Copy number variants (CNVs) are a substantial source of human genetic diversity, influencing the variable susceptibility to multifactorial disorders. Schizophrenia is a complex illness thought to be caused by a number of genetic and environmental effects, few of which have been clearly defined. Recent reports have found several low prevalent CNVs associated with the disease. We have used a multiplex ligation-dependent probe amplification-based (MLPA) method to target 140 previously reported and putatively relevant gene-containing CNV regions in 654 schizophrenic patients and 604 controls for association studies. Most genotyped CNVs (95%) showed very low (<1%) population frequency. A few novel rare variants were only present in patients suggesting a possible pathogenic involvement, including 1.39 Mb overlapping duplications at 22q11.23 found in two unrelated patients, and duplications of the somatostatin receptor 5 gene (SSTR5) at 16p13.3 in three unrelated patients. Furthermore, among the few relatively common CNVs observed in patients and controls, the combined analysis of gene copy number genotypes at two glutathione S-transferase (GST) genes, GSTM1 (glutathione S-transferase mu 1) (1p13.3) and GSTT2 (glutathione S-transferase theta 2) (22q11.23), showed a statistically significant association of non-null genotypes at both loci with an additive effect for increased vulnerability to schizophrenia (odds ratio of 1.92; P=0.0008). Our data provide complementary evidences for low prevalent, but highly penetrant chromosomal variants associated with schizophrenia, as well as for common CNVs that may act as susceptibility factors by disturbing glutathione metabolism.
Recent studies have highlighted structural variants as a largely under-explored source of human genetic variation1 that may be a common underlying factor in human disease.2, 3 Such variation includes insertions, deletions, inversions, duplications and translocations of DNA sequences, and encompasses copy number differences also known as copy number variants (CNVs).1, 4 CNVs are likely to contribute significantly to human diversity and disease susceptibility, a postulate further substantiated by the enriched CNV affecting genes involved in drug and hormone metabolism, immune response, as well as in the detoxification of environmental and dietary pro-carcinogens.5, 6 Reported examples include low copy numbers of the CCL3L1, FCGR3B and DEFB4 genes, which are associated with increased susceptibility to AIDS, immunologically mediated glomerulonephritis and Crohn's disease, respectively.7, 8, 9 In addition, evidence is accumulating that multiple rare CNVs contribute to the genetic component of vulnerability to neuropsychiatric conditions, such as autism or schizophrenia.10 Some karyotypic and submicroscopic CNV abnormalities associated with schizophrenia have been described, such as the 3 Mb hemizygous microdeletion at 22q11 causing velo-cardio-facial syndrome (VCFS).11, 12, 13 Other chromosomal aberrations have been detected in affected individuals with schizophrenia and other psychiatric disorders,14 leading to the discovery of potential disease-causing mutations in DISC1, PDE4B15 and NPAS3.16
The increasing use of genome-scan array technologies has revealed several new CNVs at different loci in schizophrenic patients. In an initial report, four aberrant regions containing genes involved in neuronal function were observed in 12 schizophrenic patients,17 although these results were not reproduced later by using an independent method.18 Variation of multiple loci were reported in the study of a small sample of 30 patients without enough control data.19 The lack of reproducibility emphasizes the importance of reliable methods for CNVs validation before full-scale association studies are carried out.18 Despite this controversy, there is a rapidly growing number of well-defined CNVs associated with schizophrenia with low prevalence, but high penetrance. Hemizygous deletions of varying sizes affecting the CNTNAP2 gene at 7q34–36.1 were found in three unrelated individuals with a complex phenotype of schizophrenia, epilepsy and cognitive impairment.20 Overall, thirteen aberrations with two of them likely pathogenic, a 2p16.3 deletion disrupting the NRXN1 gene in affected siblings and a de novo 15q13.1 duplication spanning the APBA2 gene, were found among 93 schizophrenic patients.21 Furthermore, three recent studies have established that rare de novo germline mutations contribute to schizophrenia vulnerability in sporadic cases and that rare genomic lesions at many different loci can account, at least in part, for the genetic heterogeneity of this disease, including recurrent deletions at 1q21.1, 15q13.3 and 15q11.2.22, 23, 24 It is interesting to note that genes involved in pathways important for brain development are over-represented among those disrupted by structural variants in patients, suggesting that rare mutations affecting those genes contribute to schizophrenia.25 Therefore, the role of structural genomic variation in the development of schizophrenia, either directly causing the disease or acting as susceptibility factor, is a topic of considerable interest.
To further understand the contribution of this type of genomic variation to human diversity and complex diseases, we have designed a multiplex ligation-dependent probe amplification (MLPA) based method to target previously reported and putatively relevant gene-containing CNV regions in a high-throughput manner. The approach allowed to carry out large-scale association studies with 140 selected CNVs in 654 schizophrenic patients and 604 controls. Overall, eight variants were detected only in patients, two of them in more than one case (>1 Mb duplications at 22q11.23 and duplications of the somatostatin receptor 5 gene (SSTR5) at 16p13.3), which were further characterized by array technology. Among the few CNVs with relatively high population prevalence, variation at the GSTM1 (glutathione S-transferase mu 1) and GSTT2 (glutathione S-transferase theta 2), encoding two members of a superfamily of proteins that catalyse the conjugation of reduced glutathione (GSH) to a variety of electrophilic and hydrophobic compounds, were found associated to the disease. Genotypes containing GSTM1 and the GSTT2-specific copy were found significantly associated with an increased vulnerability to schizophrenia with additive effect. Our results indicate that rare but highly penetrant CNVs can be responsible for a reduced number of schizophrenia cases, whereas common CNVs, such as those affecting the glutathione S-transferase (GST) genes, may influence a risk of developing the disease.
Materials and methods
Unrelated patients (n=654; 451 male patients and 204 female patients, mean age=46.52±17.08 years) were recruited from in-patient and out-patient services in two different psychiatric hospitals (Consorci Hospitalari Parc Taulí, Sabadell, Spain and Institut Universitari de Psiquiatria Pere Mata, Reus, Spain). All patients had a consensus diagnosis from two senior psychiatrists according to several semi-structured interviews for Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) criteria. Concordant values between psychiatrists varied between 0.6 and 0.95. Unrelated subjects (n=604; 339 male individuals and 265 female individuals, mean age=41.42±12.83 years) from the same geographical region (Catalonia, Spain) and free of signs of neuropsychiatric illness were used as controls. Additional details of the selection and recruitment of patients and controls have already been reported.26 Family relatives of several probands with CNVs were also recruited for this study. Isolation of DNA was carried out from peripheral blood following standard procedures. All samples were obtained under institutional review board-approved informed consent.
Selection of gene containing CNVs
The database of genomic variants (DGVs, http://projects.tcag.ca/variation)27 was used to select gene-containing CNV regions reported in more than one individual or by more than one method, including those with functional candidate genes for schizophrenia. A subset of variants was selected from our data generated by bacterial artificial chromosome (BAC)-array comparative genomic hybridization (aCGH) experiments.28 Given the putative limitation of the MLPA method to quantify variation at complex or multiple copy loci, we only targeted genes located in theoretical single-copy loci in the reference genome, avoiding those overlapping with segmental duplications (SDs). For the sequence-related GST genes, we designed copy-specific probes with mismatches in the ligation nucleotides. Supplementary Table S1 shows a complete list of the selected genes.
Multiplex-ligation PCR amplification
We successfully combined 140 probes in four MLPA panels in a dual-colour strategy previously described29 (Supplementary Figure S1). Oligonucleotides were designed following MRC-Holland (Amsterdam, The Netherlands) recommendations (http://www.mlpa.com/files/protocol_synthetic_probe_design.pdf) and are listed in Supplementary Table S1. Additional probes designed to assess gene copy number changes at the 22q11.22-23 region in further detail are shown in Supplementary Table S2 and located in Figure 1. For targets that presented highly similar sequences with other loci or pseudogenes (such as GSTM1, GSTT1 and GSTT2), we designed a copy-specific MLPA probe by selecting a nucleotide at the probe ligation site that was only present in the desired gene copy. GSTT2 is a single copy gene in the Celera human genome assembly, whereas there is a second copy per chromosome in the reference National Center for Biotechnology Information (NCBI) human genome sequence (GSTT2B). Mismatch nucleotides were detected by sequence alignment between the different homologous sequences. A mismatch at the ligation site is usually enough to avoid ligation and therefore, PCR amplification from other loci.30
The MLPA reactions were carried out essentially as described previously,30 with slight modifications to allow for double colour detection.29 We used the relative peak height (RPH) method recommended by MRC-Holland, as previously described, for data analysis.31, 32 Theoretically, heterozygous deletions and duplications showed a relative peak height of approximately 0.5 and 1.5, respectively. Owing to the high frequency of deletions at the GSTM1, GSTT1 and GSTT2 targeted loci, copy number detection at these positions was normalized independently. Individuals with peaks corresponding to two copies were visually identified and consequently used as reference for the remaining samples.
To validate the assay reproducibility and genotype assignment MLPA reactions from 150 random samples were obtained in duplicate and independently analysed in a blind fashion by two investigators (BR-S and AB). There was full concordance in genotype assignment both between experiments and between investigators.
A total of five microsatellite markers distributed along 22q11.22-23 (D22S0232i, D22S0055i, D22S0163i, D22S0184i, and D22S0153i) were genotyped by PCR and subsequent electrophoretic analysis (GeneScan; Applied Biosystems, Foster, CA, USA) in first-degree relatives of the schizophrenic patients carrying the genomic duplication found by MLPA (see Supplementary Table S3 for primer sequences, PCR conditions and detection methods and Figure 1 for location).
To confirm and better characterize the different variants detected by MLPA we used two commercial arrays, either an oligoarray for CGH (Agilent 44 K or 244 K, Santa Clara, CA, USA) or a single-nucleotide polymorphism (SNP)-array for comparative intensity analysis (Illumina 370Duo, San Diego, CA, USA). We performed the experiments and the subsequent analyses of aCGH as previously described in detail.33 For the SNP-array, 2 μg of genomic DNA were hybridized onto the HumanCNV370-DNA Analysis BeadChip (Illumina, Inc) using the Infinium assay following the manufacturer's recommendations. We then applied the PennCNV tool34 for fine-mapping of CNVs from high-density SNP genotyping data.
Fisher's exact test for count data, Pearson's χ2 test and Bonferroni's adjustment for multiple comparisons were applied when appropriate for measuring significant differences in gene copy number frequencies between patients and controls.
Genomic DNA CNV
Change in copy number by MLPA was only detected by 28 of the 140 (20%) probes targeting candidate genes and loci. Thus, 112 probes (80%) did not detect variation in the studied individuals. Considering both types of variation (gains or losses) in the entire sample, only 7 loci (PPYR1, DDT, CYFIP1, CYP2E1, PCDHA9, GSTM1 and CXCL12) showed copy number changes with frequency above 1%. CNV at 15 loci (11%) was observed in both patients and controls, whereas six rare variants were identified only in subjects from the control group and eight rare variants were found only in the patient group (Table 1 and Supplementary Table S4).
Rare variants found in patients
The eight rare variants found exclusively in patients consisted all in gains involving the CABIN1, SNAP29, NDNL2, WWOX, ZNHIT3, PRKRIP1, MYOM2 and SSTR5 genes. They all were considered as potential candidate rearrangements to play a role in the disease and were subsequently validated and characterized by using array technologies, except in two cases with no additional DNA available. The main findings are summarized in Table 2.
Two of the rearrangements were found in more than one schizophrenic patient. In two unrelated cases (patients 1 and 2), the duplication affected two closely located probes on chromosomal band 22q11.23, targeting the DDT and CABIN1 genes. A specific MLPA assay spanning more genes in the region showed a common duplicated interval (Figure 1, Supplementary Figures S2 and S3) containing 27 genes (Supplementary Table S5). The study of parents and sibling of patient 1 by MLPA and microsatellite markers showed that the duplication was a de novo event that had occurred in the paternally inherited chromosome (Supplementary Figure S2). The inferred haplotypes using the sibling's sample to define the phase showed that the rearrangement was produced by intrachromosomal recombination (Supplementary Figure S4). Additional validation was obtained with the Illumina SNP-array that confirmed duplications of at least 1.87 Mb in patient 1 and 1.33 Mb in patient 2, with identical telomeric breakpoints and different centromeric breakpoints, all likely located in flanking segmental duplication clusters (DCs) (Figure 1). A finer calculation of the size of the duplications was predicted based on the homology between the common distal DC1868 and two proximal DCs (DC1837 for patient1 and DC1846 for patient2) that likely mediated the rearrangement by non-allelic homologous recombination (NAHR) allowing to infer a 1.39-Mb overlapping interval in both patients (Supplementary Table S6). Duplications of this large region have never been reported as polymorphisms in the DGV. It is interesting to note that patient 1 showed an additional gain type CNV at 22q11.21, 1.5-Mb distal to the de novo duplication at 22q11.23, showed by the SNAP29 MLPA probe and validated by SNP-array as a 367.5-kb segment containing eight genes (Table 2 and Supplementary Table S5). This variant was also found in his epileptic mother as well as in the patient's brother who also carries a diagnosis of schizophrenia, but was not part of the initial study because related individuals were excluded.
The second recurrent rearrangement found in three patients was a 5-kb duplication at 16p13.3 containing the entire SSTR5 gene that was validated by SNP-array in patient 8. At this locus, only deletion-type CNVs have been reported in DGV and we also found a control individual with deletion, but none with duplication. Parental samples were not available to determine whether the SSTR5 rearrangements were de novo or inherited in the three cases.
A gain of the NDNL2 MLPA probe at 15q13.1 in patient 3 was not found in the parental samples. The de novo duplication was defined by the SNP-array as 594 kb in size containing the NDNL2 gene and a portion (the first three exons) of the overlapping KIAA0527 gene (Table 2 and Figure 1). The WWOX MLPA probe variation in patient 4 was also validated by SNP-array that showed a 508-kb duplicated segment including four exons in the middle part of the WWOX gene. The observed gain of ZNHIT3 probe in patient 5 was inherited from his unaffected father. The size of the duplication on the 244 K Agilent oligo-aCGH was 59 kb, containing entirely the ZNHIT3 gene and disrupting the MYO19 gene. It is interesting to note that there are some paternal relatives affected by diverse psychiatric disorders (Supplementary Figure S5). As mentioned, the rearrangements in patients 6 and 7 containing the PRKRIP1 and MYOM2 genes, respectively, could not be further characterized due to the lack of enough DNA for array-based experiments.
The analysis of the genomic context surrounding these rare CNVs found in schizophrenic patients showed the presence of SD at one or both breakpoints in six of the ten duplications found, suggesting that NHAR is the most common mutational mechanism.
Table 3 contains the most relevant clinical features of the patients who presented rare genomic aberrations and Supplementary Figure S5 shows the pedigree and additional information from relatives of patients 1 and 5.
Association of common CNVs and schizophrenia
Among the relatively common CNVs analysed in the initial survey, significant differences in frequencies between patients and controls were observed only for GSTM1. The frequency of non-null genotypes (1, 2 or 3 gene copies) was significantly increased in patients with respect to controls (P-value=0.004, OR=1.41, CI=1.12–1.77) (Table 1).
Given the complex structure of the 22q11.22-23 region surrounding CABIN1 and DDT with multiple SD and harbouring evolutionarily duplicated GST genes, we applied the specific MLPA assay that included GSTT1 and GSTT2-specific probes to the entire sample to assess for more subtle rearrangements in the region and/or disease association with common variants. Remarkably, GSTT2 non-null genotypes (GSTT2*1/0 and GSTT2*1/1) were also associated with an increased risk of developing schizophrenia with an odds ratio of 1.405 (CI=1.07–1.84; P=0.01). However, no significant association of GSTT1 copy number with schizophrenia was found in our sample (Table 1). The combined analysis of GST-genes detected an additive effect of GSTM1 and GSTT2 non-null genotypes in the susceptibility to schizophrenia with an odds ratio of 1.92 (CI=1.32–2.78; P=0.0008) (Table 4). After Bonferroni's correction for multiple testing, the disease association of GSTM1 and the additive effect of GSTM1 and GSTT2 remained significant (P<0.006).
Supplementary Figure S6 shows the boxplots of GSTM1 and GSTT2 relative peak height ratios obtained by MLPA versus the copy number genotype assigned together with an example of the corresponding electropherogram peak pattern observed.
We describe here the application of a rapid and economic MLPA-based method to study a set of 140 gene sequences located in previously described CNVs in a large sample of individuals comprising control subjects and schizophrenic patients. An important proportion (80%) of the loci studied showed no variation in any of the individuals (n=1259). This low detection yield of copy number changes by MLPA in loci previously described as polymorphic CNVs may be because of one or several of the following reasons: (i) the different methods for CNV analysis may have different false positive and/or false negative rates; (ii) the designed MLPA probes targeting a small single-copy gene sequences of 50–80 bp might lie outside the actual variable region; (iii) the frequency of those variations may indeed be very low, under 1 in 1000 individuals in specific populations. In any case, our data confirm the previous belief that the size and frequency of the CNVs represented in the public databases are likely overestimated, with entries not fully validated and based on experimental data that do not allow their correct definition. In fact, the concordant rate for CNV calling of two different platforms (BAC vs. SNP arrays) was less than half (43%) when studying the same individuals.35 Given the low population frequency observed for most putatively relevant CNVs, genotyping strategies with PCR-based methods such as MLPA will be needed for large-scale association studies in different populations. Despite its great utility and the ongoing efforts to improve its accuracy, the DGV should not yet be used to define population frequencies of specific CNVs in the substitution of experimental control population, as we have shown in this study.
Several chromosomal aberrations have been identified by cytogenetic and molecular techniques as a possible cause of schizophrenia, including large rearrangements, inversions and balanced translocations.22, 23, 24, 25, 36 The recurrent microdeletion at 22q11.21, causing the DiGeorge and velo-cardio-facial syndromes, is associated with schizophrenia in 30–40% of cases.22, 23, 24, 25, 36 More recently, the emergent use of whole genome-scan platforms has facilitated the detection of smaller genomic structural mutations associated with schizophrenia, in general rare variants at very low frequencies.22, 23, 24, 25, 36 A few of those variants are also recurrent, such as the deletions at 15q11.2 including the CYFIP1 gene (0.5 Mb, 0.55% of patients vs. 0.19% of controls),24 and the deletions at 15q13.3 (2.2 Mb, 0.17% of patients vs. 0.1% of controls) and 1q21.1 (1.3 Mb, 0.23% of patients vs. 0.02% of controls), and can be associated with other phenotypes with incomplete penetrance, including mental retardation, congenital malformations and epilepsy.23, 24 Architectural features of the genomic regions with complex SD that predispose to misalignment and non-allelic homologous recombination may explain the recurrent nature of these rearrangements. We only targeted two of those four regions in our MLPA panel, with the GNB1L (22q11) and CYFIP1 (15q11.2) probes. We did not detect any rearrangement in the 22q11.21 DiGeorge/velo-cardio-facial syndrome critical region in patients or controls, most likely because patients with this diagnostic suspicion had been previously studied on clinical grounds and excluded from this study. We did detect 15q11.2 deletions although the frequency was not significantly different between groups, in contrast with the previous report.24 Duplications at 15q11.2 were also observed with identical frequencies in both controls and patients.
We have also found a few novel and rare CNVs exclusively present in patients, two of them in more than one unrelated individual, which were considered as potentially pathogenic. The duplications at 22q11.23 found in two unrelated patients overlapped a 1.39-Mb interval in a region of significant linkage to schizophrenia and had occurred de novo at least in the case with available parents for study. Of the 27 genes included in the common duplication that, if dosage sensitive, may contribute to the disease phenotype, two of them are good functional candidates: CABIN1 and ADORA2A. Calcineurin binding protein 1 (CABIN1) binds specifically to the activated form of calcineurin to inhibit calcineurin-mediated signal transduction in multiple tissues including brain, and also interacts with amphiphysin (AMPH), a protein associated with the cytoplasmic surface of synaptic vesicles.37 The adenosine A2A receptor (ADORA2A) is one of several receptor subtypes for adenosine and a major target of caffeine, abundant in basal ganglia, vasculature and platelets. A mouse model overexpressing ADORA2A showed an upregulation in the prefrontal cortex, suggesting a contribution to memory dysfunction in depression.38 When co-expressed in the same neuron, adenosine A2A receptor and dopamine D3 receptor form A2A/D3 heteromeric receptor complexes, in which A2A receptors antagonistically modulate both the affinity and the signalling of the D3 receptors.39, 40, 41 As the D3 receptor is one of the therapeutic targets in schizophrenia, the A2A/D3 receptor interactions have been proposed as an alternative antipsychotic target.41, 42
Interestingly, a second 367-kb duplication spanning eight genes on chromosome 22 was also found in patient 1, centromeric to the de novo duplication. This CNV was inherited from his epileptic mother and was also present in his schizophrenic brother. Two genes within the interval are good functional candidates: SNAP29 codes for a protein of the SNARE family that plays a role in synaptic vesicle function, whereas PI4KCA regulates the biosynthesis of phosphatidylinositol 4,5-bisphosphate, also implicated in synaptic vesicle function as well as in signal transduction. Given the replicated association of the 22q11 region with schizophrenia, both SNAP29 and PI4KCA polymorphisms have been analysed in targeted studies showing stronger evidence for PI4KCA as a susceptibility gene for the disorder.43, 44
The other recurrent CNV found was a gain of the SSTR5 MLPA probe that was further characterized in patient 8 as a small duplication containing the entire SSTR5 gene coding for the somatostatin receptor 5. This receptor binds somatostatin to inhibit the release of several hormones and secretory proteins throughout the body, including the nervous system. Remarkably, SSTR5 interacts physically through hetero-oligomerization with other G-protein-coupled receptors, specifically the dopamine D2 receptor (DRD2) located on postsynaptic dopaminergic neurons and involved in reward-mediating mesocorticolimbic pathways.45 Polymorphisms at the SSTR5 gene have been associated with bipolar affective disorder,46 whereas polymorphisms at the DRD2 gene have been found significantly associated with schizophrenia following meta-analysis of several case-control studies.47 In addition, variation in the DRD2 gene could partially explain variation in the timing of clinical response to antipsychotics in the first episode of schizophrenia.48 Therefore, it is tempting to speculate that the gain of SSTR5 copy number may increase the susceptibility to schizophrenia by altering the stoichiometric composition of the different subunits and therefore, the function of these hetero-oligomeric G-protein-coupled receptor complexes.
The de novo ∼600-kb duplication found in patient 3 at 15q13.1 containing the NDNL2 and KIAA0574 genes overlaps with a larger 1.4-Mb duplication spanning the same and two additional genes, APBA2 and TJP1, also reported in a patient with schizophrenia.21 NDNL2 encodes a necdin-like protein of ubiquitous expression, whereas KIAA0574 encodes a putative transmembrane protein of unknown function. This interval is also included in the 3.95-Mb deletion found in a mentally retarded patient.49 Partial rearrangements of the region have also been reported, including duplication of the APBA2 gene in an autistic patient,50 duplication of the TJP1 gene in one control subject and deletion of the TJP1 gene in a schizophrenia case.23 This region is flanked by SD, known as breakpoints BP3A and BP4, whose presence probably explains the regional instability, including inversions in the control population.
The possible link between the remaining unique CNVs and schizophrenia is less clear. The 16q23.1 gain in patient 4 only affects and may disrupt the middle portion of the WW domain containing oxidoreductase (WWOX) gene, coding for a protein that belongs to a family of conserved proteins found in all eukaryotes that play diverse regulatory roles in a wide variety of cellular functions, such as protein degradation, transcription and RNA splicing. The duplication at 17q12 containing the zinc-finger, HIT type 3 (ZNHIT3), and disrupting the myosin XIX (MYO19) genes in patient 5 was inherited from the unaffected father. Apparently, there is a family history of psychosis in two relatives in the paternal family, but we did not have access to samples to establish whether the CNV was also present in the affected relatives. The extent of the rearrangement in patients 6 and 7 could not be determined. However, the gain in 7q22.1 containing the PRKR interacting protein 1 gene (PRKRIP1), a protein involved in cytokine-mediated biological processes,51 could also include other genes such as ORAI2, a calcium release-activated calcium modulator which may influence signal transduction, in case the rearrangement was mediated by the highly homologous flanking SD. The duplicated probe in patient 7 involves the myomesin 2 gene (MYOM2), coding for an integral protein of the major structure of sarcomeres, with no apparent link to the disease.
The CNVs associated with schizophrenia described here and those previously reported might have relatively high penetrance but are individually rare, contributing each to the aetiology of just a few patients or even a single case. In agreement with this postulate, there is a strong familial component of schizophrenia and related disorders in several of the cases with rare CNVs (Table 3). However, in a multifactorial disorder, more common variants of susceptibility with lower relative risk are expected. Among the few relatively common variants genotyped, significant association of CNV with the disorder was detected at two genes coding for GSTs, GSTM1 and GSTT2, but not GSTT1. Individuals carrying at least one copy of the GSTM1 or the GSTT2-specific gene copies (GSTT2B was not genotyped), showed an increased risk of developing schizophrenia (OR ∼1.4 in both cases), with additive effects for having gene copy of both (OR=1.92; P=0.0008). Genetic polymorphisms and copy number at several GST genes have been analysed in case-control studies in different populations with contradictory results. The analysis of the genetic polymorphisms in GSTT2 (Met139Ile variant) did not reveal association,52 whereas association of the null GSTM1 genotype with an increased risk for the disease was reported in Japanese53 and Korean populations,54 which represent the opposite effect to that observed in our study. A second study in Japanese population showed no difference between controls and patients when the GSTM1 deletion was analysed.52 Our results for GSTT1 variants are in agreement with the two studies carried out in Japanese population revealing similar frequencies of GSTT1 null genotype in the patients and control groups,52, 55 but in contrast to others describing association of null genotype with a significantly reduced risk of developing schizophrenia.56 Although the discrepancies observed could be because of population differences, the strength of our results is based on a much larger dataset compared with previous reports.52, 53, 54, 55, 56
The presence of more GST gene copies or specific GST gene variants in patients could account for an increased enzyme activity, leading to an imbalance between levels of reduced and oxidised form of GSH (GSSG). Dysfunction of GSH metabolism has been suggested to be a risk factor for the disease.57 GSH and its oxidized dimer constitute the most important defence against oxidative stress and reactive oxygen species in both peripheral organs and brain.58 GSH levels have been found significantly decreased in the cerebrospinal fluid of drug-free patients with schizophrenia as compared with those in controls, and non-invasive proton magnetic resonance spectroscopy showed a significant reduction of GSH in the medial prefrontal cortex of patients.59 A negative correlation between GSH levels in the posterior medial-frontal cortex and the severity of negative symptoms in schizophrenic patients has also been reported.55 GSTM1 and GSTT2 are members of a superfamily of proteins that catalyse the conjugation of reduced GSH to a variety of electrophilic and hydrophobic compounds. GSH has also been postulated to act as neuromodulatory and neurotransmitter60, 61 with both reduced and oxidized forms preventing the excessive pathological dopamine release in the striatum,58 that would lead to degenerative processes in dopaminergic terminals resulting finally in the loss of connectivity.
Given the evidence for replicated association of the 22q11.23 region with schizophrenia19, 56, 62, 63, 64, 65 an alternative hypothesis to the direct effect of GSTT2 copy number could be that the regional structural variation would influence the risk of schizophrenia by altering regulatory elements for other genes in close proximity, such as the previously mentioned candidates CABIN1 and ADORA2A. In fact, this genomic region shows a high density of SD and CNV that may influence the expression of surrounding genes as demonstrated in other conditions.28, 66 However, the finding of disease association with GST genes located elsewhere in the genome, strongly suggests a direct effect of GSTT2 variation on the risk of schizophrenia.
Owing to the role of GSH in drug resistance metabolism, variation at the GST genes might also influence patients’ response to therapies. In fact, the long-known antioxidant role of GSH is being reconsidered for clinical use in different oxidative stress states, including neuropsychiatric disorders with demonstrated excitotoxic neuronal damage.67 A more detailed profiling of CNVs and functional variants in all genes involved in GSH metabolism might provide additional alleles or allelic combinations associated with disease susceptibility, and determine potential therapeutic targets aimed to preserve GSH balance in schizophrenia.
The rare chromosomal aberrations found in our cohort of patients with schizophrenia support previous findings, suggesting the need for alternative approaches to gene discovery in schizophrenia.25, 68 In addition to the dominant common disease/common allele model, which posits that schizophrenia is caused by combinations of common alleles that each contribute a modest effect, some mutations predisposing to schizophrenia appear to be highly penetrant, individually rare and of recent origin, even specific to single cases or families. The finding of patients with new structural genomic aberrations can define novel candidate loci for the disease.68 Given that the identification of rare variants in few cases may not conclusively establish causal links to illness unless large cohorts are analysed, robust and affordable methods such as MLPA can be used as an alternative to whole genome analyses in the search of candidate genes contained in structural variants, as well as for confirmation. The increasing list of candidate genes affected by dosage and potentially associated to the disease, provides clues for several key neurodevelopmental pathways that can be targeted for further investigations, representing a fundamental progress in psychiatric genetics.
We are indebted to patients and relatives for their support. We thank the Barcelona and Santiago de Compostela CEGEN units, especially Anna Puig, Anna Carreras and Mònica Bayés for their help with DNA-plate preparation and fragment analysis as well as Dr Andrés Medrano for his helpful comments. This work was funded by the Spanish Ministry of Health (PI070539 to LAPJ, RETIC G03/184 to XE, AC, LAPJ and EV, and PI050842 to EV), the Spanish Ministry of Education and Science (SAF2005-01005 to XE), the EU FP6 (037627) and Genoma España to LAPJ and XE. B Rodríguez-Santiago is supported by a postdoctoral fellowship of the Fondo Investigación Sanitaria (FIS CD06/00019).
About this article
Supplementary Information accompanies the paper on the Molecular Psychiatry website (http://www.nature.com/mp)