Polygenic risk for obsessive-compulsive disorder (OCD) predicts brain response during working memory task in OCD, unaffected relatives, and healthy controls

Alterations in frontal and parietal neural activations during working memory task performance have been suggested as a candidate endophenotype of obsessive-compulsive disorder (OCD) in studies involving first-degree relatives. However, the direct link between genetic risk for OCD and neuro-functional alterations during working memory performance has not been investigated to date. Thus, the aim of the current functional magnetic resonance imaging (fMRI) study was to test the direct association between polygenic risk for OCD and neural activity during the performance of a numeric n-back task with four working memory load conditions in 128 participants, including patients with OCD, unaffected first-degree relatives of OCD patients, and healthy controls. Behavioral results show a significant performance deficit at high working memory load in both patients with OCD and first-degree relatives (p < 0.05). A whole-brain analysis of the fMRI data indicated decreased neural activity in bilateral inferior parietal lobule and dorsolateral prefrontal cortex in both patients and relatives. Most importantly, OCD polygenic risk scores predicted neural activity in orbitofrontal cortex. Results indicate that genetic risk for OCD can partly explain alterations in brain response during working memory performance, supporting the notion of a neuro-functional endophenotype for OCD.


Scientific Reports
| (2021) 11:18914 | https://doi.org/10.1038/s41598-021-98333-w www.nature.com/scientificreports/ the current study is part of a larger project, and other results of this project have been published elsewhere 14,[25][26][27] . All participants gave written informed consent after receiving written and verbal information about the study, and received a monetary compensation for their time.
Genotyping and polygenic risk scores. DNA samples from blood (n = 112) or saliva (n = 7) were genotyped using the Infinium Global Screening Array (Illumina) at the LIFE and BRAIN GmbH Bonn, Germany. Genotype quality control was done using Plink-1.9, and R (version 3.5.1). We checked the data for sex inconsistencies and grossly failing markers (call rate < 0.5). Individuals with a call rate of < 0.95 were removed. The heterozygosity rate for each subject was calculated; outliers (± 3 SD from the mean heterozygosity rate) were identified and removed. On marker level, SNPs were removed if at least one of the following conditions was true: significant difference of missing rate between cases and controls, call rate < 0.95; deviation of Hardy-Weinberg equilibrium (p < 1 × 10 -6 ); and minor allele frequency < 0.05 (computed separately in cases and controls). Furthermore, all A/T or C/G SNPs were removed. A genetic relatedness check was done and none of the OCD patients were related to any of the HC participants. To check and correct for population stratification (i.e. allele frequency differences between cases and controls due to systematic ancestry differences), principal component analysis was performed including all SNPs 28 . This method uses genome-wide genotype data to estimate principal component axes that can be used as covariates in subsequent association analyses to control for spurious associations due to ancestry differences. Thus, the first two principal components were included as covariates in all analyses that involved polygenic risk scores. The genotyped data were imputed on the Michigan Imputation Service using the 1000 Genomes Phase 3 (Version 5) reference panel. Low quality (INFO-score < 0.5) and rare (MAF < 0.01) variants were removed from the imputed data set, leaving 4,869,847 variants. The polygenic risk scores for each participant of our study were computed using PLINK 29 . For the polygenic score calculation, we used summary statistics from the Psychiatric Genomics Consortium (PGC) genome-wide association studies (GWAS) for OCD 5 as a discovery sample. The number of risk alleles carried for each selected SNP (i.e., 0, 1, or 2) was weighted by the log(OR) provided by the PGC GWAS, and averaged across all SNPs. 5381 SNPs were selected that were significant at a significance threshold of p < 0.01. This significance threshold was chosen as these top-ranked SNPs were suggested to have an important role in gene expression in the brain and possibly in the etiology of OCD 30 . Absolute values of polygenic risk scores were z-transformed for further analyses. Note, that in our fMRI sample, OCD polygenic risk scores could be obtained from 119 subjects (43 HC, 32 REL, 44 OCD).
N-back paradigm during fMRI. We used the same task setup as reported previously 14 : Sixteen blocks (4 blocks of each 0-, 1-, 2-, and 3-back) were presented in different pseudo-randomized orders. The working memory load condition of each block was indicated by a cue displayed 2 s before the block started. In each block, 16 randomly generated digits from 0 to 9 were presented in the center of a black screen one at a time for 500 ms with an interstimulus interval of 900 ms; the occurrence of 5 target stimuli was pseudo-randomized. Targets were defined as re-occurrence of a number previously presented 1, 2, or 3 trials before (1-, 2-, or 3-back condition).
In the 0-back condition, the target was defined as the digit '0' . The participants were instructed to press a button with their right thumb when they recognized a target. After each block, a white fixation cross was presented in the center of a black screen for 4 s. Every fourth block, the fixation cross was presented for 14 s. The total task duration was 9:00 min. Before the fMRI session, two practice sessions of the n-back task were performed outside the scanner to familiarize participants with the task. The n-back task was presented using MR image processing and analysis. All fMRI analyses were carried out with SPM12 (revision 6906; Wellcome Trust Centre For Neuroimaging, London, UK). After correction for head motion and computation of a mean EPI image, the T1w image was co-registered to the mean EPI image and normalized (by integrating information of the T2w image) into the spatial standard space as defined by the template of the International Consortium for Brain Mapping (http:// www. loni. ucla. edu/ ICBM/). None of the participants had to be excluded due to excessive head movements. Spatial transformations as estimated during the segmentation procedure were applied to EPI images. EPI images were resampled into isotropic voxels with an edge length of 2 mm and spatially smoothed with an isotropic Gaussian kernel of 8 mm full width at half maximum 14 .
Estimation of BOLD effects in n-back. The working memory experiment was analyzed within the framework of the General Linear Model (GLM). As described in previous work 14 www.nature.com/scientificreports/ sors of interest and all other experimental conditions (cue, button presses, and the six rigid body realignment parameters) as regressors of no interest. The GLM was fitted voxel-wise into the filtered time series using the restricted maximum likelihood algorithm as implemented in SPM12. Three contrasts of interest were built: 1-back > 0-back, 2-back > 0-back, and 3-back > 0-back. On the second level, a random effects model as implemented in the GLM_Flex_Fast4 toolbox http:// mrtoo ls. mgh. harva rd. edu/ index. php? title= GLM_ Flex) was applied for a repeated measures ANCOVA with the between-subjects factor group (OCD vs. REL vs. HC), the within-subjects factor working memory load (1 > 0-back vs. 2 > 0-back vs. 3 > 0-back), and the covariate age. Age was included as a covariate to control for age differences between groups (see Table 1) and because of its effect on working memory performance and brain response 32,33 . Whole brain analyses of the group by working memory interaction effects were thresholded at p < 0.05, family-wise error (FWE) at cluster-level. Analyses were performed for the whole brain, restricted to gray matter according to the tissue probability map thresholded at 0.3 as implemented in SPM12. We used a Monte Carlo simulation correction (10,000 iterations) with an initial voxel-wise threshold of p < 0.001 (http:// afni. nimh. nih. gov/ pub/ dist/ doc/ progr am_ help/ 3dClu stSim. html). Clusters with a minimum cluster size of 48 voxels yielded a cluster-level FWE threshold of p < 0.05 and are described in the results section and in Table 2.
To assess the direct effect of OCD polygenic risk scores on BOLD response during n-back, a random effects model was applied to run a repeated measures ANCOVA with the within-subjects factor working memory load (1 > 0-back vs. 2 > 0-back vs. 3 > 0-back), and the covariates OCD polygenic risk score, the first two population structure principle components, and age. For this analysis, clusters with a minimum cluster size of 56 voxels yielded a cluster-level FWE threshold of p < 0.05.
Statistical analyses of working memory performance, polygenic risk, and group status. Group differences in working memory performance were analyzed using a group (OCD vs. REL vs. HC) by working memory load (0-vs. 1-vs. 2-vs. 3-back) analysis of covariance (ANCOVA) model with the covariate age. To test associations between OCD polygenic risk scores and group status, an ordinal logistic regression was conducted. Linear regression analyses were performed to test associations between OCD polygenic risk scores and working memory performance. Note that, in line with previous reports including polygenic risk scores in related participants 34,35 , population structure covariates were included 28 in all analyses that involved polygenic risk scores.

Results
Behavioral results n-back. The three (group) by four (working memory load) ANCOVA with the covariate age of the n-back performance revealed a significant interaction of group by working memory load (F(6, 372) = 2.46, p = 0.024, partial η 2 = 0.038) and a significant main effect of age (F(1, 124) = 19.54, p < 0.001, partial η 2 = 0.136), as well as a significant interaction of working memory load by age (F(3, 372) = 10.17, p < 0.001, partial η 2 = 0.076). Performance values are shown in Table 1, estimated marginal means of the ANCOVA model with the covariate age are shown in Fig. 1. Post-hoc two-sample t tests of this ANCOVA model indicated significant differences only in the 3-back condition between patients with OCD and HC (t(94) = 2.14, p = 0.035); and between REL and HC (t(75) = 2.35, p = 0.021). These results show that, when age is taken into account, HC show a higher performance in 3-back compared to both patients with OCD and REL. All other t tests were not significant (all p-values > 0.08).
FMRI results during n-back. As shown in Fig. 2A, whole-brain analyses of the three (group) by three (working memory load) interaction revealed significant interaction effects in two large and four smaller clusters of the fronto-parietal working memory network (p < 0.05, FWE cluster-corrected). The highest t-values and largest cluster extents were found in left superior/inferior parietal lobule (left SPL/IPL, t = 9.30) and right inferior parietal lobule (right IPL, t = 9.15). The activation patterns (see Fig. 2B,C) indicate that both the OCD and the REL groups showed reduced activations for 2-and 3-back compared to the healthy control group. See Table 2  OCD polygenic risk scores and BOLD response. As shown in Fig. 3A, the whole-brain ANCOVA with the within-subject factor working memory load and the covariates OCD polygenic risk scores, the first two population structure principle components, and age, revealed a significant effect for OCD polygenic risk scores www.nature.com/scientificreports/ in the right medial orbitofrontal gyrus (MNI coordinates 6 54 -8; t = 11.47, cluster size: 98 voxels, p < 0.05 FWE cluster-corrected). As shown in Fig. 3B, this finding indicates that higher OCD polygenic risk scores predicted an increase in BOLD response during 2-and 3-back but not during 1-back.

Discussion
Results reflect that patients with OCD and unaffected REL showed performance decrements in working memory updating as well as reduced BOLD responses in right IPL, left SPL/IPL, bilateral DLPFC, left PMC, and left IFG during 2-and 3-back. Most importantly, OCD polygenic risk scores predicted BOLD response in medial OFC, indicating that higher genetic risk for OCD led to increased activity in medial OFC during the performance of a demanding working memory task. Behavioral results are in line with previous reports of OCD-related impairments in executive functions (for review see 36,37 ) and more specifically in working memory updating (for review see 38 ). Our and previous studies that included both patients with OCD and unaffected REL, showed that executive dysfunctions can be seen not only in patients suffering from OCD, but also in subjects with a genetic risk for OCD 39,40 , thus supporting the notion of executive dysfunction being a candidate endophenotype of OCD.
The fMRI results from the group by working memory load analysis indicated that both patients with OCD and unaffected REL show reduced neural activity in the fronto-parietal working memory network 41 during high task demand. Together with a marked performance decrement, these findings suggest an impaired functioning of the working memory system at high working memory load. It seems that these dysfunctions become visible when updating is required and may be due to inefficient strategies 38 . The strongest effects and largest cluster extents were found in the bilateral IPL/SPL. IPL was found to play an important role in working memory updating specifically involving selective attention, working memory rehearsal, and capacity 42,43 . As shown in the recently published mega-analyses from the ENIGMA-OCD working group 44 , reduced cortical thickness in bilateral IPL was the main OCD-associated structural brain imaging finding in their sample of 1498 adults with OCD. Thus, we had expected that IPL would also show altered activations during working memory performance as shown previously 17,45,46 .
In line with previous studies 17,45,47 , we also found altered activations in bilateral DLPFC, a region that has been associated with executive components of working memory (e.g. distractor resistance, updating, action selection 48 ). However, in contrast to several previous studies that reported fronto-parietal hyperactivations during working memory performance 17,45,47 , we found mainly decreased activations at 2-and 3-back. These hypoactivations in bilateral IPL and DLPFC together with marked performance decrements in 3-back in OCD and REL, may indicate that compensatory attempts fail at high working memory load as suggested by models of an inverse U-shaped relationship between working memory load and BOLD responses 31,49,50 . These models indicate that impairments in the working memory system can be related to a left-ward shift of this function, showing relative hyperactivations at lower working memory load and hypoactivations at higher working memory load, as reported in OCD before 14,15 . This concept has been described in terms of a reduced fronto-parietal adaptability to increasing working memory load 14 and may partially integrate different findings of previously reported hyper-and hypoactivations.
Crucially, polygenic risk for OCD predicted BOLD response during n-back performance in medial OFC, reflecting that neural activity in OFC increased with increasing genetic risk for OCD. Therefore, carrying a higher genetic risk to develop OCD seems to affect the medial OFC functioning during the execution of a demanding working memory task. It is important to note that this analysis is agnostic to phenotypical information such as OCD symptoms or OCD diagnosis. Thus, these results suggest that alterations in medial OFC functioning may play a role in the etiology of OCD as opposed to being a consequence of the OCD phenotype. Our study expands the literature on orbitofronto-striatal dysfunctioning that has previously described in terms of increased activity and connectivity of the OFC in the OCD phenotype 51-53 by providing evidence for an association with the genetic risk for OCD. While previous genetic studies have reported effects of gene variants on OFC morphometry 20,21 , the current study showed that genetic risk for OCD may affect OFC functioning during a cognitively demanding task (n-back).
While reduced activity in lateral frontal areas and SPL/IPL was shown in the group by working memory load analysis for both OCD and REL, a correlation between SPL/IPL activity and OCD polygenic risk scores was only found in a small cluster that did not survive FWE-correction in the second model including OCD polygenic risk scores as predictors of BOLD response. Thus, the genetic contribution to alterations in lateral frontal cortex and IPL/SPL is less clear.
While OCD polygenic risk scores were significantly associated with group status, it explained only a relatively small portion of variance (R 2 = 0.043). Since models including the factor group status also rely on information on OCD phenotype, diverging results between the two reported fMRI analyses are not surprising. They may reflect the importance of both genetic and environmental factors in the etiology of OCD, as well as limitations of the current mainly symptom-based classification systems for mental disorders 54 .
Since medial OFC is not considered to be a core region that is recruited during working memory performance in healthy subjects 41,48 , an increased neural activity in this region deviates from normal working memory-related activation patterns and has been associated with an OCD-related over-monitoring 16 . Over-monitoring that is applied during high working memory demand may become an inefficient strategy, eventually leading to impaired working memory performance.
Together with our findings of reduced lateral fronto-parietal functioning in OCD and REL of OCD-patients, results seem to point to a suggested OCD-related imbalance in cortico-basal ganglia-thalamo-cortical (CBGTC) circuits 52 . Thus, the over recruitment of a "limbic" CBGTC circuit involving medial OFC 55,56 may interfere with the fronto-parietal functioning required for high working memory performance 17  www.nature.com/scientificreports/ The novel finding of the current study is a direct association between polygenic risk for OCD and neural activity in medial OFC, providing a new puzzle piece for the understanding of the etiology of OCD.
Some limitations of the study need to be noted. Groups differed in age, however, results were corrected for age differences and both behavioral and fMRI results remained significant when controlling for age. Since we aimed to investigate a naturalistic patient sample, the OCD group was relatively heterogeneous regarding medication, symptom dimensions, and comorbidity. Also, relatively large age differences add another level of heterogeneity to the sample. Our findings suggest that alterations in neural activity during working memory performance may apply to OCD patient populations in everyday care and are not restricted to highly selected study populations. Even though our sample size is the largest sample of OCD-patients and REL performing a working memory task during fMRI measurements to date, the sample is small for polygenic risk score analyses. Thus, our results need to be interpreted as preliminary and require replication in larger samples that would also facilitate further analyses such as comparisons between specific symptom dimensions or comorbidities.
Taken together, the results of the current study suggest that both patients with OCD and unaffected REL show a reduced activity in the fronto-parietal working memory network at high working memory load accompanied by deficient performance. The magnitude of genetic risk for OCD predicted the intensity of neural activity in the medial OFC agnostic to information on OCD symptoms or OCD phenotype, thus supporting the concept of a neuro-functional endophenotype for OCD.