the transition to modern H. sapiens. Within this region of homology, a gene pair (protocadherin X/Y) coding for cell surface adhesion molecules has been identified (60) that has been subject to accelerated evolution (16 amino acid changes in the Y and five in the X sequence) in the hominid lineage by contrast with the relative stability of the sequence on the X in earlier primate evolution. This gene pair therefore is a candidate determinant of cerebral dominance for language.

Of particular interest is the status of protocadherin X/Y (and other genes within this region) with respect to X inactivation. Genes that are also present on the Y chromosome are protected from this epigenetic process, a protection that can be understood as aimed at achieving gene dosage equivalence in males and females. Therefore, whereas in all other mammals protocadherin X may be expected to be inactive on the inactive X chromosome, in females in H. sapiens a new situation pertains. The mechanism by which protection from X inactivation is brought about is obscure, but it must be supposed to depend on an interaction between X and Y chromosomes, and this can only take place in male meiosis (61). If it depends on pairing of X-Y homologous sequences, then clearly the direction of the sequence on the Y and therefore the paracentric inversion will have significance. Whatever influence the presence of protocadherin Y may have had on the inactivation status of protocadherin X before the paracentric inversion, that influence will have been greater after this event. Thus, the orientation of the gene sequence on the Y must be supposed to have a key influence on the balance of gene expression in the two sexes in modern H. sapiens relative to earlier hominid species.

A Solution to the Problem of Psychosis?

Conventional wisdom is that the genes for psychosis are autosomal, but no such genes have been found. There are sex differences to be explained; the earliest onsets are predominantly in males and late onsets in females. There is also a same-sex concordance effect—a tendency for affected pairs of relatives (e.g., siblings) to be more often of the same sex than would be expected by chance (62, 63). This effect, noted by Penrose (64) and Rosenthal (65), is predicted for a gene present on both X and Y chromosomes. Interestingly, the effect is at least as strong in affective as in schizophrenic psychoses (66). There is evidence that the structural changes in schizophrenia interact with asymmetry and sex (67–69). Thus it is plausible that the primary change lies in the variation associated with an X- and Y-linked determinant of the relative rates of development of the two hemispheres. (The X chromosome was omitted from Badner and Gershon’s meta-analysis of bipolar disorder [13], Lewis et al.’s meta-analysis of schizophrenia [14], and Segurado et al.’s meta-analysis of bipolar disorder [15].) If, as argued here, the variation relating to psychosis is epigenetic, then this variation will be expected to arise in meiosis, the point at which a preexisting imprint is removed and a new one, depending on the interaction of the grandpaternal and grandmaternal Xs in female meiosis and the grandpaternal Y and the grandmaternal X in male meiosis, is imposed. In each case the interaction may be assumed to be a function of individual variations in DNA structure in adjacent genomic sequences. Because such sequences are shared by monozygotic twins more than by dizygotic twins and siblings, increased rates of concordance for psychosis will be expected in the former, as observed in the twin studies (9) notwithstanding that linkage is absent.

Given the generation of the Xq21.3/Yp regions of homology in the course of hominid evolution and the nature of the gene pair’s protein products as pathfinder molecules, protocadherins X and Y must be considered leading candidates. Sequence variation in this gene pair between individuals is low and has been shown to be unrelated to psychosis (70). The epigenetic status of this pair and other genes within the transposed region thus becomes a focus of interest (71). The fact that monozygotic twins often differ for handedness and its structural correlate, asymmetry of the planum temporale (72), suggests that for the development of cerebral dominance for language, epigenetic factors are of primary significance in determining individual differences. The alternative to the search for the multiple genes of small effect that are so elusive to the linkage approach therefore is that the relevant variation is epigenetic and that it relates to the most characteristically human dimension—hemispheric dominance for language.