Several large CNVs have previously received strong replicated support for association with schizophrenia: deletions at 1q21.1 (International Schizophrenia Consortium, 2008; Stefansson et al., 2008
), 2p16.3, affecting the NRXN1
gene (Kirov et al., 2008, 2009b; Rujescu et al., 2009
), 3q29 (Mulle et al., 2010; Levinson et al., 2011
), 15q11.2 (Stefansson et al., 2008; Kirov et al., 2009a
), 15q13.3 (International Schizophrenia Consortium, 2008; Stefansson et al., 2008
), 17p12 (Kirov et al., 2009a
), 17q12 (Moreno-De-Luca et al., 2010
) , 22q11.2 (the Velo-Cardio-Facial Syndrome (VCFS) region) (Bassett and Chow, 2008; International Schizophrenia Consortium, 2008; Stefansson et al., 2008; Levinson et al., 2011
) and duplications at 16p11.2 (McCarthy et al., 2009
) and 16p13.1 (Ingason et al., 2011
). These studies used a total of between ~ 5000–11,000 cases and ~ 37,000–50,000 controls, and produced strong statistical support for association with disease, with p-values of between 4.3 × 10− 4
and 3.8 × 10− 25
. The CNVs have very low frequencies in controls, and further studies are required both to confirm several of these associations and to establish more accurately their frequencies in patients and in the population of people without neuropsychiatric illness, thus enabling more accurate estimates of effect sizes of risk. Although the numbers of controls were much larger than the ones used in the current study, almost all previous findings were heavily supported by data on > 30,000 controls from a single country (Iceland), therefore a comparison with another large sample of controls from another country appeared to be justified.
Here we studied a sample of 10,259 individuals from the UK, affected with several common non-neuropsychiatric disorders. This dataset allowed us to assess the frequency of these CNVs in an independent sample of comparison subjects. These individuals have not been screened for the presence of schizophrenia or other CNV-related neurodevelopmental disorders. Nevertheless, the prevalence of schizophrenia is only about 1% in the general population, and therefore the use of unscreened reference samples is not expected to make an appreciable difference to the results. Even if 100 of the ~ 10,000 controls have schizophrenia (~ 1%), these will not contribute more than two or three CNVs in total from those listed in , as their combined frequency in cases is ~ 2%. Furthermore, the presence of people with psychiatric disorders in this reference set would only reduce the strength of association between these CNVs and schizophrenia, therefore the current study provides a conservative estimate of this association. Another ~ 3000 control subjects that are part of the WTCCC study have been used in the identification of several of these loci, and are therefore not included in the current analysis (Kirov et al., 2009a; McCarthy et al., 2009
). The combined burden of the examined loci in in these ~ 3000 controls does not significantly differ from that in the 10,259 non-psychiatric reference group (Yates corrected χ2
p-value = 0.92).
Overall six of the loci remain significantly more common in the subjects affected with schizophrenia, compared to the WTCCC reference set (1q21.1, 3q29, 15q11.2, 15q13.1, 16p11.2 and 22q11.2). This is despite the fact that the new comparison sample is ~ 4 times smaller than the previously used combined control samples. This provides strong additional support that these CNVs are true risk factors for schizophrenia. For the remaining three loci (16p13.1, 17p12 and 17q12) the differences were not statistically significant (, column 7). When we compared the rate of the studied CNVs in the non-psychiatric WTCCC reference set with the rate in the previous controls (, column 4) we observed statistically significant higher frequencies in the WTCCC reference set for the following loci: 15q11.2, 16p13.1, and 17p12. This is most notable for the 16p13.1 duplications (Results
), with a frequency of 0.24% among the WTCCC reference set, almost an identical frequency to that reported for schizophrenia in previous studies (0.28%). This suggests the possibility the original finding might have been a false-positive.
The difference between the control/ reference sets, and the observations that the rates of CNVs at 16p13.1, 17p12 and 17q12 were not statistically different between schizophrenia cases and the non-psychiatric WTCCC reference set suggest that either these CNVs also increase the risk to develop some of the disorders that comprise the WTCCC sample, or that the odds ratios in the original studies were inflated, or even false-positives, or that the use of the smaller set of controls here is underpowered. In we present the numbers of the pathogenic CNVs observed in each disease cohort that comprise the WTCCC reference set in order to examine if these CNVs also increase the risk to develop some of the non-psychiatric disorders. Type 2 diabetes cases had the highest rate of schizophrenia-related CNVs, a cumulative rate of 1.2%. This is intriguing given strong epidemiological evidence for an increased rate of several chronic somatic disorders among patients with schizophrenia, including Diabetes 1 and 2 types, cerebrovascular disorders and myocardial infarction. The cumulative rate of these CNVs in the rest of the cohorts ranged from 0.58 to 0.8. It is of note that the lowest cumulative CNV burden is found for rheumatoid arthritis (0.58%), as there is evidence of inverse co-morbidity between schizophrenia and connective tissue disorders, most notably rheumatoid arthritis (Mors et al., 1999; Laursen et al., 2011
). This raises the intriguing question as to whether CNVs are partially responsible for this inverse comorbidity; answering this question will require larger studies of these CNVs in rheumatoid arthritis.
The decrease of the estimated odds ratios for several loci in the new analysis could be also due to the bias known as the “winner's curse” (Zollner and Pritchard, 2007
). This arises because if it is underpowered, the first study reporting an association tends to be performed on a sample where the odds ratio happens to be inflated, as explained elsewhere (Craddock et al., 2008
). Nevertheless, the comparisons of the frequencies of these CNVs in cases and in the overall ‘control’ populations (those previously used + the new WTCCC reference set) remain highly significant for each locus, with only minor fluctuation of the significant levels of up to one order of magnitude ().
Rate of CNVs in previous studies on schizophrenia and in all control samples (combined previous controls + WTCCC reference set).
We will now discuss how the current results relate to previous findings for each locus.
4.1. 1q21.1 deletion
This 1.4 Mb deletion contains 11 genes and was found to be associated with schizophrenia in the three largest CNVs surveys (International Schizophrenia Consortium, 2008; Stefansson et al., 2008; Levinson et al., 2011
). It has also been implicated in developmental delay, dysmorphic features, cardiac abnormalities, and learning disability (Christiansen et al., 2004; Brunetti-Pierri et al., 2008; Mefford et al., 2008; O'Donovan et al., 2008b
). Our results confirm their extreme rarity in people who are not affected with neuropsychiatric disorders and give further support for the association with schizophrenia (p = 3.2 × 10− 5
4.2. 3q29 deletion
Deletions of 3q29, spanning 1.4 Mb and 21 genes, were first reported to be associated with schizophrenia by Mulle et al. (2010)
. Like many of the other schizophrenia associated CNVs, this variant has been also found to be associated with intellectual disability and autism (Ballif et al., 2008; Mulle et al., 2010
). No deletions were observed in the WTCCC reference set, providing further support that this is a risk factor for susceptibility to schizophrenia (p = 0.001).
4.3. 15q11.2 deletion
The first report of an association between the 0.6 Mb deletions at this locus that span four genes, and schizophrenia was made by Stefansson et al. (2008)
. The data in for this locus were combined from the Kirov et al. review and the study by Levinson et al. (Kirov et al., 2009a; Levinson et al., 2011
). The rate of this deletion in the new WTCCC subjects is 0.39%, a higher rate than in the previous controls (0.24%), but lower than among cases (0.57%), thus providing independent support for the association (p = 0.05). Other studies give further support for the role of this CNV in neuropsychiatric disorders (Doornbos et al., 2009; Mefford et al., 2009
), with the highest rate of this deletion (1%) found in patients with idiopathic generalised epilepsy (de Kovel et al., 2010
). The prevalence and phenotypic manifestations of this CNV need further investigation.
4.4. 15q13.3 deletion
The association with schizophrenia of this 1.5 Mb deletion that spans seven genes, is supported by the three largest studies in schizophrenia (The Molecular Genetics of Schizophrenia (Levinson et al., 2011
), the International Schizophrenia Consortium (International Schizophrenia Consortium, 2008
) and Stefansson et al. (2008)
) as reviewed by Levinson et al. (2011)
. The rate of the 15q13.3 deletion in the new WTCCC reference set is very low at 1:2500, confirming the role of this CNV in schizophrenia (p = 0.001). Similar to other loci implicated in schizophrenia, this deletion is associated with other neuropsychiatric phenotypes. Sharp and colleagues described a syndrome presenting with mental retardation, seizures and variable facial dysmorphic features (Sharp et al., 2008
). The deletion is found at ~ 1% among patients with idiopathic generalised epilepsy (Dibbens et al., 2009; Helbig et al., 2009
) and at 0.17% among individuals suffering with mental retardation, developmental delay and autism (Miller et al., 2009
4.5. 16p11.2 duplication
A large meta-analysis observed this 0.6 Mb duplication containing 26 genes at a rate of ~ 1:300 schizophrenia patients and ~ 1:3500 controls (McCarthy et al., 2009
). The rate in the new WTCCC reference set (~ 1:2500) is very similar to that among previous controls, and provides strong support for the association with schizophrenia (p = 3.9 × 10− 6
). The same duplication is also a risk factor for autism, while the reciprocal deletion of this locus is associated with autism and developmental delay (McCarthy et al., 2009
). The reciprocal deletion is recognised as one of the most common molecular risk factor for autism, but it does not appear to be a risk factor for schizophrenia (McCarthy et al., 2009
). In the current study, four WTCCC reference subjects carried the deletion.
4.6. 16p13.1 duplication
This 1.2 Mb duplication spanning eight genes was first implicated as a risk factor for autism (Ullmann et al., 2007
) and mental retardation (Mefford et al., 2009
). Ingason et al. observed a 3-fold excess of duplications and deletions in cases with schizophrenia compared to controls (Ingason et al., 2011
). combines their results with those from the International Schizophrenia Consortium (2008)
, resulting in a 3-fold increase in cases: ~ 1:350 patients carry the duplication, vs. ~ 1:1000 controls. In the current study we found a high rate of this duplication in control subjects (~ 1:400), which is quite close to that observed in schizophrenia cases (~ 1:350). As noted above, this at least raises the question whether the first observation was a false-positive result. The carriers in the present study did not come predominantly from a single phenotype that can explain the increased rate (). It is therefore not clear why there is such a strong difference between the rate in the WTCCC reference set and that in previous controls (0.24% vs. 0.1%, Fisher's exact test p = 0.001). One possibility is that the lower rate in the previous findings was due to exclusion of subjects that had known neuropsychiatric disorders in the study by Ingason et al. (2011)
, as further studies also support the pathogenic role of this CNV in such disorders. Thus, Mefford et al. detected duplications in 1.1% of individuals with unexplained intellectual disability (Mefford et al., 2009
), de Kovel et al. found deletions at this locus in 6/1234 epilepsy patients (0.5%) and in 2/3022 controls (0.07%), OR = 7.4 (de Kovel et al., 2010
), and Williams et al. found an excess of duplications (0.84%) in attention-deficit hyperactivity disorder cases (Williams et al., 2010
). Perhaps even more intriguing is the strong association of this CNV with aortic dissections (Kuang et al., 2011
). The duplication is under strong selection pressure, indicating that it is pathogenic (Rees et al., 2011
). Further studies on cases, controls and other phenotypes are needed to resolve the role of this CNV that appears to be pathogenic in general, but not clearly associated with schizophrenia.
4.7. 17p12 deletion
This 1.4 Mb deletion causes the neurological disorder hereditary neuropathy with pressure palsies (HNPP, MIM #162500) (Chance et al., 1993
), while the reciprocal duplication causes Charcot–Marie–Tooth disease Type 1A (CMT1A, MIM #118220) (Lupski et al., 1991
). The deletion was found to be 10 times more common in schizophrenia cases than in controls (0.16% vs. 0.015%, p = 5 × 10− 5
) (Kirov et al., 2009a
). These results are based on a smaller sample size than any of the other loci in . The current study finds a relatively high rate of this deletion in the WTCCC reference set: 7 out of 10,259 individuals, a 4-fold higher rate than in previous controls (p = 0.02). This is still much lower than the rate in schizophrenia patients, however the new findings are not statistically significant (p = 0.11) and weaken the evidence from the previous association. The association still remains highly significant in the overall analysis ().
4.8. 17q12 deletion
Deletions of 1.5 Mb at this region have been associated with schizophrenia (Moreno-De-Luca et al., 2010
), as well as with autism/neurocognitive impairment (Loirat et al., 2010; Moreno-De-Luca et al., 2010
). One of the 15 genes within the locus, HNF1B
, has been implicated in the renal cysts and diabetes (RCAD) syndrome (Bingham and Hattersley, 2004
). This CNV can present also with various medical conditions (macrocephaly, characteristic facial features, genitourinary tract anomalies, recurrent infections and diabetes) (Moreno-De-Luca et al., 2010; Nagamani et al., 2010
). Interestingly one of the two WTCCC subjects harbouring this deletion is from the type 2 diabetes cohort. The current results weaken the evidence for this CNV as a susceptibility factor for schizophrenia, although it remains significant in the combined analysis () and should be tested in further studies.
4.9. 22q11.2 deletion
Deletions of 2.4 Mb at this locus constitute a known genomic disorder, 22q11.2 Deletion Syndrome, also known as Velo-Cardio-Facial Syndrome (VCFS) or DiGeorge Syndrome (Shprintzen et al., 1978
). This was the first CNV shown to increase the risk for developing psychosis (Murphy et al., 1999
). Levinson et al. reported a frequency of ~ 0.32% among 11,400 schizophrenia cases, while no control with the full deletion was found in any control samples () (Levinson et al., 2011
). No deletions were observed in the WTCCC reference set either, confirming once again the status of this locus as a highly penetrant risk factor for schizophrenia (lower confidence interval for the OR = 44.2, p = 1.1 × 10− 27
) and other neurodevelopmental disorders.
The current study provides further support for the involvement of deletions at 1q21.1, 3q29, 15q11.2, 15q13.3, 22q11.2 and duplications at 16p11.2 as risk factors for schizophrenia. The evidence implicating deletions at 17p12 and 17q12 and especially duplications at 16p13.1 is weaker and needs further study. All of these loci have also been shown to increase risk for other neuropsychiatric disorders, indicating that these rare, large DNA variants are not phenotype-specific. Furthermore, most carriers of these CNVs do not develop schizophrenia (i.e. they have a limited penetrance for this disorder of 2–7.4%), with the exception of the deletion at 22q11.2 that has an estimated penetrance of 55% (Vassos et al., 2010
An important limitation of our study is that most of these CNVs cause other medical disorders and therefore could be found among cohorts recruited for other conditions. This could explain why the rate in the WTCCC comparison subjects was higher than in previously analysed controls for several of these loci. While the overall burden of these CNVs appears to be the same in the WTCCC non-psychiatric reference set and ~ 3000 controls (see above), this study has no statistical power to estimate if any individual non-psychiatric disorder in the WTCCC has an increased burden of any single CNV locus discussed here. Judging from the experience in schizophrenia, such associations have to be tested on > 5000 cases from each phenotype.