Unsuppressed Striatal Activity and Genetic Risk for Schizophrenia Associated With Individual Cognitive Performance Under Social Competition

Abstract

Background and Hypothesis

Social competition affects human behaviors by inducing psychosocial stress. The neural and genetic mechanisms of individual differences of cognitive-behavioral response to stressful situations in a competitive context remain unknown. We hypothesized that variation in stress-related brain activation and genetic heterogeneity associated with psychiatric disorders may play roles towards individually differential responses under stress.

Study Design

A total of 419 healthy subjects and 66 patients with schizophrenia were examined functional magnetic resonance imaging during working memory task including social competition stressors. We explored the correlation between stress-induced brain activity and individual working memory performance. The partial least squares regression was performed to examine the genetic correlates between stress-related activity and gene expression data from Allen Human Brain Atlas. Polygenic risk score (PRS) was used to assess individual genetic risk for schizophrenia.

Study Results

Greater suppression of bilateral striatal activity was associated with better behavioral improvement in working memory manipulation under social competition (left: r_Pearson = −0.245, P = 4.0 × 10⁻⁶, right: r_Pearson = −0.234, P = 1.0 × 10⁻⁵). Genes transcriptionally related to stress-induced activation were linked to genetic risk for schizophrenia (P_FDR < 0.005). Participants with decreased accuracy under social competition exhibited higher PRS of schizophrenia (t = 2.328, P = .021). Patients with schizophrenia showed less suppressed striatal activity under social stress (F = 13.493, P = 3.5 × 10⁻⁴).

Conclusions

Striatal activity change and genetic risk for schizophrenia might play a role in the individually behavioral difference in working memory manipulation under stress.

Introduction

Psychosocial stress has long been implicated in the etiology and pathophysiology of mental health conditions that now pose a great threat to public health.¹ Social defeat stress, driven by the evaluation when compared with others, is one of the primary sources of psychosocial stressors for human.^2,3 However, individuals differ in their susceptibility to social stress, including coping style and behavioral response.^4,5 For instance, someone was spurred by the social competition to do their best, while someone was hindered in their performance.^6,7 Altered sensitivity of stress and maladaptive coping were also suggested to be associated with the risk of developing psychosis, like schizophrenia.⁸

The brain, as the key organ of stress processes, determines the individual stressful experience and behavioral response.⁹ Neuroimaging evidence suggested that cortico-striatal regions were involved in mediating stress reactivity.^10,11 An early study pointed out the effect of social comparison on activity in the ventral striatum.¹² Greater striatal activation was also observed when people won over their counterparts compared with winning in isolation.¹³ The striatal activity driven by differential responses to winning versus losing was found to be dependent on interpersonal peer relationships during the competitive process.¹⁴ Stress could also affect the prefrontal cortex and the associated learning and memory processes, including working memory.¹⁵ Recent empirical evidence suggested competition might hinder performance on a memory task.⁷ Performance-related deactivation in the medial prefrontal cortex during social competition was observed during working memory maintenance.¹⁶ Although many studies have explored the neural response to social stress, the underlying neural mechanism interpreting the individual differences of stress susceptibility remains unclear.

Differential susceptibility (including coping style and behavioral response) was reported to be modulated by genetic factors. Twin studies confirmed the heritability of the ability to adapt to stress, and multiple studies using candidate genes or genome-wide association strategies identified some potential loci associated with specific traits related to stress, ie psychiatric diseases and resilience behaviors,¹⁷ by mediating the relationship between stress exposure, dopaminergic functioning, and psychosis.¹⁸ Genetic alterations and their interaction with stress exposure were demonstrated to be necessary to shape stress-vulnerable phenotypes.^19,20 For example, the diathesis-stress model stated that individual genetic variants regulated the susceptibility to environmental influences, which were initially developed to explain the cause of schizophrenia²¹ and adapted for depression.²²

Recent advances in transcriptome-neuroimaging association studies^23,24 give us the chance to bridge genetic and neuroimaging data via transcriptional dataset using Allen Human Brain Atlas (AHBA) and partial least squares regression (PLSR) analysis. This combined analysis of neuroimaging and transcriptional data have been applied in studies of depression and schizophrenia to provide new insight into how genes may drive structural or functional brain changes related to specific mental disorders.^25–27

Here, we proposed a hypothesis that variation in stress-related striatal activation and fronto-striatal circuity, as well as genetic heterogeneity associated with psychiatric disorders, may play roles towards individually differential responses to the cognitive process under stress. We tested this hypothesis in two independent healthy cohorts and a clinical validation cohort of schizophrenia. A newly designed working memory functional magnetic resonance imaging (fMRI) task with counterbalanced competition and noncompetition blocks was used to explore the stress-induced performance-related brain activity; the transcriptome analysis was performed to identify genes transcriptionally related to stress-induced activation changes, and the target genes were subjected to disease-related and functional annotations; polygenic risk score (PRS) was used to evaluate genetic heterogeneity individually (figure 1).

Methods

Participants

We recruited 604 healthy participants in Beijing, of which 522 for the discovery cohort and 82 for the independent validation cohort. The clinical validation cohort included 110 patients with schizophrenia (SCZ). All participants were of Chinese Han ancestry, 18- to 45-year-old and right-handed. The inclusion/exclusion criteria of participants were outlined in Supplementary Methods 3. This study was approved by the Institutional Review Boards of Peking University Sixth Hospital.

Functional MRI Task Design

We adopted a task as numerical working memory including social competition in half the trials in a mixed block/event-related design^28–32 (figure 2A). Briefly, the paradigm included two blocks (“non-competition” block and “competition” block) with 82 trials. Each block contained 41 trials, that is, 14 working memory maintenance trials, 14 working memory manipulation trials, and 13 control trials (Supplementary Methods 1). During competitive trials, participants were led to believe that they were playing against a competitor with similar age and gender, and were judged as the winner or loser based on their reaction speed and accuracy. To induce social competition stress, participants received more negative feedback of “you lost” (5/7) regardless of the actual performance. During the noncompetitive trials, participants played with no competitors and received neutral responses. Subjects performed two runs (10 min each run) counterbalanced for trial and stimuli presentation order for 20 min in total.

The accuracy percentage changes (ΔAccuracy) were defined as the indicator of working memory-related cognitive performance to stress. We further defined individual participants according to ΔAccuracy as stress-thriving (ΔAccuracy > 0, accuracy increased at least 7.14%), stress-insensitive (ΔAccuracy = 0, accuracy unchanged), and stress-vulnerable (ΔAccuracy < 0, accuracy decreased at least 7.14%).

Neuroimaging Data Acquisition and Preprocessing

The detailed fMRI parameters and preprocessing, as well as the flow chart of participant inclusion/exclusion, were described in Supplementary Methods 2 and 3.2. Finally, a total of 485 participants were included (discovery cohort: N = 353, healthy validation cohort: N = 66, schizophrenic validation cohort: N = 66).

Brain Activation Analysis

The contrasts of interest were brain activation of working memory manipulation under noncompetition and competition with the six parameters of head motion as covariates. “Competition vs. non-competition” activation difference maps were generated to indicate the stress effect. Mean condition-specific regional responses were determined using one-sample t-test (P < .05 family-wise error [FWE] corrected, and cluster size > 100) in three task conditions, ie, competition, non-competition, and competition vs noncompetition.

Whole-brain Linear Regression Analysis of Brain Activation

The individual “competition vs. non-competition” activation maps were used to conduct a second-level linear regression model. The ΔAccuracy was inputted as the dependent variable, age, gender, educational attainment, and RT change were used as covariates. Activation maps under competition and noncompetition conditions were also performed whole-brain linear regression analysis with their accuracy rates as the dependent variables. Psycho-physiological interaction analysis was performed to identify brain regions that showed significantly different covariation with accuracy-related brain activities (Supplementary Methods 5).

Gene Expression Data Preprocessing

We processed the AHBA dataset²³ of two donors with bilateral hemispheres according to the newly published guide³³(Supplementary Methods 6). Finally, a bilateral brain-wide gene (10 437) × sample (1211) matrix was obtained. Brain activation survived under the “competition vs. non-competition” condition was defined as stress-induced activation changes. Tissue samples whose MNI coordinates were no more than 4 mm away from any voxel within stress-induced region were included, which filter out 342 of 1211. The mean t-value of stress-induced activation was computed within the sphere drawn with the MNI coordinates of each sample as the center and 4 mm radius as the radius.

Partial Least Squares Regression (PLSR) Analysis

The gene expression matrix and t-statistics of stress-induced activation map were recognized as the predictor and response variables respectively. The first component of the PLS (PLS1) typically provided the most explanatory power with optimal low-dimensional representation. We tested the statistical significance of the variance explained by PLS1 utilizing 10 000 permutations. Then, the PLS1 weights for each gene were z-transformed based on standard errors obtained from 10 000 bootstrap replications.³⁴ The Benjamini and Hochberg FDR method was used to correct for multiple testing.

Enrichment Analysis

The PLS1 genes were input into the ToppGene tool (https://toppgene.cchmc.org/) for enrichment analysis of human diseases based on the DisGeNET database.³⁵ Besides, we obtained the differential gene expressions (DGE) in postmortem brain tissue measurements of messenger RNA based on the case-control study of SCZ, bipolar disorder (BD), autism spectrum disorder, major depression disorder, and alcoholism.³⁶ Spearman's correlation analysis was performed between the DGE values and PLS1 gene weights.²⁵ A multi-gene-list meta-analysis²⁵ was performed for pathway analysis based on the gene lists obtained from discovery and healthy validation cohorts through the Metascape tool.

Computation of Schizophrenia Polygenic Risk Score (SCZ-PRS)

Details of the genotyping and quality control were described in Supplemental Methods 4 and figure S1. We used the PRSice-2³⁷ to compute SCZ-PRS based on the largest GWAS study to date of East Asian participants³⁸ with a significant level of P ≤ .05, which discriminated individuals with and without schizophrenia effectively.

Results

Demographic and Task Performance

In the discovery cohort, 353 adult participants were included. The accuracy rate under competition condition was higher than that under non-competition condition (paired t-test, manipulation: 83.1% vs 85.4%, t = 3.49, P = 5.37 × 10⁻⁴). When dividing the entire cohort into three groups according to ΔAccuracy, the RT was significantly different among groups and the “stress-vulnerable” group showed the longest RT. There was no significant between-group difference in gender distribution, age, and educational attainment (table 1) as well as the working memory capability assessed by Spatial Span of Wechsler Memory Scale (56.7 vs 57.5 vs 55.6, F = 1.005, P = .367).

In the healthy and clinical validation cohort, patients with SCZ performed worse behaviorally in all the task conditions compared with healthy controls (HCs) (table 1). Marginal significant difference of accuracy was observed between competition and noncompetition in the healthy (paired t-test, 84.5% vs 87.2%, t = 1.989, P = .051), but not in patients with SCZ (paired t-test, 66.6% vs 65.7%, t = −0.442, P = .660).

Discovery Cohort

The Brain Activity of Working Memory Manipulation.

In the working memory manipulation task, under both competition and non-competition conditions, a broad fronto-parietal network including regions in the prefrontal cortex, parietal lobe, temporal lobe, insular cortex, and striatum was robustly activated, whereas the medial frontal cortex, middle temporal gyrus, posterior cingulate cortex, parahippocampal gyrus, and hippocampus were deactivated (figure 2B)

For stress-induced changes, the thalamus, parahippocampal gyrus, hippocampus, precuneus, and cerebellum showed increased activity; the deactivations were observed in the bilateral dorsal striatum, inferior frontal gyrus, and inferior parietal lobule (P < .05, whole-brain FWE corrected, cluster size > 100, figure 2B, supplementary table S1).

Brain Activity Associated With the Individual Performance.

The linear regression analyses revealed that ΔAccuracy was significantly and negatively correlated with bilateral dorsal striatum activation changes during working memory manipulation (P < .05, whole-brain FWE corrected, figure 3A, Supplementary table S3; left: r_Pearson = −0.245, P = 4.0 × 10⁻⁶, right: r_Pearson = −0.234, P = 1.0 × 10⁻⁵, figure 3B). The activities of the bilateral dorsal striatum were also negatively correlated with the accuracy rate during non-competition and competition conditions (Supplementary figure S2). The functional connectivity between the bilateral striatum and left dorsolateral prefrontal cortex was found to be increased under social stress but did not show significant correlations with individual cognitive performance (supplementary results 3).

Repeated ANOVA revealed a significant interaction effect of group (stress-thriving, stress-insensitive, and stress-vulnerable) × stress (competition and noncompetition) for the striatal activity (left: F_(2,350) = 8.41, P = 2.7 × 10⁻⁴, right: F_(2,350) = 6.23, P = 2.2 × 10⁻³). The stress-thriving group showed the largest suppressed activation, whereas the stress-vulnerable group showed the smallest (figure 3C). Noting the increased hippocampal activation under stress, we extracted the brain activity within bilateral hippocampus respectively to assess their effects on the memory process. However, there was no significant difference among three groups.

When dividing the whole sample into two groups according to striatal feature, ie, participants with decreased striatal activity under stress were defined as “suppressed” group, and the others were defined as “unsuppressed” group, the “suppressed” group showed significantly higher accuracy during competition condition (t = 2.584, P = .020, supplementary table S4), but not in noncompetition condition (t = 0.693, P = .489).

Gene Translationally Related to Stress-induced Activation. The PLS1 of PLSR model explained up to 22.42% of the variance of stress-induced activation changes, which captured the greatest fraction of total gene expression variance (permutation test, P < .001). Notably, PLS1 weighted gene expression map (a weighted sum of 10 437 genes) shown a significantly positive correlation with the stress-induced activation changes (r_Pearson = 0.474, P < .001; figure 3D). The positive correlation indicated that genes positively (or negatively) weighted on PLS1 were overexpressed in regions with increased (or decreased) activation induced by stress.^25,26 We ranked the normalized weights of PLS1 and observed 1660 genes with significantly positive Z-scores (Z > 0, P_FDR < 0.05), which were denoted as the PLS+ gene set, and 2733 genes with significantly negative Z-scores (Z < 0, P_FDR < 0.05), which were denoted as PLS1− gene set.

Genes Transcriptionally Related to Stress-induced Activation Were Linked to Genetic Risk for Schizophrenia. In the disease-related enrichment analysis, 2733 PLS-genes were mainly enriched for neuropsychiatric disorders, and schizophrenia was the most significant term at different thresholds (figure 3E, supplementary table S7). Besides, we found 440 genes overlapped between PLS1− and genes with significantly differentially downregulated expression in postmortem brain tissue of SCZ (hypergeometric test, P < .001, supplementary figure S7).³⁶ A significantly positive relationship was observed between the weights and DGE values of these genes (r_Spearman = 0.205, P < .001, figure 3G). Similar results also appeared in BD (r_Spearman = 0.241, P = .006, supplementary figure S6), but not in the other disorders. No significant correlations were observed between PLS1+ genes and psychotic disorders (supplementary table S8 and figure S8).

To further test the hypothesis that stress-induced activation changes would be associated with SCZ-risk, we focused on the 108 SCZ-risk loci reported by Psychiatric Genomics Consortium (PGC).³⁹ Among them, 43 were linked to a single gene,⁴⁰ and 23 of 43 were shared with all PLS1 loading genes (supplementary table S9). The average Z-score across these 23 genes was calculated and found to be significantly negatively weighted on PLS1, compared to 10 000 sets of 23 randomly selected genes with permutation tests (P = .007; figure 3F).^34,40 When we determined the mean weight of 12 BD-risk genes, no significantly negative weight was observed (supplementary figure S6).

SCZ-PRS Related to Individual Performance and Stress-induced Activation.

There were significant differences in SCZ-PRS between the stress-thriving group and the stress-vulnerable group (one-way ANOVA, F = 3.531, P = .030). Specifically, the stress-thriving group with increased accuracy exhibited lower SCZ-PRS than the stress-vulnerable group with decreased accuracy (figure 3I). The SCZ-PRS also showed a positive correlation with the stress-induced activation of right striatum (peak [30 −10 2], r_Pearson = 0.146, P = .007) that negatively correlated with ΔAccuracy (figure 3J, supplementary table S3). Age and gender were added as covariates in the partial correlation analysis.

Healthy Validation Cohort

Suppressed Striatal Activity Under Stress Related to Individual Performance.

In the healthy validation cohort, we also observed suppressed striatal activations under stress (P < .05, whole-brain FWE corrected, figure 1C, supplementary table S2). Linear regression analysis between ΔAccuracy and stress-induced brain activity didn’t yield significant findings, likely due to the much smaller sample size compared with the healthy discovery cohort. However, when dividing the whole sample into two groups according to striatal feature (striatal activity at peak [24 0 −4]), the “suppressed” group showed a higher accuracy rate under competition condition (t = 2.238, P = .029, supplementary table s5 and figure S3), but not noncompetition condition, which was consistent with the discovery cohort.

Genetic Risk for Schizophrenia Related to Stress-induced Activation Changes.

In the healthy validation cohort, PLS1 explained 29.21% of the variance of stress-induced activation changes (permutation test, P < .001), and we found 3111 PLS1− and 2718 PLS1+ genes. The enrichment analysis of PLS1− genes showed that schizophrenia was the most relevant disease (supplementary table S7). Futhermore, the correlation between the PLS1− gene weights and DGE values in SCZ was also confirmed (r_Spearman = 0.134, P = .005).

As for the functional pathway enrichment, we picked representative genes (Z < −5, all p_FDR < 0.05) in two PLS1− gene lists for multi-gene-list meta-analysis.²⁵ The representative PLS1− genes of discovery and healthy validation cohorts were highly overlapped (OR = 11.35, P < .001). Additionally, we identified shared enrichment pathways, including oxytocin which is associated with the response to stressful or social stimuli⁴¹ and social affiliation⁴² (supplementary figure S9). The same method was applied to PLS1+ genes (supplementary figure S10).

In addition, HCs exhibited lower SCZ-PRS than SCZ (t = 35.03, P = 3.42 × 10⁻⁶³, figure 3I). The SCZ-PRS was positively correlated with the striatal stress-induced activation (peak [24 0 −4], r_Pearson = 0.271, P = .046, figure 3J) when age and gender were added as covariates, same as the discovery cohort.

Schizophrenic Validation Cohort

No Stress-induced Suppressed Activity of Striatum Under Competition.

The whole-brain analysis in patients with SCZ revealed that stress-induced activation only centered in the cerebellum (P < .05, cluster-level FWE corrected, supplementary table S10). The bilateral dorsal striatum activity was not suppressed under social stress, as expected.

When comparing brain activity in the dorsal striatum between healthy and schizophrenic validation cohort, HCs exhibited stronger suppressed striatal activity (two-way repeated-measures ANOVA, F = 10.809, P = .0013, figure 3H). Moreover, when we tested group differences across the whole brain, the right striatum also exhibited a significant interaction effect of diagnosis (SCZ and HC) × stress (noncompetition and competition), although with a loose correction of P < .001 (peak at [26 −6 0], F = 12.840, cluster size = 22, supplementary figure S11).

Discussion

In the present study, we aimed to explore the underlying neural mechanism of individual differences in working memory-related cognitive processes under social competition. We found that the variation of striatal activation played a role in these individual differential responses. That is, greater suppression of bilateral striatal activation was associated with better behavioral improvement in working memory when faced with social stress. Genes transcriptionally related to stress-induced activation changes were associated with SCZ risk, and genetic heterogeneity could partially explain the differential susceptibility to social stress, which was also validated in patients with SCZ.

It is presumed that stress causes a shift from flexible “cognitive” memory to more rigid “habit” memory.⁴³ Striatum-dependent habit learning was more engaged under stress at the expense of prefrontal cortex-dependent cognitive system.^44–47 Accumulating evidence also suggested that stress might affect dorsal striatum-dependent memory.⁴⁸ In our study, individuals with stronger suppression of the dorsal striatum tend to perform better under social stress in working memory tasks, which might result from the suppression of habit memory systems to inhibit stimulus-response learning. The flexible cognitive memory system like the hippocampus, which showed increased engagement under social stress, didn’t exhibit individual differences. Therefore, the ability to inhibit striatum-dependent habit memory during stress appears to affect individual task performance under social competition.

Genetic alterations were reported to be necessary to shape stress-vulnerable phenotypes.^19,20 Our transcriptome-neuroimaging association study identified a specific gene set enriched in schizophrenia. Furthermore, people with a high genetic risk were inclined to be vulnerable to environmental stressors, which manifested as worse working memory performance under social competition with less suppression of striatal activation. The diathesis-stress model of schizophrenia proposes that schizophrenia develops due to stress exposure acting on a preexisting genetic vulnerability.⁴⁹ People might be more susceptible to stress sensitization through the reduction of homeostatic control of midbrain dopaminergic neurons.^18,50 Here, we speculated that the activation of bilateral striatum might act as a brain endophenotype under stress in working memory tasks. Poor stress adaptability and abnormal striatal activity present simultaneously in patients with SCZ due to the disrupted dopamine-stress regulations.⁵¹ As for healthy adults, people with suppressed striatal activity were more inclined to show lower vulnerability (or higher stress adaptability).

The neural mechanism of social stress affecting cognitive function and its potential molecular genetic mechanism supported the social defeat hypothesis of schizophrenia to a certain extent. The close link between brain features of psychosocial stress susceptibility and genetic risk for schizophrenia suggests that the vulnerable trait to acute stress might lead to progressive dysregulation and the onset of psychosis, which perhaps results from the sensitization of dopamine system affected by both genes and environmental factors. Managing the environmental stressors and, even more importantly, learning to use a variety of coping styles, according to the stressor and context could be useful strategies for improving individual mental health. The coping styles could be considered as candidate endophenotypes for psychiatric disorders and interact with exposure to stress.⁵² Reducing negative coping, eg, through stress reduction training,⁵³ developing psychosocial treatments by challenging defeatist performance beliefs⁵⁴ may be effective intervention strategies to prevent the development of mental disorders and improve functional recovery of patients.

Several limitations of our study still need to be clarified. First, the sample size of two validation cohorts was relatively small. Our conclusions would be strengthened by a larger sample size of validation cohorts. Second, the remarkable public AHBA gene data and the neuroimaging data were collected from different participants and the AHBA only included data of bilateral hemisphere for two participants. The lack of alternative gene expression datasets with a similar spatial resolution limited our ability to validate the findings in an independent transcriptomic dataset. Third, we did not include the determination of related hormone levels^55,56 and heart rate as the objective assessments under social stress in the task. Additional research is required to clarify the relationships among cognitive performance, brain activity, hormone signings, and genetic variations, which may help to form a more integrated network.

In conclusion, individual differences in behavioral response during working memory manipulation were robustly associated with striatal activation changes during social stress. The transcriptome analysis suggested that schizophrenia-related genes were strongly linked to stress-induced activation changes. Higher risk burden of schizophrenia might play a role in stress-induced individually behavioral differences and striatal activity.

Acknowledgments

We extend our gratitude to all subjects who participated in this study. The authors have declared that there are no conflicts of interest in relation to the subject of this study.

Funding

This work was supported by the National Natural Science Foundation of China (81771443, 82071505, 81825009, 81901358, 81221002, 81361120395, 31771186, 82001416), the National Key R&D Program of China (2016YFC1307000), Chinese Institute for Brain Research at Beijing (2020-NKX-XM-12), Chinese Academy of Medical Sciences Research Unit (2019-I2M-5-006), Scientific and Technological Project in Sichuan Province (2020YFSY0042), the US National Institutes of Health (R01MH101053).

References

Author notes