CNVs in neuropsychiatric disorders

Abstract

Over the last few years at least 11 copy number variations (CNVs) have been shown convincingly to increase risk to developing schizophrenia: deletions at 1q21.1, NRXN1, 3q29, 15q11.2, 15q13.3 and 22q11.2, and duplications at 1q21.1, 7q11.23, 15q11.2-q13.1, 16p13.1 and proximal 16p11.2. They are very rare, found cumulatively in 2.4% of patients with schizophrenia and in only 0.5% of controls. They all increase risk for other neurodevelopmental disorders, such as developmental delay and autism spectrum disorders, where they are found at higher rates (3.3%). Their involvement in bipolar affective disorder is much less prominent. All of them affect multiple genes (apart from NRXN1) and cause substantial increases in risk to develop schizophrenia (odds ratios of 2 to over 50). Their penetrance for any neurodevelopmental disorder is high, from ∼10% to nearly 100%. Carriers of these CNVs display cognitive deficits, even when free of neuropsychiatric disorders.

Introduction

Until 2008, the only confirmed genetic risk factor in schizophrenia (SZ) was an ∼2.3 Mb deletion on 22q11.2 (1) that had previously been shown to cause a severe syndromal disorder with various presentations, resulting in a number of disease entities associated with it: DiGeorge syndrome, Shprintzen syndrome and velocardiofacial syndrome (2,3). This deletion can now be called a copy number variation (CNV). CNVs are chromosomal deletions and duplications that range in size from kilobases to megabases of DNA sequence, and are not detectable by conventional karyotyping. Many more large and rare CNVs have been found to cause diseases, which have been called genomic disorders, a term introduced by Lupski (4). The term indicates that such conditions result by rearrangements of the genome, rather than by changes at the DNA basepair level. This type of CNV is typically caused by the presence of region-specific, highly repetitive DNA sequences, called low copy repeats (LCRs) (5), which are DNA segments previously duplicated during evolution. Recombination between adjacent and homologous LCRs can occur and leads to deletions or duplications of the DNA stretches between the repeats. This mechanism is termed non-allelic homologous recombination (NAHR) and is responsible for most of the CNVs discussed in this paper. Such CNVs are also called recurrent, as they tend to form at the same chromosomal positions, flanked by the LCRs. Other mutational mechanisms can lead to non-recurrent CNVs (that have different breakpoints) (6). The number of CNV regions known to cause genomic disorders might include between 70 and 120 regions (7,8). They can explain the genetic causes in some 5–15% of cases suffering with intellectual deficit, developmental delay (DD), autism spectrum disorders (ASD), congenital malformations (CM) and various medical conditions (8). This paper focusses only on the subset of CNVs that cause SZ and on their overlap with other psychiatric and neurodevelopmental disorders.

CNVs in schizophrenia

With the advent of microarrays and the availability of very large cohorts of patients and controls, it became possible to interrogate the whole genomes of thousands of SZ cases and controls for the presence of CNVs that increase risk for SZ. This lead to an explosion of findings since the year 2008. The first papers confirmed the role of the 22q11.2 deletion and also implicated deletions at 1q21.2, 15q11.2 and 15q13.3 (9,10). Further reports followed and unlike most previous studies in psychiatric genetics, these were supported by almost universal replication (11–15). This success in replication might have been due to the high mutation rates of these CNVs and the negative selection operating against them, which result in similar rates of these CNVs in different populations, which are not influenced much by genetic drift. Their high odds ratios for disease also provided the statistical power to detect the associations. Our latest review of the topic (16) concluded that 11 CNVs can be confidently regarded as risk factors for SZ, as the statistical significance of the findings survives multiple correction for genome-wide testing of 120 CNV loci, counting deletions and duplications at the same locus as separate CNVs (8). These are deletions at 1q21.1, NRXN1, 3q29, 15q11.2, 15q13.3 and 22q11.2, and duplications at 1q21.1, 7q11.23, 15q11.2-q13.1, 16p13.1 and proximal 16p11.2 (Table 1 and Fig. 1).

Table 1.

Frequencies of selected CNVs in control groups, subjects with DD/ASD/CM, SZ and BD

Locus	Size (kb)	N genes	CNV frequency in %				ORs for SZ (95% CI)
			Controls (47,686–81,821)	DD/ASD/CM (6,623–33,226)	SZ (12,029–21,450)	BD (4,288–9,129)
1q21.1 del	820	11	0.021	0.29	0.17	0.033	8.3 (4.6–15)
1q21.1 dup	820	11	0.037	0.2	0.13	0.099	3.45 (1.9–6.2)
NRXN1 exonic del	Variable	1	0.020	0.18	0.18	0	9.0 (4.4–18.3)
3q29 del	1610	21	0.0014	0.061	0.082	0.025	57.6 (7.6–438.4)
WBS dup	1400	28	0.0058	0.12	0.066	0	11.3 (2.6–49.9)
15q11.2 del	290	4	0.28	0.81	0.59	0.17	2.2 (1.7–2.7)
PWS/AS dup	3610	13	0.0063	0.25	0.083	0	13.2 (3.7–46.8)
15q13.3 del	1350	7	0.019	0.26	0.14	0.043	7.5 (4.0–14.2)
16p13.11 dup	790	8	0.13	0.3	0.31	0.011	2.3 (1.6–3.4)
16p11.2 dup	560	26	0.030	0.28	0.35	0.13	11.5 (6.9–19.3)
22q11.2 del	1240	40	0.00	0.54	0.29	0.012	NA (28.2–∞)

The frequencies are given as % of carriers. The numbers in brackets under each group refer to the numbers of individuals used for these estimates. They differ according to the locus; details for each locus are available from the source reviews used for the table: Rees et al. (16) for SZ and controls, Kirov et al. (17) for DD/ASD/CM and Green et al. (18) for BD.

Figure 1.

Frequencies of selected CNVs in control groups, subjects with DD/ASD/CM, SZ and BD. The y-axis shows the frequency of the CNV in %. Black: DD/ASD/CM, red: SZ, blue: BD, yellow: controls.

The table demonstrates the extreme rarity of these CNVs, even in patient populations, with each CNV being found in much less than 1%. This required very large sample sizes for achieving the necessary statistical power. Between ∼10 000 and 20 000 SZ patients were included in these studies (Table 1), necessitating the formation of large consortia and collaborations. These CNVs have strong effects on increasing the risk for SZ, with odds ratios (ORs) between 2 and over 50. As expected, CNVs with higher risk to develop SZ (higher ORs), are rarer (Fig. 2), as highly pathogenic mutations are eliminated from the population faster due to lower fecundity among their carriers, resulting in lower allele frequencies in the population. (SZ patients have reduced fecundity as discussed below.)

Figure 2.

Correlation between population rates and ORs for SZ. The axes are on a logarithmic scale.

CNVs in other neuropsychiatric disorders

While only 11 CNVs can be confidently regarded as risk factors for SZ, many more are involved in the pathogenesis of DD/ASD and CM (e.g. 7,8,19). For example, the CNVs causing Prader-Willi, Angelman, Williams-Beuren and Smith-Magenis syndromes were not found in SZ cohorts. As this review focuses on psychiatric disorders, only the role of the 11 SZ-associated CNVs listed in Table 1 will be discussed, while DD/ASD cohorts will only be used for comparison. Almost simultaneously with the identification of deletions at 1q21.1 and 13q13.3 as risk factors for SZ, other teams implicated their role in DD and ASD (20,21). This trend continued with the implication of 16p11.2 duplications in DD (22) and NRXN1 deletions in ASD (23). These results came as somewhat of a surprise to psychiatric geneticists, as a genetic link between DD/ASD and SZ was not obvious from the evidence provided by family or twin studies. It is now clear that all 11 SZ-associated CNVs from Table 1 are also risk factors for DD/ASD/CM. In fact, for many of them the frequencies are even higher in these disorders, compared with SZ (17).

In contrast, studies on bipolar affective disorder (BD) found much less evidence that these CNVs (or in fact, any CNVs) played any significant role, with the exception of duplications at 16p11.2 (14,18,22,24). These negative findings were also surprising to most researchers, as BD and SZ have always been known to share a substantial genetic component (25); many families have members affected with both BD and SZ, and many patients share symptoms of both conditions, making the distinction between the two diagnoses somewhat arbitrary and occasionally very difficult. With hindsight, it has been known for a long time that SZ patients suffer from substantial cognitive deficits, while BD patients are much more cognitively preserved and tend to make full recoveries between episodes. Perhaps genetic variants that cause cognitive problems (like the CNVs) should have been predicted to play a smaller role in BD.

The comparisons between the rates of these CNVs in different neuropsychiatric disorders are easily recognizable in Figure 1. The highest rates for most CNVs are found in the group of DD/ASD/CM, followed by SZ. BD patients have similar rates to controls and sometimes even lower rates. Cumulatively, these CNVs are found in 3.3% of DD/ASD/CM cases (using the data summarized by Kirov et al. (17)), 2.4% of SZ cases and 0.5% in both BD cases and the control populations.

Cognition and CNVs

Carriers of these CNVs are at an increased risk of having cognitive deficits. Stefansson et al. (26) demonstrated that even phenotypically healthy carriers of CNVs have cognitive deficits (i.e. these deficits are not secondary to their psychotic illness or to the effect of medication). All these CNVs affect multiple genes (except NRXN1), and some effect on brain function from the dysregulation of multiple genes should be expected. This effect on cognitive function is consistent with the known cognitive problems experienced by SZ patients who have been shown to have, on average, one standard deviation lower cognitive performance, compared with the general population (27,28). A small percentage of this can be accounted for by the role of large and rare CNVs. Our group is analysing data that show that SZ patients with one of these CNVs have even worse cognitive performance than SZ patients without such CNVs (James Walters, personal communication). On the other hand, BD patients have much more subtle cognitive problems (if any), consistent with a lower rate of pathogenic CNVs in this disorder.

CNVs are produced by de novo mutation events and filtered out by natural selection

We and others have shown that a large proportion of the 11 CNVs associated with SZ are produced by de novo mutations. About 5% of SZ probands have a de novo CNV (29,30). Most of them are produced by NAHR (5), as they are flanked by LCRs. The presence of flanking LCRs results in high mutation rates of these CNVs (6,31), which range between ∼1 : 4000 and 1 : 20 000 live births. These high mutation rates should result in much higher rates of these CNVs in the general population, compared to the observed rates. It is clear that they must be eliminated from the population by negative selection. Indeed, many studies have shown that SZ patients have lower fecundity (32–34), probably secondary to problems in forming relationships. Carriers of some of these CNVs have indeed been shown to have fewer offspring (26,35) (Table 2).

Table 2.

Observed fecundity among carriers, predicted selection coefficients and penetrance estimates

Locus	Fecundity	s	Penetrance (%)
			SZ	DD/ASD/CM	Combined
1q21.1 del	0.71	0.26	5.2	35	40
1q21.1 dup	1.02	0.23	2.9	18	21
NRXN1 exonic del	0.83	0.23	6.4	26	33
3q29 del	–	0.83	18	53	71
WBS dup	–	0.61	6.0	44	50
15q11.2 del	0.92	0.09	2.0	11	13
PWS/AS dup	0.69	0.5	4.2	54	58
15q13.3 del	–	0.31	4.7	35	40
16p13.11 dup	0.91	0.13	2.2	8.4	10.6
16p11.2 dup	0.89	0.33	8.0	26	34
22q11.2 del	0.79 (0.28)	0.8	12	88	100

s, selection coefficients as reported in Kirov et al. (17).

Fecundity of CNVs carriers is according to Stefansson et al. (26), except for 22q11.2 deletion, where (in brackets) is shown the result reported by Costain et al. (35) on 141 Canadian adults with 22q11.2 deletions. This correlates better with the estimated high selection coefficient for this CNV.

Fecundity and s should show an inverse relationship if they are correlated (high s should result in low fecundity).

We have attempted to estimate the strength of the negative selection (s) operating against these CNVs, based on the ratio between de novo CNVs and all observed CNVs among carriers (de novo + transmitted), assuming that the rate of these CNVs remains the same from generation to generation (17,31). The selection coefficients are very strong, ranging between 9% for deletions at 15q11.2 to 83% for deletions at 3q29 (17,31) (Table 2). Our theoretical estimates for the selection coefficients (s) operating against these CNVs correlate well with the observed fecundity of carriers of the CNVs (Table 2). The observed fecundity rates tend to be higher than what would be expected according to our estimations of s, possibly because they are based mostly on healthy carriers, while the estimates of s are based largely on populations of persons affected with SZ/DD and ASD (17,31). The low frequencies of these CNVs in the population are maintained by a balance between new mutations and the strength of the negative selection, while new mutations are eliminated after a very small number of generations (31). Not surprisingly, CNVs conferring higher risk for SZ have lower rates in the general population (Fig. 2).

Penetrance of CNVs for neuropsychiatric disorders

The SZ-associated CNVs from Table 1 are not fully penetrant, unlike mutations that cause Mendelian disorders. Geneticists and genetic counsellors could benefit if they have more accurate estimates of their penetrance for different disorders (i.e. what is the risk for a carrier to develop a disorder) (36). The first attempt to estimate the penetrance for SZ for some of these CNVs was made by Vassos et al. (37) who found fairly small estimates of between 2 and 7% (except for the 22q11.2 deletion), suggesting that they are not very high-risk variants. However, this does not account for their effect on DD/ASD/CM. The penetrance for these disorders was estimated for a small subset of these CNVs by Rosenfeld et al. (38) and was found to be much higher, at 10–62%. As carriers of these CNVs can develop any of these conditions (DD/ASD/CM in young age and SZ during adolescence or adulthood), it is necessary to estimate the penetrance of CNVs for any serious neurodevelopmental disorder. The combined group of DD/ASD/CM is several times more prevalent in the general population than SZ, and these CNVs have higher rates among the group of DD/ASD/CM. These two factors result in much higher penetrance estimates for the latter conditions (17). The combined penetrance for these CNVs is therefore quite substantial, varying from ∼10% to nearly 100% (Figure 3 and Table 2). These are serious effect sizes, conferring substantial adverse effects on health.

Figure 3.

Penetrance of CNVs for SZ (red) and for the group of DD/ASD/CM (black).

We have considered if we have over-estimated the penetrance of these CNVs because the patients who take part in such studies might be a more affected sub-group. We have repeated the estimates, using population frequencies of 0.5% for SZ and 2% for the combined group of DD/ASD/CM (instead of 1 and 4%, respectively, as in the original publications). We also used another study reporting slightly lower rates of these CNVs among DD/ASD/CM (39). The penetrance estimates are only slightly reduced after such strong corrections (data not presented) and still lead to the same conclusions. On the other hand, the penetrance of these CNVs might be actually higher, if we include more subtle outcomes, such as mild cognitive problems, as even healthy carriers of these CNVs have subtle cognitive deficits, as shown by Stefansson et al. (26).

If a CNV confers a high risk to develop one of these disorders (i.e. has a high penetrance), then we would expect to find high negative selection operating against it. Figure 4 (based on the data shown in Table 2) shows that this is indeed the case, producing a striking correlation. We should point out that the selection coefficients and frequencies are based on different datasets, so the strong correlation between them provides further independent support that these conjectures are correct.

Figure 4.

Correlation between the overall penetrance of CNVs for any neurodevelopmental disorder and their selection coefficients (s). The numbers are based on those in Tables 1 and 2.

Conclusions

The discovery of CNVs that increase risk for SZ has revolutionized the genetics of SZ. Researchers now have risk loci of large effect that can guide research into the molecular mechanisms resulting in psychosis and will hopefully result in new treatments. Genetic counselling of carriers of CNVs is now possible and psychiatrists are beginning to debate the various ethical and practical implications of this new knowledge (36). Carriers of these CNVs can develop various medical problems and a better knowledge about these risks should also improve the care of such patients. Further CNVs that increase risk for SZ and other psychiatric disorders are also likely to be identified when larger studies are completed.

Conflict ofInterest statement. None declared.

Funding

This work was supported by the Medical Research Council, London, grants G0800509 and G0801418.