Cross-ancestry and cross-disorder transferability of polygenic risk scores for obsessive-compulsive disorder

Abstract

Most genome-wide association studies (GWASs) of obsessive-compulsive disorder (OCD) have been conducted in European populations, limiting the understanding of OCD genetics across populations. Here, we investigated the cross-ancestry and cross-disorder transferability of European-derived polygenic risk scores (PRSs) for OCD in the first East Asian OCD cohort. We performed a GWAS in 532 Korean OCD cases and 4376 controls, followed by PRS analyses using European GWASs for OCD and 10 psychiatric disorders. The European-derived OCD PRS was significantly associated with Korean OCD status (best-fit R² = 1.17%). In multivariable models, PRSs for OCD (odds ratio [OR] = 1.28, 95% confidence interval [CI] = 1.13–1.45), schizophrenia (SCZ; OR = 1.17, 95% CI = 1.03–1.32), and anorexia nervosa (AN; OR = 1.17, 95% CI = 1.03–1.32) were independently associated with OCD. A stepwise multi-PRS model including PRSs for OCD, AN, SCZ, post-traumatic stress disorder, anxiety disorder, attention deficit hyperactivity disorder, and bipolar disorder (BD) yielded the highest predictive performance (incremental R² = 2.99%). Where available, we evaluated PRSs derived from East Asian, European, and combined populations. For BD and SCZ, PRSs derived from cross-ancestry meta-analysis GWASs showed the greatest predictive performance for OCD, outperforming single-ancestry PRSs (incremental R² = 2.84 and 1.28%, respectively). Our findings demonstrate cross-ancestry transferability of OCD polygenic risk and highlight substantial genetic overlap between OCD and other psychiatric disorders, particularly SCZ and AN. Together, they underscore the importance of including diverse ancestral populations in psychiatric genetics and suggest that transdiagnostic polygenic frameworks may provide deeper insight into the biological basis of OCD.

Introduction

Obsessive–compulsive disorder (OCD) is a severe and disabling clinical condition, with a worldwide lifetime prevalence estimated at 1.3% [1]. Based on substantial evidence that OCD is heritable [2, 3], several genetic studies, including genome-wide association studies (GWASs) have been conducted to elucidate the genetic predisposition of OCD [4, 5]. Recent OCD GWAS meta-analyses, including 2,098,077 European individuals (53,660 cases and 2,044,417 controls), have substantially advanced our understanding of its genetic architecture, identifying 30 independent genome-wide significant loci [6]. However, almost all studies on OCD GWAS have been performed in European populations, potentially limiting the generalizability of the findings to other ancestral groups.

Several methodological improvements for understanding the genetic diversities based on ancestry of disease have emerged [7, 8]. However, the scarcity of non-European ancestry data remains a major issue. This gap has led to growing interest in GWAS investigating the genetic correlations of psychiatric disorders across ancestries [9–11]. A few previous studies have reported cross-ancestry meta-analyses involving East Asian and European populations, particularly for schizophrenia [12] and depression [13]. However, no OCD GWAS had been conducted in East Asian populations, and the genetic relationship between OCD in East Asian and European populations remains poorly understood.

Polygenic risk scores (PRSs) have emerged as a powerful tool to assess genetic susceptibility to diseases [14]. In cross-ancestry settings, they can also be used to examine the transferability of polygenic signals across populations [15]. Despite these advances, PRS applications across diverse ancestries remain limited, particularly for psychiatric disorders such as OCD. Investigating whether PRSs derived from European data are transferable to non-European populations, such as those of East Asian descent, could provide insights into both shared and distinct genetic factors influencing OCD across ancestries. Furthermore, a cross-disorder PRS approach [16] may help identify genetic overlaps between OCD and other psychiatric disorders.

This study presents the first OCD GWAS conducted in a Korean population and evaluates whether the genetic architecture of OCD demonstrates cross-ancestry and cross-disorder transferability of findings from previous European GWASs. We estimated genetic correlation across ancestral groups and performed PRS analyses to examine whether European-derived PRSs for OCD and other major psychiatric disorders are associated with OCD status in the Korean population. In addition, for psychiatric disorders with publicly available East Asian GWAS data, we compared the predictive performance of ancestry-specific and cross-ancestry PRSs.

Materials and methods

Study population

Yonsei cohort

A total of 540 patients with OCD and 220 healthy controls were enrolled. Patients with OCD were recruited from a specialized outpatient OCD clinic at Severance Hospital, Yonsei University Health System, Seoul, Republic of Korea. All patients had been referred from primary care for the evaluation or treatment of OCD and were evaluated by expert psychiatrists using the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I) to confirm the presence of current or past psychiatric disorders. All 220 control participants also underwent SCID-I assessments by expert psychiatrists to confirm the absence of any current or past psychiatric disorders. Demographic and clinical information was collected for all participants. Ethnicity was determined based on both participants’ self-report and genetic principal component analysis, with all individuals clustering tightly together. Patients were excluded if they met any of the following criteria: a severe medical condition, a neurological disorder, or a diagnosis of a schizophrenia-spectrum disorder, bipolar I disorder, substance use disorder, or intellectual disability. Patients with OCD and other comorbid DSM-IV Axis I disorders were not excluded if obsessive-compulsive symptoms were the primary reason for treatment. All participants provided written informed consent in accordance with procedures approved by the Severance Hospital Institutional Review Board (IRB No. 4-2015-0655). Eight patients with OCD were excluded from the analyses because of missing age information (n = 7) or discordant sex information (n = 1).

KoGES Ansan-Ansung cohort

To increase statistical power by expanding the control group, we included additional control individuals from the Ansan-Ansung cohort of the Korean Genome and Epidemiology Study (KoGES) [17], a population-based cohort. Because direct information on OCD diagnosis was not available in the KoGES dataset, we excluded participants with potential indicators of psychiatric symptoms—such as self-reported insomnia or daily alcohol consumption of 40 g or more—as a proxy approach to reduce the likelihood of including individuals with undiagnosed psychiatric conditions. Although a direct variable indicating the presence or absence of OCD was not available in the KoGES dataset, the population-based nature of the cohort and the relatively low prevalence of OCD in the general population suggest that the vast majority of included individuals are unlikely to have OCD. This approach has been similarly applied and justified in prior psychiatric genetics studies utilizing large population-based cohorts for control group construction. As a result, 1337 individuals out of 5493 were excluded, and 4156 individuals from the KoGES Ansan-Ansung cohort were included. The acquisition and use of the KoGES dataset were approved by the Severance Hospital Institutional Review Board (IRB No. 4-2024-1156).

Genotyping and quality control

The Yonsei cohort samples and KoGES Ansan-Ansung control samples were genotyped using the Korean Biobank Array designed by the Korean National Institute of Health, Republic of Korea, and manufactured by Affymetrix (Thermo Fisher Scientific) [18]. Genotype data from both cohorts underwent stringent quality control (QC) procedures in accordance with their respective criteria prior to imputation. In the Yonsei cohort, low-quality single-nucleotide polymorphisms (SNPs) in any batch with Hardy-Weinberg Equilibrium (HWE) P-value < 1×10⁻⁴, or call rate < 95% were excluded. SNPs with a minor allele frequency (MAF) < 1% were also excluded. In the KoGES Ansan-Ansung cohort, both variant-level and sample-level QC procedures were applied. Samples were excluded if they had a call rate < 97%, excessive heterozygosity, first- or second-degree relatedness, or sex discordance. At the variant level, SNPs were excluded if they had an HWE P-value < 1×10⁻⁶ or call rate < 95% in any batch. After genotype QC, we merged the Yonsei and KoGES Ansan-Ansung cohorts, retaining only variants present in both datasets. Phasing was performed using Eagle v2.42, and genotype imputation was conducted using Minimac4 [19] with the Haplotype Reference Consortium (HRC) r1.1 2016 reference panel [20]. We then filtered variants with an imputation quality score (INFO) < 0.8 or a MAF < 0.5%. After all QC procedures, a total of 4908 samples and 5,585,816 variants were retained for our downstream analyses.

Association analysis

We performed a GWAS for OCD in the Korean population using the generalized linear model implemented in PLINK v2.0 [21]. The analysis was adjusted for sex, age, and the top 10 principal components (PCs) of genetic ancestry. For cross-ancestry GWAS meta-analysis for OCD combining Korean and European populations, we applied an inverse-variance-weighted fixed-effect model with genomic control correction implemented in METAL 2020-05-05 [22]. European OCD GWAS summary statistics were obtained from a publicly available dataset provided by Strom et al. [6]. Before conducting the cross-ancestry meta-analysis, allele frequencies in each GWAS dataset were compared with ancestry-matched reference panels from the 1000 Genomes (1KG) Project to identify potential strand inconsistencies. Variants showing large allele frequency differences between datasets ( > 20%) were excluded. After filtering, 5,585,314 variants remained in the Korean GWAS and 6,789,622 variants remained in the European GWAS, resulting in a union set of 8,042,322 variants. We additionally performed a sensitivity analysis restricted to the intersection variant set (n = 4,332,614). Genome-wide significance was defined as P-value < 5 × 10⁻⁸.

SNP-based heritability

We estimated the single-nucleotide polymorphism (SNP)-based heritability of OCD on the observed scale using linkage disequilibrium score regression (LDSC) v1.0.1 [23]. For the LDSC analysis, we used HapMap3 SNPs and pre-computed linkage disequilibrium (LD) scores from the East Asian 1KG Project phase 3 data. To account for case-control imbalance, we transformed heritability from the observed scale to the liability scale, assuming an OCD population prevalence of 2%. The significance of SNP-based heritability was evaluated by calculating a z-score as the heritability estimate divided by its standard error, and the corresponding two-sided P-value was derived assuming a standard normal distribution.

Cross-ancestry genetic correlation

We computed a cross-ancestry genetic correlation for OCD between Korean and European populations using Popcorn v1.0 [24]. We used the ‘genetic effect’ mode to estimate the genetic effect correlation for OCD between Korean and European populations. For this analysis, we used pre-computed LD score files derived from the East Asian and European 1KG Project phase 3 data, available from the Popcorn software GitHub repository (https://github.com/brielin/Popcorn).

Polygenic risk score analysis

To perform cross-ancestry and cross-disorder PRS analyses, we obtained European GWAS summary statistics for various psychiatric disorders (sample sizes ranging from n = 13,677 to n = 2,000,702) (Supplementary Table 1) [6, 25–38]. PRSs were generated using PRSice-2 v2.3.3 that employs the pruning and thresholding approach [39]. SNPs were clumped at an LD cutoff of r² = 0.1 with a 1000-kb clumping window, and PRSs were calculated across multiple P-value thresholds. Across the range of P-value thresholds, the PRS explaining the largest portion of the variance (R²) of OCD was considered as the best-fit model.

As a complementary analysis, we additionally generated PRSs using PRS-CS-auto v1.0.0, a Bayesian regression framework that automatically learns the global shrinkage parameter from the data using a fully Bayesian approach without requiring an external validation dataset [40]. For PRS-CS analyses, we used the European 1KG Project LD reference panel.

To evaluate the performance of PRS on predicting OCD, we performed logistic regression analyses. A baseline regression model was established with covariates including age, sex, and the top 10 PCs. Then, each PRS was added to the baseline model to construct a full model. By comparing the full model with the baseline model, we assessed the gain in the R² explained by including the PRS in the model. We calculated the incremental McFadden’s adjusted R²; if this value is positive, it indicates that adding the PRS explains more variance in OCD status. Associations across multiple P-value thresholds within each psychiatric disorder were interpreted at the nominal significance level (P-value < 0.05). To account for multiple testing across the psychiatric disorders evaluated, Bonferroni correction was applied based on the number of PRSs tested, resulting in an adjusted significance threshold of P-value < 0.0045 (0.05/11).

Results

Genome-wide association analysis

We conducted a GWAS for OCD in the Korean population (532 cases and 4376 controls), with sample characteristics summarized in Supplementary Table 2. After performing variant QC to ensure data quality, we included 5,585,816 autosomal SNPs in the analysis. Associations of SNPs with OCD were examined; however, no variants reached genome-wide significance (P-value < 5 × 10⁻⁸) (Supplementary Figs. 1 and 2). The strongest association was observed for rs58318383 on chromosome 17 (P-value = 4.44 × 10⁻⁷), which is located near the RPTOR gene.

We performed a cross-ancestry GWAS meta-analysis to increase discovery power by combining Korean and European populations. The analysis included 1,143,014 individuals and a union set of 8,042,322 variants. We identified one genome-wide significant locus on chromosome 8 within the intronic region of TSNARE1, indexed by rs13262595 (effect allele A; β = 0.064, s.e. = 0.0114, P-value = 1.85 × 10⁻⁸; Supplementary Figs. 3 and 4). This locus corresponds to the previously reported 8q24.3 OCD locus identified in the European-ancestry OCD GWAS by Strom et al. [6]. The association at this locus remained genome-wide significant in the sensitivity meta-analysis restricted to variants present in both the European and Korean GWASs.

Heritability and cross-ancestry genetic correlation

To assess the genetic contribution to OCD, we estimated SNP-based heritability. The heritability estimate on the liability scale was 0.18 (standard error [s.e.] = 0.15, P-value = 0.2514) for OCD in the Korean population. We estimated genetic correlation of OCD between Korean and European populations (genetic effect correlation = 1.34, s.e. = 2.87, P-value = 0.9049). However, neither estimate reached statistical significance, likely reflecting the limited statistical power of the Korean OCD GWAS.

Association between European-derived PRS for OCD (PRS-OCD) and OCD

We computed PRS-OCD from European OCD GWAS summary statistics across multiple P-value thresholds, ranging from 5 × 10⁻⁸ to 1, and then assessed its association with OCD status in the Korean population. We observed a significant association between European-derived PRS and OCD at several P-value thresholds (P-value = 1 × 10⁻⁵, 1 × 10⁻⁴, 0.001, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, and 1) (Fig. 1H). The best-fit PRS was observed at a P-value threshold of 1 (R² = 1.17%) in the Korean population.

Fig. 1. PRS prediction of OCD in the Korean cohort based on European GWASs of psychiatric disorders.

Each panel shows the proportion of variance in OCD status explained by PRS derived from European GWAS summary statistics for each psychiatric disorder. A–K show PRSs for ADHD, AN, ANX, ASD, BD, DEP, HS, OCD, PTSD, SCZ, and TS, respectively. The x-axis indicates the P-value thresholds used to select the variants from the European GWAS, and the y-axis indicates Nagelkerke’s pseudo-R² from logistic regression models. Bars indicate Nagelkerke’s pseudo-R² values, and their colors indicate statistical significance: gray bars represent non-significant associations, while shades of blue represent significant associations, with darker shades corresponding to smaller P-values (i.e., stronger statistical significance). P-values are displayed above each bar. All reported P-values are two-sided. Abbreviations: PRS polygenic risk score, GWAS genome-wide association study, ADHD attention deficit hyperactivity disorder, AN anorexia nervosa, ANX anxiety disorder, ASD autism spectrum disorder, BD bipolar disorder, DEP depression, HS hoarding symptoms, OCD obsessive-compulsive disorder, PTSD post-traumatic stress disorder, SCZ schizophrenia, TS Tourette syndrome.

Association between PRSs for other psychiatric disorders and OCD

To investigate associations between OCD and PRSs for other psychiatric disorders, we conducted cross-ancestry, cross-disorder PRS analyses of 10 psychiatric disorders: attention deficit hyperactivity disorder (ADHD) [25], anorexia nervosa (AN) [26], anxiety disorder (ANX) [27], autism spectrum disorder (ASD) [28], bipolar disorder (BD) [29, 36], depression (DEP) [30, 37], hoarding symptoms (HS) [34], post-traumatic stress disorder (PTSD) [31, 38], schizophrenia (SCZ) [32], and Tourette syndrome (TS) [33]. When examining the results for each psychiatric disorder individually, PRS for ADHD (PRS-ADHD; Fig. 1A), AN (PRS-AN; Fig. 1B), ANX (PRS-ANX; Fig. 1C), ASD (PRS-ASD; Fig. 1D), BD (PRS-BD; Fig. 1E), DEP (PRS-DEP; Fig. 1F), PTSD (PRS-PTSD; Fig. 1I), and SCZ (PRS-SCZ; Fig. 1J) showed nominally significant associations with OCD (P-value < 0.05) at multiple P-value thresholds in the PRSice-2 framework. However, PRS for HS (PRS-HS; Fig. 1G) and TS (PRS-TS; Fig. 1K) did not show nominally significant associations at any P-value threshold.

Multivariable model for OCD using each PRS

We performed logistic regression analyses to evaluate whether PRS for each psychiatric disorder was associated with OCD. In these models, each psychiatric disorder PRS was included individually as the predictor of interest, adjusting for age, sex, and the top 10 PCs. Bonferroni correction was applied to account for multiple testing across the 11 psychiatric disorder PRSs (adjusted significance threshold: P-value = 0.0045 [0.05/11]). PRS-OCD, PRS-AN and PRS-SCZ remained significantly associated with OCD after Bonferroni correction (Supplementary Table 3). Higher scores on all three PRSs were significantly associated with higher odds of OCD (Fig. 2 and Supplementary Table 3). PRS-OCD, PRS-AN, and PRS-SCZ were associated with 34, 24, and 23% higher odds of OCD, respectively (PRS-OCD: odds ratio [OR] = 1.34, 95% confidence interval [CI] = 1.19–1.51, P-value = 1.65 × 10⁻⁶; PRS-AN: OR = 1.24, 95% CI = 1.10–1.39, P-value = 0.0006; PRS-SCZ: OR = 1.23, 95% CI = 1.09–1.38, P-value = 0.0009). Notably, PRS-AN and PRS-SCZ showed associations with OCD similar to that observed for PRS-OCD, and all three remained in the multivariable model.

Fig. 2. Association between psychiatric disorder PRSs and OCD risk in the Korean cohort.

Logistic regression analyses were conducted to assess the association between each psychiatric disorder PRS and OCD risk in the Korean cohort. Each PRS was included individually in the model, adjusting for age, sex, and the top 10 principal components (PCs). The x-axis indicates each psychiatric disorder PRS, and the y-axis indicates the corresponding odds ratio (OR) for OCD. Points represent the ORs, and vertical lines indicate the 95% confidence intervals (CIs). Point colors denote statistical significance after Bonferroni correction (P-value < 0.0045): gray indicates non-significant associations, while colored points indicate significant associations, with distinct colors assigned to each psychiatric disorder. Abbreviations: PRS, polygenic risk score; ADHD attention deficit hyperactivity disorder; AN anorexia nervosa; ANX anxiety disorder, ASD autism spectrum disorder, BD bipolar disorder, DEP depression, HS hoarding symptoms, OCD obsessive-compulsive disorder, PTSD post-traumatic stress disorder, SCZ schizophrenia, TS Tourette syndrome.

To further determine whether these associations were independent of other PRSs, we conducted a multivariable logistic regression analysis including all 11 psychiatric disorder PRSs simultaneously in the model, together with age, sex, and the top 10 PCs. In this joint model, PRS-OCD, PRS-AN, and PRS-SCZ remained independently associated with OCD at the nominal significance level (P-value < 0.05; Table 1), suggesting a potential cross-ancestry genetic overlap across OCD, AN, and SCZ.

Table 1.

Multivariable logistic regression model for OCD using polygenic risk scores from 11 psychiatric disorders.

Polygenic risk score	OR (95% CI)	P-value
OCD	1.281 (1.133–1.448)	7.8 × 10−5
ADHD	0.904 (0.8–1.021)	0.1042
AN	1.166 (1.031–1.318)	0.0143
ANX	0.854 (0.756–0.965)	0.0115
ASD	1.004 (0.889–1.134)	0.9449
BD	1.138 (1.006–1.287)	0.0398
DEP	1.08 (0.95–1.229)	0.2403
HS	1.089 (0.964–1.23)	0.1693
PTSD	1.1 (0.968–1.249)	0.1435
SCZ	1.166 (1.03–1.319)	0.0149
TS	1.059 (0.939–1.195)	0.3514

ORs, 95% CIs, and P-values were estimated using a joint multivariable logistic regression model including PRSs for 11 psychiatric disorders simultaneously, with adjustment for age, sex, and the top 10 principal components. The estimated effects and corresponding P-values represent the association of each PRS with OCD conditional on the other psychiatric disorder PRSs included in the model. All reported P-values are two-sided.

ADHD attention deficit hyperactivity disorder, AN anorexia nervosa, ANX anxiety disorder, ASD autism spectrum disorder, BD bipolar disorder, CI confidence interval, DEP depression, HS hoarding symptoms, OCD obsessive-compulsive disorder, OR odds ratio, PTSD post-traumatic stress disorder, SCZ schizophrenia, TS Tourette syndrome.

Risk Stratification of OCD by PRS quartiles

We categorized individuals into four groups based on PRS quartiles to further assess OCD risk stratification, using PRS-OCD, PRS-AN, and PRS-SCZ. Using the bottom quartile as a reference, we compared it with the other quartiles. For both PRS-OCD and PRS-AN quartiles, we observed a stepwise increase in the odds of OCD from the second lowest to the highest quartile (Fig. 3A, Fig. 3B, and Supplementary Table 4). Individuals in the highest PRS-OCD quartile had 2.34-fold higher odds of OCD (95% CI = 1.64–3.34, P-value = 2.69 × 10⁻⁶) compared to those in the lowest quartile, while those in the highest PRS-AN quartile had 1.52-fold higher odds of OCD (95% CI = 1.09–2.13, P-value = 0.0141). For PRS-SCZ, the OR for the highest PRS-SCZ quartile compared to the lowest quartile was 1.82-fold (95% CI = 1.29–2.58, P-value = 0.0007; Fig. 3C and Supplementary Table 4). Overall, all three PRS stratification analyses indicated that the odds of OCD were 1.52–2.34-fold higher in the highest PRS quartile compared to the reference.

Fig. 3. Risk stratification of OCD by PRS quartiles for OCD, AN, and SCZ.

Logistic regression models were used to estimate the odds ratio (OR) for OCD across quartiles of PRSs for OCD, AN, and SCZ, adjusting for age, sex, and the top 10 principal components (PCs). A–C show risk stratification based on PRSs for OCD, AN, and SCZ, respectively. The x-axis indicates PRS quartiles (1 = lowest, 4 = highest), with the lowest quartile serving as the reference group. The y-axis indicates the OR for OCD in each quartile compared to the reference. Points represent the ORs, and vertical lines indicate 95% confidence intervals (CIs). Abbreviations: PRS polygenic risk score, AN anorexia nervosa, SCZ schizophrenia.

Best model to predict OCD

To evaluate the regression models for predicting OCD, we first established a baseline model that included age, sex, and the top 10 PCs, and then assessed incremental R² to quantify the added explanatory power when including each psychiatric disorder PRS. To account for the increased explanatory power due to the addition of more PRSs, we used McFadden’s adjusted R², which corrects for the number of variables in the model. Among individual PRSs added to the baseline model, PRS-OCD showed the greatest incremental R² (1.54%), while PRS-ASD, PRS-ADHD, and PRS-TS did not improve model performance, with incremental R² values close to zero (Fig. 4).

Fig. 4. Incremental adjusted R² for OCD prediction by PRS models.

Incremental McFadden’s adjusted R² values from logistic regression models were used to assess the additional variance in OCD status explained by incorporating PRSs for individual psychiatric disorders. The x-axis indicates the psychiatric disorder PRS added to the model, and the y-axis indicates the corresponding incremental McFadden’s adjusted R² (%). Each bar represents the change in adjusted R² when a given PRS is added to a baseline model including age, sex, and the top 10 principal components (PCs). Bar colors reflect the degree of variance explained, with darker bars indicating greater explanatory power. The final two bars represent a model including all PRSs (ALL) and a model selected via stepwise regression (Stepwise). Incremental McFadden’s adjusted R² (%) values and corresponding P-values are displayed above each bar. P-values were obtained from two-sided likelihood ratio tests comparing logistic regression models with and without the corresponding PRS. Negative values indicate no improvement in model performance relative to the baseline model. Abbreviations: PRS polygenic risk score, ADHD attention deficit hyperactivity disorder, AN anorexia nervosa, ANX anxiety disorder, ASD autism spectrum disorder, BD bipolar disorder, DEP depression, HS hoarding symptoms, OCD obsessive-compulsive disorder, PTSD post-traumatic stress disorder, SCZ schizophrenia, TS Tourette syndrome.

A model including all 11 psychiatric disorder PRSs yielded an incremental R² of 2.71%, corresponding to a 1.76-fold increase relative to the best single-PRS model (PRS-OCD; incremental R² = 1.54%). Finally, we constructed a stepwise logistic regression model to identify the combination of psychiatric disorder PRSs that maximized predictive performance. Using bidirectional stepwise regression, the optimal model included seven PRSs (OCD, AN, SCZ, PTSD, ANX, ADHD, and BD), achieving the highest incremental R² of 2.99% (Fig. 4 and Supplementary Table 5). Overall, we confirmed that the model including all psychiatric disorder PRSs outperformed single-PRS models, and the stepwise model with seven PRSs showed the best performance in predicting OCD.

As a complementary analysis, we additionally evaluated PRS performance using PRS-CS, a Bayesian regression method. Predictive performance varied across psychiatric disorders and PRS construction methods, with PRSice-2 generally outperforming PRS-CS across most traits (Supplementary Fig. 5).

Comparison of within-ancestry and cross-ancestry PRS performance

For psychiatric disorders with publicly available East Asian GWAS summary statistics [12, 29, 41], we compared the predictive performance of ancestry-specific and cross-ancestry PRSs for OCD. Consistent with the results obtained using European-derived PRSs, East Asian-derived PRSs for BD and SCZ were significantly associated with increased odds of OCD, whereas the East Asian-derived PRS for DEP did not show a significant association (P-value ≥ 0.05).

We further calculated the incremental R² for each PRS model and compared the results between European- and East Asian-derived PRSs. For BD, PRS derived from East Asian GWAS showed greater predictive performance for OCD than those derived from European GWAS (incremental R² = 1.99% in East Asian vs. 0.32% in European; Table 2 and Supplementary Fig. 6). For SCZ, the East Asian- and European-derived PRSs showed comparable predictive performance, with a slightly higher incremental R² for the East Asian-derived PRS (incremental R² = 0.6607% in East Asian vs. 0.6564% in European; Table 2 and Supplementary Fig. 6). In contrast, for DEP, in which the East Asian-derived PRS did not show a nominally significant association with OCD, the European-derived PRS demonstrated better predictive performance than the East Asian-derived PRS (Table 2 and Supplementary Fig. 6).

Table 2.

Predictive performance of East Asian- and cross-ancestry meta-analysis derived cross-disorder PRSs for predicting OCD.

Polygenic risk score	Reference	Ancestry	Sample size	Incremental adjusted R2 (%)	P-value
BD	O’Connell et al., 2025	East Asian	80,204	1.9869	5.22E-08
DEP	Meng et al., 2024	East Asian	286,890	−0.0334	0.2153
SCZ	Lam et al., 2019	East Asian	58,140	0.6607	0.0008
BD	O’Connell et al., 2025	European	840,309	0.3182	0.0113
DEP	Adams et al., 2025	European	2,000,702	0.3465	0.0090
SCZ	Trubetskoy et al., 2022	European	130,644	0.6564	0.0008
BD	O’Connell et al., 2025	Combined (East Asian, European)	920,513	2.8387	1.19E-10
DEP	Meng et al., 2024;Adams et al., 2025	Combined (East Asian, European)	2,287,592	−0.0645	0.2937
SCZ	Lam et al., 2019;Trubetskoy et al., 2022	Combined (East Asian, European)	188,784	1.2773	8.76E-06

Each row represents a separate single-PRS prediction model in which one East Asian-, European-, or cross-ancestry meta-analysis-derived PRS was added to the baseline logistic regression model including age, sex, and the top 10 genetic principal components. Incremental adjusted R² (%) represents the increase in McFadden’s adjusted R² obtained by adding the corresponding PRS to the baseline model. P-values were obtained from two-sided likelihood ratio tests comparing logistic regression models with and without the corresponding PRS.

PRS polygenic risk score, OCD obsessive-compulsive disorder, BD bipolar disorder, DEP depression, SCZ schizophrenia.

In addition, we performed cross-ancestry meta-analyses by combining East Asian and European GWASs and derived PRSs from the resulting meta-analysis GWASs to predict OCD risk. For both BD and SCZ, PRSs derived from the cross-ancestry meta-analysis showed the highest predictive performance (BD: incremental R² = 2.84%; SCZ: incremental R² = 1.28%; Table 2 and Supplementary Fig. 6). In contrast, the PRS derived from the cross-ancestry meta-analysis for DEP did not show a nominally significant association with OCD (Table 2 and Supplementary Fig. 6).

Discussion

Our GWAS of OCD in the Korean population did not identify any genome-wide significant loci and a meta-analysis of Korean and European OCD GWASs identified one locus on chromosome 8. This locus, indexed by rs13262595, had previously been reported in the published European-ancestry OCD GWAS [6]. Because our cross-ancestry meta-analysis used the publicly available European OCD GWAS summary statistics excluding 23andMe participants, it had lower discovery power than the full GWAS reporting 30 loci and identified only one locus. In addition, the significant association of European-derived PRS-OCD with Korean OCD status provides PRS-based evidence for shared polygenic liability between European and Korean populations. Moreover, PRSs for other psychiatric disorders, including SCZ and AN, derived from European samples, were significantly associated with OCD in the Korean cohort. These findings highlight the cross-ancestry and cross-disorder nature of the genetic liability to OCD.

To the best of our knowledge, this is the first GWAS of OCD in an East Asian population. The Korean OCD cohort analyzed in this study has the strength of including clinically confirmed diagnoses based on expert psychiatrists’ assessments within a genetically homogeneous population [42]. The SNP-based heritability was estimated at 0.18; however, this estimate had a large s.e. and was not statistically significant and therefore should be interpreted cautiously. The most recent large-scale OCD GWAS [6] reported SNP heritability estimates of 0.067 overall, 0.16 in clinical subgroups, and 0.13 in comorbid subgroups. Thus, our estimate appears broadly consistent with estimates from clinically ascertained or comorbid OCD subgroups. Given the relatively small sample size of this Korean OCD cohort, this finding should be regarded as preliminary and requires replication in larger East Asian samples.

In this study, the genetic correlation between Korean and European OCD could not be reliably estimated using Popcorn, primarily because the Korean OCD GWAS had insufficient power and yielded imprecise heritability estimates. The estimated genetic correlation exceeded 1 and had a large s.e., indicating that the estimate was highly unstable and should be interpreted cautiously. We therefore regard the cross-ancestry genetic correlation analysis as inconclusive and place greater emphasis on the PRS analyses, in which the European-derived PRS-OCD was significantly associated with OCD status in the Korean sample. Similar cross-ancestry PRS transferability has also been reported for SCZ in a Korean population [43]. However, larger OCD samples will be required to more reliably determine the extent to which European-derived PRSs capture OCD liability in Korean individuals.

Importantly, cross-ancestry transferability may be influenced not only by ancestry-related differences in allele frequencies and LD structure, but also by phenotypic, ascertainment, and sociocultural heterogeneity across populations. Prior cross-cultural work suggests that sociocultural context can shape perceptions of OCD, help-seeking behavior, stigma, and explanatory models of symptoms, although the extent to which these factors specifically affect Korean clinical OCD samples requires further investigation [44, 45]. For example, compared with respondents in the USA and Western Europe, respondents in East Asia were more likely to blame individuals with OCD and to recommend that they not seek help from others, whereas respondents in Western countries showed more favorable views of psychosocial explanations and interventions [45]. Culture may also influence the clinical prominence or recognition of particular symptom dimensions across samples [44, 46]. Such non-genetic heterogeneity could attenuate PRS transferability even when part of the underlying biological liability is shared. Future cross-ancestry studies with harmonized phenotyping and culturally informed assessment will be important to disentangle biological from ascertainment-related sources of heterogeneity.

Consistent with previous evidence for cross-disorder genetic overlap involving OCD [47–50], our PRS analyses suggest that this overlap is also detectable across ancestries in a Korean OCD cohort using European-derived psychiatric disorder PRSs. Among the cross-disorder PRSs examined, European-derived PRS-AN and PRS-SCZ remained significantly associated with OCD status after Bonferroni correction in logistic regression models adjusted for age, sex, and the top 10 PCs. These associations are consistent with previous studies reporting shared genetic liability between OCD and AN, including a high genetic correlation between the two disorders and cross-disorder PRS associations across OCD- and AN-related symptom dimensions [50, 51]. Shared genetic liability between OCD and SCZ has also been reported in prior PRS and genetic-correlation analyses [47, 48].

The absence of significant associations for PRS-HS and PRS-TS should be interpreted cautiously. Both HS and TS are clinically and nosologically related to OCD. Hoarding disorder is classified within obsessive-compulsive and related disorders in DSM-5 [52]; however, the GWAS summary statistics for HS used in the present study were derived from symptom-based measures rather than clinically diagnosed hoarding disorder. Therefore, PRS-HS may not fully capture the genetic liability to clinically defined hoarding disorder. Similarly, TS and OCD have historically been examined jointly in GWASs, and previous genetic studies have suggested partially shared genetic liability between these conditions [53, 54]. The lack of significant PRS-TS association in our study may therefore reflect limited statistical power rather than the absence of shared genetic architecture. In particular, the discovery GWAS sample sizes for HS and TS were smaller than those for several other psychiatric disorders included in our PRS analyses, which may have reduced PRS prediction accuracy and cross-ancestry transferability.

In PRS-based models for predicting Korean OCD status, PRS-OCD showed the strongest predictive performance among individual PRSs, which is expected because it is disorder-matched and most directly captures OCD-related polygenic liability. The higher predictive performance of the all-PRS and stepwise models compared with PRS-OCD alone suggests that PRSs for genetically related psychiatric disorders may add predictive information by capturing shared transdiagnostic genetic liability that is not fully captured by the current European OCD GWAS-derived PRS. The best-performing model included PRS-OCD, PRS-AN, PRS-SCZ, PRS-PTSD, PRS-ANX, PRS-ADHD, and PRS-BD, while excluding PRS-ASD, PRS-HS, PRS-TS, and PRS-DEP. These findings highlight that the common genetic risks across phenotypes with shared biological underpinnings may contribute to improving risk prediction for complex disorders, consistent with recent studies [55–57]. However, prior research on psychiatric disorders has provided only limited evidence supporting the predictive advantage of multi-PRS models. A study involving patients with mood disorders, ANX, ADHD, ASD, and substance use disorders found that using all PRSs as predictors increased the model’s power compared to using a single-PRS, although the improvement was not statistically significant [58]. In addition, a cross-disorder PRS composed of ASD, ADHD, SCZ, DEP, and BD did not show better prediction performance for psychiatric disorders compared to single-disorder PRSs for SCZ, BD, or DEP [59]. Taken together, our findings support the use of multi-PRS frameworks as a research approach for capturing known cross-disorder polygenic overlap in OCD, although their predictive value should be validated in larger independent cohorts.

For psychiatric disorders with publicly available East Asian GWAS data, we further compared the predictive performance of PRSs derived from East Asian, European, and cross-ancestry meta-analysis GWASs for predicting OCD risk. For BD and SCZ, PRSs derived from East Asian GWASs showed better or comparable predictive performance for Korean OCD than PRSs derived from European GWASs, suggesting that ancestry-matched PRSs may improve predictive accuracy relative to PRSs derived from different ancestral populations. Furthermore, PRSs derived from cross-ancestry meta-analysis GWASs demonstrated the highest predictive performance for both BD and SCZ, outperforming single-ancestry PRSs. This pattern suggests that ancestry matching and discovery GWAS sample size may jointly influence PRS performance. Ancestry-matched PRSs may benefit from more similar LD structure and allele-frequency patterns, whereas cross-ancestry meta-analysis PRSs may gain additional power by combining East Asian genetic information with larger European discovery samples when genetic effects are sufficiently shared across populations. In contrast, for DEP, the European-derived PRS showed better predictive performance than the East Asian-derived PRS, likely reflecting the substantially larger sample size and greater statistical power of the European GWAS, together with the highly polygenic architecture of DEP. Collectively, these findings suggest that integrating genetically distinct ancestral populations may partially improve PRS performance in a trait-dependent manner. For OCD, we could not directly evaluate East Asian-derived or cross-ancestry meta-analysis-derived OCD PRSs because no independent East Asian OCD GWAS summary statistics are currently available. Although we additionally evaluated ancestry-matched and cross-ancestry PRSs where publicly available East Asian GWAS data were available, the present analyses still relied primarily on European-derived summary statistics. Future studies incorporating larger independent East Asian datasets, particularly for OCD, are therefore essential to validate these patterns and determine whether ancestry-matched or cross-ancestry OCD PRSs can further improve prediction in Korean and other East Asian populations.

Several limitations should be considered when interpreting the results of this study. First, the relatively small sample size of our Korean OCD cohort had limited statistical power to detect genome-wide significant loci. To the best of our knowledge, this study represents the only OCD GWAS conducted in an East Asian population to date, underscoring the need for future efforts to assemble larger and more diverse cohorts within this ancestry group. Second, this study was conducted on Korean individuals with OCD, therefore, there may be limitations in generalizing the results to the entire East Asian population. Third, our study was unable to determine the specific genetic mechanisms underlying the genetic transferability between OCD and other psychiatric disorders. Future studies investigating shared biological pathways among OCD, AN, and SCZ may provide deeper insights into the biological basis of these cross-disorder genetic associations.

This study is the first GWAS of OCD in an East Asian population, demonstrating cross-ancestry and cross-disorder genetic transferability between European and Korean populations. PRSs derived from European OCD, AN, and SCZ were significantly associated with Korean OCD risk, indicating substantial genetic overlap among these psychiatric disorders. The enhanced performance of the multi-PRS model suggests shared genetic predispositions across ancestries and diagnoses. These findings highlight the potential clinical utility of cross-disorder PRS models in improving OCD risk prediction in diverse populations.

Author contributions

CIP and JK wrote the manuscript as co-first authors. CIP, JIK and SJK acquired the data. JK, S-HJ, BK, SP and H-HW contributed in data analyses. WM provided scientific input on data analysis and interpretation of the results. H-HW and SJK conceived and designed the study and wrote the protocol as the co-corresponding authors. All authors contributed significantly to the study and approved the final manuscript.

Funding

This study was conducted with bioresources from the National Biobank of Korea, the Center for Disease Control and Prevention, Republic of Korea (24110701-01-01). This work was supported by a National Research Foundation of Korea (NRF) grant funded by the Korean Government (RS-2021-NR056913 and RS-2023-00209077 to S.J.K.), and by a grant of Korean ARPA-H Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health and Welfare, Republic of Korea (RS-2025-25455095 to H.-H.W.).

Data availability

The individual-level data analyzed in this study are not publicly available due to ethical and legal restrictions related to participant privacy and data-sharing agreements. Access to these data may be granted upon reasonable request to the corresponding authors and subject to approval by the relevant institutional review boards. European GWAS summary statistics used in this study are publicly available from the original publications.

Competing interests

The authors declare that there are no conflicts of interest in relation to this study.

Ethics declaration

This study was approved by the Severance Hospital Institutional Review Board (IRB No. 4-2015-0655 and 4-2024-1156) and was conducted in accordance with the Declaration of Helsinki.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41398-026-04395-7.