Copy Number Variations in Schizophrenia: Critical Review and New Perspectives on Concepts of Genetics and Disease

CNVs may be detected by targeted or genome-wide methods. The latter include karyotype with a lower resolution of at best 5–10 megabases (Mb) of DNA (a typical band of a chromosome) and multiple genome-wide DNA microarrays with coverage and resolution that vary according to the individual technology platform, probe set, and methodology. Most microarrays use SNP-based or array comparative genomic hybridization (aCGH) methods. The former require algorithms to assist in determining copy number, and the latter rely on one or more "hybridization reference" genomes, the choice of which can influence results. These methods vary in their resolution and ability to determine the break points and extent of individual CNVs. There is also rapid progress to using whole-genome ("next-generation") sequencing methods (18). Targeted methods include those often used in clinical laboratories that detect microdeletions or microduplications associated with genomic disorders, such as fluorescence in situ hybridization (FISH) using specific probes and molecular cytogenetic methods, and microarrays using various probe-screening sets that offer increasingly extensive coverage. Quantitative polymerase chain reaction (qPCR) is another targeted method often used to confirm a CNV identified by using array-based methods and to quickly screen large sample sets (1).

Much remains to be discovered about the mechanisms giving rise to CNVs and the pathways from CNVs to phenotypic expression recognizable as illness. Repeated elements in the genome have been implicated in many but not all CNV break points (8, 19). Mechanisms include meiotic unequal crossing over, or nonallelic homologous recombination, mediated by flanking repeated sequences or segmental duplications, and nonhomologous DNA repair mechanisms such as nonhomologous end joining or microhomology-mediated end joining (19). If DNA repair mechanisms during mitotic divisions of germ cells are involved, this could lead to new mutations arising in sperm through adulthood, possibly associated with late paternal age, or in maternal grandchildren, given the timing of these divisions in fetal development for females (19).

While CNVs make a significant contribution to observable variability between individuals, there is not always a straightforward correlation between gene dosage and expression (20, 21). Larger CNVs on average will encompass more genes, but they may also contain regulatory elements for genes and they can in addition exert effects on chromatin structure. Therefore, CNVs can have long-distance effects on expression of genes in neighboring chromosomal regions, even up to 1 Mb away from a particular structural variant (20–22).

There are numerous complexities in phenotypic categorization, CNV detection and characterization, and statistical analyses that need to be considered in CNV-disease associations. Schizophrenia has strong evidence for neurodevelopmental origins (23, 24). Thus, schizophrenia may be expected to have CNV associations similar to those in disorders such as autism, where specific CNVs and conceptually similar but microscopically visible chromosomal anomalies detectable by karyotype are present in a substantial minority (approximately 10%) of cases (25–28). Hallmarks of developmental changes, e.g., mental retardation, birth defects, dysmorphic facies, and/or childhood-onset schizophrenia, could indicate individuals with schizophrenia for whom the a priori likelihood of having associated CNVs may be elevated. Sampling strategies can significantly influence the observed prevalence of CNVs in both individuals with the disorder and comparison subjects. In general, no or low prevalence in comparison subjects would be expected for highly penetrant CNVs and higher prevalence in comparison subjects may be expected with less-penetrant CNVs. Likewise, familial forms of schizophrenia may be less commonly associated with certain CNVs than "sporadic" forms of illness, depending on the likelihood of de novo mutations and penetrance of the disease for the specific CNV. Ideally, one would be able to assess all individuals with rare CNVs of interest to determine the full phenotypic spectrum and overt or cryptic relatedness to others who carry the same CNV.

In contrast to SNPs, for which one can expect a 99% sensitivity and specificity of detection of the targets on the microarray, the complex nature of CNVs (Figure 1) and related experimental issues do not allow the same level of confidence for CNV detection (1). Uncertainty about the presence, extent, and break points of CNVs affects analyses of the genes that may be involved. In general, with current commercial microarrays large CNVs (>100 kb) are more reliably detectable, have greater overlap between platforms, and are easier to assign pathological status, particularly if they arise as de novo mutations.

Recurrent de novo CNV events and/or multiple rare CNVs that show convergence on a single gene or set of genes (or networks or pathways of genes) strengthen the association with disease. Although SNP-based arrays have poor coverage of segmental duplications, large CNVs in intervals between such complex repeat regions are readily detectable (29). Sources of possible laboratory- and/or array-based variability include variable probe performance and measurement error, batch effects and "noise" from cell line artifacts, and/or poor-quality DNA samples. Platforms vary with respect to coverage across the genome; more recent arrays with 1 million or more SNPs increase detection down to CNVs spanning less than 20 kb. Smaller CNVs, which constitute the majority of those detected by using SNP-based platforms, are usually inherited (11). It is as yet unclear whether smaller CNVs will have properties similar to those of large CNVs with respect to association with diseases like schizophrenia.

Common CNVs (allele frequency >5%) have properties similar to those of SNPs, comprise the majority of CNV differences between individuals, and are almost always inherited (8, 11). Most are in strong linkage disequilibrium with SNPs and thus may be analyzed with standard genetic association methods. Genuine associations of CNVs with disease may be difficult to determine for rare events with low prior probability, however. Accurate estimates of such associations will depend on an individual with a rare CNV being reported only once or, if included in different studies, clearly demarcated as such. Parental data, i.e., phenotypes and genotypes, are valuable, although not always available given the late age at onset of schizophrenia. Data on family members carrying the CNV provide information relevant to estimates of penetrance and variable expression. Case-control studies rely on the comparison groups being free of disease and unrelated to the affected individuals. The relatively high prevalence of schizophrenia in the general population and reduced penetrance and/or variable expression of genetic variants present further challenges to interpreting CNV results. Given all of these factors, prevalence figures and odds ratios will be very approximate estimates at this early stage of CNV studies. Independent replication is essential.

Up to 18 subjects with schizophrenia who had 1q21.1 deletions, originating from eight sites, were reported in three studies involving over 6,000 patients with schizophrenia (Table 1) (14, 16, 17). There were up to 15 schizophrenia subjects from 10 originating sites reported with 15q13.3 deletions in the studies reported by Stefansson et al. (16) and the International Schizophrenia Consortium (17) (Table 1). Both studies indicated that the proportion of subjects with schizophrenia and these CNVs was significantly different from the proportion of subjects with these CNVs in the comparison groups used. As expected for case-control studies with no capability of following up subjects, limited data are available on the phenotype or de novo versus inherited status of the individuals from either the schizophrenia or comparison groups with these deletions.

Results for the 15q11.2 locus are less clear. In one study (16), the 471-kb CNV microdeletion at 15q11.2 was reported, in fairly large numbers, in both the schizophrenia subjects and comparison subjects with unspecified disorders, including autism (Table 1). While the difference was reportedly significant, the effect size appeared smaller than that for the 1q21.1 and 15q13.3 deletion CNVs (16). A later publication that included data from the International Schizophrenia Consortium study supported the possibility that CNVs at the 15q11.2 locus are risk factors for schizophrenia with reduced penetrance (47). Other recurrent CNVs flanked by segmental duplications previously reported in autism and/or developmental delay (25–27, 48–50) that may also be risk factors for schizophrenia with reduced penetrance include 400–600-kb 16p11.2 deletions and duplications (14, 16) and duplications at 16p13.1 (1.16 Mb) (47, 51), 1q21.1 (1.4 Mb) (16, 51), and 22q11.2 (3–4 Mb) (47); the latter two are reciprocal to higher-penetrance deletions at the same loci.

To address the hypothesis that a general mechanism of increased spontaneous copy number mutations could be involved in causing schizophrenia, one would like to know the answer to the question, Is the overall prevalence of CNVs higher in schizophrenia than in the general population? Some of the studies reviewed suggest that this is the case, but to interpret the findings reported one needs to consider the methods used. As may be expected for such different sampling, molecular, and analytic methods (Table 1), there were no consistent CNV prevalence data reported in all six studies that would allow direct comparisons, for either the schizophrenia or the comparison groups; thus, they are considered individually.

The International Schizophrenia Consortium study considered CNVs that were more than 100 kb long and found at a frequency of less than 1% in the comparison groups used (17). The CNVs discovered were 1.15-fold more common in the total schizophrenia group than in the total comparison group, in which the base rate was 0.99 CNVs per subject. Multiple subanalyses were presented. A 1.67-fold increase in schizophrenia over the total comparison group was the largest reported, obtained by restricting the analysis to loss CNVs more than 500 kb in size that had the lowest base rate (0.03) per subject in the comparison group (17). Not surprisingly, especially given that the large, gene-rich 22q11.2 deletions, as well as 1q21.1 and 15q13.3 deletions, were included (Table 2), the schizophrenia group was reported to have on average 3.57-fold more genes involved in CNVs than the comparison group. Individual genome-wide CNV data were not provided.

The multisite study of Stefansson et al. (16) covered 66 CNVs that had been identified as de novo events in genome-wide surveys of Icelandic samples of parents and offspring from 7,718 parent-offspring pairs or trios. Samples from this genetic isolate had been assembled for various genetic studies of medical and psychiatric disorders, such as autism, but excluding schizophrenia. This strategy thus enhanced for recurrent CNVs with variable expression. Genome-wide CNVs, except for the 66 selected CNVs, were not reported, precluding comparisons with data from other populations.

Walsh et al. (14) considered CNVs greater than 100 kb that had not been described in the literature or in public databases as of November 2007 and that affected genes. They reported a threefold greater proportion with these CNVs in their forensic and adolescent groups with schizophrenia (22 of 150 subjects) than in the composite comparison subjects used (13 of 268 subjects). Inheritance or de novo status of the CNVs was unknown. The schizophrenia group excluded subjects with 22q11.2 deletions but included a subject with a sex chromosome anomaly, 47,XYY. This report also included a study using a different array platform and another group comprising 83 subjects with childhood-onset schizophrenia. This group excluded nine additional subjects with chromosomal anomalies or 22q11.2 deletions (32). For the same categories of CNVs, this group showed a 2.6-fold greater proportion (23 of 83) than did the parental comparison subjects used (10 of 77). It is not clear why the proportion of individuals studied who had CNVs longer than 100 kb was lower in the comparison group for the forensic and adolescent subjects than in both the subjects with childhood-onset schizophrenia and the parental comparison group, but array differences and small study group sizes could be factors. Two (2.4%) of the CNVs in patients with childhood-onset schizophrenia were de novo changes. The expected de novo CNV mutation rate is approximately 1% at this resolution of analysis in comparison populations. Some individual genome-wide CNV data were provided.

Xu et al. (15) considered rare de novo CNVs of all sizes and prevalences and reported an eightfold greater number of these CNVs per subject in the Afrikaner subjects with sporadic schizophrenia (15 of 152), including three with 22q11.2 deletions, than in the comparison group used (two of 159). In contrast, there was a barely significant difference between groups with respect to inherited CNVs (46 of 152 versus 32 of 159). With few exceptions, individual data on genome-wide CNVs were not provided, precluding comparisons with data from other populations.

Using a study group of similar size to the groups in the studies by Walsh et al. (14) and Xu et al. (15), Shi et al. (30) reported no significant difference in the proportion of subjects with rare (<1%) CNVs larger than 100 kb between individuals with schizophrenia (109 of 155) and a comparison group (132 of 187). Results were similar whether or not the CNVs involved genes (30). No individual genome-wide CNV data were provided.

Our study (34) focused on subjects with 22q11.2 deletions and included all other CNVs in the analyses. There were no significant differences between subjects with and without schizophrenia or between those with 22q11.2 deletions and unaffected parents on any CNV-related measure, regardless of size, rarity, or de novo status. The genome-wide CNV results were also similar to comparable CNV data available from large general population samples (34). Moreover, there was no evidence of an increase in de novo CNVs as an underlying mechanism for occurrence of the 22q11.2 deletion. This study provided individual CNV data for all subjects studied.

In addition to the major CNVs just discussed, the studies reviewed provided some initial evidence for recurring smaller CNVs that may implicate individual candidate genes for schizophrenia (Table 2). Some methodological issues, however, obscure the published findings.

Although the data are preliminary and involve few individuals, CNV studies provide some evidence to support the neurexin superfamily, specifically the NRXN1 and CNTNAP2 genes, as potentially important in schizophrenia. These genes cover large genomic extents, over 1 Mb for NRXN1 and over 2 Mb for CNTNAP2. Recurring CNVs may more likely affect such genes than genes of smaller size. Altogether, there appear to be 18 or 19 individual subjects with schizophrenia reported in a total of six articles to have 2p16.3 loss CNVs involving NRXN1 (44), providing some evidence that the NRXN1 gene may be involved in schizophrenia. As is the case with the larger CNVs associated with schizophrenia, the phenotype of CNVs involving these genes embraces other conditions, including autism and epilepsy (25–27, 46). In the case of CNTNAP2, Tourette's syndrome, obsessive-compulsive disorder, and ADHD may also be part of the expression profile (55, 56). The relevance of these findings awaits further data.

Relatively few investigators have yet reported their complete CNV data sets or released their raw data, minimizing the opportunity for comparing and contrasting data or for meta-analyses. As a result, some commonalities (or differences) in the data may be missed. We anticipate that the results presented are just the beginning of a rich literature that will broaden our understanding of structural variants as risk factors for schizophrenia and their possible role in genetic interactions.

We propose some recommendations (Table 3) that will help in the evaluation of CNV results and appreciation of their implications. Methodological and reporting issues significantly limit the interpretations possible from the genome-wide CNV studies reported to date. This review highlights ascertainment and phenotypic details and subject overlap that hampered adjudication of results. More rigorous phenotyping is key to genetic discoveries and to understanding implications of potential discoveries, including those related to CNVs (1, 2, 57). As for most mutations, expression associated with CNVs is variable and severe phenotypes are usually the first to be discovered (Figure 2). Early penetrance estimates for specific CNVs may therefore be higher than those available after further study, and more low-penetrance CNVs will be reported over time (47, 51). Apart from 22q11.2 deletions (34), little has been reported about the chromosomal parent-of-origin status of de novo deletions. This may be particularly important for imprinted regions, such as 15q11–q13, where a specific phenotype may be observed only in a maternally (or paternally) transmitted chromosome. There is a paucity of data on parents and other family members of individuals reported to have these CNVs, leaving the inheritance or de novo status and full range of expression largely unknown. Differing methods were used to determine rates of novel CNVs, de novo CNVs, rare CNVs, and genes, including CNS genes, that were disrupted, involved, and/or affected (Table 1). Differences in array platforms and algorithms used between study groups and between studies further limit interpretation of results. With few exceptions (14, 34), there were no or only limited validation studies of CNVs (Table 3) reported as recurring and/or of interest. The studies to date mostly involve Caucasians, and the effects of ethnicity on the prevalence of CNVs are largely unknown (4; also see the Database of Genomic Variants: http://projects.tcag.ca/variation/). Despite the limitations, however, the replicated findings provide new perspectives for the genetics of schizophrenia and genetic concepts of neuropsychiatric disease, and they have important clinical implications.

What is the general relevance of these CNV findings for clinicians? The 22q11.2, 1q21.1, and 15q13.3 deletions are the only major genetic risk factors identified to date for schizophrenia—and for which clinical genetic assessment and testing are widely available. The relatively modest prevalence of these anomalies suggests to us that, at this time, expensive genomic investigations are not warranted for all individuals with schizophrenia. Clinicians should, however, have a raised index of suspicion for genomic disorders or de novo CNVs in individuals with schizophrenia who have significant learning difficulties, dysmorphic facial features, birth defects, and/or unprovoked seizures (58). Family history, e.g., in offspring, of birth defects, dysmorphic features, developmental delay, and/or autism should also prompt consideration of a genomic disorder. If a syndromic form of schizophrenia is suspected, a referral to a genetics specialist would usually be recommended for diagnostic assessment (58).

What is the clinical relevance of these CNV findings for patients determined to have one of these major CNVs? For 22q11.2 deletions, the relevance is clear: detection significantly changes genetic counseling and anticipatory care from that for other patients with schizophrenia (58). For the 1q21.1 and 15q13.3 deletions, much less is known and there are challenges to providing genetic counseling for individuals with emerging genomic disorders (1, 57). However, detection of these anomalies would be clinically relevant to individuals with these CNVs, their families, and clinicians.

For the patient, genetic counseling would include the 50% risk of transmitting the deletion at each pregnancy, with the caveat that the severity of the phenotype cannot be predicted (58). As for 22q11.2 deletions, prenatal detection would be available (58). Parents of affected individuals should be tested because the deletion may be inherited from a parent. Specific anticipatory care recommendations for 1q21.1 and 15q13.3 deletions await further clinical data and, preferably, larger studies of population comparison subjects. However, history and physical examination, including neurological examination, echocardiogram, and abdominal ultrasound to investigate organ involvement, and routine blood work would also appear warranted (58). For lower-penetrance CNVs, many of which may be inherited from apparently unaffected parents, genetic counseling and clinical recommendations would be even less certain at this time (1, 57). In all cases, a genetics specialist would have the most up-to-date knowledge about these issues.

In addition to being directly relevant to clinical practice, these CNV findings challenge some widely held beliefs and point to new research strategies. The data explode a common myth related to the diagnostic specificity of genetic findings. While there has been growing acceptance of the genetic relatedness of schizophrenia and bipolar disorder, there may be some resistance to accepting that expression of an individual CNV may also include autism and other developmental disorders (Figure 2, Table 2). The CNV results suggest a broader neuropsychiatric spectrum of phenotypes. In fact, there is little evidence in medicine for diagnostic purity associated with individual genetic changes. Genetic heterogeneity (many different genetic variants leading to the same phenotype), reduced penetrance (presence of the genetic variant not always expressed as the full disease), and variable expressivity (the same genetic variant leading to various phenotypic expressions) are the norm in human diseases and are apparent with the CNVs associated with schizophrenia. Studies of familial connections between schizophrenia and mental retardation (59, 60) and studies of childhood-onset schizophrenia, a rare clinical subtype in which comorbid autism spectrum disorders are common, also support a neuropsychiatric spectrum involving these developmental conditions (24). On the other hand, the data suggest that although there is some overlap, the pattern of associated CNVs may differ between schizophrenia and autism (Figure 2). For example, while autistic features are commonly reported in children with 22q11.2 deletions who receive psychiatric assessments, 22q11.2 deletions are rarely reported in subjects with autism who are studied (25–27). Also, current data suggest that 1q21.1 duplications are more often present than 1q21.1 deletions in autism (26, 61), whereas the reverse may be the case for schizophrenia (16, 51).

With respect to psychiatric diagnosis, these CNV findings suggest that, as for the dementias, the schizophrenias are beginning to yield to classification based on major causal factors. Comparable to an Alzheimer-type dementia associated with a beta-amyloid precursor protein (APP) gene mutation, a schizophrenia that is related to a 22q11.2 deletion (and perhaps a 1q21.1 or 15q13.3 deletion) may be considered a subtype with distinct management implications.

The findings may also prompt consideration of historical concepts of schizophrenia as an "epiphenomenon" of mental retardation. In fact, schizophrenia is not associated with most forms of mental retardation. The majority of individuals with 22q11.2 deletions, including those with schizophrenia, do not have mental retardation (34, 62), and this will probably also be the case for 1q21.1 and 15q13.3 deletions and lower-penetrance CNVs. Nonetheless, dual-diagnosis populations with schizophrenia and mental retardation are likely to be enriched for these recurrent CNVs, providing a much greater window on the etiology of pfropfschizophrenie than have chromosomal abnormalities detectable by karyotype.

Recurrent rare CNVs such as the 22q11.2, 1q21.1, and 15q13.3 deletions may also provide some concrete proof of a long-suspected mutational mechanism in schizophrenia. Over 50 years ago, geneticists Lionel Penrose (63) and Jan Böök (64) proposed that elevated rates of new mutations were likely in schizophrenia. Decreased reproductive fitness in schizophrenia (negative selection) in the face of a steady population prevalence of schizophrenia supports this possibility (65). The mechanism of the recurrence of the associated major CNVs, involving segmental duplications and nonallelic homologous recombination (3, 19), can account for the persistence of these mutations despite the biological disadvantage (reduced fitness) associated with their full expression. It is tantalizing to speculate about accumulating rare CNVs in schizophrenia and CNVs disrupting multiple genes in relevant pathways (14), but these possibilities remain to be proven. Certainly, the relationship of CNVs to reproductive fitness, natural selection, disease prevalence, and human evolution is of intense interest in genetics (9).

Multiple independent reports of CNVs overlapping specific genes in patients with schizophrenia may show convergence on individual genes and/or pathways of interest that could assist in understanding pathogenesis. The mechanism of action of CNVs on the expressed phenotype is hypothesized to include effects of the copy number change, e.g., dosage effects, on genes within the CNV and perhaps extending as far as several megabases adjacent ("position" effect). There may be disruption of genes at the CNV break points. Sequence changes, e.g., on the intact chromosome in the case of deletions, may also be a factor. It is important to recognize that copy number gains such as duplications can affect gene expression, and in many cases these effects are similar to the effects of deletions involving the same chromosomal region.

In contrast to the situations for autism and other neuropsychiatric disorders, identifying rare mutations in single genes implicated by multigene CNVs remains a hypothetical possibility for schizophrenia. Could family studies be redesigned to help localize such rare mutations in CNV-related single genes? The phenotypic spectrum associated with the CNVs may provide a clue. One could predict that inherited (familial) forms of "pure" schizophrenia would differ in associated regions from those identified by de novo CNV mutations. The findings of a meta-analysis of linkage studies of familial schizophrenia are consistent with this prediction (66). Rare families with both schizophrenia and autism or mental retardation segregating (59, 60) may represent a better strategy for subtypes of schizophrenia that are associated with CNVs and/or with mutations in key genes affected by such CNVs.