De Novo Mutation in Schizophrenia

Abstract

Several studies in the last 5 years have shown that newly arising (de novo) mutations contribute to the genetics of schizophrenia (SZ). This will replenish genetic variants removed by natural selection and could, in part, explain why SZ prevalence has remained stable in the general population despite low fecundity. The strongest evidence to date for the association between SZ and de novo mutation comes from studies of de novo copy number variation (CNV), where the rate of de novo CNV mutation is shown to be increased in cases when compared with controls, and genes disrupted by these mutations are enriched for those encoding proteins involved in synaptic function and development. Previous estimates have shown high levels of negative selection operating against SZ associated CNVs, and we provide an updated estimate of these levels of selection using the most recently published data. Recent studies involving next-generation sequencing technology have provided preliminary evidence that de novo single-nucleotide mutations might also increase risk of SZ. However, these are very small in scale, and the results can only be considered as preliminary.

Recent studies have supported the view that genetic susceptibility to schizophrenia (SZ) involves a spectrum of risk alleles, common and rare, with effect sizes ranging from small to large, but each allele contributing only a small fraction to the total population variance.¹ Such a view is consistent with population genetic theory and the so-called mixed model of SZ; it has also received compelling support from recent genomic studies.² Genome-wide association studies have identified a number of risk loci at genome-wide levels of significance as well as evidence for a substantial burden of common risk loci. Moreover, these recent findings suggest genetic overlap with bipolar disorder, which has traditionally been assumed to be genetically distinct from SZ. Genome-wide studies of at least one class of relatively uncommon variant, submicroscopic chromosomal abnormalities often referred to as copy number variations (CNVs), suggest that these individually confer relatively high risks of SZ (ORs 3–30) compared with common risk alleles (ORs < 1.3).^3–5 There is evidence both for an increased burden of large rare CNVs in SZ and that risk is conferred by a number of specific CNV loci. Many of these CNVs have also been implicated in autism, mental retardation, epilepsy, and other neurodevelopment disorders.

SZ is associated with substantially reduced fecundity,⁶ with fertility rates estimated at less than half that of healthy controls.^6,7 As a result, genetic variants that confer high risk to SZ might be expected to be subject to strong negative selection and will be eventually filtered out from the population. It follows that if genetic variants with large effect sizes make an important contribution to SZ, the apparent stability in population prevalence⁸ could be explained by new (de novo) mutations that replenish those lost by selection (ie, mutation-selection balance)⁹ though other explanations are possible. De novo mutations appear to make a significant contribution to other neuropsychiatric disorders that show reduced fecundity, such as autism¹⁰ and intellectual disability,¹¹ and this raises the question of whether and to what extent they might contribute to SZ.

De Novo CNVs

Recently, direct evidence that de novo mutation plays a role in SZ has come from studies of CNVs in parent-proband trios.^12–14 The first such study found the rate in cases (10%) to be ∼8 times that of controls (1.3%).¹⁴ This increase was only observed in cases without a family history of the disorder. Two subsequent studies have also found an elevated rate in cases, but estimates have been lower at 5.1%¹² and 4.5%.¹³ The de novo rate in one of these studies¹² was again higher when restricting analysis to probands with negative family history (5.5%), but the difference between familial and nonfamilial SZ was not significant. De novo rates in controls are consistently reported as being lower, at between 1% and 2.2%.^12,13

Evidence from case-control studies now supports association between SZ and CNVs at more than 10 genomic loci (table 1),^3,15 and de novo occurrence has been observed at all strongly supported SZ CNV loci.¹⁵ Moreover, in the largest study of de novo CNVs in SZ to date in over 600 parent-proband trios, 8 of the 34 de novo events occurred at known SZ CNV loci,¹² which is further support that they are highly enriched for pathogenic loci. Most of these confirmed SZ CNV loci are flanked by large repetitive regions of DNA that have high homology to each other, called segmental duplications, making these regions susceptible to copy number alteration by mechanisms of nonallelic homologous recombination and explaining their high mutation rate.

Table 1.

Odds Ratios, Frequency in Schizophrenia (SZ) and Selection Coefficients for Associated SZ CNV Loci

CNV Locus	OR in SZ vs Controls	Frequency in SZ (%)	Selection Coefficient (de novo ratio)
1q21.1 del	9.2	0.18	0.33
NRXN1 exonic del	6.93	0.16	0.22
3q29 del	49.5	0.097	0.84
15q11.2 del	2.2	0.57	0.11
15q13.3 del	8.3	0.19	0.37
15q11–13 dup	7.3	0.05	0.54
16p13.1 dup	2.1	0.28	0.18
16p11.2 dup	9.8	0.3	0.39
17q12 (31.8–33.3Mb) del	18.4	0.06	0.89
22q11.2 del	(44.2–∞)	0.31	0.80

Note: Del, deletion; Dup, duplication; CNV, copy number variation.

The effect sizes for SZ of the associated CNVs are high (ORs between 3 and >30).³ We recently estimated that the strength of selection (s) operating against them is also very strong, ranging between 0.11 and 0.89 (table 1).¹⁵ To put these rates of selection into perspective, a mutation with s = 0.01 remains in the population for ∼100 generations, whereas a mutation with s = 1 is either lethal or leaves carriers incapable of reproducing.⁹ Based on the selection coefficients, the average number of generations a de novo CNV from the list in table 1 will pass through before it is removed from the population can be estimated. Given the high associated individual risks and the known reduced fecundity of SZ, it is not surprising that risk CNVs are quickly removed from the population, persisting on average between 1.2 (3q29 deletion) and 5.1 generations (15q11.2 deletion) following de novo mutation.¹⁵ The CNVs with higher levels of s have lower population frequencies and are maintained at their respective frequencies by high locus specific mutation rates of between 1.7 × 10⁻⁵ and 1.4 × 10⁻⁴.¹⁵ These locus specific mutation rates are similar to those observed for other pathogenic genomic rearrangements.¹⁶

The odds ratios and frequencies in table 1 were taken from Grozeva et al,³ apart from NRXN1 and 15q11-13 duplications, which were taken from Rees et al.¹⁵ Selection coefficients are based on Rees et al¹⁵ and have been updated with recently published data.

Larger CNVs are more likely to be pathogenic as they tend to disrupt one or more genes, and CNV gene content and size are correlated with the strength of negative selection operating against them.¹⁵ Indeed, the best-supported CNV locus in SZ is 22q11.21 where large, generally 2.7 MB deletions, disrupt 40 genes and are known to cause velocardiofacial syndrome. Up to 30% of carriers of these deletions are known to suffer from psychosis.⁴

Given that the frequency of inherited mutations is critically dependent upon selection, it is to be expected that they will be on average smaller than de novo mutations, and this is indeed what is observed.¹² Although larger CNVs have greater potential to be pathogenic, de novo CNVs between 10 and 100 kb have been detected at an increased rate of 1.7% in cases compared with 0.7% in controls, suggesting that susceptibility to SZ is not limited to large (>1 Mb) de novo CNVs.¹³

SZ has been associated with increased paternal age, and this has been attributed to increased de novo mutation with age in the male germline.¹⁷ In one study of de novo CNVs in SZ, parental origin was determined for 21 of 34 de novos, of which 14 occurred paternally and 7 occurred maternally, though this difference did not achieve statistical significance.

As SZ susceptibility variants that confer large effects on risk are present at low-population frequencies, limitations in sample size usually mean studies are underpowered for gaining significant association for a single locus that would survive correction for multiple testing. This in turn limits the biological insights that can be drawn from studies of individual CNVs, as does the fact that most span multiple genes. Seeking to overcome these limitations, researchers have sought to draw inferences from analysis of gene sets representing biological pathways that are disrupted by CNVs. One study using manually curated gene sets based on proteomic studies found that case de novo CNVs were significantly enriched for genes encoding members of the postsynaptic density proteome, specifically those involved in postsynaptic signaling complexes and synaptic plasticity.¹² A second study using independent samples and methods¹³ identified several functional categories (defined by the gene ontology project) to be enriched among case de novo CNVs, and a subsequent analysis that tested the enrichment of these groups in a large case-control dataset showed enrichment of genes encoding synaptic proteins and neurodevelopment.

De Novo Single-Nucleotide/Indel Mutations

Research on de novo point mutations and small insertion-deletions (indels) has been more limited, but with the development of next-generation sequencing technology, such studies are likely to increase in scale in the very near future. However, de novo single-nucleotide variants (SNVs) of possible relevance to SZ have been reported in single-gene sequencing studies¹⁸ and more recently in moderately large targeted resequencing studies.¹⁹ A few whole-exome sequencing studies have also been reported but these are very small in scale and the results can only be considered as preliminary (table 2).^20,21

Table 2.

Summary of Current De Novo Sequencing Studies in Schizophrenia (SZ)

			De Novo Mutations
Author	Study	Samples	Number of SZ De Novos	Number of Probands With De Novos	Missense/Nonsense	Silent	Intronic	Indels
Gauthier et al18	Single-gene SHANK3	185 SZ proband-parents 285 control-parents	2	2	1/1
Awadalla et al19	Deep resequencing of 401 synapse expressed genes	142 ASD proband-parents143 SZ proband-parents285 control-parents	8	7	2/2	2	1	1
Girard et al20	Whole-exome sequencing	14 SZ proband-parents	15	8	11/4
Xu et al21	Whole-exome sequencing	53 SZ proband-parents22 control-parents	38	27	32/0	2		4

Note: ASD, Autism Spectrum Disorder.

Gauthier et al¹⁸ sequenced the synaptic scaffolding protein SHANK3 in 185 SZ offspring-parent trios and 285 controls because de novo SNVs in this gene had previously been associated with autism spectrum disorder (ASD). This study found one nonsense and one missense de novo SNV in cases and none in controls. In another report, Awadalla et al¹⁹ presented sequencing data from 401 synapse expressed genes in 142 unrelated ASD, 143 SZ and 285 population control offspring-parent trios. The neutral SNV mutation rate, as estimated from the number of de novos occurring in apparently nonfunctional DNA, was reported to be 1.36 × 10⁻⁸. From those data and taking into account the expected elevated mutation rates at CpG sites, the authors under the null hypothesis expected to see 1.3 de novo SNVs in a total of 96 Mb of “functional” sequence in nonfamilial ASD/SZ probands. However, 6 nonsynonymous de novo SNV mutations were found. The nonsynonymous de novo mutations found in SZ patients from this study include one missense and one nonsense SNV in SHANK3, one missense SNV in GRIN2B, one nonsense SNV in KIF17, and a de novo indel within the synaptic gene NRXN1, thus providing additional evidence to previous CNV studies implicating NRXN1 with SZ.²² While these findings are certainly of interest, it should be recalled that these genes were selected on the basis of being expressed in synapses, and further work is required before we can conclude that the specific mutations are pathogenic.

In contrast to the above candidate gene-based approaches, whole-exome sequencing allows for an unbiased genome-wide scan of coding sequence variants. It is estimated that each human genome has around 600 rare missense mutations,²³ and the expected number of genome-wide de novo mutations per new born is between 50 and 100, of which only 0.86 change the amino acid sequence.²⁴ Around 20% of these amino acid changing mutations are estimated to be strongly deleterious, ∼53% mildly deleterious, with the remaining having little to no phenotypic effect.²³ Girard et al²⁰ sequenced the exomes of 14 SZ trios and found 15 de novos in 8 probands. In this study, no single gene was hit by multiple de novos, and none of the genes hit by de novos had previously been associated with SZ. All 15 mutations were nonsynonymous and interestingly, 4 were nonsense mutations, which is more than expected suggesting that some may be of relevance to disease. A comparison with de novo rates observed in other studies showed the de novo rate to be significantly higher in these 14 SZ trios (P < .02).²⁰ However, this increased rate should be viewed with caution due to the small sample size in both the cases (14 trios) and in the comparison studies (3 families sequenced in 2 studies and an estimation of de novo rate based on a database of de novo mutations in monogenic disorders).

In the second SZ exome–sequencing study published to date, 53 sporadic cases, 22 unaffected controls, and their parents were studied.²¹ Here, the observed rate of case de novos did not differ from that reported in previous studies. Thirty-four de novo SNVs were found in 27 patients, of which 32 were nonsynonymous. None of the variants that occurred as de novo mutations in cases were found in any of the 22 controls or in any available previous sequencing study. The ratio of nonsynonymous to synonymous mutations for de novo SNVs was found to be highly increased at 16:1, while the ratio of nonsynonymous to synonymous mutations in novel (not reported in dbSNP132) and private inherited variants was 1.61:1 and 1.69:1, respectively. This ratio in case de novos was significantly higher (P < .0003, 10-fold enrichment).²¹ A total of 7 de novo SNVs were found in the 22 unaffected controls that were sequenced by the same methods. The ratio of nonsynonymous to synonymous de novo mutations in controls was 1.33:1, again showing that case de novo SNVs are enriched for being nonsynonymous. However, this high level of enrichment might not be replicated when larger studies are available, and the expected ratio for de novos is 2.2:1, as estimated from theoretical studies.²³ Biologically important sequences are generally highly conserved throughout different species, and variants in conserved sequence are more likely to have an adverse effect on protein function. Through the use of bioinformatic tools, the authors showed that, in comparison to inherited SNVs, de novo SNVs tended to involve highly conserved sequence and therefore have a greater potential for being pathogenic.

Conclusions

There is now substantial evidence supporting the role of de novo CNV mutation in the pathogenesis of SZ and increasing evidence that de novo SNV mutation might also play a role. It is clear from the high heritability of SZ and from genomic findings in relation to both common and rare variants that inherited genetic variation plays a key role in the disorder. While the extent of involvement of de novo mutation is currently unknown, the evidence for its involvement offers a new approach to detecting potentially pathogenic variants and may help to explain how the prevalence of SZ remains stable in the general population given a reduction in fitness.

The advent of next-generation sequencing approaches has meant that whole-exome studies aimed at detecting de novo mutations are currently feasible and affordable and whole-genome sequencing approaches will soon be so. An important challenge will be how to establish that de novo events are pathogenic given that they are expected to be rare and that the extent of enrichment for de novo SNVs seems less so than for large CNVs. This will require follow-up resequencing studies in large samples and, based upon experience with CNVs, will be aided by detailed bioinformatic analyses of the impact of mutations on protein function and of the convergence of mutations on specific biological pathways.

Funding

This work was supported by a Medical Research Council (UK) programme (G0800509) and Centre grant (G0801418) and a Medical Research Studentship to ER.