This review critically summarizes the neuropathology and genetics of schizophrenia, the relationship between them, and speculates on their functional convergence. The morphological correlates of schizophrenia are subtle, and range from a slight reduction in brain size to localized alterations in the morphology and molecular composition of specific neuronal, synaptic, and glial populations in the hippocampus, dorsolateral prefrontal cortex, and dorsal thalamus. These findings have fostered the view of schizophrenia as a disorder of connectivity and of the synapse. Although attractive, such concepts are vague, and differentiating primary events from epiphenomena has been difficult. A way forward is provided by the recent identification of several putative susceptibility genes (including neuregulin, dysbindin, COMT, DISC1, RGS4, GRM3, and G72). We discuss the evidence for these and other genes, along with what is known of their expression profiles and biological roles in brain and how these may be altered in schizophrenia. The evidence for several of the genes is now strong. However, for none, with the likely exception of COMT, has a causative allele or the mechanism by which it predisposes to schizophrenia been identified. Nevertheless, we speculate that the genes may all converge functionally upon schizophrenia risk via an influence upon synaptic plasticity and the development and stabilization of cortical microcircuitry. NMDA receptor-mediated glutamate transmission may be especially implicated, though there are also direct and indirect links to dopamine and GABA signalling. Hence, there is a correspondence between the putative roles of the genes at the molecular and synaptic levels and the existing understanding of the disorder at the neural systems level. Characterization of a core molecular pathway and a ‘genetic cytoarchitecture’ would be a profound advance in understanding schizophrenia, and may have equally significant therapeutic implications.
To the molecular psychiatrist, Alzheimer's disease and schizophrenia provide an interesting contrast. In the former there is, by definition, a diagnostic neuropathology, which is unequivocal, quantifiable, and correlates with the clinical severity of the disorder. Substantial elements of the genetic architecture are known, with autosomal dominant mutations in several genes, and a major influence of the apolipoprotein E4 allele on disease susceptibility. A cardinal neurochemical deficit, in cholinergic transmission, has been identified and characterized. With this fundamental knowledge in place, significant strides have been made in its pharmacotherapy, crucially moving the field forward from therapeutic nihilism. The discovery that familial Alzheimer's disease is caused by mutations in genes which impact on β-amyloid trafficking and metabolism led to the advancement of the β-amyloid hypothesis1 and thence β-amyloid targeted treatments, which hold the promise of retarding or reversing the disease.2 These developments have together revolutionized geriatric psychiatry and engendered an optimism which could not have been foreseen a decade ago.
By comparison, the understanding of schizophrenia remains rudimentary. Neuropathological findings are controversial and not diagnostically useful; loci and genes have been difficult to identify and replicate. Equally, treatment has improved only incrementally. In the absence of definitive genes or pathogenic molecular mechanisms, it is not surprising that there is no equivalent of the secretase inhibitors and vaccines being developed in Alzheimer's disease. However, the situation seems about to change; indeed, it may be changing already. The purpose of this review is to provide a critical update on the neuropathology and genetics of schizophrenia, and to consider how they may intersect in the pathogenesis of the disorder. Relatively more space is devoted to the genetics because of dramatic recent developments. We postulate that the genes predispose, in various ways but in a convergent fashion, to the central pathophysiological process: an alteration in synaptic plasticity, especially affecting NMDA receptor (NMDAR)-mediated glutamatergic transmission, that disrupts neural microcircuits involved in higher-order cortical function, particularly executive processing.
Neuropathology of schizophrenia: a summary
The cumulative literature, including several meta-analyses, disproves the null hypothesis that there is no neuropathology of schizophrenia, at least at the macroscopic level. Recent reviews should be consulted for details and full citations;3,4,5,6,7,8 here, we summarize the major findings and cite only a selection of papers.
Along with ventricular enlargement, there are small but significant reductions in brain volume9,10 and weight.11 Imaging studies particularly implicate the hippocampus,12,13 association neocortex (prefrontal and superior temporal),4,6,14 and thalamus.15 There are also abnormalities reported in a diverse range of other parameters, including cortical thickness,16 cortical gyrification,17,18 hippocampal shape,19,20 and cerebral asymmetry.21,22 Volumetric differences are seen in first-episode and drug-naïve patients,23,24,25 and some exist before the onset of psychosis26 and occur in at risk and unaffected relatives.27,28,29 Thus, there is evidence of a neuropathology intrinsic to the disease process, part of which may be related to genetic predisposition rather than to the illness itself.30
While the demonstration that schizophrenia is beyond doubt a brain disease has been of fundamental importance,31,32 it is also important not to over- or misinterpret these data.33 Firstly, there are incomplete, inconsistent, and even contradictory reports for many findings: for example, concerning the brain structures most affected and the clinicopathological correlations;34,35,36 also, the question of progression or variation in the changes during the course of illness and their seemingly counterintuitive clinical implications.37,38,39 Secondly, the magnitude of change in each parameter is usually small, and concerns group means with considerable overlap between schizophrenia and comparison groups. Third, when another disease group is available for a direct comparison, the findings are rarely specific. Overall, therefore, schizophrenia cannot be considered to have a clear or ‘diagnostic’ neuropathological signature. This may of course change in the future as new methods are applied. On the other hand, it seems more likely that the macroscopic differences, and the histological abnormalities to be mentioned below, will prove to be downstream or tangential manifestations of the core neurobiological phenotype, viz. the genetically influenced molecular disruption of neural circuits subserving particular neurofunctional domains. We return to this issue later.
An important negative observation is that schizophrenia is not associated with an increased frequency of Alzheimer's disease40,41,42 nor other recognized neurodegenerative disorders, nor astrogliosis.43 This applies even in schizophrenics with dementia,41,42 unless there is a coincidental pathology (eg infarction).44 These negative findings mean that the cognitive impairments of schizophrenia, increasingly viewed as core features and therapeutic targets, are not explained in conventional neuropathological terms.45,46 Also, their absence, by default, gives support to a developmental origin of the neuropathology, and constrains theories of schizophrenia as a progressive disorder, at least in a classical neurodegenerative sense.38,47
Robust positive findings have been harder to come by, in part because the studies have been smaller and fewer, and it remains the case that no single abnormality can be considered wholly established. The most intriguing and potentially most notable histological observations are those of aberrantly located or clustered neurons, especially in lamina II of the entorhinal cortex,48,49,50,51 and in the neocortical white matter,52,53,54,55,56,57 since these kinds of abnormality are strongly indicative of an early neurodevelopmental anomaly affecting neuronal migration, survival, and connectivity.58,59 The number of positive reports means that these findings cannot be dismissed, and should be actively considered as candidate neuropathological features of the disorder. However, neither can they yet be accepted uncritically, as there are methodological limitations, negative studies, and the positive studies disagree as to the nature of the alterations.57,60,61,62,63
There are several other histological findings to note that lack the strong neurodevelopmental implication of aberrantly located neurons, but which are reasonably well replicated and together may provide clues about the nature of the disorder and may relate more directly to its genetic origins. First, the cell bodies of pyramidal neurons in the hippocampus and neocortex are smaller in many64,65,66,67,68,69 though not all70,71 studies. Smaller perikarya are probably a correlate of a less extensive or less active axodendritic tree which the neuron has to support.72 Second, consistent with this interpretation, the same neuron populations have fewer dendritic spines and reduced dendritic arborizations, as assessed using Golgi stains73,74,75,76 and molecular markers such as MAP2 and spinophilin,77,78 though inconsistencies (or anatomical heterogeneity) exist here as well.79,80 There are also reductions in several presynaptic markers of pyramidal and other neurons, with decreased expression of genes such as synaptophysin, SNAP-25 and complexin II.81,82 Third, the density of some interneurons, especially parvalbumin-immunoreactive cells, and their synaptic projections are reduced.83,84 Fourth, though there is no overall change in the number of neurons in the cerebral cortex85 or hippocampus,86,87 the thalamus may have fewer neurons, notably in the mediodorsal nucleus and pulvinar.88,89,90,91,92,93 However, the mediodorsal findings also exemplify the frustration of research in this field: despite the five positive reports,88,89,90,91,92 two comparable subsequent studies have been resoundingly negative.94,95 Fifth, a reduction in the number and function of oligodendrocytes is becoming apparent from ultrastructural,96 morphometric97,98 and microarray99,100 studies. Given the role of oligodendrocytes in myelination and other aspects of neuronal and synaptic integrity,97,101,102 their involvement in schizophrenia is likely to be inextricably linked to the neuronal alterations mentioned, and to the functional consequences thereof, but at present it is not clear which are primary changes and which are secondary. Moreover, oligodendrocytes are an especially vulnerable cell population to many insults, and molecular abnormalities of oligodendrocytes are reported in association with many CNS disorders not involving myelin directly. A similar interpretational query affects another neuropathological theme that has emerged recently, that of mitochondrial and metabolic involvement in schizophrenia, for which there is also morphological,103 biochemical,104,105 and molecular106,107 evidence.
Additional conclusions regarding the post-mortem data are facilitated by consideration of imaging and other in vivo findings. For example, smaller neurons with less extensive (or less active) arborizations may well explain the reduced N-acetylaspartate (NAA) signal in schizophrenia, which is used as a marker of neuronal ‘integrity’ in proton magnetic resonance spectroscopy studies108 and which is also decreased in post-mortem tissue.109 In tandem with the apparent decreases in some presynaptic and glial populations, these morphometric changes could also contribute to decreased regional brain volumes and cortical thickness, via a reduced neuropil.110 The fact that volumetric and NAA deficits are seen in first episode and medication-free subjects111,112 is important as it supports the assumption that the neuropathological observations—all of which are based on chronic, medicated patients—are not merely a consequence of the illness or its treatment. The latter risk is, in any event, often exaggerated; there is little evidence that antipsychotics cause the morphometric and molecular alterations (except within basal ganglia), and often quite strong evidence that they do not.113,114 A similar reassurance applies to autopsy delay effects.115 Instead, relatively neglected factors such as smoking,116,117 substance misuse,118 and premortem events,115,119,120,121 may be greater confounders.
Synaptic connectivity in schizophrenia
In summary, rather than having a distinctive signature, the neuropathology of schizophrenia seems to consist of quantitative alterations in various normal parameters of neural microcircuitry, ranging from the dendritic tree to the cell body and axon to the synaptic terminal, and including associated glial elements. In this respect, the neuropathology may be viewed as representing the structural anlage of the functional ‘dysconnectivity’ which is prominent in pathophysiological models of the disorder.3,122,123,124,125,126,127,128 A more focused variant of these models is the concept of schizophrenia as a disorder of the synapse.129,130,131 Another view of altered connectivity emphasizes more the ‘cabling’ itself, based on the recent evidence for white matter abnormalities emerging from MRI and microarray studies.132,133
Formulations of this kind are inevitably weak and imprecise, since the nature of the pathology is only partially known, and since any severe brain disorder is almost bound to affect, one way or another, most neuronal, glial, and synaptic populations. Nevertheless, there are some intriguing clues. First, with regard to synaptic pathology, the few electron microscopy data suggest that it is only partly morphological, in the sense of a loss or other visible abnormality of synaptic terminals, the rest being presumably ‘molecular’; that is, affecting the composition, activity, or plasticity of the synaptic machinery.81,82 Parenthetically, this continuum of conventional neuropathology with biochemical and functional indices (eg receptor densities, growth factor abundance, fMRI signal, etc) is also seen in Alzheimer's disease,134,135 and in normal synaptic plasticity,136 and will prove relevant when we address how the genes may operate. Second, though the neurochemical phenotype of connections affected in schizophrenia is unclear, several types are clearly involved,137 including glutamatergic ones in the hippocampus82,138,139 and cerebellum,140 and alterations in some GABAergic as well as glutamatergic synaptic populations in dorsolateral prefrontal cortex (DPFC);83,141,142 changes in cortical dopaminergic innervation143 and signalling79,144 are also apparent. The glutamate data are noteworthy as they add to the increasing focus on this transmitter and its interaction with dopamine,145,146,147,148,149,150,151 and will also be seen to be genetically pertinent. Third, concerning the timing of the synaptic pathology, there is altered expression in schizophrenia of several ‘developmental’ genes such as DLX1,152 reelin,153 and semaphorin 3A154, and correlations between their expression and that of other synaptic markers.57,153 These findings provide some neuropathological support for a developmental basis to schizophrenia, perhaps via effects on synaptogenesis or synaptic pruning; however, they should not be overinterpreted, since every gene implicated in schizophrenia probably plays some role in brain development and, conversely, the very fact that so-called developmental genes continue to be expressed in adulthood implies that they have ongoing functions (which may or may not be the same as those during maturation).
The current gaps in knowledge make a more robust and fine-grained understanding of the neuropathology of schizophrenia desirable, and further efforts to this end continue in many laboratories, including ours. However, the subtlety of the pathology, and the intrinsic conceptual and practical limitations of post-mortem brain research, mean that this approach is unlikely to yield a full explanation. Certainly in isolation it will not explain the cause of the pathology, which ultimately is the major goal. To do so requires identification of the susceptibility genes, since these presumably are pathogenic, at least partly, via their influence on the formation, maintenance, and activity of the brain systems that underlie the disorder. We therefore revisit the neuropathology after discussion of the leading genetic candidates. First, however, we consider briefly some of the key issues which have affected the field of schizophrenia genetics, and which significantly influence interpretation of the recent data.
The search for schizophrenia genes
Twin studies show unequivocally that schizophrenia is predominantly a genetic disorder, with estimates of heritability of risk of around 80%.155,156 Hence, while factors other than DNA sequence variation are important too, identification of the genes responsible for this high heritability will be critical to understanding the disease, and even the environmental and epigenetic factors may be difficult to unravel without parsing based on genotype. Family studies show that simple major gene effects are unlikely; instead, polygenic models, that is, the effects of multiple risk genes acting additively or multiplicatively, provide the best explanatory fit.157 Thus, like cancer, diabetes, and heart disease, schizophrenia is a complex genetic disorder, not characterized by a single causative gene and not showing simple patterns of inheritance (though there may be rare examples of such families). Similarly, its genes will each account for only a small increment in risk (eg no greater than a three-fold risk elevation in siblings),158 are likely to be modified by other genes, including protective ones, both additively and epistatically, and also to show environmental modification. The genetic architecture of susceptibility is almost certainly heterogeneous, meaning that no particular constellation of genes will be characteristic of most ill individuals. Furthermore, the same causative allele(s) may have a variable phenotype depending on genetic background.
Much of the scepticism about finding genes for schizophrenia, and indeed the difficulties encountered by researchers, was fuelled by a failure to appreciate the implications of the assumptions above, and there continue to be substantial and unresolved issues.159 One concerns the interpretation of the existing evidence for linkage based upon the 20 genome-wide scans reported to date (references available on request to DRW). These studies have together implicated much of the genome, and in only four studies has any region reached accepted levels of statistical significance (6p22–24;160 8p21–22;161 1q21–22;162 and 10q25.3–q26.3163). Each of these regions, however, has been decidedly unremarkable in other studies. There have been many explanations offered for these apparent inconsistencies, including lack of power to detect the weak linkages expected in complex genetic disorders, and problems related to sample ascertainment biases, diagnostic imprecision, and genetic heterogeneity.
Two recent meta-analyses have sought clarification by combining the results of the genome-wide studies. Badner and Gershon164 combined the uncorrected P-values for markers clustered around a region showing evidence for linkage in any study (defined as a nominal pointwise P<0.01). After correcting for region size and marker number, but not sample size, three loci reached genome-wide significance: 8p, 13q, and 22q. They concluded that these are valid linkage regions, likely containing one or more susceptibility genes. Lewis et al165 applied a within-study rank-order analysis of published and some unpublished datasets and produced rather different results. Their approach is less dependent on models of inheritance and heterogeneity and even to marker order; instead, they lumped marker data into 30 cM bins across the genome (120 bins in all) and ranked each bin within a study based on the most significant evidence for linkage of any marker in that bin. The bin rankings were averaged across all the studies and permutation tests used to reject the null hypothesis that rank and relative order for any given bin were random. Under stringent criteria, genome-wide significance was found only for 2p12–22.1, a region that had not achieved even nominal significance in any single study. Using a more liberal statistical threshold, several other loci emerged, viz., in decreasing order of significance, 5q, 3p, 11q, 2q, 1q, 22q, 8p, 6p, 20p, and 14q. The authors concluded that multiple regions of the genome were likely to contain genes for schizophrenia; an optimist would also note that 8p and 22q were identified in both meta-analyses.
These meta-analytic approaches amplify signals that appear to generalize across multiple samples (even if none are particularly impressive in a single study) and reduce strong signals that are unique to only a few samples. The ultimate validation of these results will be the evidence supporting the genes themselves. It is worth noting, however, that of the genes identified so far as promising candidates, none map to 2p and only COMT and neuregulin map to loci common to both meta-analyses. The well-described weakness of linkage as a strategy for mapping genes of small effect158 raises the possibility that families in positive linkage studies may be segregating uncommon large effect genes. Alternatively, each of the positive linkage regions may contain multiple susceptibility genes that individually account for small increments in risk across populations (ie odds ratios <2 for a specific gene), but together account for linkage across families that are heterogeneous for risk genes in these regions. Consistent with this latter scenario is evidence that genes associated with schizophrenia that do not map to linkage regions show similar genotypic relative risks as genes that do (see below).
The inconsistencies and uncertainties regarding the linkage findings have generated two schools of thought: a sceptical one, which doubts all results and views the strategy as nonproductive,166 and an optimistic one,159 which sees the results as being consistent with predictions of weak effects of multiple genes and with genetic heterogeneity, and which looks forward to ever larger-scale studies. In our view this is a stale debate since ultimately validation comes from the discovery of a gene or genes within the linkage region. Hence, it is time to move away from solely statistical arguments to directly test the importance of specific loci or genes.167 Armed with the public genome databases, researchers can now identify candidate genes or expressed sequences within the linkage regions and, in tandem with available functional information, make a case for gene identification. This has led to testing of specific allelic variants, usually single-nucleotide polymorphisms (SNPs), within or adjacent to a gene. Demonstration of an increased frequency of an allele in schizophrenics compared to a suitable comparison group constitutes evidence of genetic association. Such association, if not an artefact (eg of multiple testing, genotype errors, or population stratification), means that the SNP is either causative or is in linkage disequilibrium (LD) with a variant that is. Association, however, does not mean that the gene has been found. It may also be that the detected association is in fact a proxy for a functional variant in a nearby gene. A precedent for this is lactose intolerance, a genetic disorder characterized by decreased lactase expression in the gut. The gene abnormality, however, is a polymorphism located in another gene that encodes a regulatory element impacting on lactase expression.168 If we knew nothing about the biochemistry of lactose intolerance, we might be misled by the various functions of the other gene. This means that allelic association per se is not evidence of gene identification. A further caveat about the association data is that the risk alleles related to schizophrenia vary between studies; in other words, though association to several genes is well replicated, specific alleles are not. The implications of this are unclear; while it has been argued that the inconsistencies reflect haplotype differences between populations related to ancestral background,169,170 this largely post hoc interpretation will not be confirmed or refuted until the causative mutations are identified. In any event, the essential point is that the end game in identifying susceptibility genes for polygenic disorders like schizophrenia will not come from statistics, either linkage or association, but will require biological evidence that the risk variant impacts on the pathogenesis of the disease.167,171 This end game involves a convergence of analyses at many levels, including functional studies of gene variation, animal and cell model systems, and molecular neuropathology in human brain. Indeed, some of the post-mortem findings in schizophrenia (eg synaptic markers) may serve as molecular phenotypes to validate the pathogenic implications of susceptibility alleles.
We now consider the genes recently reported to be associated with schizophrenia, focussing on those that are positive in at least three published independent samples, adopting the criterion for robustness recommended by Lohmueller et al.172 Several other genes are also mentioned which have not yet reached this arbitrary threshold, recognizing that not all will stand the test of time. For each gene, the genetic evidence is considered along with what is known of its function and expression in the brain, and how it may be affected in schizophrenia. Table 1 summarizes the genes to be considered along with our opinion as to the current strength of the evidence rated in four domains. We do not cover some earlier genetic associations (eg HTR2A, DRD2, DRD3), reviewed elsewhere.159,173,174,175
The genetics, expression, and biology of schizophrenia susceptibility genes
Catechol-O-methyl transferase (COMT)
Catechol-O-methyl transferase (COMT) is the most plausible of the susceptibility genes a priori, because of its role in monoamine metabolism, and because the main genetic variant being associated with schizophrenia is functional. Its candidacy is furthered by its mapping to 22q11, implicated in both meta-analyses,164,165 and hemideletion of this region produces velocardiofacial syndrome (VCFS), a condition associated with manifold increased risk of schizophrenia-like psychoses.176 We therefore consider COMT first and in detail, even though the statistical evidence for its association with schizophrenia is not as strong as for some of the other genes.
Identified in 1958,177 COMT catalyses the methylation of catechols, such as dopamine, norepinephrine, and catecholoestrogens.178,179 COMT exists in membrane-bound (MB) and soluble (S) forms, which differ by the presence of a 50 amino-acid signal anchor in MB-COMT. In peripheral tissues and in rodents, S-COMT predominates, but in human brain it is MB-COMT.180 MB-COMT has much greater affinity for dopamine than S-COMT, but a lower Vmax,181 suggesting that brain COMT has high efficiency but low capacity, that is, suited to neurotransmission. COMT is expressed as two mRNAs, of 1.5 and 1.3 kb, corresponding to two start codons and two promoter elements within a 27 kb sequence.180 The longer transcript can give rise to MB- and S-COMT, whereas only S-COMT is produced from the shorter transcript; the latter is rare in human brain.180,182,183 Early studies suggested that COMT, especially S-COMT, was a glial enzyme, but subsequent analyses show that COMT is expressed primarily in neurons, and is much more abundant in prefrontal cortex and hippocampus than in striatum or in brainstem dopamine neurons,178,183,184 supporting conclusions from pharmacological studies that COMT inactivates catechols at postsynaptic sites.185 The distribution, together with other data,186,187 implicates COMT in cortical interneuronal monoaminergic signalling, especially dopamine. However, the precise distribution of COMT remains unclear; for example, whether COMT can ‘see’ synaptic dopamine; whether and where S and MB forms are differentially expressed, and the role, if any, of COMT within dopaminergic neurons.178,188 Similarly, there is still uncertainty about the role of COMT in metabolism of norepinephrine and other catechols in brain. These issues may limit the conclusions that can be drawn regarding the cellular and molecular basis for the effect of COMT variation on cortical function and schizophrenia, discussed below.
In 1978, a trimodal distribution of peripheral COMT enzyme activity was reported,189 consistent with inheritance of two codominant alleles and three genotypes. Grossman et al190 identified a SNP in exon 4, a G–A substitution changing valine (val) to methionine (met) at position 108/158, that accounts for the observation; the polymorphism is here denoted as val158met. The amino-acid change impacts on the stability of the enzyme, such that val-COMT has significantly lower enzyme activity than met-COMT.181,191 Both S- and MB-COMT are affected. Early studies suggested that the alleles account for up to four-fold variation in enzyme activity,181,191 but this was determined at nonphysiological temperatures; more recent data show closer to a two-fold variation in activity in human brain192 and other tissues.193
Its functionality has led to the val158met polymorphism being extensively investigated in schizophrenia, with more than 15 association studies reported. The results are decidedly mixed. There are at least eight studies claiming evidence for association to the val allele, but as many with negative results, although only one with association to the met allele (refs on request to DRW). A recent meta-analysis was inconclusive, but showed that any effect would be small (odds ratio 1.2–1.4).194 The reasons for the inconsistencies are not straightforward. Most studies are small and underpowered to reject association, since the allele frequency difference between patients and controls, even in the positive studies, is only ∼5–8%. Moreover, the allele frequency shows a marked variation between populations,195,196 and so occult population structure (ethnic stratification) could easily obscure small case–control differences, as could ascertainment biases hidden in incomplete characterization of controls. To help circumvent these problems, seven studies have reported transmission of COMT alleles within families. In five,197,198,199,200,201 greater transmission of val158 alleles was found, which reached nominal significance in four studies,197,199,200,201 with a similar odds ratio in the fifth.198 Although the data in total show that COMT by itself contributes at most a very small increase in genetic risk for schizophrenia, the chances of finding a significant positive association to val-COMT in eight independent samples (four from family studies and four from case–control studies) is slim, especially given the prior linkage data and biological plausibility. Thus, it is likely that the COMT val158met allele is part of the complex risk architecture of the illness.
One of the early surprises in the COMT literature was the more frequent association of schizophrenia with the val not the met allele. Val association has also been reported in a large population study of schizophrenia spectrum personality traits.202 The classic dopamine hypothesis of schizophrenia would have predicted the association would be with the low activity (hence higher dopamine) met-COMT. Egan et al200 offered a potential explanation for this apparent inconsistency and for the mechanism of the val158 association. They found that val-COMT was associated with abnormal prefrontal cortical function, relative to met158, as measured by cognitive tests and with fMRI activation, even in normal subjects. Additional evidence of an association of the val158 allele with poorer prefrontal function has emerged from a number of sources, including several other studies of cognition,203,204,205 fMRI,206 and EEG.207 Moreover, performance on comparable tasks in rodents is improved, together with increased frontal cortex dopamine release, by COMT inhibition,187 and COMT knockout mice reportedly have enhanced memory (see Weinberger et al,208). The implication that COMT impacts critically on dopaminergic transmission and associated functions in the prefrontal cortex is consistent with the anatomical and pharmacological data mentioned; moreover, COMT knockout mice show gene dose-dependent increases in dopamine186 and dopamine turnover209 in prefrontal cortex compared with striatum; larger cortical than striatal effects are also seen in studies using COMT inhibitors. The anatomical selectivity can be explained by the fact that dopamine transporters, which play the principal role in inactivating synaptic dopamine in the striatum, are at very low abundance in the prefrontal cortex,210,211 and play virtually no role in cortical dopamine inactivation.212 The absence of dopamine transporters gives COMT (and norepinephrine transporters213), and the variation in COMT activity associated with the val158met genotype, particular impact on cortical dopamine signalling. It is interesting to note that in dialysis studies of COMT inhibition187 and in studies of COMT knockout mice,208 changes in norepinephrine levels were not seen, suggesting that COMT is less important for norepinephrine flux, perhaps because of the role of norepinephrine transporters in prefrontal cortex. In the context of the dopamine hypothesis of schizophrenia, therefore, COMT appears especially relevant to the cortical deficits and their putative basis in dopamine hypofunction. Certainly, the range of prefrontal abnormalities associated with the val allele (eg poorer executive cognition, cortical processing inefficiency, and an abnormal P300 evoked EEG response), are all found more commonly in schizophrenia and in their first-degree relatives, suggesting that the effect of val-COMT is qualitatively isomorphic with the pattern of prefrontal deficits that characterize the risk biology of the disease.208 In passing, it is worth pointing out that if variation in COMT is linked more strongly with cognitive intermediate phenotypes rather than with the schizophrenia syndrome itself, it may partially explain the inconsistent results of the genetic association studies based on standard diagnostic criteria.
Although the emphasis has been on the frontal cortex, COMT genotype may also have influences elsewhere. For example, there has been considerable interest in the possibility that abnormal prefrontal cortical function in schizophrenia (particularly reduced cortical DA), could have downstream implications for regulation of brainstem DA activity (specifically, increased activity).122 Akil et al214 measured tyrosine hydroxylase mRNA in brainstem dopamine neurons of brains from 23 normal subjects, using its abundance as a reflection of cortical excitatory drive on the dopamine neurons. The brain specimens were genotyped for the val158met polymorphism. Consistent with predictions from the animal literature,208 val-COMT was associated with higher tyrosine hydroxylase expression, especially in neurons projecting to the striatum and amygdala. The val158 allele may, therefore, not only bias directly towards diminished prefrontal function, but also indirectly to disinhibited mesencephalic dopamine activity. COMT may therefore contribute, along with other mechanisms,122,151,215,216,217,218 to both cortical dopamine deficiency and mesolimbic hyperdopaminergia in schizophrenia.
Most COMT association studies have focused on the val158met polymorphism because of its known biochemical correlate and the increasing evidence for its effects on brain function. However, other genetic variants may also be relevant. Shifman et al219 reported that two common SNPs, one upstream and the other in or near the 3′ untranslated region (3′-UTR), were associated with schizophrenia in a large sample of Israelis of Ashkenazi descent. They claimed that the effect of these SNPs was much greater than that of val158met, though, in fact, the odds ratios did not differ significantly between the three polymorphisms. Moreover, haplotypes containing either or both risk alleles in combination with val158 were highly significantly associated with schizophrenia in this population (P=9.5 × 10−8), more so in women than men (see below). This level of significance reflects the large sample sizes (>3000), as the odds ratios for the SNPs and haplotypes were all ∼1.3–1.6. Although the increase in risk from this haplotype was small, its high frequency in the population translates into a high attributable risk; it was estimated that 32.5% of their female schizophrenic population would not be ill had they not inherited this haplotype.219 However, the findings have yet to be replicated, and one study201 and two unpublished ones find no enhanced risk of these other SNPs or haplotypes (see Owen et al159 and BS Kolachana, KE Straub, MF Egan and DR Weinberger, unpublished). Furthermore, no functionality of these SNPs was found in terms of gene or protein expression or enzyme activity, as measured in DPFC of over 100 subjects (G Chen, B Lipska, J Kleinman, DR Weinberger, unpublished). Nevertheless, the observations of Shifman et al219 together with other considerations discussed below,196,220 suggest that the val158met allele alone may not capture the complexity of the genetic regulation of COMT activity. It is conceivable that the inconsistencies in the association of val-COMT with schizophrenia, and with prefrontal performance, will be clarified by a more detailed analysis of combinations of functionally interacting alleles. For example, the val158 effect may only increase risk of the disease in individuals who carry other COMT alleles that exaggerate its biological effect, or who carry modifying alleles in other genes impacting upon dopamine signalling.
It is unclear whether the effects of COMT genetic variation on schizophrenia and prefrontal function are confounded (enhanced or reduced) by differential allelic expression. Two studies have found no evidence that val158met impacts upon level of COMT mRNA.221,222 These negative findings apply both to subjects with schizophrenia and controls. However, Bray et al220 found that COMT SNPs, including val158met and two SNPs from the Shifman et al,219 study did alter expression in human brain homogenates, with the high-risk alleles being associated with ∼20% less COMT mRNA. The relatively higher expression of met-COMT, if also present in terms of translated protein, would tend to counterbalance the greater activity of val-COMT. These results, however, are contradicted by evidence that met-COMT protein was associated with reduced immunoreactivity in transfected cell lines and in liver biopsies compared with val-COMT.193 These apparently inconsistent results are yet to be reconciled; in any event, enzyme activity is ultimately the critical COMT parameter, and other measures (such as mRNA or protein level) should be interpreted with caution. The evidence to date indicates that only the val158met variant has a clear effect on enzyme activity.
Another potential complicating factor in COMT studies is gender. There is evidence that COMT shows stronger association with schizophrenia in females,219 a more abnormal phenotype in female COMT knockout mice,186 and ∼30% lower peripheral COMT activity in women.178,179 The basis of the gender differences may reside in transcriptional regulation via estrogen response elements in the promoter.196,223 However, whether COMT expression is sexually dimorphic is unclear, and there is no indication that the influence of val158met on prefrontal function differs between men and women. The role of COMT in metabolizing catecholoestrogens, and the association of the met158 allele with cancer risk in women, may also be relevant to the gender differences.179
COMT expression in schizophrenia has been studied in DPFC, and shows only minor alterations. COMT mRNA221,222 and protein (E Tunbridge and PJ Harrison, unpublished) abundance are unchanged, consistent with earlier negative enzyme activity data.224 However, the mRNA distribution may be altered, with relatively greater expression in the deep than superficial laminae compared to controls.221 The importance of this finding is unknown.
Straub et al225 reported that variation in dysbindin (DTNBP1), a conserved 140 kb gene, was associated with schizophrenia in the Irish families that had shown linkage to 6p24–22.160 They identified and genotyped 17 SNPs, mostly intronic, in a 670 kb region flanked by two microsatellite markers that capped one of their two 6p linkage peaks. Several of the SNPs showed association to schizophrenia, both narrow and broad diagnoses, in a transmission disequilibrium analysis of 270 families (consisting of only 60 fully typed trios). A three-marker haplotype showed highly significant association to a broad region of the gene (from introns 2 to 7), with no significant association to SNPs outside the gene. A further analysis of the same sample narrowed the high-risk haplotype block to a 30 kb segment from within introns 2–5.226 In both analyses, the high-risk haplotype was rare, present in less than 7% of the sample. This means that if this haplotype contains the causative mutation in this population, and even if it is highly penetrant, it accounts for a very small percentage of cases.
Associations between dysbindin and schizophrenia have subsequently been reported in a number of studies. Schwab et al170 studied 203 families from Germany, Hungary, and Israel (including 150 trios), some of whom had previously shown linkage to 6p, and found association to several of the SNPs identified by Straub et al, and to a six-marker haplotype block spanning introns 2–5. However, the alleles associated with schizophrenia were the common variants, that is, the opposite result to the association with the rare alleles seen in the Irish sample. This is not easily explained by ethnic diversity as the origins of these populations (excepting perhaps the Israelis) are not distinct; the authors suggested that the Irish mutation may have emerged independently on a more recent ancestral background. Further complexities have emerged from other case–control studies. Three samples from England and Ireland were all negative for the original SNPs, but showed association to different SNPs, implicating potentially different functional elements within the gene.227,228 Another positive report has come from a Han Chinese family-based sample, typed for seven SNPs spanning the Straub et al225 panel from introns 1 to 9, and reporting significant overtransmission of a common five marker haplotype.229 Again, the original three marker haplotypes positive in the Irish study were not overtransmitted in this sample, and the authors did not report their individual SNP analyses. Van den Bogaert et al230 typed four of the most positive SNPs from the earlier studies, spanning introns 3 and 4, and an additional marker contained within the high-risk haplotypes, and examined three case–control samples, from Germany (418 cases, 285 controls), Poland (294, 113), and Sweden (142, 272). These had to be studied separately because variable allele frequencies were found between the control groups, starkly illustrating the potential for stratification errors even in studies of European populations. The German and Polish samples proved negative, but one SNP in the Swedes showed the rare allele slightly enriched in the patients, and the five marker haplotype was also positive. When only the cases with a positive family history were analysed, the associations were stronger, and the five marker haplotype showed a six-fold increased risk effect (3% frequency in controls, 18% in patients). Again, however, the allelic composition of this haplotype was distinct from earlier reports. The authors argued that dysbindin may be associated with familial schizophrenia in particular, although this was not the case in the other studies. Further evidence for dysbindin as a susceptibility gene for schizophrenia has come from a large study of Bulgarian parent–proband trios, which found highly significant association with the common alleles of two SNPs from the original225 study, as well as with several undertransmitted 2-, 3-, and 4-marker haplotypes; however, the haplotypes and individual SNP alleles were not consisent.231 Finally, in the Weinberger lab at NIMH, associations with dysbindin have been found in two datasets. In ∼200 American-Caucasian family trios, a SNP in intron 4 showed strong association, with overtransmission of the common allele and multimarker haplotypes containing this allele, which is also contained within the risk haplotype seen in the Irish and German samples (RE Straub, MF Egan and DR Weinberger, unpublished). In a second family sample of ethnically heterogeneous individuals from the NIMH Genetics Initiative (NIMHGI) dataset, association was found for this SNP and for several others, but the overtransmitted alleles were not the same. Notably, several SNPs were associated with intermediate cognitive phenotypes related to genetic risk for schizophrenia, especially IQ and also working and episodic memory;232 also, a ‘protective’ dysbindin haplotype has been associated with higher educational attainment.228 This suggests that, as with COMT and some of the other genes to be discussed, variation in dysbindin may prove to be related to cognitive aspects of schizophrenia as well as (or more so than) to the core syndrome itself.
In summary, there is considerable evidence that genetic variation in dysbindin is associated with schizophrenia, but striking inconsistency in the high-risk alleles and haplotypes across various populations, even those of similar geography and ancestry. This conundrum is unlikely to be resolved until causative mutations are identified. None have yet emerged despite extensive resequencing.228,233 The evidence so far suggests that there may be true allelic heterogeneity, that is, a number of mutations have emerged independently that have caused subtle but common pathophysiological effects on dysbindin function. There are many precedents for this; in cystic fibrosis, for example, over 100 causative mutations in the same gene have been identified.
Dysbindin is a ∼50 kDa protein originally cloned from a yeast two-hybrid screen of binding partners of α- and β-dystrobrevin (to wit, dystrobrevin binding protein), which are components of the dystrophin-associated protein complex (DPC) in the neuromuscular junction and brain, respectively.234 Dystrophin mutations cause several forms of X-linked muscular dystrophy. In this context, the associations between dysbindin SNPs and cognitive domains in schizophrenia are noteworthy, as cognitive deficits are classical features of Duchenne's muscular dystrophy; interestingly, this disease also has a neuropathology reminiscent of schizophrenia, with a fronto-temporal distribution, cortical heterotopias, and reduced dendritic arborization of pyramidal neurons.235 Since the DPC is concentrated at the postsynaptic density (PSD),236 dysbindin is thought be involved in one or more PSD functions, which include trafficking and tethering of receptors (including NMDA, nicotinic, and GABAA receptors) and signal transduction proteins.237,238 However, a substantial fraction of dysbindin occurs presynaptically.234,239 As the DPC is absent from this compartment, presynaptic dysbindin may well associate with different proteins and play different roles. Dysbindin has a widespread distribution in the brain, being expressed by many neuron populations, including pyramidal neurons in the hippocampus and DPFC, and also in substantia nigra and striatum.80,225,234,239
Dysbindin expression is decreased in schizophrenia. In DPFC, this has been shown both for protein240 and mRNA.80 The latter study also found that one of the dysbindin SNPs was associated with less mRNA expression; this observation may be related to the finding that dysbindin alleles are differentially expressed in heterozygotes,241 suggesting that cis-acting regulatory elements, including possibly the intronic variants associated with schizophrenia, could represent a mechanism for the association. Dysbindin immunoreactivity also has been reported to be reduced in the hippocampus in schizophrenia, with the decrease occurring presynaptically.239 Moreover, the affected pathways were excitatory, and so this study raises the possibility that dysbindin might contribute to the hippocampal glutamatergic synaptic pathology of schizophrenia mentioned above.242 Indeed, preliminary evidence suggests that overexpression of dysbindin increases glutamate release by pyramidal neurons in culture, possibly because of a role in vesicular trafficking243 Finally, it is important to note that existing anatomical and schizophrenia data are based on ‘pan’-dysbindin mRNA probes and antibodies which overlook the possibility that dysbindin isoforms may be differentially expressed in the brain, or differentially altered in schizophrenia.
Neuregulin 1 (NRG1)
In an important paper consisting of linkage, association, and animal modelling, Stefansson et al244 reported evidence that NRG1 is a susceptibility gene for schizophrenia. Starting with a small sample of 33 families and 105 affected subjects in Iceland, they performed a microsatellite-based whole genome scan and found suggestive linkage at 8p12–21, near the 8p region highlighted in prior scans.164,165 Using high-resolution genetic and physical mapping techniques, they focused on a 5 cM region around their best marker and identified two large risk haplotypes, one of which was found in seven families and the other in two (of the 33 linkage families). The region shared by the nine families defined a DNA block of 600 kb, which contained the 5′ domain of the NRG1 gene. This region and the entire NRG1 gene was further explored, using molecular and informatics strategies and extensive resequencing to uncover 1200 SNPs, including 15 that are nonsynonymous. Genotyping of 58 of these SNPs in 478 patients and 394 controls (and 121 SNPs in a subsample) revealed a seven-marker core haplotype spanning a 290 kb block that was highly significantly associated with schizophrenia (P<6.7 × 10−6). The core haplotype extended from the first intron of NRG1 to far upstream of the transcription start codon, but included the first exon of the full-length transcript. Despite the statistical significance, the core haplotype was relatively uncommon, found in approximately 7.5% of controls and 15% of their patients. Thus, it accounted for a small (∼10%) incremental risk across the Icelandic population, and probably does not account entirely for the 8p linkage signal in the population. Interestingly, none of the individual SNPs were associated with schizophrenia nearly as significantly as the haplotype, suggesting that none of the identified variants are functional polymorphisms. It is assumed that the risk haplotype is tagging an ancestral block of DNA that carries the schizophrenia risk allele(s) accounting for the association. The authors stated that they are resequencing the full 290 kb region in search of causative alleles but, as yet, none have been reported.
Because association does not establish causation, and because the risk haplotype was largely upstream of the coding sequence, the sequence variation associated with schizophrenia could be involved in the action of another gene (c.f. lactase deficiency, above). Hence, the authors also studied transgenic mice to see whether disruption of NRG1 mimics phenomena associated with schizophrenia.244 NRG1 knockout mice are nonviable, but NRG1+/− heterozygotes expressing 50% of normal overall levels of NRG1 were created by inserting a stop-codon in the trans-membrane domain (exon 11). The mice developed normally but were hyperactive when exposed to novel environments, and were abnormal in the prepulse inhibition of startle paradigm (PPI). Clozapine ameliorated the hyperactivity but not the PPI deficit. There was also a small reduction in whole brain NMDAR binding sites. Similar but less severe abnormalities were observed in mice heterozygous for a knockout of ErbB4, a receptor mediating postsynaptic effects of NRG1. These findings, together with those in another hypomorph mouse245 provide some support for the plausibility of NRG1 as a schizophrenia gene. However, it is interesting to note a potential inconsistency between the genetic findings and the mouse model: the core risk haplotype implicates NRG1 type II (see below), which is the only isoform not disrupted by the trans-membrane domain construct used to create the animal model.
Following the original report, evidence of association between NRG1 and schizophrenia has emerged in multiple populations. Using the same SNPs, Stefansson et al246 genotyped 609 patients and 618 controls from Scotland and found an increased frequency in schizophrenia of the same haplotype (P=0.0003, one-tailed), and with a similar relative risk to that of the Icelandic sample. Several of the individual SNPs were also significantly more frequent in patients, but with a lesser relative risk than the haplotype. Three markers within the Icelandic core haplotype were also typed in 709 unrelated cases and 710 blood bank controls from England and Ireland.247 These data are much less compelling, but still suggestive; each individual marker was decidedly negative (all P>0.75) but the three marker haplotype showed weak association with schizophrenia (P<0.04, one-tailed). When only the 141 subjects with an affected first-degree relative were included, the frequency of the high-risk haplotype was enhanced from 9.5% in the whole sample to 11.6% (compared with 7.5% in controls; P<0.02, one-tailed)—somewhat akin to the dysbindin association in the Swedish sample mentioned. However, after this parsing procedure, the nonfamilial sample, which presumably is similar to the Scottish singleton sample246, was not associated with the NRG1 haplotype. Yang et al248 conducted a family-based association analysis of 248 Han Chinese trios, typing only three SNPs, viz. the most significant single Icelandic SNP in the 5′ upstream sequence, a nonsynonymous SNP in the second exon, and a SNP in the 5th intron. They found 50% overtransmission (P<0.005) of the same allele that was positive in Stefansson et al,244 and association at both other SNPs. Although these SNPs span over a million bases and are not clearly in LD, these investigators also showed strong association to several fairly common computational haplotypes made up of these three markers. In a separate Chinese sample of 540 patients and 279 controls from Shanghai, Tang et al249 typed 13 microsatellites spanning a 540 kb region around the 5′ end of NRG1. They found evidence for association with individual markers and with four and five marker haplotypes in the region of the Icelandic core haplotype, though different alleles were associated with increased risk in this sample. Another study, from Ireland, found association of several NRG1 markers to schizophrenia in 243 cases compared to 222 controls.250 The strongest association was to a haplotype within intron 1 which overlapped with, but was not identical to, that found by Stefansson et al244,246 A further Chinese study reports strong association with another closely overlapping haplotype, though with a different set of SNPs, in both a family sample (184 trios, 138 sib pairs) and a case–control population of 298 patients and 336 controls.251 The most recent report also shows association with 5′ SNPs in NRG1 in Han Chinese (369 patients, 299 controls, and 352 family trios).252
These results, notwithstanding two negative studies,253,254 argue convincingly that NRG1 is a likely susceptibility gene for schizophrenia. The associations are clustered in two regions, one in the 5′ regulatory domain of the gene, and one further downstream. The pattern suggests that NRG1, like dysbindin, manifests allelic heterogeneity with respect to risk for schizophrenia. The occurrence of multiple risk haplotypes, spanning distant functional elements of the gene, enhances the likelihood that NRG1 is the gene, and not a coincidental neighbour. However, Corvin et al250 noted that the variants associated with schizophrenia in their study occurred close to an expressed sequence tag (EST) cluster located within the large first intron. The function of the EST is unknown, but it is possible that variation in this expressed fragment, or in a cryptic exon, might contribute to the associations at the NRG1 locus.
How might variation in NRG1 impact on susceptibility for schizophrenia, and how can definitive evidence for its involvement be produced? As with the other genes, substantial additional information beyond statistical association is needed, including identification of functional sequence variants, knowledge of the normal expression and roles of the many NRG1 isoforms in the brain and, critically, demonstration of what is actually different about the biology of NRG1 in schizophrenia. Regarding expression, NRG1 mRNA and protein is detected in neurons of many areas of developing and adult human brain, including hippocampus, cerebellum, neocortex, and some subcortical nuclei, with immunoreactivity observed in cell bodies, dendrites, and axonal projections, depending on the neuron population concerned.255 In rat, the various NRG1 isoforms, discussed below, show differing cellular and regional expression profiles, highlighting how complex will be its full characterization in human brain.256 One study of NRG1 expression in schizophrenia has been reported, using real-time PCR to quantify the three major NRG1 mRNA isoforms in DPFC.257 It found a small increase in the type I isoform, and weak evidence of a change in the relative abundance of the isoforms; neither finding was related to subjects’ genotype at the two SNPs most strongly associated with schizophrenia in the original study.244 There is preliminary evidence confirming upregulation of type I NRG1, in another brain series (A Law, B Lipska, CS Weickert, PJ Harrison, DR Weinberger and J Kleinman, unpublished), and more extensive studies of NRG1 expression in schizophrenia and its relationship to genotype are underway.
Regarding what is known about NRG1 function, a large literature has emerged over the past decade, showing its involvement in remarkably diverse aspects of developmental biology, both in the brain and peripherally. A seminal review should be consulted for details.258 NRG1 is a member of the neuregulin family, comprising NRG1-4, identified in 1992 from a search for proteins that interacted with cell cycle signalling pathways. NRG1 is a huge gene (1.4 MB) and is really a family in itself, giving rise to at least 15 distinct peptides, derived from three principal isoforms, types I, II, III; a type IV has recently been reported too.259 The gene contains multiple regulatory elements and promoter sequences, and the isoforms reflect differing transcription initiation sites and alternative splicing. (A range of other names for NRG1 isoforms exist but are no longer useful; for example, the type II isoform was also called glial growth factor, which is misleading as glial differentiation is primarily subserved by type III.) All NRG1 isoforms contain an epidermal growth factor (EGF)-like motif that is critical for cell–cell signalling, and most are trans-membrane proteins. In the best-described mode of NRG1 signalling, proteolytic cleavage of NRG1 releases the N-terminal part including the EGF domain, which interacts with a membrane-associated ErbB-type tyrosine kinase receptor on the recipient cell—for example, postsynaptic neuron or glial cell. NRG1-ErbB receptor interaction leads to receptor dimerization, tyrosine phosphorylation and activation of downstream signalling pathways.260,261 Many other modes of NRG1 signalling may also occur. For example, type III NRG1 can interact with postsynaptic ErbB receptors, while still tethered to the presynaptic membrane, and in addition to forward signalling, cleavage of the type III intracellular domain sends retrograde signals to the nucleus and regulates gene expression within the NRG1-expressing cell.262 The large number of NRG1 signalling mechanisms and isoforms parallel the range of its functional effects.258,260,263,264,265,266 These include neuronal and glial functions implicated in schizophrenia, ranging from development (eg neuronal migration, axon guidance, synaptogenesis, glial differentiation, myelination), to neurotransmission and synaptic plasticity (eg recruitment of nicotinic, GABA, and NMDA receptors, long-term potentiation). It is a reasonable if vague working hypothesis that the genetic risk for schizophrenia associated with NRG1 is mediated by a molecular ‘bottleneck’ in NRG1 signalling that alters, probably to a small degree and in a temporally, spatially, and isoform-limited fashion, the efficiency of NRG1 effects on neural development and plasticity. It is impossible at this stage to predict which of the specific NRG1 functions within these broad domains is most relevant to its involvement in schizophrenia.
Regulator of G-protein signalling 4 (RGS4)
Mirnics et al267 compared gene expression profiles from DPFC in brains from five schizophrenia-control pairs. RGS4 was the only transcript consistently reduced, out of 7800 sampled. They expanded the sample to include five more pairs and performed in situ hybridization in DPFC, and in visual and motor cortices. In all locations, nine of the 10 patients had reduced expression of RGS4 mRNA. Noting that RGS4 maps to 1q21–22, the locus with the strongest single study whole genome linkage finding,162 Chowdari et al268 typed 13 SNPs across a 300 kb segment spanning the gene in three independent trio-type datasets: one from Pittsburgh (93 Caucasian trios), another from New Delhi (269 trios) and the third from the NIMHGI (39 full trios). Subpopulations of sib-pairs were also analysed in these samples for allele sharing. Despite the relatively small sample sizes, weak evidence for association to RGS4 was found in each of the populations, though not in an allele-consistent manner, and reaching significance for individual SNPs only in the NIMHGHI sample. In general, association was present for a haplotype block stretching from intron 1 to approximately 9 kb upstream of the transcription start site. However, in the two American samples, the significant alleles and haplotypes differed, and there was no significant association in the Indian sample, just a trend for one of the NIMHGI haplotypes. The Cardiff group have reported modest association to two of the RGS4 SNPs in their dataset of 709 cases and 710 controls,269 as did Morris et al270 in 196 subjects with schizophrenia and 231 controls.
Overall, the genetic data for RGS4 are suggestive, but it is unclear whether the positive results seen with linkage, association, and expression represent a convergence that greatly strengthens the candidacy of RGS4 as a susceptibility gene, or whether they are at least partly coincidental. For example, the positive genetic association to RGS4, which carries a low relative risk of schizophrenia, is unlikely to relate directly to the reduction in RGS4 mRNA, which was seen in nine out of 10 subjects.267 RGS4 expression is decreased in Alzheimer's disease, which is not associated with the gene,271 illustrating that the two findings are not necessarily linked. Also, no potential coding SNPs have been found in a resequenced sample or in cases from the expression series,268 leaving the SNPs without obvious functional correlates. On the other hand, RGS4 certainly has some biological plausibility as a schizophrenia gene. It is the most brain-enriched of the 19 human RGS transcripts,272 and is abundant in the cerebral cortex, with much lower levels in thalamus and basal ganglia.273 RGS4 is a GTPase activator which desensitizes Gi/o and Gq and so negatively modulates G protein-mediated signalling via some dopamine, metabotropic glutamate, and muscarinic receptors.274,275 RGS4 is involved in neuronal differentiation276 and is under dopaminergic regulation.277
Disrupted-in-schizophrenia 1 (DISC1)
St Clair et al278 described a large Scottish family in which a balanced translocation involving chromosomes 1 and 11 (1;11)(q42.1;q14.3) was strongly linked to psychopathology including schizophrenia, depression, and mania. The 1q breakpoint was cloned and found to involve two genes, called DISC1 and DISC2; the latter did not encode a protein, but may be an inhibitory RNA regulator of DISC1.279 Evidence for linkage of schizophrenia to 1q42 has been reported in three samples.280,281,282 In the Finnish samples,280,282 the marker with the highest LOD score mapped to DISC1. Moreover, in the original Scottish kindreds, translocation carriers (ie cytogenetically abnormal but without a psychiatric phenotype) have a reduced P300 amplitude,283 a physiological EEG trait manifested by patients with schizophrenia. In another report, a microsatellite marker near DISC1 was associated with impaired spatial working memory.284 These various pieces of circumstantial evidence support a possible role for DISC1 in susceptibility to schizophrenia, but chromosomal rearrangements can disrupt large regions, and the linkage and cognitive associations may be reflecting variation in another gene on 1q.
Direct evidence relating DISC1 to schizophrenia has come from two out of three studies. The negative study found no association with four DISC1 SNPs in a Scottish population of 267–328 schizophrenics and 426–726 unrelated controls.285 Hennah et al286 performed a family-based association study of 28 SNPs in 450 Finnish families and found a three marker haplotype spanning intron 1 to exon 2 that was significantly undertransmitted to female probands. They identified two other significant haplotypes as well, but could not exclude linkage as the basis of the finding because association was no longer significant when the families from the previous positive linkage analysis were excluded. Interestingly, one of the undertransmitted haplotypes is in another 1q42 gene, TRAX, which may coalesce with DISC1 by intergenic splicing and be expressed as a fusion protein under certain conditions.280,287 The third haplotype, which was the only overtransmitted one, spanned a 10 kb region including exon 9. This region is near the chromosomal breakpoint in the Scottish translocation families. However, this finding also was not significant in the nonlinked Finnish families. The association with an undertransmitted haplotype is difficult to interpret, and might suggest that not inheriting certain haplotypes is in some way protective against schizophrenia. In the third DISC1 study, Callicott et al288 typed 12 SNPs spanning the gene, including several from the earlier studies, in the Weinberger lab Caucasian ‘quad’ dataset (260 families containing index cases, usually one unaffected sib, and parents) and in an NIMHGI-derived family dataset (67 Caucasian and 51 African-American families). They found association (P<0.005) to a coding SNP in exon 11 (cys704ser) that had been negative in both earlier studies. A three-marker haplotype from intron 8 to exon 10, including the overtransmitted ser704 allele, was also positive (P=0.005 global, P<0.002 specific). A trend (P=0.06) for overtransmission of the same haplotype was found in the Caucasian NIMHGI sample, and significant association to a SNP in intron 3 was also found in the African Americans. These investigators also found relationships between the cys704ser polymorphism and intermediate phenotypes related to schizophrenia, with the ser704 allele being associated with reduced hippocampal grey matter volume and NAA signal, and abnormal engagement of the hippocampus during several cognitive tasks as assayed with fMRI. These convergent data implicate DISC1 in genetic risk for schizophrenia and suggest that the mechanism may involve hippocampal development and function; as such they also illustrate how genes may implicate the various parameters of neuropathology outlined above. However, the evidence for DISC1 association with schizophrenia is not yet conclusive, and the causative variant(s) await identification.
DISC1 is a complex gene with protean but poorly understood implications for development and plasticity. The gene has 13 exons spanning over 200 kb, and encodes a protein of 854 amino acids. It is associated with numerous cytoskeletal proteins involved in centrosomal and microtubule function, and with cell migration, neurite outgrowth, and membrane trafficking of receptors and possibly mitochondrial function.289,290,291,292,293 Different domains of DISC1 interact with distinct families of proteins involved in these various functions, so allelic heterogeneity could impact differentially on DISC1 function. For example, regions of the gene downstream from the translocation breakpoint, including exon 10, are critical for DISC1 binding to neurite outgrowth factors (eg NUDEL and FEZ1) and transcription factors (eg ATF4/CREB2), and for the normal intracellular distribution of DISC1.291 DISC1 mRNA expression is highest prenatally, at least in the mouse,292 and there is a splice variant expressed during human fetal development which alters exon 10.279 Consistent with its potential influence on hippocampal structure and function,284 DISC1 expression is prominent in limbic structures.294,295 The distribution of DISC1 between cell types and compartments has not been described in detail, but is said to be localized in mitochondria, as well as in cytoplasm, nuclei, neurites, and the plasma membrane.289 DISC1 expression in schizophrenia has not yet been reported.
Metabotropic glutamate receptor-3 (GRM3; mGluR3)
After dopamine, glutamate is arguably the neurotransmitter system most implicated in schizophrenia,145,146,147,148,149,150,151 and glutamate receptor genes have been favoured by many investigators.296,297 However, despite several isolated weak associations, only a type II metabotropic receptor, GRM3 (mGluR3), meets the criterion172 of association in three independent studies. GRM3 maps to 7q21–22, not a locus highlighted in either meta-analysis.164,165 The first, equivocal, report of GRM3 association with schizophrenia came in a German study. In 265 patients and 227 controls, an increased frequency of a SNP in exon 3 was found; however, the same paper also reported a nonreplication in a second case–control population (288 patients, 162 controls), and in 128 family trios.298 Fujii et al,299 in a case–control analysis of six SNPs in 100 Japanese patients and 100 controls, found association to another SNP in intron 3, and to various two and three marker haplotypes containing this SNP, especially those spanning introns 3–5. Egan et al300 genotyped seven common SNPs, including the positive ones from the earlier studies, in the Weinberger lab dataset consisting then of 217 mostly Caucasian-American families, and also the Caucasian (n=67) and African-American (n=51) family subsets of the NIMHGI. There was significant association to a SNP in intron 2, and trends for association with the positive SNPs from the earlier studies, together with strong association (P<0.0001) to common three and five marker haplotypes that included them. Trends for association (P=0.03–0.07) were also found for SNPs in the NIMHGI samples. Egan et al300 also searched for evidence that the intron 2 variant associated with schizophrenia either causes or monitors a change in GRM3 function. They found that the allele was associated, even in normal individuals, with impairments commonly seen in schizophrenia, including poorer episodic memory and attention, abnormal prefrontal and hippocampal activation with fMRI, and reduced prefrontal NAA signal measured with MR spectroscopy. Also, in post-mortem human DPFC, there was weak evidence for genotype effects on GRM3 expression, and a strong inverse association between the high-risk GRM3 allele and mRNA for the glial glutamate transporter EAAT2. These findings suggest direct and indirect influences of GRM3 variation upon the regulation of synaptic glutamate, consistent with the known roles of GRM3 in such processes.301,302
Other aspects of GRM3 biology strengthen its candidacy as a schizophrenia gene. It is a heteroceptor modulating serotonin and dopamine transmission and associated effects.301,303,304 Type II metabotropic glutamate receptor agonists (GRM2 and 3) block the behavioural and cognitive effects of NMDAR antagonism.305 The peptide neurotransmitter N-acetylaspartylglutamate (NAAG; which is hydrolysed to produce NAA and glutamate), is itself a GRM3 agonist and has NMDAR activity.306 These findings link GRM3 with models of schizophrenia centred around NMDAR transmission and NAAG.146,147,149,307,308 Consistent with these roles, GRM3 is expressed in many neuron populations, with a predominantly presynaptic localization, as well as being expressed in astrocytes and oligodendrocytes.309,310,311,312,313,314,315 Thus, convergent data implicate GRM3 as a schizophrenia susceptibility gene and suggest that the mechanism involves an alteration in prefrontal and hippocampal glutamate neurotransmission and the functioning of these regions. However, further positive associations and identification of a functional variant that convincingly explains the association are needed. Schizophrenia itself does not affect GRM3 mRNA or protein levels in the DPFC,312,314 or GRM3 mRNA in thalamus316 or hippocampus (B Lipska, DR Weinberger and J Kleinman, unpublished). However, these studies have not yet considered the possibility of cell-specific alterations or alternative splicing. Moreover, studies of GRM3 protein are confounded by antibody crossreactivity with GRM2. It would thus be premature to conclude that GRM3 involvement in schizophrenia is not reflected, at least partly, by altered expression.
G72 (and DAAO)
The novel gene G72 was cloned from a 5 MB ‘gene desert’ in the 13q linkage region after construction of a dense LD map of SNPs across the region.317 Following annotation and in vitro translation, the gene was shown to encode a 150+ amino-acid protein with little evolutionary homology, and to be part of a larger gene on the opposite DNA strand, called G30. Several SNPs and haplotypes in this region were found to be associated with schizophrenia in a French-Canadian case–control sample, and one of these SNPs also showed association in a Russian case–control sample.317 Biochemical experiments revealed that G72 protein activated a second protein, D-amino acid oxidase (DAAO), which was known to be involved in the metabolism of D-serine, an agonist at the glycine modulatory site of the NMDAR.318 This made both genes attractive glutamatergic candidates for many of the same reasons mentioned with regard to GRM3. Furthermore, the authors reported that four SNPs in DAAO (located at 12q24) were themselves associated with schizophrenia in the French-Canadian sample, along with some indication that SNPs in DAAO and G72 might act in combination to influence schizophrenia risk.317 Association of both genes to schizophrenia was confirmed, in a study of seven G72 SNPs and three DAAO SNPs, in a German case–control study of 299 patients and 300 controls.319 However, the alleles were not the same as those reported by Chumakov et al,317 the strongest association for G72 was with a different, and under-represented, haplotype, and the DAAO associations were with the opposite alleles. This mixture of replication in a broad sense (ie to variation in the gene) but nonreplication in detail (ie to a particular allele or haplotype or to risk v. ‘protective’ haplotypes) is perplexing and reminiscent of some of the data with other genes. However, further evidence for G72 as a risk factor for schizophrenia has emerged. A Han Chinese study320 of G72 in 233 trios found significant overtransmission of two of the original SNPs, and a haplotype including all three SNPs, consistent with the original report.317 Similarly, a small study has reported association of several G72 SNPs and haplotypes, but not those in DAAO, with childhood-onset schizophrenia.321 Thus, in total, the genetic data are suggestive for G72 (and it passes the ‘three replications’ threshold172), though much less so for DAAO (which does not). In the Weinberger lab schizophrenia datasets, Goldberg et al322 found no significant associations to 11 SNPs in G72 nor to five in DAAO, but they did observe that the positive SNPs in the study by Chumakov et al317 are associated in the predicted direction with cognitive and physiological abnormalities related to prefrontal and hippocampal function in schizophrenia. This suggests that G72 gene variation may show greater penetrance for, or be more directly related to, such intermediate phenotypes and that G72 may militate towards the emergence of schizophrenia via disruption of function in these cortical systems (see below), akin to the findings for GRM3, COMT, and DISC1.
DAAO is localized in peroxisomes in astrocytes and some neurons in the rat brain,323 but otherwise little is known about G72 or DAAO expression. There is a preliminary report of an equivocal increase of DAAO mRNA in DPFC in schizophrenia.324
In this section, we briefly review some other genes which do not yet meet the ‘three replication’ criterion, but for which there is sufficient evidence to merit mention, and which have neurobiological plausibility.
Calcineurin is a multifunctional calcium-dependent serine/threonine phosphatase, known to be centrally involved in many aspects of synaptic plasticity, on both sides of the synapse.325,326,327 It has particular roles in glutamate and dopamine signalling and their interactions, including regulation of DARPP32, a molecular node of convergence between D1 and NMDAR signalling pathways.328,329 Interestingly, calcineurin appears to be absent from many inhibitory neurons.330 Based on evidence that calcineurin knockout mice exhibit deficits in behavioural and pharmacological assays used as animal models of aspects of schizophrenia (eg motor and social behaviour, PPI, responses to NMDAR antagonists),331 Gerber et al332 searched for associations between schizophrenia and variations in four calcineurin subunit genes in two family trio datasets. In an American sample of 210 trios consisting in part of the ethnically diverse NIMHGI dataset, weak evidence was found for association to two of 16 SNPs in the gamma catalytic subunit (PPP3CC), and to common two and five marker haplotypes including these SNPs. In a replication sample of 200 trios from South Africa, described in minimal detail, there was a trend for overtransmission of the same five marker haplotype, with a similar odds ratio (∼1.3). Calcineurin is another attractive candidate gene because of its functional roles,325,326,327,331 and given the location of the PPP3CC subunit gene close to NRG1, at 8p21, a locus highlighted in the meta-analyses.164,165 Moreover, calcineurin mRNA and protein expression is decreased in schizophrenia in the hippocampus (SL Eastwood and PJ Harrison, unpublished). However, calcineurin expression is dynamic and altered in many disease and experimental states, and so this observation only incrementally adds to the candidacy of the gene (c.f. RGS4), certainly until the association with schizophrenia is replicated, and the functional genetic variants identified.
The α7 nicotinic receptor gene (CHRNA7) is implicated in schizophrenia by considerable, albeit partially circumstantial, evidence. This includes the modulation by nicotine of attentional and sensory processing, such as those assayed by the P50 response, and by the striking association between smoking and schizophrenia.333,334 Direct evidence for involvement of CHRNA7 in schizophrenia comes from the University of Colorado group, who first reported that abnormalities of the P50 response were linked to a region on 15q13–14 containing CHRNA7; this region had also shown suggestive linkage to schizophrenia in a few studies.335 The same group later found that combinations of SNPs in the CHRNA7 promoter region were associated in a case–control analysis with the abnormal P50 phenotype and possibly with schizophrenia; several of these SNPs appeared functional in an in vitro gene expression assay.336 CHRNA7 is involved both pre- and postsynaptically in modulating dopamine and glutamate signalling,337,338,339,340 and is recruited to the synapse by NRG1.341 It is expressed widely,342 and prominently in inhibitory interneurons.333,339,343 CHRNA7 expression, measured as immunoreactivity or binding site densities, is reduced in schizophrenia in several areas, including hippocampus,344 thalamus,345 frontal cortex,346 and cingulate cortex.347 The biological candidacy of CHRNA7 is therefore impressive, but its impact on genetic risk remains uncertain prior to independent replication. Genetic analyses of CHRNA7 are complicated by partial duplication of the gene.348
Proline dehydrogenase (PRODH2) was identified as a possible susceptibility gene after extensive association analysis of SNPs in genes within a 1.5 Mb block of the VCFS deletion region on 22q11. This revealed several missense SNPs within a locus containing PRODH2, which were associated with schizophrenia in an American sample of 107 trios, part of the population from the study of calcineurin (above) and consisting primarily of the NIMHGI datasets.349 A three marker haplotype spanning three 3′ exons, but seemingly independent of the missense SNPs, was also positive in an ethnically diverse childhood-onset schizophrenia sample (29 trios) and in a small adult case–control dataset from South Africa. In the two adult samples, the power of the association was enhanced by restricting the analysis to cases with earlier ages of onset, and the odds ratio for overtransmission of the high-risk haplotype in these cases was unusually high (>4). Liu et al350 then demonstrated that PRODH2 has a duplicated pseudogene about 1.5 Mb downstream which contains a few of the missense SNPs they had associated with schizophrenia, raising the interesting question of whether the variation in PRODH2 was the result of conversion between the real and pseudogene sequences. Also, the congenital syndrome of hyperprolinemia, caused by PRODH2 deficiency, may be weakly associated with schizophrenia.351 Some functional data support PRODH2 as a candidate gene, notably from the PRODH knockout mouse,352 which has PPI deficits and decreased levels of glutamate and GABA in some brain regions, consistent with the role of the enzyme in regulating proline levels, which in turn influence glutamate metabolism and release.353,354 However, the status of PRODH2 as a schizophrenia susceptibility gene remains equivocal, as association has not been replicated, either in a Chinese population (albeit only one of the positive SNPs349 was studied),355 or in a Japanese sample,356 or in a large UK/Irish case–control group, which included early onset cases and several cases of VCFS with psychosis, along with 55 Bulgarian trios.357 These authors subsequently sequenced the entire gene in 14 patients and failed to reveal any additional associated SNPs.358 PRODH2 mRNA is unaltered in the DPFC in schizophrenia.222
Emamian et al359 used a combination of experiments to implicate Akt1 (protein kinase B) as a susceptibility gene. Starting with the broad concept that kinases and phosphatases are candidate genes, they measured the abundance of several such proteins in lymphocytes from patients and controls. After finding Akt1 to be consistently altered (reduced) in schizophrenia, they showed that this also occurred in the hippocampus and frontal cortex of two case–control series, and was accompanied by decreased phosphorylation of glycogen synthase kinase 3β (GSK-3β), a target of Akt1 and a molecule of prior interest in schizophrenia.360 Genetic association was then found between an Akt1 haplotype and schizophrenia in 268 affected families, with the risk haplotype being associated with lower Akt1 expression in lymphocytes. Finally, Akt1 knockout mice were shown to be more sensitive to amphetamine-induced PPI disruption. The Akt1 study359 serves as another good illustration of how putative susceptibility genes for schizophrenia may be identified using an hypothesis-driven strategy combined with convergent, multifaceted experimental evidence. However, the Akt1 association with schizophrenia remains to be confirmed, and caution must be exerted when postulating the biological mechanism explaining the association, as Akt1 has multiple and diverse functions.361
How does genetic variation confer susceptibility to schizophrenia?
Simplistically, genetic variation affects disease susceptibility in one of two ways. Either it changes the structure of the encoded protein (eg by an amino-acid substitution or frame-shift mutation) or it alters the expression of the gene (eg by altering some parameter of transcription or translation) and thereby the amount or distribution of the protein. Both processes, which in reality are not mutually exclusive, ultimately exert their effects by affecting the function of the protein. The COMT val158met polymorphism is a prime example of the former type. Variation in the other susceptibility genes may also come into this category, but this would require that the SNPs currently associated with the disease, virtually all of which are noncoding, are acting as markers for coding variants; this will become increasingly unlikely as the genes are resequenced more extensively in affected individuals and as no unique transcripts or protein isoforms are found. Instead, it seems likely that many of the associations, if genuine, come into the category of altering expression of the gene. This is a more complex and subtle manifestation of genetic predisposition, and may be exerted in many ways. Promoter variants can markedly impact on transcriptional activity of the gene.362,363,364,365 Intronic SNPs can also affect transcription, or alter splicing or mRNA stability and thus the relative abundance and proportions of isoforms;363,366 the differing functions ascribed to NRG1 isoforms mentioned above illustrates the potential pathogenic as well as physiological consequences of this.367 SNPs in the 3′ UTR may alter mRNA stability and thence translation.368,369 Even conservative exonic SNPs, usually considered functionless, can alter mRNA structure and translation.370,371 Finally, the effects of individual SNPs cannot be studied alone, since they are influenced by haplotype background.372 As if these molecular complexities were not enough, studies cannot rely solely on in vitro or animal models, because some aspects of gene regulation may be unique to the human brain in vivo, and so must include post-mortem research with all the practical problems that entails. It must also be borne in mind that altered expression of susceptibility genes need not occur only because of genetic variants that are associated with the disease. The susceptibility genes may encode molecules that represent convergent nodes in signalling pathways that can be affected via numerous other entry points, including variations in other genes that feed into these pathways, and lead to compensatory or secondary changes. Finally, it is worth reiterating that the relationship between genotype and phenotype may also be complicated by epigenetic factors (heritable factors without sequence variation, eg, changes in DNA methylation and chromatin structure), that regulate gene activity and which have been advocated to be important in schizophrenia.373,374,375
Overall, therefore, understanding in molecular terms how genetic variation confers susceptibility to schizophrenia may prove a deceptively difficult task: first the ‘true’ risk variant(s) in each gene must be identified, and then the way in which a variant alters the function of the encoded protein must be established. Even when this has been achieved, the picture will still be incomplete, since there will likely be important gene–gene, gene–environment and protein–protein interactions to be studied, not to mention the protean effects on downstream molecular and neural system processes. The discovery of schizophrenia susceptibility genes may well be looked back upon as a relatively trivial task compared to the subsequent elucidation of how they operate.
Schizophrenia as a complex genetic disorder of cortical microcircuits
Undaunted by the preceding paragraph, we turn finally to perhaps the most interesting question of all: what does the identity of the genes tell us about the nature of schizophrenia? While any answer is in the realm of speculation, we would opine, parsimoniously, that the susceptibility genes influence brain function directly, in a way which is consistent with existing neurobiological understanding of schizophrenia and with the neuropathological clues mentioned above, and, furthermore, that the genes may confer susceptibility by converging on a shared pathophysiological process. Specifically, it has been noted376 that most if not all the susceptibility genes impact upon the molecular biology of the synapse, in keeping with the view of schizophrenia as a disorder of synaptic signalling. The additional genes reported since this proposal was made (calcineurin, Akt1, GRM3) are consistent with it. The genetic influences on the synapse include effects on receptors (eg GRM3, CHRNA7, G72), signal transduction (RGS4), and regulation of plasticity and synaptogenesis (NRG1, dysbindin, DISC1, calcineurin). Figure 1 is a highly simplified schematic of these relationships. Glutamatergic synapses and processes appear to be particularly affected,376,377 notably NMDAR signalling, which is influenced in one way or another by most if not all of the current catalogue of putative susceptibility genes (Figure 2). However, there are also direct links for several of the genes with dopaminergic (viz. COMT) and GABAergic systems,333,378,379 and the latter may itself harbour susceptibility genes. Thus, when considered along with the manifold functional and anatomical connections between these transmitter systems, it is already clear that there is no one single target upon which the genes act, nor any one cell type or compartment wherein they are expressed. In turn, therefore, although the genes currently appear to disproportionately impact glutamate and NMDAR, the genetic basis of schizophrenia is most unlikely to reside in any one transmitter or receptor, and the disease will not prove reducible to, or be genetically explained by, a single molecular processs or neurotransmitter signalling system.
The evidence that schizophrenia susceptibility genes affect diverse synaptic processes suggests that it will not be synapses per se but the neural circuits in which they participate which will prove to be the appropriate explanatory level to understand how the genetic influences operate. This is consistent with the view that the disorder is fundamentally one of abnormal information processing at the highest level, and such abnormalities are probably best understood in terms of malfunction of cortical microcircuits.218 In other words, the real locus of genetic convergence, if there is one, is downstream of any specific molecule or synaptic event per se, and resides in some integrative activity or emergent property of the circuits subserving the core cognitive elements affected in schizophrenia, for example, by impairing the signal-to-noise ratio and decreasing the efficiency of information processing.218,380 This makes it conceivable that various combinations of susceptibility genes can converge on synaptic processing in these microcircuits to effect a common pattern of dysfunction and emergent symptoms, though the specific combination of genes and possibly alleles can vary across ill individuals.
Returning to the other main theme of this review, viz. the neuropathology of schizophrenia, the susceptibility genes have implications for how the data should be interpreted. If the genes are, in one way or another, influencing the properties of certain neural circuits, then the morphological alterations are likely a secondary manifestation (both in terms of causal sequence and importance) of this developmental and dynamic shift in the normal regulation of synaptic connectivity and activity. There are already many examples of an overlap between the susceptibility genes and reported pathological features. For example, the alterations in dendritic spines in schizophrenia may reflect the effects of mutiple genes on diverse aspects of NMDA signalling, or perhaps, the close relationship between calcineurin and spinophilin,329 or that between Akt1 and reelin.381,382 Similarly, the influences of NRG1 signalling on oligodendrocyte differentiation383 and interneuron development384 might contribute to the morphometric alterations seen in these cell populations in schizophrenia. The same principle would be predicted to govern the relationship with pathology for subsequent susceptibility genes which are discovered (eg under the 2p locus peak),165 and also to apply in other disorders sharing susceptibility genes and/or neuropathological features with schizophrenia, for example, bipolar disorder,100,385,386 autism,387 and mental retardation syndromes.388,389 By the same token, the clinical correlates of the genetic neuropathology of schizophrenia seem more likely to be features which are observed across broad diagnostic categories and which are stable across time (eg neurocognitive impairments) rather than the specific and fluctuating psychotic symptoms of the syndrome.
Having indulged in this speculation upon how the genes might drive the pathogenesis of schizophrenia, we finish with some due caveats. First, the ideas are vague, lacking the molecular elegance of disease models built around a specific gene or biochemical mechanism. Second, it is hard to envisage that any brain disease, or any gene expressed in the brain, could entirely lack effects on one or other aspect of neural plasticity and functioning, and so the proposals are inherently superficial. Third, the fact that no abnormal proteins have been identified, and the genetic variants are common (and vary between studies), suggests that the disease process will prove, in molecular terms, to be subtle, complex, and fundamentally a quantitative trait rather than a qualitative abnormality—just as is the case for the genetics and the neuropathology. These issues mean that though a web can be woven which ties all the putative schizophrenia genes to synaptic plasticity, NMDAR signalling, etc, it is not clear whether this is what makes these genes relevant to schizophrenia. It also means that much more evidence will be needed, from genetic, neuropathological, and experimental approaches, before this or any other incisive molecular pathophysiological model can be established. On the other hand, at this stage of research, the notion that the neurobiology of schizophrenia might be reducible to one or a few common pathways may be a useful starting point, to be refined or refuted as research progresses. A heuristic value has certainly been apparent for the aforementioned β-amyloid hypothesis of Alzheimer's disease, which has been central in focusing pathophysiological and pharmacotherapeutic strategies, even while debate continues regarding its detailed specification2 and even its validity.390 Finally, we note that despite its unequivocal genes and overt neuropathology, even Alzheimer's disease research is not immune from the genetic uncertainties being faced in schizophrenia.391
Schizophrenia continues to lack a diagnostic neuropathology, convincing causative genetic mutations, and even unequivocally replicated associations with the same alleles or haplotype within each gene. Nevertheless, the weight and convergence of evidence of susceptibility genes for schizophrenia cannot be dismissed. Even if some genes prove to be false positives, others will remain and provide important insights into the pathogenesis and pathophysiology of psychosis. Genes represent mechanisms of disease, and in a field previously based on phenomenology, this is a sea change in the science of schizophrenia. Future neuropathological investigations can now take genetic background into account, while studies relating genetic variation to schizophrenia and its intermediate phenotypes will be complemented by investigations characterizing the distribution, abundance, and potential modifications of the gene products. Together these approaches should allow identification of the molecular and cellular mechanisms that link the susceptibility genes to the neurobiology, both structural and neurochemical, although the scale of this task should not be underestimated. In the process, the research will determine the extent to which the genes operate in a convergent way, whether they do in fact impact primarily on synaptic plasticity in the service of microcircuit information processing, and it will allow this suggestion to be mechanistically specified. Both the genes themselves and the biochemical pathways in which they participate will be attractive, though not necessarily tractable, therapeutic targets. What cannot be disputed is that the discovery of susceptibility genes for schizophrenia changes the research landscape and its horizons profoundly and permanently.