Characterizing Rare DNA Copy-Number Variants in Pediatric Obsessive-Compulsive Disorder RH = Rare Copy-Number Variants in OCD

Abstract

Objective:

Pediatric obsessive-compulsive disorder (OCD) is a common neuropsychiatric disorder for which genetic factors play an important role. Recent studies have demonstrated an enrichment of rare de novo DNA single nucleotide variants in OCD cases compared to controls, and larger studies have examined copy-number variants (CNVs) using microarray data. Our study examines rare de novo CNVs using whole-exome sequencing (WES) data to provide additional insight into genetic factors and biological processes underlying OCD.

Method:

We detected CNVs using whole-exome DNA sequencing (WES) data from 183 OCD trio families (unaffected parents and children with OCD) and 771 control families to test the hypothesis that rare de novo CNVs are enriched in OCD cases compared to controls. Our primary analysis used the eXome-Hidden Markov Model (XHMM) to identify CNVs in silico. We performed burden analyses comparing individuals with OCD vs. controls and downstream biological systems analyses of CNVs in probands with OCD. We then used a second algorithm (GATK-gCNV) to confirm our primary analysis.

Results:

Our findings demonstrate a higher rate of rare de novo CNVs detected by WES in individuals with OCD (0.07 CNVs per proband) compared to controls (0.005) (corrected rate ratio = 11.7 95% CI, 3.6–50.0, p = 4.00×10⁻⁶). We confirmed this enrichment using GATK-gCNV. The majority of these rare de novo CNVs in OCD cases are predicted to be pathogenic or likely pathogenic, and an examination of genes disrupted by rare de novo CNVs in OCD cases finds enrichment of several gene-ontology sets.

Conclusion:

This study shows for the first time an enrichment of rare de novo CNVs detected by WES in OCD, complementing previous larger CNV studies and providing additional insight into genetic factors underlying OCD risk.

INTRODUCTION

Obsessive-compulsive disorder (OCD) is a neuropsychiatric condition with an estimated prevalence of 1–3% worldwide. It is characterized by disabling obsessions (intrusive, unwanted thoughts, sensations, or urges) and compulsions (ritualized, repetitive behaviors that are difficult to control).¹ Although serotonergic antidepressants and behavioral therapies have been used to treat OCD for several decades, a sizeable proportion of patients do not find relief from these treatments, and OCD tends to persist and become chronic.^2–5 Twin studies have demonstrated substantial heritability of pediatric OCD, with estimates around 45–65%.^1,6–9 Family studies examining multigenerational pedigrees affected by OCD or non-twin siblings and relatives of affected individuals, along with population-based studies, also point to a significant genetic contribution and indicate that the pediatric-onset form of the disorder is particularly heritable.^10–12 Despite evidence for a significant genetic contribution to OCD pathogenesis, only a few risk genes have been identified. A limited understanding of OCD etiology remains a barrier to developing more effective therapeutic options. For this reason, there is a great incentive to study the molecular basis of the disorder.

In recent years, there has been progress in identifying genes and genetic variation relevant to OCD. Studies examining whole-exome DNA sequencing (WES) of individuals with OCD (probands) and their parents have demonstrated an increased rate of rare gene-damaging de novo sequence variants^13–15 and implicated putative risk genes.¹⁴ These rare de novo variants arise spontaneously due to DNA replication errors and are not inherited from parents. Several genome-wide association studies (GWAS) of OCD have been conducted to examine the contribution of common variation.^16–21 Recently, the largest GWAS to date in OCD (combined pediatric and adult-onset) found genome-wide significant associations at approximately 30 independent genetic loci, including one previously identified locus.²²

Examination of copy-number variants (CNVs; deletions or duplications of DNA sequences over one kilobase in length) has shown promise in the study of neuropsychiatric disorders. Recent studies of CNVs in Tourette’s disorder (TD), which frequently co-occurs with OCD and is believed to have significant genetic overlap, have demonstrated an increased rate of rare CNVs in participants with TD compared to controls.^23,24

Recent large-scale studies have examined CNVs in OCD using microarray data. A recent case-control study found an increased rate of large, rare CNVs in OCD compared to controls.²⁵ Other studies have shown that OCD cases harbor a significantly higher rate of large rare (>500 kilobases) deletions overlapping regions implicated in other neurodevelopmental disorders and have reported enrichment of rare CNVs affecting genes related to neurological function.^26–28 Other microarray studies have found potentially damaging and clinically significant CNVs in OCD.^29,30 However, enrichment of de novo CNVs in OCD has not yet been demonstrated. While these microarray studies have provided important insights into larger CNVs, examination of smaller variants using whole-exome sequencing approaches may reveal additional contributions to OCD risk.

The few studies investigating CNVs in OCD have used microarray data, which are limited in resolution by the density of their DNA probes, usually down to a lower limit of 50 to 100 kilobases (kb).³¹ In contrast, high-throughput sequencing approaches like WES can detect small-to medium-sized CNVs, which are more frequent than large CNVs.^13,32 In autism spectrum disorders (ASD), rare exonic deletions of 1–30 kb size have been estimated to contribute to disease risk in up to 7% of cases, suggesting the potential clinical relevance of smaller exonic CNVs. Furthermore, unlike large CNVs that often contain multiple genes, small exonic CNVs typically affect just one gene, making them potentially useful for risk gene discovery and downstream biological systems and pathway analyses.³²

In this study, we apply a computational approach to detect rare de novo CNVs in WES that could impart a previously undetected contribution to OCD pathogenesis and provide new insights into the genetic landscape and underlying biology. We compare the rates of rare and rare de novo CNVs in OCD probands to those of unaffected individuals and explore the significance of these variants in downstream analyses. Based on prior findings investigating rare CNVs in other neuropsychiatric disorders and results with other variant types detected by whole-exome sequencing in OCD, we expected to find an increased rate of rare de novo CNVs in OCD compared to control parent-child trios.

METHOD

Data Collection and Processing

Participant recruitment, sample collection, and whole-exome DNA sequencing (WES) were performed as described in our previous report.¹⁴ In brief, we generated WES data from peripheral blood DNA of 686 individuals from parent-child OCD trios recruited in Toronto, Canada; São Paulo, Brazil; and New Haven, CT, USA; and from a separate Tourette International Collaborative Genetics (TIC Genetics) study that included patients with both OCD and chronic tics.^33,34 To enrich for de novo genetic variants, inclusion criteria required that the proband has early-onset OCD and that the parents are unaffected (see Supplement 1, available online, for additional assessment and inclusion details). We excluded 20 samples from incomplete trios, leaving 222 complete parent-child trios. These samples were sequenced at the Yale Center for Genome Analysis (YCGA) using the NimbleGen SeqCap EZ Exome v2 (109 trios) or MedExome (113 trios) capture (Roche NimbleGen, Madison, WI) and the Illumina HiSeq 2000 platform (74-bp paired-end reads) (Illumina, San Diego, CA). These data were compared to WES from peripheral blood DNA in 2,562 parents and control siblings (854 parent-child trios) without OCD from the Simons Simplex Collection (SSC),³⁵ sequenced at YCGA using NimbleGen SeqCap EZ Exome v2 capture and the Illumina HiSeq 2000 platform. These WES data were aligned using a well-validated analysis pipeline following Genome Analysis Toolkit (GATK) Best Practices guidelines.³⁶ Sequence reads obtained from WES were aligned to the GRCh37 human reference genome using the Burrows-Wheeler Aligner, PCR duplicates were marked using Picard’s MarkDuplicates tool, and a binary alignment map (BAM) file containing the aligned exome data was generated.^37,38 Coverage depth and other sample metrics are included in Table S1, available online.

CNV Calling With XHMM

DNA sequencing read depths were calculated for 3,228 individuals (222 OCD and 854 control parent-child trios) using GATK’s DepthOfCoverage tool. Calls were generated using eXome-Hidden Markov Model (XHMM), a statistical package to detect CNVs from normalized read-depth data from targeted sequencing.³⁹ XHMM default quality control measures, described in Supplement 1, available online, excluded nine individuals; additional outlier trios were excluded based on principal component analysis of sequencing quality metrics as described in our previous report¹⁴ (Table S1, available online). Our primary analysis used a 6-exome target minimum threshold for CNV calling. Target intervals with extreme GC content and low sequence complexity were excluded (Supplement 1, available online). We excluded four control trios with probands found to screen positively for OCD based on family medical history data collected during interviews with Simons Foundation clinical staff.³⁵ For inclusion in subsequent analyses, we retained from our sample set 183 OCD trios (117 male probands, 66 female) and 771 control trios (354 male probands, 417 female) (Table 1; Table S1, available online).

Table 1:

Baseline Demographic Characteristics of Case and Control Samples

Characteristic		OCD cases	Controls
Sex	Female	66 (36%)	417 (54%)
	Male	117 (64%)	354 (46%)
Reported race	American Indian	1 (0.5%)	2 (0.3%)
	Asian	2 (1%)	30 (4%)
	Black	0	2 (0.3%)
	Pacific Islander	0	0
	White	157 (86%)	587 (76%)
	More than one race	6 (3%)	63 (8%)
	Other	0	42 (5%)
	Unknown	17 (9%)	45 (6%)

Note: Numbers are after quality control. Cases are children with OCD. Controls are unaffected children (siblings) from the Simons Simplex Collection.

We then used a previously described protocol³⁹ using PLINK and PLINK/Seq to classify rare CNVs (frequency <0.1% among all parents in the sample set) in the children as transmitted (inherited) or de novo. PLINK/Seq quality thresholds for de novo calls retained variants with a high probability of a CNV in the child and a high probability of no CNV in the parents (Supplement 1, available online).³⁹ We discarded two additional outlier OCD trios (one male proband, one female) with excess CNV calls in the proband. The final CNV set is provided in Table S2, available online. We used the XHMM visualization function to inspect plots of XHMM-normalized read depths across all samples for de novo CNV calls in OCD probands (Figure S1, available online).

CNV Burden Analysis

For our primary analysis, we calculated the burden of CNVs in cases and controls using two different approaches. First, we calculated the number of rare de novo CNVs per individual. We compared these per-individual CNV rates in our OCD samples to rates in SSC controls using the rateratio.test R package.⁴⁰ Second, to ensure any burden difference was not driven by multiple CNVs in an individual, we calculated the proportion of individuals in the case and control groups with at least one rare de novo CNV call. We compared these group proportions using a Fisher exact test (one-tailed). Based on previous studies of rare genetic variation in OCD and related childhood-onset conditions,^{14,15,23,32,41} we hypothesized that both the rate and proportion of rare de novo CNVs would be significantly greater in OCD cases versus controls. Rates and proportions were calculated together and separately for deletions and duplications.

To account for differences in mean sequencing coverage between cases and controls (Table S1, available online), we calculated the number of callable base pairs per trio using the GATK DepthOfCoverage tool.³⁶ Callable bases were defined as those with a sequencing depth of at least ten reads in all three family members at that genomic position, as XHMM was set to require a minimum average read depth of 10x across samples to retain an interval in the analysis. The ratio of callable bases between cases and controls was used to correct the rate ratio of CNVs between the two groups. We did not compare CNV lengths between cases and controls because the start and end points (breakpoints) of CNVs called from WES are uncertain, as they may fall outside the targeted exomic intervals.

We used X-CNV to predict the pathogenicity of de novo CNVs in cases and controls.⁴² To address potential batch effects, sex-based differences, and differences between capture libraries, we conducted additional analyses detailed in Supplement 1, available online; Table S3, available online; Table S4, available online.

Given our sample size of 183 OCD trios and 771 control trios, we estimated statistical power to detect differences in de novo CNV rates between groups. While our study was adequately powered to detect large effects in overall burden, we acknowledge limited power for analyses of specific CNV types and gene-level associations. We, therefore, focus primarily on aggregate burden differences and consider gene-level findings as preliminary.

Exploratory Gene Ontology, Spatiotemporal Expression, and Overlap Analyses

We used AnnotSV to identify genes overlapping rare de novo CNVs in individuals with OCD based on National Center for Biotechnology Information reference sequence (NCBI RefSeq) human genome hg19 transcripts.⁴³ These genes were used as the input gene list for downstream gene ontology and spatiotemporal expression analyses. Metascape was used to quantify gene function enrichment with ontology terms from several knowledgebases.⁴⁴ All known genes in the human genome were used as background when calculating enrichment factors (the ratio between the observed counts and the counts expected by chance) and an associated p-value. Spatiotemporal expression analyses using these genes were conducted using the Cell-type Specific Expression Analysis (CSEA) tool.⁴⁵ To assess convergence among CNVs detected in this study with variants reported in the literature, we used Gene4Denovo, a comprehensive integrated database of de novo mutations from multiple published WES and whole-genome sequencing studies.⁴⁶ Finally, we used CNVRuler, a software tool for association analysis of CNV regions.⁴⁷

CNV Calling with GATK-gCNV

To confirm our primary burden analysis, we next performed CNV calling from the same case and control WES datasets using the GATK-gCNV tool.⁴⁸ Detailed methods are described in Supplement 1, available online. In brief, models were generated for cases and controls and used to call CNVs in “cohort mode.” Exome targets, samples, and variants were filtered according to the recommended parameters.⁴⁸ We retained rare (frequency < 0.1% in the parent sample set) autosomal calls that met the recommended quality score thresholds. Calls in children overlapping calls in their parents by less than 30% were classified as de novo.

RESULTS

CNV Mutation Rates and Burden Analysis

For our primary analysis (XHMM, minimum 6-exome target) in our sample of OCD probands, while our sample size resulted in wide confidence intervals, we detected a rare de novo CNV rate of 0.07 per individual, 95% confidence interval = 0.04–0.12. We observed a significant enrichment of rare de novo CNVs in OCD probands compared to controls (0.07 vs. 0.005 per individual, 95% confidence interval = 0.001–0.01), corrected rate ratio = 11.7, 95% confidence interval = 3.56–50.0, p = 4.00×10⁻⁶). This difference appears to be driven by a significantly increased rate of rare de novo deletions, but not duplications, in OCD compared to controls (Table 2).

Table 2:

Number and Rate of Rare Copy-Number Variants (CNV) Calls in Obsessive-Compulsive Disorder (OCD) Cases and Controls

Variant class	Type	CNV count (CNVs per individual)		Rate ratio (95% confidence interval)	Corrected rate ratio (95% confidence interval)	p
		OCD 181 trios	Control 771 trios
All rare	All	49 (0.27)	184 (0.24)	1.14 (0.81–1.56)	1.04 (0.74–1.43)	0.44
	DEL	21 (0.12)	60 (0.08)	1.49 (0.86–2.49)	1.36 (0.79–2.27)	0.14
	DUP	28 (0.15)	124 (0.16)	0.96 (0.62–1.46)	0.88 (0.56–1.33)	0.31
De novo	All	12 (0.07)	4 (0.005)	12.8 (3.90–54.4)	11.7 (3.54–49.7)	4×10 −6
	DEL	9 (0.05)	0 (0)	Inf (8.42-Inf)	Inf (7.69-Inf)	6×10 −7
	DUP	3 (0.02)	4 (0.005)	3.20 (0.47–18.9)	2.92 (0.43–17.3)	0.16

Note: Rates of de novo CNVs per individual are higher in OCD probands compared to unaffected controls. Boldface p values are statistically significant. DEL = deletion; DUP = duplication.

There was no statistically significant difference between the rate of all rare (de novo and inherited) CNVs per individual in OCD probands compared to controls (0.27 vs. 0.24, corrected rate ratio = 1.04, 95% confidence interval = 0.75–1.44, p = 0.42). Similarly, there were no significant differences when comparing rates of all rare deletions and duplications separately (Table 2). The ratio of the sum of callable bases across samples in controls compared to cases, 0.913 (Table S1, available online), was used to adjust the rate ratios for differences in sequencing depth.

We also examined the proportion of individuals with at least one rare CNV to ensure any burden difference was not driven by multiple CNVs in an individual. Consistent with the above findings, a significantly greater number of OCD individuals have at least one rare de novo CNV compared to control subjects (p = 1×10⁻⁴), and this difference appears to be driven by deletions (p = 2×10⁻⁶). There is no significant difference between OCD cases and controls in the proportion of individuals with at least one rare CNV (p = 0.36) (Table 3).

Table 3:

Number and Proportion of Individuals With at Least One Rare Copy-Number Variants (CNV)

Variant class	Type	Individuals with at least one rare CNV (proportion)		Fisher exact p
		OCD (181 trios)	Control (771 trios)
All rare	All	41 (0.23)	162 (0.21)	0.36
	DEL	19 (0.11)	57 (0.07)	0.12
	DUP	23 (0.13)	112 (0.15)	0.33
De novo	All	9 (0.050)	4 (0.005)	1×10 −4
	DEL	8 (0.044)	0 (0)	2×10 −6
	DUP	2 (0.011)	4 (0.005)	0.32

Note: The proportion of children with at least one rare de novo CNV is enriched in OCD cases compared to controls. Boldface p values are statistically significant. DEL = deletion; DUP = duplication.

Given the high co-occurrence of TD in our sample, we assessed whether the known increased rate of de novo CNVs in TD could drive our results. Removing 87 individuals with OCD known to have co-occurring TD or a chronic tic disorder (including 42 from TIC Genetics), we still saw a significant burden difference for the rate of de novo CNVs (0.043 in cases vs. 0.005 in controls, corrected rate ratio = 8.25, 95% confidence interval = 1.54–44.3, p = 0.004).

In our sample, 12 rare de novo CNVs were detected among nine OCD cases. In total, these 12 CNVs overlap 25 genes (Table 4). Of the 12 de novo CNVs in OCD cases, 3 were classified as pathogenic and 6 as likely pathogenic by X-CNV. In contrast, none of the 4 de novo CNVs in controls are classified as pathogenic or likely pathogenic (Table S5, available online).

Table 4:

Rare De Novo Copy-Number Variants (CNVs) Detected in Individuals With Obsessive-Compulsive Disorder (OCD)

Sample (Site)	Sex	Other clinical diagnoses	Chromosome: start-end	Type	Size	Gene(s) impacted	Pathogenicity
OCD8134.p1 (São Paulo)	Male	TD, ADHD, SAD	2:26587170–26611971	DUP	24,801	SELENOI	Likely benign
			2:167055182–167085482	DEL	30,300	SCN1A-AS1; SCN9A	Pathogenic
			2:170657471–170681107	DUP	23,636	METTL5; SNORD3K; SSB	Likely benign
			3:130305350–130318654	DEL	13,304	COL6A6	Pathogenic
OCD8168.p1 (TIC Genetics)	Male	TD	2:112536252–112545897	DEL	9,645	ANAPC1	Likely pathogenic
OCD8046.p1 (Toronto)	Female	MDD	7:64389157–64439914	DEL	50,757	ERV3–1-ZNF117; ZNF117; ZNF273	Likely pathogenic
OCD8129.p1 (São Paulo)	Male	Dysthymia	8:101718922–101730116	DEL	11,194	PABPC1	Likely pathogenic
OCD8171.p1 (TIC Genetics)	Male	TD, ADHD	12:69124890–69279669	DEL	154,779	CPM; LOC100130075; MDM2; NUP107; SLC35E3	Likely pathogenic
OCD8221.p1 (New Haven)	Male	TD, ADHD	15:29367124–30092905	DUP	725,781	APBA2; FAM189A1; LOC100130111; NSMCE3; TJP1	Uncertain
OCD8205.p1 (TIC Genetics)	Male	TD	18:45368198–45423127	DEL	54,929	SMAD2	Pathogenic
OCD8067.p1 (Toronto)	Male	None	19:40540487–40541842	DEL	1,355	ZNF780B	Likely pathogenic
OCD8013.p1 (New Haven)	Male	None	19:40580517–40581701	DEL	1,184	ZNF780A	Likely pathogenic

Note: Pathogenicity is predicted by X-CNV. Site refers to the site of sample collection. ADHD = attention-deficit/hyperactivity disorder; DEL = deletion; DUP = duplication; MDD = major depressive disorder; SAD = separation anxiety disorder; TIC Genetics = Tourette International Collaborative Genetics Study; TD = Tourette’s disorder.

Gene Ontology Enrichment Analysis

For our list of 25 genes overlapping de novo OCD CNVs (the subclass of CNVs with the highest rate ratio in the burden analysis), we identified six ontology terms with a p-value < 0.01, an enrichment factor > 1.5, and association with at least three genes from the input list. Enrichment was found for the following ontology terms: cell cycle (p=9.37×10⁻⁵; SMAD2, MDM2, NUP107, and ANAPC1); SUMO E3 ligases (p=3.05×10⁻⁴; MDM2, NSMCE3, and NUP107); and nucleobase-containing compound transport (p=4.96×10⁻⁴; SSB, SLC35E3, and NUP107). Other significant enrichment included noncoding RNA processing, herpes simplex 1 virus infection, and axon guidance (Table S6, available online).

Spatiotemporal Gene Expression Analysis

We performed an expression analysis of the 25 genes overlapping de novo OCD CNVs across brain regions and developmental time periods using CSEA. No expression networks were significantly enriched for our list of genes. Of note, three genes were contained in the Cortex Early Fetal network (NUP107, ZNF780A, ZNF117) and Striatum Early Fetal network (NUP107, ZNF273, ZNF117) (Table S6, available online).

Overlap with Other Datasets

We did not find any overlap between genes intersecting de novo CNVs in our OCD cases (Table 4) and genes containing predicted deleterious de novo single-nucleotide variants (SNVs) or insertion-deletions (indels) in prior WES studies of OCD.^14,15 We also found no overlap with genes containing rare CNVs from prior OCD microarray studies.^26–28 We then looked for evidence of an association between OCD and CNV regions by combining our rare CNV calls with those previously published. Using CNVRuler,⁴⁷ we found one region (chr16:14938694–18165043) with nominal statistical evidence of association (p=0.04), but this did not remain significant after accounting for multiple testing (Table S7, available online). Finally, using Gene4Denovo⁴⁶ to query our list of 25 genes (Table 4), we identified that the gene ZNF273 was reported to have a de novo intronic SNV in an individual with OCD⁴⁹ and a de novo stopgain SNV in an individual with TD.²³ Also, the gene ZNF780B was reported to have a de novo nonsynonymous missense SNV in an individual with TD.²³ Neither of the two individuals with these CNVs in our OCD group had co-occurring TD. Finally, the gene PABPC1 was reported to harbor six deleterious missense variants in individuals with ASD⁵⁰; the individual in our group with this CNV did not have co-occurring ASD (Table 4).

GATK-gCNV Analysis

To assess whether our primary finding of increased de novo CNVs in OCD was robust to different CNV calling methods, we called CNVs from the same WES case and control datasets using the GATK-gCNV algorithm. As shown in Table S8, available online, and Table S9, available online, the relative rates of rare CNVs called by GATK-gCNV were consistent with findings from XHMM CNV calls. We still see significantly increased rates of all rare de novo CNVs and rare deletion CNVs at both 3- and 6-exome target thresholds. Calls are listed in Table S10, available online.

DISCUSSION

In summary, this study demonstrates for the first time an enrichment of rare de novo CNVs detected by whole-exome sequencing in OCD cases compared to controls, complementing previous larger-scale CNV studies using microarray data. Our findings suggest that this type of genetic variation contributes to OCD pathogenesis and provides additional insight into genetic factors underlying OCD. Specifically, we observed that the proportion of individuals with and per-individual rate of de novo CNVs were significantly greater in individuals with OCD compared to controls. This finding is consistent with studies of other types of de novo sequence variation in OCD^14,15, with studies of de novo CNVs in related childhood-onset neuropsychiatric conditions^23,32,41, and with a recent large-scale study of OCD cases and controls supporting a larger role for genetic loci across the entire allele frequency spectrum influencing risk for OCD.⁵¹

Our findings about rare de novo CNVs detected by WES complement recent larger studies examining CNVs in OCD using microarray data. While microarray studies provide robust evidence for the role of large CNVs in OCD, our WES approach allows the detection of smaller exonic variants that may be missed by microarray platforms. The convergence of evidence from both approaches strengthens the case for CNV involvement in OCD pathogenesis. While our sample size limits our ability to make definitive claims about specific genes or pathways, as reflected in the wide confidence intervals in our analyses, our findings add to the growing evidence for the role of rare variation in OCD risk.

Although the confidence intervals reported in this study are wide given the sample size, our estimated rate of rare de novo CNVs (~7%) detected in our sample of OCD cases is slightly higher than rates previously observed in ASD (~4–5%)^41,52 and TD (3.4%).²³ Conversely, the rate of rare de novo CNVs (0.5%) in our control sample set is lower than rates previously reported in whole-exome sequencing of larger control samples (~1.5%).^23,53 Difference from previously reported values in the literature likely reflect a higher margin of error given the smaller sample size in this study, along with differences in variant calling and filtering methodology. Future studies in larger samples would more accurately estimate the absolute rate of de novo CNVs in OCD probands.

Our results suggest that the enrichment of de novo CNVs observed in OCD cases versus controls is driven by an increased burden of de novo deletions. A similar finding of a robust increased burden of deletions and nonsignificant enrichment of duplications previously has been demonstrated in WES data for ASD cases versus controls,³² and an enrichment has been shown for large, rare deletions in OCD.²⁵ Furthermore, we found that nine of the 12 de novo CNVs identified in OCD cases were predicted to be pathogenic or likely pathogenic, while all four de novo CNVs in the controls were predicted to be benign. This suggests that studies of de novo CNVs in larger OCD cohorts may provide new insights into genetic risk factors and the underlying biology of OCD in a similar manner previously successful in other neuropsychiatric conditions.

Identifying genes disrupted by de novo CNVs in OCD could point to genetic risk factors and biological mechanisms underlying the disorder. Exploratory gene-ontology analyses of genes disrupted by de novo CNVs demonstrate that the most highly associated terms are related to cell cycle, SUMOylation (post-translational protein modification by addition of a Small Ubiquitin-related MOdifier protein),⁵⁴ and nucleoside transport. The association with the cell cycle terms is consistent with findings that many genes related to neurodevelopmental disorders play a role in neural stem cell proliferation and differentiation and that particular genes are associated both with neurodevelopmental disorders and with cancers.⁵⁵ Another nominally significant ontology term from our gene set, axon guidance, was among the most significantly enriched terms in a prior OCD CNV microarray study.²⁸ Expression analysis of genes harboring OCD de novo CNVs shows some degree of convergence in the early fetal cortex and striatum, though not statistically significant. Two genes containing OCD de novo CNVs have been implicated in TD, and one gene has been implicated in ASD. These CNVs were found in individuals without these co-occurring disorders.

While previous CNV studies in OCD have examined microarray data,^26–28 this study demonstrates that CNV calling approaches can also be successfully applied to WES, as has been done for other neuropsychiatric disorders.^23,32 Unlike microarrays, WES covers only the coding regions of the genome and cannot be used to detect portions of CNVs in noncoding regions. However, we expect CNVs involving gene coding regions to directly impact gene dosage and, therefore, may have greater clinical significance. In addition, the interpretation of most noncoding variation remains challenging.

Several of the OCD cases with de novo CNVs have other co-occurring psychiatric conditions, notably TD, as do many individuals with OCD. The proband from family OCD8134 was previously found to have a de novo, germline, predicted-damaging missense SNV in the CHD8 gene.¹⁴ CHD8 is a known high-confidence risk gene for ASD, though there is no history of ASD or intellectual disability in the patient or any close family members.¹⁴ These cases highlight the challenges in teasing apart the contribution of genetic variants to OCD and to co-occurring conditions. Given recent evidence that OCD and TD have overlapping genetic etiologies, future risk gene analyses in OCD should examine the overlap with genes implicated in TD.^14,23 Future efforts to recruit patients with OCD and without co-occurring disorders may help isolate potential OCD-specific genetic risk. Further, deep phenotyping of enrolled patients would allow for the interrogation of genetic factors that could contribute to the clinical heterogeneity of OCD.

This study has several potential limitations that should be considered. First, because our trios comprise parents without OCD, our sample is not optimized to detect pathogenic inherited variants that could contribute to OCD risk. However, our inclusion strategy attempted to enrich for de novo variants, and we expect that de novo and inherited variants contributing to a disease phenotype converge on common biological networks and processes.⁵⁶ In this way, studying both simplex and multiplex families has synergistic potential to enhance our understanding of the biology of OCD. Furthermore, the focus on de novo variants in simplex families with an affected child and unaffected parents affords greater statistical power to implicate risk genes in a disorder. Second, the case and control samples were sequenced at separate times. We attempted to correct for potential batch effects in our burden analyses (see Methods and Supplement 1, available online), and we are reassured by similar rates of all rare CNVs in cases and controls using these methods. Third, our estimated CNV rates and enrichment effect sizes have wide confidence intervals due to our limited sample size. Initial studies in several neuropsychiatric disorders have used comparably sized parent-child trio samples to compare rates of rare de novo CNVs in autism (n=195),⁵⁷ schizophrenia (n=177),⁵⁸ bipolar disorder (n=185),⁵⁸ Tourette’s disorder (n=148),⁵⁹ and attention-deficit/hyperactivity disorder (n=305).⁶⁰ These early studies provided an impetus for subsequent larger studies, which have generated more precise CNV rates. Similarly, our initial finding of enrichment of rare de novo CNVs in OCD provides the rationale to pursue future larger studies leveraging this class of genetic variation to identify risk genes and to estimate their contribution to OCD more precisely. Fourth, there is a lack of ancestral diversity in this study. Because de novo mutations occur spontaneously, we do not expect de novo CNVs to cluster in different genomic regions based on ancestry. Moreover, the rates of de novo variants are similar across ancestries.⁶¹ Still, inherited variants may differ in families of different ethnicities, and genomic research and studies of OCD historically have had an underrepresentation of minority participants.^62–64 Future studies of rare variation in OCD would benefit from the inclusion of more diverse samples. Fifth, our sample set has an underrepresentation of female individuals, which precludes any informative sex-stratified CNV rate estimates or comparisons. Our post hoc exploratory analysis of cases versus controls for male and female individuals suggests the possibility that the overall increased rare de novo CNV rate in OCD may be driven by the male subset, but this will require further investigation with sufficient numbers of both sexes (Supplement 1, available online). Finally, our OCD group contained some probands who were adults at the time of sample collection. While we required an OCD diagnosis or symptom onset before age 18 for inclusion, this may be subject to recall bias.

This study did not find any gene-level recurrence of de novo CNVs or overlap with prior microarray studies of CNVs in OCD, highlighting the need for future studies with larger sample sizes. As methods continue to evolve, future studies may also benefit from comparing the results of various CNV calling tools for whole-exome and whole-genome sequencing data in OCD. As we continue to sequence more OCD trios, we will be able to leverage information about rare CNVs to detect additional OCD risk genes. In ASD, it has been demonstrated that statistical models can incorporate small de novo deletions along with SNVs and indels for enhanced risk gene prediction.⁴¹ A similar approach for combining different types of DNA variation to implicate risk genes in OCD may provide jumping-off points to guide later studies examining molecular mechanisms of the disorder. Ultimately, these mechanistic studies will point to potential new therapeutic targets and will allow for the development and testing of crucial new treatment options for individuals with OCD.

While our study represents one piece of evidence examining CNVs in OCD, future studies would benefit from systematic integration of CNV data across multiple sources, including published OCD studies and large biobanks. Such integrative analyses could provide increased power to detect CNV-phenotype associations and better characterize the contribution of different types of CNVs to OCD risk. However, these analyses will need to carefully address methodological challenges, including harmonization of CNV calls across different platforms (microarray vs. WES), standardization of phenotype definitions, and careful consideration of technical batch effects. Our CNV calls are available to facilitate such future integrative analyses.

Data Sharing:

All sharable data has been provided in the manuscript and supplement.