Schizophrenia: an example of complex genetic disease

Schizophrenia is a complex multifactorial and polygenic disease. Although family, twin and adoption studies have consistently suggested that genetic factors play an important aetiological role in schizophrenia, until now, very little is known about the nature and number of these genetic factors. The genetic basis may involve at least several genes of only mild influence with interactive effects, but the relationship between observed genetic risk factors and specific DNA variants or protein alterations has not so far been identified.

Linkage analysis, which seeks to localize a chromosomal region associated with the transmission of a disease within families, and ultimately the responsible gene, may be very difficult in multifactorial diseases such as schizophrenia. During recent years, numerous linkage claims have been made with moderate Lod scores. In many cases in which an attempt has been made to replicate the linkage claim, the original positive finding has not been replicated. Linkage analysis in schizophrenia is problematic for several reasons: (1) uncertainty about clinical status in relatives; (2) genetic heterogeneity may be substantial; (3) penetrance of susceptibility genotypes may be incomplete (the genotype, by interacting with the environment (either biological or psychosocial) produces either a clinical phenotype meeting the diagnostic criteria for the disorder, a sub-clinical phenotype or a symptom-free phenotype); (4) phenocopies are subjects who may meet the diagnostic criteria but do not have the genotype of the disorder; (5) misspecification of the genetic model may strongly affect the results of multipoint linkage; and (6) there are few large families with multiple affected individuals (Gershon and Cloninger 1994). Although sibling pairs linkage studies avoid some of the problems encountered in family-based linkage analysis, such as the estimation of penetrance, the method has its own limitations and is most powerful under conditions of moderate genetic heterogeneity. Then, a large number of families has to be studied before a reliable linkage finding can be obtained and replicated.

Linkage or association studies may be conducted using either the whole genome or candidate genes. Indeed in this latter approach, genes of neurobiological interest are first examined at the DNA level to search for variants (whenever possible, functional variants).

Large case–control studies, consisting of unrelated schizophrenic patients and ethnically similar (to avoid stratification bias), unrelated controls, have been used in order to search for an association between a variant allele and the disease. Family-based strategies, which are using the Transmission Disequilibrium Test (some alleles may be preferentially transmitted from parents to affected children), are less likely to have population stratification problems. However, candidate gene association studies, and mostly whole genome association studies, which are using multiple Single Nucleotide Polymorphisms located throughout the genome, have led to false results because of multiple testing and the presence of variable extents of linkage disequilibrium. In addition, the great variation across studies in the statistical tests used, in the phenotypes and in the polymorphisms or haplotypes used, make the comparison between studies difficult. Furthermore, some studies report negative studies with the disease itself but positive associations to subgroups based on sex, clinical symptoms, etc. These studies suffer from multiple testing and can lead to many false-positive results. Finally, in many cases negative studies are not published. A consensus definition of replication for genetic association, in a complex disease with allelic heterogeneity, is difficult to achieve. Ioannidis et al. (2006) and DeLisi et al. (2006) have recently published a list of recommendations for reporting associations in genetic diseases.

Susceptibility genes or loci have been identified on the basis of linkage and association studies (NRG1 or neuregulin 1 gene located at Ch 8p13, DTNBP1 or dysbindin gene located at Ch 6p22 (Williams et al. 2005), G72/G30 or d-amino acid oxidase activator gene region in Ch 13q). All these genes may be involved in glutamatergic signalling through NMDA receptors. However, no consistent biological hypotheses have been reported until now. Significant associations of variants located on these latter susceptibility genes have also been reported with bipolar disorders. Following these recent results, the validity of the Kraepelinian diagnostic distinction between schizophrenia and bipolar disorder is challenged. Indeed, there is increasing evidence for an overlap in genetic susceptibility across the traditional classification categories. For example, the NRG1 gene might contribute to a range of psychosis-related phenotypes (schizophrenia, bipolar disorders with manic or mood incongruent psychotic features, schizotypal personality traits or even psychosis that occurs in some patients with Alzheimer's disease (Harrison and Law 2006)).

Attempts have been made to decrease the complexity of genetic analysis by using more elementary measurable phenotypes associated with the disease (called intermediate phenotypes or endophenotypes). The number of genes required to produce variations in these traits may be fewer than those involved in producing a psychiatric diagnosis entity, ideally they would have monogenic roots. Neuropsychological, neurophysiological, biochemical or neuroanatomical measurements have been proposed as endophenotypes. To meet endophenotype criteria, candidate markers have to be: (1) heritable; (2) primarily state-independent; (3) associated with illness in the population; (4) to co segregate with the disease within families; and (5) to be found in affected, as well as unaffected family members, at a higher rate than in the general population (for review, see Gottesman and Gould 2003). In addition, these endophenotypes may be used for classification or diagnosis (Louchart-de la Chapelle et al. 2005) and in the development of animal models. The auditory-evoked P50 paradigm was described by Freedman and colleagues to measure deficit in sensory-motor gating in schizophrenic patients. Animal and human studies have suggested a role for septohippocampal cholinergic activity (involving the α7 subunit of nicotinic receptors) in sensory gating. Smoking affects sensory gating. Several studies have shown that promoter variants (in the α7 gene) or variants located in the α7-like gene are associated with P50 inhibition deficits (Raux et al. 2002; for review, see Leonard and Freedman 2006). This paradigm has also been used to identify a potential susceptibility locus for schizophrenia on chromosome 15, a region where the gene for the α7 nicotinic receptor is located. Recent reports also link this locus to nicotine addiction in schizophrenic subjects.

Taking into account the low correlations between genotype and phenotype and considering the difficulties encountered with linkage analyses in complex diseases, several teams have chosen alternative strategies for the identification of candidate genes. In a given proband, when a cytogenetic event is necessary and sufficient to predispose to a psychiatric phenotype (e.g., psychosis), molecular cytogenetics is a powerful alternative strategy to nominate candidate genes. Thus, the DISC 1 gene and the PRODH gene have been identified as candidate genes for schizophrenia. The discovery of DISC1 gene on chromosome 1q42 has its roots in a balanced translocation (1:11) that segregated with schizophrenia, bipolar disorder and recurrent major depression. However its biological function is still unknown (for review, see Porteous et al. 2006). In the same way, DNA sequence variations within the 22q11 velo-cardio-facial syndrome chromosomal region are likely to confer susceptibility to psychotic disorders. Indeed, there is an increased prevalence (25–30%) of schizophrenia among patients with the 22q11 deletion syndrome. Moreover, this region has been implicated in schizophrenia by linkage studies. The PRODH gene, which encodes the mitochondrial enzyme proline dehydrogenase, is located in this region. Homozygous deletion and/or missense mutations of this gene were associated with high plasma proline levels, severe mental retardation and epilepsy (Jacquet et al. 2003). The same molecular alterations of the PRODH gene (either deletion and/or variants affecting highly conserved amino acids and causing drastic reduction in enzyme activity), but at the heterozygous state, were associated with moderate hyperprolinemia in certain forms of psychosis (Jacquet et al. 2005). In addition, high levels of prolinemia observed in a subset of 22q11 deletion syndrome patients were associated with lower IQ, and, in some cases, epilepsy (Raux et al. manuscript in preparation). Interestingly, Liu et al. (2002) have reported an association between several of these PRODH variants and schizophrenia. Lastly, mice that model PRODH deficiency already exist and may help our understanding of the biological significance of this pathway. Homozygous truncating mutation of the PRODH gene in mice results in hyperprolinemia, in a dysregulation of glutamate release associated with a cortical dopaminergic hypersensitivity to amphetamine, in a deficit in prepulse inhibition (a sensory gating impairment) and in deficits in associative learning, that are associated with schizophrenia (for review, see Paterlini et al. 2005).

However, associations between one gene variation and schizophrenia are probably more complex than a linear relationship between allele dose at a given locus and schizophrenia risk. In fact, schizophrenia is a complex genetic disorder with multiple risk genes of small effect (odds ratios are often below 1.5) which might influence different components of the phenotype, none of these genes are necessary or sufficient. Thus, genetic studies should consider the potential for interactions between multiple loci at the same or at different genes (epistasis). Tunbridge et al. (2006) have recently discussed the role of additional loci in the catechol-O-methyltransferase (COMT) gene that have an impact on the enzyme's function independent of the Val158Met polymorphism. Transcriptional and behavioural interactions between the PRODH and COMT genes, in mice bearing an homozygous truncating mutation of the PRODH gene, have recently been described (Paterlini et al. 2005). In the same way, COMT polymorphisms may also interact with other risk genes such as the PRODH gene for developing psychotic symptoms in the 22q11 deletion syndrome (Raux et al. manuscript in preparation).

Epigenetics refers to DNA and chromatin modifications that play a critical role in regulation of various genomic functions. DNA methylation represents an epigenetic means of inheritance without associated DNA sequence alterations. Though the function of DNA methylation is not completely understood, it may control gene expression, chromosomal integrity and recombinational events. Epigenetics may contribute to phenotypic differences in genetically identical individuals (such as twins) or in inbred animals (Wong et al. 2005) and they may also play a role in the inheritance of complex trait diseases and traits.

DNA micro array analyses of the transcriptome provide an assessment of the expression levels of a list of genes in a given post mortem brain tissue sample. They do not require any a priori hypothesis. Combining micro array gene expression data sets with association studies have been done. RGS4, dysbindin, NRG1 genes have shown altered expression in the post mortem brain, even when the diseased subjects did not carry the susceptibility genotype. In fact, altered expression in the post mortem human brain may have a dual origin: polymorphisms in the candidate genes themselves or upstream genetic–environmental factors that converge to alter their expression level. However, cross-comparing data sets generated by different laboratories is usually challenging because of the use of different experimental designs, the diversity across and within cohorts, the limited sample sizes, the effects of disease treatments before death on gene expression, the co morbidity with other disorders, the risk of type II errors (Mirnics et al. 2006). The specificity of the findings is also questionable. In addition, the question of the primary or secondary origin of these gene expression alterations is difficult to answer. Finally, there is a low concordance between proteomic and transcriptomic data sets. Unfortunately, these important translational and post translational events are important biological regulatory mechanisms that are not assessed by any of the methods used to investigate the transcriptome.

Non-protein coding RNA (ncRNA) plays a critical role in regulating the timing and rate of protein translation. A specific class of small non-coding regulatory RNAs called microRNAs, which regulate gene expression post-transcriptionally by suppressing translation or destabilizing mRNAs, may play a role in brain development (Perkins et al. 2005).

The potential importance of these ncRNAs is suggested by the fact that the complexity of an organism is poorly correlated with its number of protein coding genes, and that in the human genome only a small percentage (2–3%) of genetic transcripts are translated into proteins.

Finally, gene–environment interactions add another level of complexity to psychiatric genetics. The impact of isolated genetic variations is likely to be dependent on the context in which it is expressed. This concept is supported by emerging data showing interactions between some candidate genes and environmental risk factors. Genes may moderate the effect of environmental pathogens on a given disorder (see below) but they may also increase the likelihood that a given subject might be exposed to this pathogenic environment (e.g., the risk of cannabis dependence may be increased by susceptibility genes to addictive disorders). A recent study, from Caspi et al. (2005), reported an interaction between the Val158Met genotype (a functional polymorphism in the COMT gene) and adolescent cannabis use in a longitudinal study of risk for developing psychosis. Cannabis users carrying the COMT Val allele showed subsequent increased risk for exhibiting psychotic symptoms and/or for developing schizophrenia-spectrum disorders. In this field, animal models (e.g., knockout mice and animal exposure to different pathogenic environments) will be very helpful to disentangle gene–environment interactions.

In conclusion we could say with Goethe that everything, including genetics, ‘is both simpler than we can imagine and more complicated than we can conceive’