The Genetic Architecture of Obsessive-Compulsive Disorder: Contribution of Liability to OCD From Alleles Across the Frequency Spectrum

Obsessive-compulsive disorder (OCD) is a serious and often long-lasting psychiatric disorder characterized by intrusive and unwanted thoughts, images, or urges (obsessions) that are typically linked to ritualized acts (compulsions) (1–4). OCD affects 1%–3% of the population, and multiple studies provide reliable evidence for a significant genetic contribution to risk (1, 3–6) as well as a role for environmental factors in risk (7, 8). The heritability of OCD, historically estimated by analysis of twin and family studies and within the context of the ACE model (A, additive genetic, also known as narrow-sense heritability; C, shared environment; and E, nonshared environment), is reported to be in the range of 35%–50% (1, 4, 8–14).

As an alternative to the analysis of recurrence risk for OCD within pedigrees, heritability can also be estimated from individuals drawn from a population who have no obvious familial relationships, as long as they have been characterized for genetic variation across their genomes. Usually, this genetic characterization employs genotypes of single-nucleotide polymorphisms (SNPs) for which alleles are common in the population. In this approach, which we will call SNP based, the central idea is that the multiplicity of SNP genotypes allows estimation of familial relationships, albeit distant ones, among subjects as well as the covariance of their phenotypes, and these are the key elements for estimating heritability. When the heritability of OCD is computed in this manner, estimates range from 25% to 43% (5, 14–16).

It is useful to compare the heritability results from family-based and SNP-based approaches. Family-based studies, being more direct, typically yield estimates of heritability with lower standard errors, whereas the inaccuracy of estimating distant relationships from genetic data tends to produce fuzzier estimates. Family-based estimates also tend to yield higher estimates of heritability because the familial covariance reflects both rare and common genetic variation, whereas SNP-based estimates mostly arise from covariance due to common genetic variants. Looking at the results summarized above, one might conclude that this is also operating for OCD, that is, that family-based studies are producing higher heritability estimates than SNP-based studies.

However, in an influential study of data from the International Obsessive-Compulsive Disorder Foundation Genetics Collaborative (IOCDF-GC) by Davis et al. (5) (1,061 case subjects, 4,236 control subjects, 373,846 SNPs), there was no evidence for heritability from SNPs with minor allele frequency (MAF) <0.05, and over 60% of total heritability mapped to the most common variants (MAF >0.3). In addition, in a meta-analysis of data from the International Obsessive Compulsive Disorder Foundation Genetics Collaborative (IOCDF-GC) and OCD Collaborative Genetics Association Study (OCGAS) (16), ∼60% of heritability was accounted for by SNPs with MAF >0.4 in both the OCGAS sample alone and in the combined sample. If this observation were true, it could have profound implications for which evolutionary forces shaped this unusual mapping of risk alleles to their population frequency distribution. For example, balancing selection, where multiple alleles are maintained in the gene pool of a population at frequencies larger than expected from genetic drift alone, may play a role in OCD.

At the same time, other studies have implicated rare variants in risk for OCD (17–20). Thus, the contribution of inherited genetic variation across the allelic frequency spectrum to the risk of OCD remains uncertain and worthy of further study, as it affects both our understanding of processes underlying OCD risk architecture and rational study design. Here, using a substantially larger sample compared with previous studies and new genetic data from the Swedish population, we estimate SNP-based heritability for OCD.

Methods

Study Population

Ethical approval was obtained from the Institutional Review Board at the Icahn School of Medicine at Mount Sinai, New York, and the Regional Ethical Review Board in Stockholm. We used Swedish OCD cases collected through two studies: the EGOS cohort (Epidemiology and Genetics of Obsessive-Compulsive Disorder and Chronic Tic Disorders in Sweden) (21) and the NORDiC cohort (Nordic OCD and Related Disorders Consortium) (22).

In the EGOS cohort, individuals who were born between 1954 and 1998 and had at least two clinical diagnoses of OCD in the Swedish National Patient Register were eligible for inclusion (21). At the Swedish site of the NORDiC cohort, individuals with OCD were recruited from specialty OCD and related disorder clinics across Sweden (22). Genotype data on the Global Screening Array (Illumina) were collected for 1,108 individuals from the EGOS cohort and 1,107 individuals from the NORDiC cohort (Table 1).

A sample of 4,738 control subjects from the LifeGene cohort was available for this study. LifeGene is a prospective population-based cohort of around 50,000 individuals in Sweden (23). The samples were available in four batches: LifeGene-EGOS (N=1,444), LifeGene-NORDiC (N=500), LifeGene-ANGI-Wave-1 (N=1,500), and LifeGene-ANGI-Wave-2 (N=1,500) (ANGI is the Anorexia Nervosa Genetics Initiative). LifeGene-ANGI control subjects were previously used in a study of anorexia nervosa (24); they were mostly females (2,935 females and 65 males), and all individuals with a diagnosis of anorexia nervosa were previously removed from this batch. All control subjects were genotyped using the Global Screening Array.

Quality Control

All OCD case subjects, LifeGene-EGOS control subjects, and LifeGene-NORDiC control subjects were genotyped in the same laboratory but in different batches. GenomeStudio’s genotyping module was used to recall genotypes on the joint data.

Quality control was first carried out on three batches of samples that may differ in key variables: 1) all case subjects, LifeGene-EGOS control subjects, and LifeGene-NORDiC control subjects, 2) LifeGene-ANGI-Wave-1 control subjects, and 3) LifeGene-ANGI-Wave-2 control subjects. We employed the following quality control steps using PLINK, version 2.0 (see Tables S1–S3 in the online supplement): individuals were removed who had a genotype non-call rate >0.05, were discrepant for nominal versus genetically determined sex, or had either low or high heterozygosity (>±3 SD from the mean); a SNP was removed if its non-call rate for genotypes was >0.05, if its MAF was <0.01, or if it had a Hardy-Weinberg equilibrium p value <0.00125. GEMTools was used to choose individuals with European ancestry where indicated (see Figure S1 in the online supplement).

We next used the tool developed by McCarthy (https://www.well.ox.ac.uk/∼wrayner/tools/) to match the SNPs to 1000 Genomes data, and Genotype Harmonizer (automatic strand alignment software) to align the different cohorts (25). After quality control, we merged the cohorts based on the set of all intersecting SNPs and performed additional quality control procedures as noted in the online supplement. The final data set included 2,090 case subjects and 4,567 control subjects, with 412,813 SNPs (Table 2).

Statistical Analysis

We used the Genome-Wide Complex Trait Analysis (GCTA) program, version 1.26.0, to estimate the genetic relationship matrix between all pairs of individuals from SNPs (26). Next, we used PLINK, version 2.0, to extract the top principal components from the variance-standardized relationship matrix (for more details, see the online supplement). We performed restricted maximum likelihood analysis, implemented in GCTA, to estimate the heritability of OCD attributable to SNP genotypes. Because the OCD diagnosis is dichotomous, we scaled the phenotypic variance to an underlying liability scale using the population prevalence of 1%, similar to our most recent estimate of population prevalence in Sweden using data from the Swedish national registers (1) (for more details, see the online supplement, where we also provide results for 2% prevalence).

To evaluate the sensitivity of estimates of SNP-based heritability to modeling approaches, we assessed the data in multiple ways. The first assessment included all affected and unaffected individuals born in Sweden, of whom most but not all were of Swedish/European genetic ancestry, and used all 405,105 high-quality genotyped SNPs for analysis. The sampling in the first assessment is consistent with our previously published family-based analyses and constitutes our primary analytical approach here. The second assessment pruned SNPs according to linkage disequilibrium to obtain a smaller set of 184,296 largely independent SNPs. The third assessment limited the sample to individuals of European genetic ancestry. The fourth assessment removed all individuals for whom there was also a fifth-degree or greater relative in the sample. The fifth assessment analyzed only pairs of affected and unaffected individuals, matched on two dimensions of genetic ancestry using the function pairmatch in the package optmatch in R (“1-to-1 fullmatch”) (see the online supplement). The sixth assessment was conducted as was done in assessment 5, using only individuals of European ancestry. Pair matching, as done in assessments 5 and 6, is a common epidemiological approach for controlling confounding (here, differences in ancestry in case subjects versus control subjects) and has been shown to be useful for genetic studies (27–29). Note that assessments 3–6 use all high-quality SNPs.

We also estimated heritability partitioned by chromosomes and MAF bins and compared the results with those from the study of the IOCDF-GC sample by Davis et al. (5). We created six MAF bins: 0.01–0.05, 0.05–0.1, 0.1–0.2, 0.2–0.3, 0.3–0.4, and 0.4–0.5. For each bin, we computed a genetic relationship matrix, and then the additive genetic variance attributed to each subset was jointly estimated with multiple genetic relationship matrices (using −mgrm in GCTA). This allows for the effects of linkage disequilibrium to be partitioned by the restricted maximum likelihood analysis.

Results

Our study population included 2,090 case subjects and 4,567 control subjects after quality control was completed. Sixty percent of the case subjects were female, and 76% of the control subjects were female (Table 1). Based on principal component analysis (see Figure S2 in the online supplement), we used the first six principal components as covariates to adjust for variation in ancestry in all heritability analyses. As a check for compatibility of cohorts, we first estimated heritability by treating EGOS and NORDiC control subjects as cases and LifeGene-ANGI control subjects as controls. Heritability was estimated at 0.0001% (SE=5%). These results show that the control cohorts were homogeneous. Next, we estimated heritability of OCD for the full sample, contrasting OCD cases to controls and yielding an estimate of 29% (SE=4%), for a population prevalence of 1%.

Technically, heritability is first estimated on the observed scale, namely, dichotomous OCD diagnosis; however, heritability on the continuous liability scale is more interpretable and so is usually reported. Heritability can be transformed from the observed to the continuous liability scales because they are functions of prevalence (30). To determine how sensitive our heritability estimate was to prevalence, we varied it between 0.5% and 3% and found heritability to vary between 25% and 38% (see Table S4 in the online supplement).

We next performed a set of sensitivity analyses by different treatments of the data, as described in the Methods section, and found the estimates to be quite robust (Table 3). Notably, although analyses suggested that EGOS and NORDiC case subjects had slightly different ancestry distributions, the results in Table 3 show that our adjustments for ancestry were sufficient to compensate for these differences (see Figures S4–S9 and Table S5 in the online supplement). In addition, we did not observe a significant difference in heritability between the EGOS and NORDiC case subjects (see Table S5).

Heritability Analysis Partitioned by MAF Bin

Having established that a substantial portion of OCD traces to common variation, we next addressed an important issue about its nature. In the study of the IOCDF-GC sample (5), it was reported that alleles with MAF <0.05 did not contribute meaningfully to the heritability of OCD (0.0001% of total heritability). To compare our results with those of Davis et al. (5), we estimated the portion of total heritability for groups of autosomal SNPs with distinct allele frequencies, grouping the SNPs into six bins based on their MAF (Figure 1; see also Table S9 in the online supplement): 0.01–0.05, 0.05–0.1, 0.1–0.2, 0.2–0.3, 0.3–0.4, and 0.4–0.5. For all the bins, we included the first six principal components as covariates and set population prevalence to 0.01. Estimates of the portion of total heritability for the bins were distributed differently between these two studies (Figure 1; see also Table S9). Curiously, although the total heritability of the first two bins (MAF <0.1) was similar across studies—that is, 2.6% in our study, compared with 4% in the study of the IOCDF-GC sample (5)—estimates for specific bins were not similar. In the Davis et al. study (5), the 0.01–0.05 MAF bin accounted for essentially no heritability (0.0001%), whereas our estimate was much larger (2.6%) (Figure 1; see also Table S9). A portion of the difference could be due to the number and nature of the SNPs falling in this bin: approximately 10 times more genotyped SNPs fell into this bin in the present study compared with the study of the IOCDF-GC sample (5) (see Table S9); however, Davis et al. also imputed genotypes for over 2 million SNPs for this bin, and those genotypes did not alter their heritability estimate in that bin (see their Table 2, in reference 5).

To investigate these differences, we estimated what the expected portion of total heritability in these bins should be. First, we observed that the percentages of the total SNPs in each bin were distributed differently in comparison to 1000 Genomes data (for SNPs with MAF >0.01) (see Tables S6 and S7 in the online supplement), which we would expect is more representative of variation in the general population. For example, 45.2% of the SNPs in our study had MAF between 0.01 and 0.05, compared with 29.5% of SNPs in 1000 Genomes data. Under the standard quantitative genetic “infinitesimal model” (also referred to as the “polygenic model”) (31), it is reasonable to assume that the effect of all risk SNPs is equal. With this assumption, we then explored various models to predict the expected heritability in each MAF bin (Figure 1; see also Tables S6–S8 and Figures S10 and S11 in the online supplement).

The model that best fit the data was one in which risk alleles were sampled in proportion to their occurrence in 1000 Genomes data, with a goodness-of-fit-adjusted R² of 0.49. Notably, and in contrast to what was observed in the study of the IOCDF-GC sample (5), the largest proportion of expected heritability was not explained by SNPs in the higher-frequency allele bins (0.3–0.4 and 0.4–0.5). In addition, we observed that SNPs with low MAF (0.01–0.05) are expected to account for 10.4% of the heritability under this model, similar to the 10% that we observed and in contrast to the IOCDF-GC sample, where low-MAF SNPs accounted for almost no heritability. These discrepancies and the smaller ones observed in our study track with sample size. For example, the size of the IOCDF-GC sample (1,061 case subjects and 4,236 control subjects) was smaller than in our study, and variance of estimates is a direct function of sample size. Combining results from both studies demonstrated strong concordance with expectation (Figure 1C). In addition, previous studies (5, 16) are likely more ancestrally heterogeneous than our Swedish sample, which can lead to increased variance.

Heritability Analysis Partitioned by Chromosomes

Under the infinitesimal model, SNPs affecting heritability of OCD (or any trait) should be scattered randomly across chromosomes, so that heritability per chromosome should track with chromosome length. This is observed in our study (Figure 2), and there is a significant correlation between heritability per chromosome and length (r=0.55, p=0.008). Chromosome 13 had the lowest heritability, significantly lower than what would be expected under the uniform distribution model.

As noted above, the noisy nature of these results can likely be attributed to the relatively small sample size for this type of analysis. We conjectured that if this were the case, and assuming that both study samples were homogeneous, combining the heritability estimates from the IOCDF-GC sample (5) and our heritability estimates, per chromosome, would produce a somewhat better fit between heritability per chromosome and length. This result is confirmed in Figure 2C,D; the fit of the regression for this weighted average heritability (weights proportional to the inverse of variance), an adjusted R² of 0.31, is better than the fit for our sample alone, with an adjusted R² of 0.27. Furthermore, note that chromosome 6, which had very low heritability in the Davis et al. study, shows reasonable heritability in both our analyses and in the combined data, again suggesting that small sample sizes are driving some of the results.

Discussion

Common genetic variation—variants shared among many individuals in a population, which are most frequently SNPs—has been found to play a role in liability for most psychiatric disorders, including OCD. Open questions remain about the impact on risk due to common variation, including how much of the heritability of OCD common variation accounts for and how it is partitioned across the frequency spectrum of alleles. These are important questions for a variety of reasons. For example, both schizophrenia and autism spectrum disorder demonstrate high heritability (32, 33), and much of the heritability traces to common genetic variation. Yet, rare variation with a damaging impact on gene function, especially de novo variation, plays a larger role in overall autism spectrum disorder risk than in overall schizophrenia risk (32, 34, 35); for example, Singh et al. (35) found, when they evaluated evolutionarily constrained genes, that de novo protein-truncating variants were four times more common among individuals with autism than among individuals with schizophrenia. This difference is critical for clinical genetics, genetic counseling, and possibly treatment. It also could be relevant for disentangling evolutionary processes underlying different psychiatric disorders, consistent with stronger natural selection on autism than schizophrenia. Finally, it would have an impact on study design (if, for example, rare variants contribute little to OCD heritability).

Here, we evaluate whether a substantial portion of the heritability of OCD traces to common variation, as it does for autism and schizophrenia, and we characterize its frequency spectrum, which is directly relevant to evolutionary processes. For example, in an early study estimating the heritability of OCD from common variation, results from the IOCDF-GC sample (5) suggested that alleles with the highest frequencies—that is, those with MAF >0.3—account for the bulk of SNP-based heritability of OCD. Similar findings were reported using meta-analysis of data from IOCDF-GC/OCGAS (16). Such a strong pattern would suggest that OCD was under strong balancing selection.

By sampling individuals with OCD from the Swedish population, as well as a larger sample of unaffected (control) individuals, we were able to address these questions. Our analyses of over 2,000 individuals diagnosed with OCD and twice as many unaffected individuals, each genotyped across their genome via >400,000 SNPs, yielded an OCD heritability estimate of 29% (SE=4%), a robust estimate (Table 3).

Moreover, when we assumed that SNPs contributed equally to risk for OCD, regardless of MAF, we obtained good fit between estimated OCD heritabilities from MAF bins of our sample and what was expected based on the distribution of MAF in 1000 Genomes data (Figure 1). SNPs affecting risk appear to be distributed at random over chromosomes, because size was a good predictor of a chromosome’s contribution to total heritability (Figure 2). Chromosome 13 showed the poorest fit to this model, which may be partially explained by its having one of the lowest gene densities (6.5 genes per Mb) among human chromosomes. All of these results fit expectations of the infinitesimal quantitative genetics model.

In terms of estimated heritability from common variation, our results compare favorably with previous studies of OCD. Published estimates of SNP-based heritability, based on different samples from different populations, range from 25% to 43% (5, 15, 16). Thus, all studies have converged on a substantial contribution of common variation to the heritability of OCD, showing notable consistency. There are some differences, however. Notably, recent studies (5, 16) suggest that SNPs with substantial frequency (MAF >0.05) contribute disproportionately to this heritability, and the contribution tends to increase with increasing MAF.

In light of our findings, we found their results intriguing: An increasing heritability associated with MAF is appealing because the contribution to heritability of any SNP of frequency p is 2p(1−p)a², where the SNP’s effect a can be assumed to be roughly equal over all SNPs under the infinitesimal model; on the other hand, it seems unlikely that low-MAF SNPs have no contribution to heritability because there are so many of them in the human genome (see Table S9 in the online supplement). Our results argue that these low-MAF SNPs do contribute to OCD heritability, that their contribution is roughly in proportion to the frequency spectrum of alleles, and that they can be assumed to be of similar effect (i.e., a) across the frequency spectrum. Thus, our results show that future studies of less common and even rare alleles will also be informative on OCD etiology, with the caveat that effects of risk alleles of very low frequency can be difficult to detect by case-control methods.

Another interesting contrast is the evidence for heritability across chromosomes. Davis et al. (5) observed essentially no heritability for OCD on chromosome 6, which encodes both the HLA and histone gene clusters, and extremely high heritability on chromosome 15. In discussing these results, the authors suggest that chromosome 15 has an outsized contribution to OCD risk and that the HLA locus is effectively excluded from OCD risk. Given the contrasting results in our study, and in our analyses combining results from both studies, we again conclude that the data are consistent with the infinitesimal model and that smaller sample sizes may account for results that diverge from expectation.

The Davis et al. study (5) involved about 50% fewer OCD cases, and the variance in any estimate is a direct function of sample size. It is also possible that the IOCDF-GC sample had a different distribution of distantly related individuals than our relatively homogeneous sample from Sweden. The accuracy of SNP-based heritability diminishes as the fraction of very distantly related pairs, relative to all relative pairs, increases. Consistent with estimates from both studies being noisy, when we combined our results with results from the IOCDF-GC sample (5) to obtain new estimates of average heritability per allele bin and heritability per chromosome, the average fit expectation was better than in either study alone.

Prior to the advent of dense genotyping, the heritability of a trait was typically estimated from its distribution within pedigrees. These kinds of studies continue to this day, in large part because they capture heritability due to both common and rare inherited genetic variation. It is thus interesting to compare our SNP-based heritability estimate from common variation, 29%, to that from Swedish families, 35%–50% (1, 4). This comparison suggests that while the majority of inherited liability for OCD traces to common genetic variation, rare variation contributes to OCD liability as well, but to a lesser degree, consistent with the findings to date regarding rare variation and risk for OCD (17–20).

The present study had strengths and limitations. We used OCD cases from the EGOS and NORDiC cohorts, the two largest OCD studies in Sweden, to examine the role of genetic and environmental factors. The EGOS cohort utilized the Swedish National Patient Register for its sampling frame, and thus it is an epidemiological cohort, minimizing selection biases, while NORDiC recruited through specialty OCD clinics across Sweden, a sampling frame more typical of case-control studies. This difference in sampling frames could introduce heterogeneity into our study. Nonetheless, when we evaluated this possibility by estimating the heritability induced by contrasting OCD case subjects from EGOS to OCD cases from NORDiC, and doing the same for control subjects, both estimated heritabilities were not significantly different from zero. Hence, while there could be subtle heterogeneity between the cohorts, it must be small. Furthermore, for both cohorts, reliance on inclusion as a result of individuals seeking care at mental health hospitals and clinics can inadvertently exclude those with milder forms of the disorder who may seek treatment from primary care providers and those who do not present to clinical services at all. If such individuals were included, and if their genetic architecture were different from that of our current OCD case sample, it would affect the estimated heritability. By restricting case subjects to individuals in Sweden, we had a genetically homogeneous sample, which minimized the risk of confounding due to population stratification and facilitated the combining of the cohorts. Nonetheless, it does limit the generalizability of our results. However, after combining our results with those of the IOCDF-GC sample (5), we observe results that fit expectation, thus suggesting that the results are likely to be relevant for most populations.

Our results provide new insights into the genomics architecture of OCD, affecting research design for genomic discovery and the ultimate clinical impact of such studies. While there is no doubt that rare and common genetic variation contributes to risk for OCD, the balance of their contributions has remained uncertain. The results from the IOCDF-GC sample and the meta-analysis of IOCDF-GC and OCGAS samples (5, 16) implied an unexpectedly large role for very common variation in OCD risk and no evidence for heritability related to rarer variation (MAF <0.05). This would be quite distinct from what is known about other psychiatric disorders, and it would be consistent with some form of evolutionary selection, such as balancing selection. Our results differ substantially from those of the earlier studies; specifically, we observed that the contribution to risk from common SNP variation follows expectations. Hence, our results do not support a role for unusual evolutionary forces playing a role in OCD risk and do support a role for rare variation in risk.

Assessing the contribution of rare variants in OCD has the additional benefit of uncovering variation of major effect, which can lead to direct insight into OCD biology and potentially pave the way for family counseling. In addition, these high-effect genes represent tools to create animal models of OCD to study pathobiology and also may represent targets for developing novel therapeutics.

As data sets become larger, risk prediction will improve, as will our ability to characterize the distribution and effects of common and rare risk variation. We conjecture that the liability arising from common and rare risk variation likely combines additively to determine risk for individuals diagnosed with OCD, similar to the risk patterns for ASD (36, 37). This knowledge can be translated into a deeper etiological understanding of OCD subtypes and their treatment and, in the future, into better predictors of OCD risk. OCD is a clinically and etiologically heterogeneous condition (38) with a complex symptom structure (39). Studies suggest that the burden of common risk alleles of OCD may differ by OCD symptom type. For example, although not yet replicated, in one study (40), compulsive symptoms rather than obsessive symptoms showed higher SNP heritability and genetic correlations with OCD. However, it is still unclear to what extent rare genetic variation, and the joint effect with common variation, may 1) differ between the subtypes of OCD, 2) interact with sex, 3) influence age at onset, and 4) affect the risk of comorbid conditions. Further research in this area could accelerate discovery of biomarkers and novel treatments, eventually helping clinicians offer patients optimal prognosis and treatment.

Pharmacogenetic studies of OCD have focused on the role of common genetic variants in treatment response (41). However, to date, no replicated significant GWAS variant has been reported for OCD—likely because of the small sample sizes—and therefore it has been challenging to contextualize the results of OCD pharmacogenetic studies. Future pharmacogenetic studies examining predictors of treatment response will likely shift focus from selected common genetic variants to genome-wide studies that also estimate how rare and common risk variants jointly affect liability (42).

The heterogeneity of OCD should always be considered in the light of psychiatric comorbidity, an approach that is facilitated in samples such as EGOS and NORDiC that are linked to national health registries. For example, in EGOS, using an epidemiological frame, approximately 40% of individuals with OCD have more than one psychiatric comorbidity, with anxiety disorders and major depressive disorder being most common (43). In addition, the severity of OCD was significantly higher in individuals with at least one additional psychiatric comorbidity compared with individuals with no psychiatric comorbidity: higher symptoms of obsessing and ordering, measured using the Obsessive Compulsive Inventory–Revised, were observed in individuals with OCD and at least one additional psychiatric comorbidity. In future studies, it will be important to investigate how the combinations of rare and common genetic variants differ in their relationship with the comorbid conditions.

In summary, our results demonstrate that the majority of inherited liability for OCD traces to common genetic variation, and the distribution of risk as a function of allele frequency is consistent with expectations of the infinitesimal quantitative genetics model. These results indicate that balancing selection or other, more complex evolutionary forces are not strongly at play in OCD. Furthermore, since the infinitesimal model assumes that genetic loci across the entire frequency spectrum influence the risk of OCD, future studies that examine the contribution of rare variation, both inherited and de novo, together with common variation, will allow a more comprehensive understanding of OCD genetic risk.