Polygenic Resilience Scores are Associated With Lower Penetrance of Schizophrenia Risk Genes, Protection Against Psychiatric and Medical Disorders, and Enhanced Mental Well-Being and Cognition

Abstract

Background and Hypothesis

In the past decade, significant advances have been made in finding genomic risk loci for schizophrenia (SCZ). This, in turn, has enabled the search for SCZ resilience loci that mitigate the impact of SCZ risk genes. We identified the first genomic resilience profile for SCZ, completely independent from known risk loci for SCZ, though it remains unclear whether resilience loci foster protection against adverse states associated with SCZ involving clinical, cognitive, and brain-structural phenotypes.

Study Design

We analyzed genomic and phenotypic data from 459 784 participants from the UK Biobank, using regression models to estimate interaction effects of resilience and SCZ risk scores on phenotypes spanning multiple scales.

Study Results

We found that resilience loci for SCZ afforded protection against lifetime psychiatric (schizophrenia, bipolar disorder, anxiety, and depression) and medical disorders (such as type 2 diabetes, cardiovascular, and digestive and metabolic disorders). Resilience loci also moderated the impact of SCZ loci, associated with protection against self-harm behavior and greater fluid intelligence scores. The main effects of resilience are also observed in the absence of a moderating effect in some instances, such as associations with larger brain structures.

Conclusions

Overall, this study highlights a wide range of phenotypes that are significantly associated with resilience loci within the general population, revealing distinct patterns separate from those associated with SCZ risk loci. Resilience loci may protect against serious psychiatric and medical outcomes, comorbidities, and cognitive impairment. Therefore, it is conceivable that resilience loci facilitate adaptive processes linked to improved health and life expectancy.

Introduction

Over the past decade, genome-wide association studies (GWASs) have found many loci contributing to the risk for schizophrenia (SCZ). These efforts led to the development of a polygenic risk score, which accounts for ~20% of the genetic liability toward SCZ.¹ The identification of reliable risk markers, in turn, paved the way for studies of the genomic underpinnings of resilience.

Resilience has been defined using various conceptualizations, drawing from perspectives encompassing genetic, epigenetic, environmental, lifestyle, and psychosocial factors.^2–7 Psychological resilience stands out as one of the earliest and foremost conceptualizations of resilience, aiming to describe protective mechanisms that enable individuals to successfully adapt to and recover from serious adversities in life.^3,6 Several psychological factors are thought to influence individuals’ resistance to stressful or traumatic situations, including personality strengths, social support, and physical well-being, among others.⁸ Three prevailing definitions outlined in the literature characterize resilience within the following frameworks: resilience as a capacity (ie, described as pre-existing resiliency factors that instill a general potential for resilience), resilience as a process (ie, resilience emerging as a dynamic interplay between protective and predisposing factors, which unfolds following exposure to adversity), and resilience as an outcome (ie, resilience factors promoting broadly protective adaptive processes among those exposed to adversity, such as the absence of psychopathology among those at exceedingly high levels of risk).³ Resilience is increasingly viewed as a general capacity that dynamically shields individuals against unfolding risk factors throughout life, fostering positive outcomes. This revised perspective transcends multiple complementary definitions of resilience.

In descriptions of the genetics of resilience, it has been suggested that certain genetic variants may act as buffers against the impact of risk genes associated with psychopathology. Recent studies of early-onset Alzheimer’s disease in a densely affected Colombian kindred carrying an autosomal dominant mutation in the PSEN1 gene, have highlighted a prominent example of genetic resilience. A rare variant in the APOE3 gene was identified in a woman who, despite having the PSEN1 mutation, staved off dementia for nearly 3 decades.⁹ This concept of genetic resilience helps explain why some individuals, despite having high genetic susceptibility, are spared from disease. While no such rare single-gene modifications influencing resilience to schizophrenia have been documented, our recent work has uncovered the first-ever evidence of resilience-conferring common genetic variants among unaffected individuals at high risk for schizophrenia.^10,11 According to this perspective, single nucleotide polymorphisms (SNPs) associated with better clinical outcomes can be classified as “resilience factors,” as they instill a capacity to enhance resilience to psychopathology by mitigating the effects of risk loci. Recently, we discovered the first-known genetic resilience profile for SCZ, which comprised thousands of protective alleles at SNPs enabling some individuals in the population to avoid developing SCZ despite carrying an elevated genetic predisposition.¹⁰ We also found a significant positive correlation between resilience and risk scores, which was seen in unaffected individuals at high genetic risk for SCZ yet absent in affected individuals with comparable genetic risk. Hence, we posited that, in unaffected individuals, genomic resilience increases with their genomic risk to counteract the effects of risk loci, thereby conferring resistance against SCZ. The discovery of the first-ever resilience score for SCZ marks a significant milestone in the emerging field of resilience genetics within psychiatry. Our current understanding of resilience genetics largely stems from previous studies of SCZ resilience and is further enriched by insights from longstanding research into psychological factors contributing to resilience. However, many aspects of SCZ resilience remain unresolved due to insufficient sample sizes, limiting power to detect resilience loci with genome-wide significance, and obscure pathways influenced by these resilience loci. While it is known that resilience loci account for a small but significant portion of the explainable variation in resilience status (0.07%), improvements in SNP-based heritability estimates for these loci could be achieved with larger sample sizes. Furthermore, no studies have yet explored the genetic overlap between SCZ resilience loci and those associated with other psychiatric or medical conditions.

This study was conceived as a crucial step toward understanding how resilience loci influence psychiatric and medical outcomes related to SCZ. Specifically, we aimed to determine whether loci associated with resilience to SCZ were associated with enhanced overall well-being, or if their effects were confined to moderating the risk of developing SCZ. Addressing these questions would help to advance our knowledge of resilience genetics and set the stage for future research, ultimately leading to more informed and effective strategies for deciphering resilience loci.

In this study, we tested the hypothesis that our SCZ resilience score would moderate the impact of SCZ risk on phenotypes and outcomes associated with schizophrenia among 487 409 participants in the UK Biobank. We analyzed extensive genome-wide SNP data in conjunction with lifetime 10th revision of the International Classification of Diseases diagnoses gathered from hospital reports, fluid intelligence scores, educational attainment, self-reported mental health outcomes, self-reported lifetime self-harm measures relevant to suicide, and neuroimaging-derived phenotypes of whole-brain and brain-regional sizes. Taken together, our study provides new knowledge about the moderating effects of resilience scores spanning multiple diagnoses and clinically relevant outcomes within a population-representative sample from the United Kingdom.

Methods

A graphical abstract is provided in Figure 1 depicting a simplified overview of the current study.

Figure 1.

A Graphical Abstract of the Current Study. We Investigated Main and Interactive Effects of Schizophrenia (SCZ) Risk and Resilience Scores on 4 Domains: (1) Cognitive Performance and Reserve Measured by Fluid Intelligence and Educational Attainment, (2) Medical Outcomes (Includes Hospital-Recorded ICD-10 Codes), (3) Self-Reported Mental Health Status and Measures of Self-Harm, and (4) Neuroimaging-Derived Phenotypes from Structural Magnetic Resonance Imaging (sMRI) Scans. Data for This Investigation was Obtained from the UK Biobank. Images for this Figure were Generated with BioRender

Datasets and Quality Control

We accessed genome-wide genotypes and phenotypic data from 487 409 participants from the UK Biobank. The genome-wide genotypic data were received post-imputation to the Haplotype Reference Consortium (HRC) panel. This was the same reference panel used to impute the genotype datasets from the Schizophrenia Working Group of the Psychiatric Genomics Consortium in which we derived the SNP weights for the SCZ risk and resilience-scoring formulae.¹ Basic quality control procedures were applied to these data to exclude uncommon variants with a minor allele frequency (MAF) of <1% and variants found to significantly depart from Hardy-Weinberg equilibrium (P ≤ 1e−06). Further filtering addressed cryptic relatedness and extreme outliers based on heterozygosity. Individuals with an ancestry-adjusted heterozygosity rate greater than 4 standard deviations from the group mean (n = 727) or those found to be closely genetically related (2nd degree or closer) were excluded from our analyses (excluded n = 26 557). We retained 459 784 participants (247 960 females) aged 37-73 years at the time of recruitment. The quality control analyses were conducted using Plink2 and KING (version 2.3.2) software. Demographics for the UK Biobank participants included in our analysis are summarized in Supplementary Table 1. We used the top 5 genetic principal components released by UK Biobank covering all retained 459 784 participants (data field: 22009) throughout our downstream analyses to account for genetic variation due to ancestry.

Polygenic Scoring

For each participant in the UK Biobank, we used existing formulae to compute genetic scores for SCZ risk and resilience, which represent the weighted sum of risk and resilience-associated effects of SNPs from independent loci. We calculated genetic scores using the software Plink2. The polygenic risk and resilience scores were calculated using specific procedures and P-value thresholds defined by previous studies.^10,12 By adhering to these established methods, we sought to ensure both reproducibility and consistency with prior literature. However, it is important to acknowledge that results might differ if different polygenic scoring methods were used.

Adhering to procedures used by the Schizophrenia Working Group of the PGC, we calculated polygenic risk scores for schizophrenia using the P + T scoring method. This approach accounts for 17.5% of the variance in liability for schizophrenia among individuals of European ancestry and also demonstrated explanatory power in other populations, including those of Latino (R² = 0.144), African (R² = 0.032), and East Asian (R² = 0.126) ancestries.¹² Although other polygenic scoring methods (eg, LDpred2, SBayesR, and MegaPRS) show relatively better performance,¹³ the P + T scoring method uniquely distinguishes between effects linked to risk loci and those associated with resilience loci. For risk-scoring, we selected SNPs that had a maximum risk-association P-value threshold of .05, which was found to maximize the explainable variation in SCZ (24 166 SNPs).^1,10 Likewise, for resilience-scoring, we selected resilience SNPs with a maximum resilience-association P-value threshold of .3, which was found to maximize the explainable variation in resilience status (35 098 SNPs).^10,11 Resilience loci were previously found by performing a genome-wide association analysis that compared allele frequencies between unaffected controls with high genetic risk scores for SCZ—specifically, those in the top decile—and SCZ cases with similarly high genetic risk scores. The polygenic resilience score was derived from a set of SNPs that were filtered to exclude any SNPs with a known association with SCZ. Additionally, we pruned away loci that were in linkage disequilibrium (LD) with known SCZ risk loci to maintain orthogonality between resilience and risk scores. The resilience scores were initially derived to reduce the potential for LD between resilience-associated SNPs and risk SNPs for SCZ by solely considering SNPs within low LD regions (R² < 0.2) of risk SNPs. Thus, these resilience scores reflect protective effects from residual genetic variation that remains independent of known risk-related genetic variants for SCZ (including the alternate alleles of risk SNPs), which would theoretically confer protection against SCZ rather than tagging risk effects. For our primary analysis, we focused on resilience scores that were derived using this initial LD pruning threshold of R² < 0.2. However, to ensure that no SNPs of very small risk effects might have unduly influenced our resilience scores, we repeated our analysis using a more stringently defined set of resilience SNPs that exhibit exceptionally low LD with known risk loci, at a threshold of R² < 0.1. We note that it is impractical to filter out variants exhibiting any non-zero LD with known risk loci for SCZ, which would effectively eliminate nearly all variants across the genome and consequently exclude genuine resilience loci. We assessed the similarity of results obtained from our models using the two separate resilience scores.

Comparison of Results Derived from P + T and PRS-CS Methods for Resilience Scoring

We included a post-hoc comparison to assess the generalizability of the P + T scoring method (described in the previous subsection) and PRS-CS with respect to findings revealed for the main effect of resilience scores and their moderating effect on SCZ risk scores.¹⁴ PRS-CS infers posterior SNP effect sizes based on external LD patterns observed in a reference panel. Reference samples from the European subsample of the UK Biobank were used in our PRS-CS analyses. We supplied the PRS-CS method with GWAS summary statistics from our previous study of resilience to SCZ.¹⁰ Known risk SNPs for SCZ (P < .05) and any SNP showing LD with a known risk SNP (R² > 0.2) were discarded from these summary statistics, leaving 521 407 SNPs to be utilized by PRS-CS. We calculated resilience scores in the full set of 459 784 UK Biobank participants with plink2 using the posterior SNP effect sizes derived by PRS-CS. We then re-analyzed all outcomes using the resilience scores obtained with the PRS-CS method and our results from these analyses were compared with our findings obtained from the P + T scoring method via Pearson’s correlation tests.

Statistical Analysis of the Moderating Effects of Resilience on Schizophrenia Diagnosis

To investigate whether resilience loci moderate the impact of risk loci in protecting against SCZ, we employed a logistic regression model to assess both the main and interaction effects of resilience and SCZ risk scores on diagnosis of SCZ based on hospital-reported lifetime 10th revision of the International Classification of Diseases (ICD-10) codes from UK Biobank participants (n = 459 784, females = 247 960) ranging from 37 to 73 years old at recruitment. We calculated Elixhauser Comorbidity Index scores,^15,16 a weighted measure of overall comorbidity rates that predict mortality,¹⁷ for each UK Biobank participant using the R package comorbidity. The diagnosis of SCZ was excluded from the calculation of the Elixhauser Comorbidity Index to avoid collinearity with the outcome measure in our statistical analysis. This comorbidity index was included as a covariate in our regression model to account for the overall health state. In addition, we included age at recruitment as a covariate, along with self-reported sex assigned at birth, and the top 5 principal components derived from genome-wide genotypes to account for differences due to ancestry.

Statistical Analyses of Lifetime Medical Diagnoses

We used logistic regression models to estimate the main and interaction effects of resilience and SCZ risk scores on hospital-reported lifetime medical outcomes, diagnoses, and comorbidities recorded in ICD-10 codes for each UK Biobank participant. To account for the relatively low prevalence of disease outcomes within the UK Biobank, which tended to skew towards older and healthier participants,¹⁸ we applied a 2-pronged approach. Several disorders and conditions of interest to our investigation had exceedingly low numbers of cases (dozens to a low hundreds), hence we applied a grouping strategy to increase the number of positive outcomes based on collections of related ICD-10 diagnoses. We referred to these downstream as “broadly defined ICD-10 diagnoses.” Specifically, we aggregated participants who had any diagnoses within a particular chapter, referring to those individuals as “cases” and those without any diagnoses within the same chapter as “controls.” An exception was made for ICD-10 Chapter XX (“External causes of morbidity and mortality”), wherein we aggregated diagnoses of participants based on 4 sub-chapters. These sub-chapters delineate various sources of morbidity or mortality as designated by ICD-10 codes starting with letters “V” (such as transportation-related accidents), “W” (such as falls or environmental exposures), “X” (such as unspecified exposures, poisoning, intentional self-harm), and “Y” (such as adverse outcomes related to medical care or sequalae of other origins). We analyzed data from a total of 17 ICD-10 chapters and sub-chapters using logistic regression models. We used logistic regression models to estimate the interaction effects of resilience and SCZ risk scores in the presence of their main effects and covariates. We specified the top 5 principal components as covariates to adjust for genetic variation due to ancestry. Additionally, we included covariates to adjust for age at recruitment, self-reported sex assigned at birth, and whether the participant had a hospital-recorded lifetime diagnosis of SCZ based on the ICD-10 coding “F20” (data field: 41202).

Additionally, we examined select ICD-10 codings that contained a minimum of 500 cases, using logistic regression models to estimate the main and interaction effects of resilience and SCZ risk scores. We refer to these downstream as “narrowly defined ICD-10 diagnoses.” This analysis included neuropsychiatric disorders (bipolar disorder [BD], depression, anxiety, substance abuse), neurodegenerative diseases (Alzheimer’s disease, vascular dementia, and unspecified dementia), chronic inflammatory, or immunologic conditions (Crohn’s disease, asthma, primary biliary cirrhosis [also known as “primary biliary cholangitis” or “fibrosis and cirrhosis of the liver”]), and metabolic conditions (type I and type II diabetes mellitus). These conditions were selected for analysis as they have been observed to have significant genetic overlap with SCZ.^19–22 We specified the top 5 principal components as covariates to adjust for genetic variation due to ancestry. Additionally, we included covariates to adjust for age at recruitment, self-reported sex assigned at birth, and whether the participant had a hospital-recorded lifetime diagnosis of SCZ based on the ICD-10 coding “F20” (data field: 41202). In each of the above-described analyses, we adjusted for multiple comparisons using the Benjamini-Hochberg false-discovery rate (FDR)-adjustment procedure. This same significance criteria was applied to all subsequent statistical analyses.

Statistical Analysis of Mental Health Questionnaires

To complement our analyses of ICD-10 clinical codes, we explored the moderating effects of resilience loci in the context of mental health self-reports. Our objective in integrating mental health self-assessments into our study was to capture potential links between resilience loci and mental or behavioral outcomes that may hold clinical relevance, even if those outcomes do not meet the criteria for a formal psychiatric diagnosis. We analyzed data obtained from a questionnaire pertaining to psychosocial factors that included mental health assessments (field ID: 100060) offered to UK Biobank participants upon enrollment, yielding a total of 15 items of mental health status completed by 447 060 participants.²³ Using logistic regression models, we estimated the main and interaction effects of resilience and SCZ risk scores on the likelihood of respondents answering “yes” to each questionnaire item. Covariates included in our regression models were the top 5 genome-wide principal components (PCs) to adjust for genetic variation due to ancestry, participants’ age at recruitment, self-reported sex assigned at birth, Elixhauser comorbidity index, and a dummy-coded variable denoting whether the participant had a hospital-recorded lifetime diagnosis of SCZ based on the ICD-10 coding “F20” (data field: 41202).

Statistical Analyses of Fluid Intelligence and Educational Attainment

We hypothesized that resilience loci might enhance fluid intelligence, which is often impaired among individuals affected with SCZ.^24–26 Fluid intelligence was assessed among UK Biobank participants at enrollment (field ID: 20016, n = 160 002) with each score representing a simple unweighted sum of the total number of correct answers from 13 fluid intelligence questions. We employed multiple linear regression models to estimate the main and interaction effects of resilience and SCZ risk scores on inverse-rank normalized scores for fluid intelligence scores while accounting for ancestry PCs, together with age at assessment, sex of participants, and a dummy-coded variable denoting whether the participant had a hospital-recorded lifetime diagnosis of SCZ based on the ICD-10 coding “F20” (data field: 41202).

Furthermore, we hypothesized that resilience loci might be associated with greater educational attainment. Exploring this hypothesis holds importance as SCZ is known to be associated with lower educational attainment.²⁷ Moreover, greater education is associated with increased levels of cognitive reserve, which protects against cognitive impairment later in life,^28–30 and reduces the risk for mortality.³¹ Educational qualifications (field ID: 6138) were collected from UK Biobank participants at the time of enrollment (n = 480 714). Following the approach described by Ge et al,³² we mapped qualification onto 1 of 7 categories defined by the 1997 International Standard Classification of Education (ISCED) of the United Nations Educational, Scientific, and Cultural Organization. We employed multiple linear regression models to estimate the main and interaction effects of resilience and SCZ risk scores on inverse-rank normalized scores for educational qualifications while accounting for ancestry PCs, age at assessment, sex, and a dummy-coded variable denoting whether the participant had a hospital-recorded lifetime diagnosis of SCZ based on the ICD-10 coding “F20” (data field: 41202).

Statistical Analysis of Self-reported Suicidality

Schizophrenia is associated with a greater lifetime risk for self-harm, suicide behavior, and suicide attempts than the general population.^33–35 Therefore, we hypothesized that resilience loci might moderate the penetrance of SCZ risk genes, protecting against suicidality. To test this hypothesis, we analyzed 2 variables linked to suicidality corresponding to self-reported lifetime measures of self-harm ideation or engagement (ie, “Ever contemplated self-harm” [data field: 20485] and “Ever self-harmed” [data field: 20480]). These variables were obtained from an online mental health questionnaire completed by UK Biobank participants at the time of enrollment, of which 157 297 participants responded. We estimated both the main and interaction effects of resilience and SCZ risk on the likelihood of respondents endorsing each of these self-harm measures. We specified the top 5 genome-wide PCs to adjust for genetic variation due to ancestry, participants’ age at recruitment, participants’ sex, and a dummy-coded variable denoting whether the participant had a hospital-recorded lifetime diagnosis of SCZ based on the ICD-10 coding “F20” (data field: 41202) as covariates in each model.

Statistical Analysis of Structural Brain Phenotypes

Prior studies have observed widespread reductions in cortical and subcortical brain region sizes among individuals with SCZ.^36,37 In light of these results, we performed an exploratory analysis to test the moderating effects of resilience loci on processed imaging-derived phenotypes (IDPs) representing structural profiles for 240 distinct brain parcellations. The IDPs were collected at a single time-point among 4 imaging centers in the United Kingdom (ie, Cheadle, Reading, Newcastle upon Tyne, or Bristol) from the year 2014 onward among a subsample of UK Biobank participants (n = 39 508). A demographic summary of the brain-imaging subsample is provided in Supplementary Table 2. Standardized cortical and subcortical atlases were used to extract left and right hemispheric measures of cortical surface area, volume, thickness, and subcortical regional brain volumes. Brain-regional phenotypes were standardized to a mean of 0 and variance of 1 prior to statistical analysis. We used linear regression models to estimate both the main and interaction effects of resilience and SCZ risk scores on brain-regional sizes. The covariates included in our regression models were: (1) the top 5 genotype-based principal components to adjust for variation in ancestry, (2) self-reported sex, (3) age at brain-imaging assessment, (4) site of imaging assessment, (5) weight, height, and body mass index (BMI) at time of imaging assessment, (6) a volumetric scaling factor for T1-head image to standard space (data field: 25000), and (7) variation in location of the scanner and head position described in three-dimensional coordinates (data fields: 25756, 25757, 25758).³⁸ The main effects of resilience and risk were also evaluated in separate models that excluded their interaction effect.

Post-hoc Statistical Analysis of Structural Brain Phenotypes and Well-being

Based on the results from our analysis, we performed a post-hoc analysis to investigate the association between self-reported mental well-being and resilience-associated brain-imaging phenotypes. We hypothesized that enhanced well-being is associated with resilience-related changes in brain structure. We computed “well-being” scores for each UK Biobank in the brain-imaging subsample (n = 39 508). The “well-being” score was derived by summing the items on the mental health assessment (field ID: 100060) for which respondents answered “No” followed by an inverse-rank transformation of raw scores. A higher “well-being” score indicates fewer self-reported mental health issues. We tested the hypothesis that enhanced well-being is mediated by brain changes associated with resilience scores using a linear regression model. This model included the main effects of resilience scores and resilience-associated brain-imaging phenotypes, along with their interaction. The covariates included in our regression models were: (1) the top 5 genotype-based principal components to adjust for variation in ancestry, (2) self-reported sex, (3) age at brain-imaging assessment, (4) site of imaging assessment, (5) weight, height, and body mass index (BMI) at time of imaging assessment, (6) a volumetric scaling factor for T1-head image to standard space (data field: 25000), and (7) variation in location of the scanner and head position described in three-dimensional coordinates (data fields: 25756, 25757, 25758).

Congruence of Effects Between the Full and Western European Subsample

We sought to determine if the main and interaction effects of resilience and SCZ risk scores observed in the full sample of UK Biobank participants held up in a subsample of individuals with a Western European background. We re-analyzed all data using a subsample of participants that have a self-reported white British, white Irish, or other white background (field ID: 21000; codes 1, 1001, 1002, or 1003). We used Pearson’s correlation tests to measure the congruence of main and interaction effects observed in the full sample compared with those observed in the Western European subsample.

Results

Across all 299 outcome measures that we analyzed, we observed strong, significant agreement between the main effects of the two resilience scores using LD R² threshold of 0.2 and 0.1 (Pearson’s r = 0.91, P = 1.79 × 10⁻¹¹⁵) (Supplementary Figure 1). Additionally, we found strong and highly statistically significant concordance between our results obtained by the P + T and PRS-CS methods with respect to the main effects of resilience scores (Pearson’s r = 0.74, P = 1.1 × 10⁻⁵³) and their moderating effects (Pearson’s r = 0.98, p = 6.7 × 10⁻²⁰⁷) (Supplementary Figure 2). Henceforth, we focus on the results from our models that used resilience scores calculated using the LD R² threshold of 0.2, which had better statistical power relative to other pruning thresholds, and was optimized to the distribution of SCZ risk SNP effects identified using the P + T scoring method.

Resilience scores showed an opposite direction of effect than SCZ risk scores for 193 of the outcomes examined. Furthermore, the main effects of resilience and SCZ risk scores were significantly negatively correlated across all outcomes (Pearson’s r = −0.36, p = 1.5 × 10⁻¹⁰) (Supplementary Figure 3). These findings lend support to our hypothesis that resilience scores moderate the penetrance of SCZ risk genes.

All the regression models in our study included the top 5 genome-wide PCs as covariates to adjust for potential confounding relationships due to genetic variation between ancestries. Our primary rationale for including PCs as covariates in our models was to adjust for possible admixture and population stratification, which could artificially inflate the effects and increase the number of false-positives. We observed widespread, significant associations between ancestry-related PCs and the outcome measures included in our study (Supplementary Figures 4–9).

Validation that Resilience Scores Protect Against Schizophrenia

Resilience scores were found to significantly interact with SCZ risk scores, with this interaction associated with a reduced likelihood of lifetime diagnosis of SCZ in the UK Biobank (Figure 2A) (interaction: odds ratio [OR] = 0.94, 95% confidence interval [CI] 0.90-0.98, z-value = −2.97, P = .0029). This finding independently validates our previously reported results,^10,11 demonstrating that higher resilience scores mitigate the impact of SCZ risk loci, thereby protecting against SCZ. Having validated this relationship, we sought to improve our understanding of resilience by exploring its associations with other medical disorders, mental well-being, education, fluid intelligence, and brain structure.

Figure 2.

Line Graphs Depicting Statistically Significant Interactive Effects of Schizophrenia (SCZ) Risk and Resilience Scores on Hospital-Reported Time Diagnoses Among UK Biobank Participants Spanning Neuropsychiatric, Immunologic, and Metabolic Disorders Based on the 10th revision of the International Classification of Diseases (ICD-10). All Records were Obtained from UK Biobank Participants at the Time of Enrollment and Reflect Lifetime Diagnostic Codes. The Results Shown in this Plot for Resilience Scores Correspond to the Scores Calculated Using Resilience-Associated SNPs that Exhibit Minimal Linkage Disequilibrium with Mild-Risk SNPs for SCZ (R² Value of 0.2 or Less). (A) Line Graphs Depicting Significant Interaction Effects of Resilience and SCZ Risk Scores on Lifetime ICD-10 Diagnosis of Schizophrenia. The y-Axis Shows the Fitted Probabilities of Schizophrenia Diagnosis. The Solid Line Denotes the Effect of SCZ Risk Among Individuals with a Resilience Score 1 Standard Deviation Above the Sample Mean. The Dashed Line Denotes the Effect of SCZ Risk at a Mean Level of Resilience. The Dotted Line Denotes the Effect of SCZ Risk Among Individuals with a Resilience Score 1 Standard Deviation Below the Sample Mean. (B) Line Graphs Depicting Significant Interaction Effects of Resilience and SCZ Risk Scores on Narrowly Defined ICD-10 Diagnoses. The y-Axis Shows the Fitted Probabilities of the ICD-10 Diagnoses. The Solid Blue Line Denotes the Effect of SCZ Risk Among Individuals with a Resilience Score 1 Standard Deviation Above the Sample Mean. The Dashed Light-Red Line Denotes the Effect of SCZ Risk at a Mean Level of Resilience. The Dotted Green Line Denotes the Effect of SCZ Risk Among Individuals with a Resilience Score 1 Standard Deviation Below the Sample Mean. (C) Line Graph Depicting the Significant Interactions of Resilience and SCZ Risk Scores on Probability of any ICD-10 Diagnoses per Chapter. The y-Axis Show the Probabilities of Chapter-Specific ICD-10 Diagnoses Among UK Biobank Participants. The x-Axis Represents the Standardized SCZ Risk Scores. Each Line Represents the Effect of SCZ Risk Scores on the Likelihood of Participants Having any ICD-10 Diagnoses Based on Levels of Resilience. The Solid Blue Line Depicts the Effect of SCZ Risk Scores Among Individuals with Resilience 1 Standard Deviation Above the Sample Mean. The Dashed Light-Red Line Depicts the Effect of SCZ Risk Scores at Mean Levels of Resilience. The Dotted Green Line Depicts the Effect of SCZ Risk Scores Among Individuals with Resilience Scores 1 Standard Deviation Below the Sample Mean. The Lines in the Plot Depict the Effects of SCZ Risk at Different Levels of Resilience. We Note that the y-Axis for all Plots was Scaled According to Each Outcome Variable to Enhance Visualization of the Interaction Effects

Resilience Scores Moderate Effects of SCZ Risk on Narrowly Defined ICD-10 Diagnoses

We found 7 narrowly defined ICD-10 diagnoses wherein resilience scores interacted significantly with SCZ risk scores, reducing the likelihood of diagnosis (FDR_P < .05) (Figure 2B, Supplementary File 1). Significant interaction effects were associated with protection against the following psychiatric diagnoses: BD (interaction: OR = 0.93, 95% CI 0.89-0.98, z-score = −2.65, FDR_P = .029), depressive episodes (interaction: OR = 0.96, 95% CI 0.95-0.98, z-value = −4.8, FDR_P = 1.1 × 10⁻⁵), anxiety disorders (interaction: OR = 0.97, 95% CI 0.95-0.98, z-value = −3.6, FDR_P = 1.4 × 10⁻³), tobacco abuse (interaction: OR = 0.94, 95% CI 0.93-0.95, z-value = −7.9, FDR_P = 5.8 × 10⁻¹⁴), alcohol abuse (interaction: OR = 0.93, 95% CI 0.90-0.96, z-value = −5.1, FDR_P = 3.9 × 10⁻⁶), and specific personality disorders (interaction: OR = 0.85, 95% CI 0.78-0.92, z-value = −3.89, FDR_P = 5.4 × 10⁻⁴). These findings suggest that resilience scores moderate the penetrance of SCZ risk loci, providing cross-protection against multiple psychiatric outcomes related to SCZ. Additionally, we found a significant interaction between SCZ risk and resilience scores, associated with protection against non-insulin-dependent (type 2) diabetes mellitus (interaction: OR = 0.99, 95% CI 0.98-0.99, z-value = −2.4, FDR_P = .048).

Resilience Scores Moderate Effects of SCZ Risk on Broadly Defined ICD-10 Diagnoses

As presented in Figure 2C, we found significant interaction effects between resilience and SCZ risk scores influencing risk for 11 categories of hospital-reported medical diagnoses, grouped by ICD-10 chapters/sub-chapters. All summary statistics from our regression models are provided in Supplementary File 1. High levels of resilience scores attenuated the penetrance of SCZ risk genes, conferring protection against the following categories of diseases and/or co-morbid outcomes:

• any mental and behavioral disorders (ICD-10 codes from Chapter V with prefix “F”) (interaction: OR = 0.96, 95% CI 0.95-0.97, z-value = −6.98, FDR_P = 4.8 × 10⁻¹¹),

• any disease of the genitourinary system (ICD-10 codes from Chapter XIV with prefix “N”) (interaction: OR = 0.986, 95% CI 0.98-0.99, z-value = −3.48, FDR_P = .0011),

• any diseases of the musculoskeletal system and connective tissue (ICD-10 codes from Chapter XIII with prefix “M”) (interaction: OR = 0.98, 95% CI 0.97-0.99, z-value = −5.8, FDR_P = 5.9 × 10⁻⁸),

• any diseases of the respiratory system (ICD-10 codes from Chapter X with prefix “J”) (interaction: OR = 0.978, 95% CI 0.97-0.99, z-value = −5.02, FDR_P = 2.8 × 10⁻⁶),

• any diseases of the nervous system (ICD-10 codes from Chapter VI with prefix “G”) (interaction: OR = 0.98, 95% CI 0.97-0.99, z-value = −3.97, FDR_P = 2.3 × 10⁻⁴),

• any endocrine, nutritional, and metabolic diseases (ICD-10 codes from Chapter IV with prefix “E”) (interaction: OR = 0.98, 95% CI 0.98-0.99, z-value = −3.9, FDR_P = 2.3 × 10⁻⁴),

• any disease of the digestive system (ICD-10 codes from Chapter XI with prefix “K”) (interaction: OR = 0.989, 95% CI 0.98-0.99, z-value = −3.2, FDR_P = .0026),

• any diseases of the circulatory system (ICD-10 codes from Chapter IX with the prefix “I”) (interaction: OR = 0.984, 95% CI 0.98-0.99, z-value = −3.85, FDR_P = 2.8 × 10⁻⁴),

• any diagnoses broadly defined as symptoms, signs, and abnormal clinical and laboratory findings, not elsewhere classified (ICD-10 codes from Chapter XVII with prefix “R”) (interaction: OR = 0.98, 95% CI 0.977-0.99, z-value = −4.4, FDR_P = 5.13 × 10⁻⁵),

• any external causes of morbidity and mortality (ICD-10 codes from Chapter XX with prefix “X”) (interaction: OR = 0.97, 95% CI 0.95-0.99, z-value = −2.77, FDR_P = .0088),

• any instances of injury, poisoning, or other consequence of external causes (ICD-10 codes from Chapter XIX with prefix “T”) (interaction: OR = 0.982, 95% CI 0.97-0.99, z-value = −3.16, FDR_P = .0026).

Resilience Promotes Greater Educational Attainment and Fluid Intelligence

We found that high levels of resilience scores moderated the penetrance of SCZ risk genes, promoting higher fluid intelligence scores (interaction: β = .024, t-value = 9.4, FDR_P = 5.01 × 10⁻²¹) and greater educational attainment (interaction: β = .017, t-value = 11.76, FDR_P = 1.3 × 10⁻³¹) (Figure 3, Supplementary File 1).

Figure 3.

Line Graphs Depicting Statistically Significant Interactive Effects of Schizophrenia (SCZ) Risk and Resilience Scores Educational Attainment and Fluid Intelligence Scores Obtained from UK Biobank Participants at the Time of Enrollment. The Results Shown in this Plot for Resilience Scores Correspond to the Scores Calculated Using Resilience-Associated SNPs that Exhibit Minimal Linkage Disequilibrium with Mild-Risk SNPs for SCZ (R² Value of 0.2 or Less). The y-Axis Shows the Fitted Probability Values for Each Outcome Measure. The Lines in the Plot Depict the Effects of SCZ Risk at Different Levels of Resilience. The Solid Line Denotes the Effect of SCZ Risk Among Individuals with a Resilience Score 1 Standard Deviation Above the Sample Mean. The Dashed Line Denotes the Effect of SCZ Risk at a Mean Level of Resilience. The Dotted Line Denotes the Effect of SCZ Risk Among Individuals with a Resilience Score 1 Standard Deviation Below the Sample Mean. We Note that the y-Axis for all Plots was Scaled According to Each Outcome Variable to Enhance Visualization of the Interaction Effects

Resilience Attenuates Risk for Self-Harm and Enhance Self-reported Mental Well-being

We found that resilience scores significantly interact with SCZ risk scores, fostering enhanced well-being for 9 mental health items spanning mood instability, anxiety, and depressive-related measures, such as anxious feelings (interaction OR = 0.99, 95% CI 0.98-0.99, z-value = −2.19, FDR_P = .048), fed-up feelings (interaction: OR = 0.97, 95% CI 0.97-0.98, z-value = −5.73, FDR_P = 5.08 × 10⁻⁷), being seen by a psychiatrist for nerves, tension, and/or depression (interaction: OR = 0.96, 95% CI 0.95-0.97, z-value = −7.01, FDR_P = 2.97 × 10⁻¹⁰), being seen by a doctor for nerves, tension, and/or depression (interaction: OR = 0.97, 95% CI 0.96-0.98, z-value = −7.3, FDR_P = 7.98 × 10⁻¹¹), as well as miserableness (interaction: OR = 0.976, 95% CI 0.97-0.983, z-value = −6.3, FDR_P = 3.1 × 10⁻⁸), to name a few (Figure 4A, Supplementary File 1).

Figure 4.

Resilience Scores Moderate the Penetrance of Schizophrenia Risk Scores, Enhancing Self-Reported Mental Well-being and Reducing Self-Harm Behaviors. A Graphical Representation of Statistically Interactive Effects of Schizophrenia (SCZ) Risk and Resilience Scores on Self-Reported Indicators of Mental Health Provided by UK Biobank Participants at the Time of Enrollment. All Measures Included in our Analysis were Scored as Binary (Yes/No) Events. The Results Shown in this Plot for Resilience Scores Correspond to the Scores Calculated Using Resilience-Associated SNPs that Exhibit Minimal Linkage Disequilibrium with Mild-Risk SNPs for SCZ (R² Value of 0.2 or Less). (A) Line Plots Depicting the Significant Interaction Effects of Resilience Scores and SCZ Risk Scores on Self-Reported Mental Health Measures. The x-Axis Represents the Standardized SCZ Risk Scores. The y-Axis Shows the Fitted Probability Values of Self-Reported Mental Health Measures for Which Resilience and SCZ Risk Scores Exhibited a Significant Interaction Effect. The Lines in the Plot Depict the Effects of SCZ Risk at Different Levels of Resilience. The Solid Line Denotes the Effect of SCZ Risk Among Individuals with a Resilience Score 1 Standard Deviation Above the Sample Mean. The Dashed Line Denotes the Effect of SCZ Risk at a Mean Level of Resilience. The Dotted Line Denotes the Effect of SCZ Risk Among Individuals with a Resilience Score 1 Standard Deviation Below the Sample Mean. (B) Line Plots Depicting the Significant Interactions of Resilience and SCZ Risk Scores on Probability of Self-Reported Suicidality Measures. The y-Axis Shows the Probabilities of Both Lifetime Engagement in Self-Harm and Ideation of Self-Harm Among UK Biobank Participants. The x-Axis Represents the Standardized SCZ Risk Scores. Each Line Represents the Effect of SCZ Risk Scores on Likelihood of Participants Any Suicidality Based on Levels of Resilience. The Solid Blue Line Depicts the Effect of SCZ Risk Scores Among Individuals with Resilience 1 Standard Deviation Above the Sample Mean. The Dashed Light-Red Line Depicts the Effect of SCZ Risk Scores at Mean Levels of Resilience. The Dotted Green Line Depicts the Effect of SCZ Risk Scores Among Individuals with Resilience Scores 1 Standard Deviation Below the Sample Mean. We Note that the y-Axis for All Plots Was Scaled According to Each Outcome Variable to Enhance Visualization of the Interaction Effects

In addition, our findings revealed a mental well-being item for which resilience scores demonstrated a significant main effect (FDR_P < .05), indicating protection against feelings of nervousness (OR = 0.97, 95% CI 0.949-0.992, z-value = −2.63, FDR_P = .025). However, the interaction effect between resilience and SCZ risk scores for these outcomes was not significant (Supplementary Figure 10 and File 1).

Resilience and SCZ risk scores showed interaction effects associated with protection against self-harm ideation (interaction: OR = 0.95, 95% CI 0.928-0.973, z-value = −4.24, FDR_P = 1.98 × 10⁻⁵) and engagement (interaction: OR = 0.92, 95% CI 0.88-0.96, z-value = −4.3, FDR_P = 1.67 × 10⁻⁵) (Figure 4B, Supplementary File 1).

Resilience Scores Promote Larger Whole-Brain and Brain-Regional Brain Sizes

No significant interaction effects between resilience scores and SCZ risk scores were seen for brain-imaging phenotypes (Supplementary File 1). However, we found 5 brain imaging-derived phenotypes for which significant main effects were observed solely for resilience scores, with SCZ risk scores showing non-significant effects. High levels of resilience were found to be associated with increased size of multiple brain regions, including thickness of the pars opercularis, thickness of the rostral middle frontal cortex (left and right hemispheres), volume of the rostral middle frontal cortex (left and right hemispheres) (Figure 5A, Supplementary File 1).

Figure 5.

A Graphical Representation of Main and Interactive Effects of Schizophrenia (SCZ) Risk and Resilience Scores on Brain Imaging-Derived Phenotypes Collected from Structural Magnetic Resonance Imaging (sMRI) of 39 508 UK Biobank Participants. The Results Shown are for sMRI Phenotypes that Exhibited a Significant Association with the Main Effect of Resilience Scores. The Results Shown in this Plot for Resilience Scores Correspond to the Scores Calculated Using Resilience-Associated SNPs that Exhibit a Maximum Linkage Disequilibrium with Mild-Risk SNPs for SCZ at an R² Value of 0.2. (A) Bar-Plots Showing the Main Effect of Resilience and SCZ Risk Scores on Brain-Imaging-Derived Phenotypes for Which a Significant Main Effect of Resilience was Observed. No Significant Interaction Effect of Resilience and SCZ Risk was Observed for any sMRI Phenotype, nor were Any Significant Associations Observed for the Main Effect of SCZ Risk Scores. The Bar-Plot on the Left Depicts the Main Effects of Resilience and SCZ Risk Scores, Excluding Their Interactive Effect, for Each of the 5 Corresponding Brain Imaging-Derived Phenotypes (y-Axis). The Main Effect of Resilience and SCZ Risk is Presented in Terms of the Odds Ratio (OR) on the x-Axis. The 95% Confidence Intervals of the OR are Shown as Error Bars in the Bar-Plot. Non-significant Main Effects (FDR_P > .05) are Denoted by a Red “x.” The Heat-Map Shown on the Right Side of the Plot Presents the Significance (FDR-Adjusted P Values, −log₁₀ Transformed) of the Main Effects of Resilience and SCZ Risk Scores in the Presence of Their Interactive Effect. Similar to the Bar-Plot, Non-significant Main, or Interactive Effects are Denoted by a Red “x.” (B) Anatomical Brain Plots Showing the Main Effects of Resilience and SCZ Risk Scores for Brain-Regional Sizes for all 240 Distinct Cortical and Subcortical Parcellations. Shading Reflects the Magnitude and Direction of the Main Effects for SCZ Resilience and Risk Scores, with Darker Blue Indicating a Stronger Positive Association (ie, Higher Scores Associated with Larger Regional Sizes) and Brighter Yellow Indicating a Stronger Negative Association (ie, Higher Scores Associated with Smaller Regional Sizes). Regions are Labeled that Show an Association with a Lifetime Hospital-Reported ICD-10 Diagnosis of Schizophrenia at a Significance Level of FDR_P < .1. A Red Asterisk (*) Denotes Regions Associated with Resilience Scores at a Significance of FDR_P < .05. (C) Scatterplots Comparing the Effects of Resilience Scores on Brain Imaging-Derived Phenotypes vs Risk Scores for Schizophrenia (Left) and Diagnosis of Schizophrenia (Right). Best-Fit Regression Lines are Shown in Each Panel. Pairwise Pearson’s Correlation Tests were Performed in Each Panel to Measure the Relationship Between the Effects Shown on the Horizontal and Vertical Axis in Each Panel. Uncorrected P Values from Pearson’s Correlation Tests are Provided in Each Panel

None of the IDPs measured in UK Biobank participants were statistically significantly associated with a lifetime diagnosis of SCZ after correcting for multiple comparisons. However, we identified 13 parcellations associated with SCZ diagnosis at an FDR_P < .1. These included reduced volumes of 10 cortical parcellations, including left hemisphere of the rostral middle frontal cortex (t-value = −3.34), left hemisphere of the cortex (t-value = −3.23), whole total gray matter (t-value −3.1), right hemisphere of the pars orbitalis (t-value = −3.09), right hemisphere of the posterior cingulate cortex, whole-brain segmentation volume excluding the ventricles, left hemisphere of the inferior temporal gyrus, left hemisphere of the precuneus (t-value = −2.9), and right hemisphere of the precuneus (t-value = −2.88). Schizophrenia was also associated with a reduction in cortical thickness of the left hemisphere of the pars opercularis (t-value = −2.85) and superior frontal gyrus (t-value = −2.79). In addition, schizophrenia was associated with a larger volume of the 3rd ventricle (t-value = 2.99).

Figure 5B provides an anatomical representation of the main effects of resilience and SCZ risk scores, illustrating that higher resilience scores were associated with widespread increases in cortical brain thickness, regional increases in surface area particularly within the frontal and temporal lobe, and increases in subcortical volume. Conversely, high SCZ risk scores were associated with widespread reductions in cortical and subcortical brain-regional sizes. Brain regions associated with a lifetime diagnosis of SCZ are presented in Figure 5B. Effects of SCZ diagnosis on IDPs were positively correlated with those observed for SCZ risk scores (Pearson’s r = 0.15, P = .017). In contrast, the effects observed for resilience scores on IDPs were negatively correlated with effects observed for SCZ risk scores for cortical surface area and thickness, and subcortical volume (Figure 5C). Effects of resilience scores on cortical volume IDPs were not correlated with effects observed for SCZ risk scores (Pearson’s r −0.043, P = .68). Similarly, negative correlations between observed between effects of resilience scores on IDPs and the effects observed for the diagnosis of SCZ (Figure 5C). However, a significant negative correlation was observed for cortical volume IDPs (Pearson’s r = −0.43, P = 7.9 × 10⁻⁶). This relationship may be driven by the larger effects of SCZ on IDPs compared to those associated with SCZ risk scores. We found that the standardized effect sizes observed for SCZ diagnosis on IDPs (mean t-score = −0.87) were significantly larger than the effects observed for SCZ risk scores (mean t-score = −0.095) (2-tailed t-value = 10.8, P = 1.2 × 10⁻²²).

In our post-hoc analysis, no significant associations were found between mental well-being scores and resilience scores, brain-imaging phenotypes, or resilience-moderated changes to brain-imaging phenotypes.

Comparison of Findings Within a Western European Subsample

We observed that the main effects of resilience and SCZ risk scores, as well as their interaction, were consistent between the full UK Biobank dataset and the Western European subsample (Supplementary Figure 11). The main effect of SCZ PRS showed the strongest concordance (Pearson’s r = 0.99, P = 7.69 × 10⁻²⁷⁴) between the full UK Biobank dataset and the Western European subsample, followed by a similarly strong agreement in the main effects of resilience scores (Pearson’s r = 0.98, P = 1.47 × 10⁻²¹⁷). The interaction of resilience and risk scores showed moderate yet highly significant agreement between the full set and the Western European subsample of participants (Pearson’s r = 0.3, P = 9.5 × 10⁻⁸). While no interaction effects remained significant after correcting for multiple comparisons in the Western European subsample, significant main effects of resilience scores were observed for 2 brain-imaging phenotypes and items related to mental health state. These effects included increased mean thickness of the left hemisphere of the rostral middle frontal gyrus (β = .054, SE = 0.016, FDR_P = .039), increased volume of the right hemisphere of the rostral middle frontal gyrus (β = 0.046, SE = 0.013, FDR_P = .021), and decreased likelihood of participants reporting feeling tense/high-strung (OR = 0.952, 95% CI 0.93-0.977, FDR_P = .021), or being seen by a doctor (OR = 0.96, 95% CI 0.94-0.98, FDR_P = .011), or psychiatrist (OR = 0.94, 95% CI 0.91-9.97, FDR_P = .011) for nerves, anxiety, tension, or depression.

Discussion

In prior studies, we posited that genomic loci associated with resilience moderate the influence of risk genes for SCZ, thereby protecting “resilient” individuals who have a high risk for schizophrenia. Our current study aimed to improve our understanding of genetic resilience for SCZ by investigating the main effect of resilience scores, as well as interactive effects of resilience and SCZ risk scores, across multiple genetic and phenotypic scales, including (1) lifetime medical diagnoses, (2) fluid intelligence and educational attainment, (3) self-reported mental well-being outcomes and measures of suicidality, and (4) anatomical measures of whole-brain and brain-regional size.

Our results showed that resilience scores exhibit significant interactions with SCZ risk scores, buffering the impact of SCZ risk genes on a range of outcomes including cognitive indicators, psychological states, and psychiatric and medical outcomes. It is remarkable that for all the interaction plots we see the quintessential signature of resilience. They show resilience protecting those at high risk for outcomes but having no effect at lower levels of risk.

From our analyses that focused primarily on ICD-10 diagnoses as outcome measures, we found that resilience scores significantly attenuated the impact of SCZ risk genes, reducing the likelihood of psychotic (schizophrenia), mood (BD and depression), substance abuse (tobacco and alcohol), and anxiety disorders among UK Biobank participants. Collectively, this is the first-ever reported evidence of genetic resilience profiles for SCZ exhibiting bona fide cross-protection across multiple lifetime neuropsychiatric outcomes.

In light of these results, it is plausible that genetic resilience architectures might mirror the pattern of shared protective genes among neuropsychiatric disorders.³⁹ Initially, resilience scores for SCZ were derived without accounting for co-morbidities or pleotropic effects. Hence, there is potential to expand and refine the SCZ resilience scores to encompass additional cross-protective effects. Such insights might only emerge through joint GWASs of multiple psychiatric disorders. Refinements to our resilience scores, such as integrating additional samples into our analyses to capture risk genes with very small effect sizes and incorporating cases with other psychiatric disorders, may lead to improved detection of moderating effects of resilience scores on shared pleotropic risk genes. Another avenue that is still unexplored is the possibility of disorder-specific resilience factors. Prior studies have identified disorder-specific genetic risk factors for SCZ and BD.⁴⁰ Hence, it is plausible that an analogous yet undiscovered pattern of disorder-specificity exists for resilience loci. Further investigations are needed to explore resilience loci that may differ across psychiatric disorders. Based on our findings, we expect that the resilience scores for SCZ are partly reflective of protective effects against transdiagnostic risk genes shared between SCZ and other mental disorders and medical comorbidities. Medical and psychiatric comorbidity data are not available for the individuals in the datasets that were used to derive the SCZ risk and resilience scores, but prior research indicates a relatively high prevalence of both physical and mental comorbidities in SCZ.^41–44 Therefore, it is likely that the genetic data used to derive the risk and resilience scores for SCZ include individuals with such comorbidities. From a methodological perspective, a possible presence of comorbidities could have influenced the resulting resilience scores, potentially capturing their protective effects against conditions commonly comorbid with SCZ. Detailed medical history data, preferably from prospective cohorts, would be paramount for investigating genetic factors of resilience that may be specific to SCZ and not confounded by comorbid conditions. To address the question of whether resilience loci associated with SCZ are specific to this diagnosis or generalizable across psychopathology, future research should focus on examining resilience loci in the context of cross-disorder relationships. Ensuring the loci used for calculating risk scores remain independent from those putatively associated with resilience is an essential component of our framework. Currently, there is a significant overlap between the list of SNPs used for calculating resilience scores for SCZ^10,11 and those in the best-available scoring formulae for cross-disorder psychopathology.³⁹ Therefore, there is strong interest in developing a resilience score that can effectively capture protective effects against transdiagnostic risk loci associated with different psychiatric diagnoses, as this could significantly advance our understanding of genetic resilience and the landscape of shared resilience factors.

Multiple studies have shown that SCZ is associated with lower rates of educational attainment, lower intelligence scores, and cognitive impairment.^27,45,46 This suggests that SCZ risk genes may influence cognitive functioning and cognitive reserve,^47,48 leading to impairment in cognitive performance and poorer clinical outcomes. Our study showed that resilience scores moderated the penetrance of SCZ risk loci, with their interaction effect promoting greater rates of educational attainment, and higher fluid intelligence scores.

We found that resilience profiles moderated the penetrance of SCZ risk loci for 9 items from the mental well-being questionnaire. Higher resilience scores protected against numerous emotional, psychological, and behavior-related outcomes such as miserableness, loneliness, risk-taking, mania, prolonged episodes of anhedonia, and prolonged depressive symptoms, and depressive symptoms severe enough to prompt a healthcare visit, to name a few. These findings lend support to the hypothesis that high levels of resilience loci may foster enhanced states of psychological resilience, improved mental well-being, and better adaptation to adversity.

Increased genetic vulnerability for SCZ was significantly associated with increased risk for ideation and engagement in self-harm behaviors, in agreement with the well-documented relationship between SCZ and increased risk for suicide attempt, self-harm, and suicide ideation.^33–35 While resilience scores did not exhibit a significant main effect on either suicidality measure, our results demonstrated a significant interaction effect between resilience and SCZ risk. The interaction revealed that high resilience scores offset the adverse effects of SCZ risk, reducing the probability of individuals engaging in or contemplating self-harm. The effect sizes of the interaction term on these self-harm measures indicated that for each unit increase in both risk and resilience scores, the likelihood of self-harm contemplation or engagement decreased by 5%-7%, respectively. This finding holds implications for suicide and psychiatric research. Yet, the number of respondents to the question about lifetime ideation or engagement in self-harm represented about one-third of the total number of participants in the UK Biobank, thus these findings will need to be evaluated in independent cohorts. The rate of attempted suicide is substantially higher among individuals with a pre-existing psychiatric disorder. Furthermore, suicide ranks among the leading causes of death globally, with rates of suicides steadily increasing over the past 2 decades.⁴⁹ Identifying risk factors for suicide serves as a critical step in accurately predicting those at elevated risk and in improving intervention and prevention. Moreover, it remains to be shown whether resilience loci foster protective effects against suicidality independent of the interaction effect with SCZ risk genes.

Due to selection biases, the UK Biobank is skewed towards older, healthier participants.¹⁸ Therefore, our study lacked sufficient statistical power to detect significant associations between resilience scores and individual ICD-10 codes, particularly those outcomes that are already relatively uncommon in the general population, but further skewed to an even rarer frequency within the UK Biobank, such as schizophrenia. Hence, our analytic strategy examined any instances of a condition within individual ICD-10 chapters, with one exception being the ICD-10 Chapter XX wherein we aggregated sub-items of a chapter, to enhance the number of positive outcomes for more cohesively defined external causes of co-morbidity and mortality. This came at a cost of granularity, as we lacked statistical power to link resilience scores to individual diseases or conditions. Yet, our results of diagnostic groups provided several important insights into types of pathology, as well as causes of morbidity and mortality against which resilience scores show protective effects. Our analysis of lifetime medical diagnoses stored in hospital-reported ICD-10 codes for UK Biobank participants showed that resilience scores significantly attenuated the probability of individuals being diagnosed with any diseases among 2 ICD-10 chapters that are chiefly relevant to brain disorders, namely: any mental or behavioral disorder, and any nervous system disorder. In addition, we discovered that resilience scores moderate the effect of SCZ risk scores, reducing the overall probability of participants being diagnosed with conditions not specifically originating in the central nervous system, namely: cardiovascular diseases, disease of musculoskeletal or connective tissues, disease of the genitourinary system, diseases of the digestive tract, and diseases of the endocrine and metabolic system. Moreover, we found that resilience scores moderated the effect of SCZ risk, reducing the probability of individuals succumbing to external causes of morbidity or mortality, which included instances of accidents, injury, assaults, self-harm, medical or surgical complications, or sequalae of other causes. These findings provide insights into the clinical relevance of resilience scores within a population-level sample. Moreover, our results show that the moderating effects of resilience scores extend beyond psychiatric and mental health outcomes. Our findings suggest that resilience loci might install a general capacity to adapt to various adverse conditions, including reducing risk for diseases that manifest in peripheral tissues and organ systems beyond the central nervous system.

A multitude of epidemiological studies show that SCZ is associated with risk for numerous serious medical co-morbidities of relevance to our key findings, including urinary infections^50,51 and sexually transmitted infections,⁵² chronic kidney disease,⁵³ gastrointestinal disease,^54,55 cardiovascular disease,^56,57 and metabolic disorders such as diabetes.⁵⁸ Metabolic dysfunction is a major clinical concern in SCZ as patients are prone to poor dietary habits, unhealthy lifestyles (such as increased smoking behavior and inactivity), and metabolic alterations resulting from the use of second-generation or atypical antipsychotic medications. Epidemiological data also highlights SCZ as a leading cause of morbidity and mortality.⁵⁹ Taken together, our results demonstrate that resilience scores reduce the probability of serious medical outcomes and co-morbidities that have been firmly linked with SCZ in prior work. This underscores the potential clinical and epidemiological significance of our results, as resilience loci may foster protection against adverse outcomes associated with reduced quality of life, increased disability, and premature mortality.

Our study also found significant associations between resilience scores and increased total gray matter and size of multiple cortical brain regions. However, no significant association was observed between brain structure and SCZ genetic liability, nor did we observe a significant interactive effect between resilience and SCZ risk in the context of brain structure. Less than 10% of UK Biobank participants have been currently assessed via structural brain-imaging, thus posing a potential limitation on the power to detect statistically significant effects. The UK Biobank dataset contains significantly fewer cases diagnosed with SCZ compared to prior brain-structural meta-analyses,^36,37 which might have reduced our statistical power to detect moderating effects of resilience scores on brain imaging-phenotypes. The age of participants may factor into our ability to detect significant effects of risk and resilience scores for SCZ on brain-imaging phenotypes with these data. We note that age-related progressive cortical thinning is possible as the average age of participants at the time of MRI assessment was 64 years. Previous studies have reported widespread age-related cortical thinning in the brain.^60,61 These potential age-related changes may have obscured or minimized the effects of genetic scores for risk and resilience to SCZ on brain structure in aging adults. While such effects might not have been found in the current data, they could potentially be observed in younger populations or different cohorts. Regardless, the main effect of resilience scores on increased total gray matter and brain-regional sizes of the cortex hints at potential protective mechanisms against SCZ genetic liability. Furthermore, resilience and SCZ risk scores exhibited distinct patterns of effects on brain structure. Higher resilience scores were associated with widespread increases in thickness and volume of the cortex, as well as selective increases in cortical surface area particularly in the frontal and temporal lobes, and selective increases in subcortical brain volumes. Conversely, SCZ risk scores were associated with a distinct profile of cortical reductions in thickness, as well as selective reductions in cortical surface area and volume, as well as selective reductions in subcortical brain volumes. While many of the brain-region-specific effects of resilience and SCZ risk scores did not reach statistical significance, their directions of effect on brain structure seem to be clearly divergent. In accordance, the Enhancing Neuro Imaging Genetics through Meta-Analysis Consortium found that individuals with SCZ, on average, show widespread reductions in cortical thickness, with multiple brain regions being affected. On average, cortical surface area is shown to be globally reduced in SCZ, with regional differences in surface area appearing comparatively small. Our findings suggest that resilience loci might foster an adaptive increase in cortical brain size to compensate for the adverse effects of SCZ risk genes. We found that resilience scores were associated with increased thickness of the pars opercularis, a region that shows impaired connectivity, reduced brain activation, and reduced size in SCZ.^62,63 Furthermore, we found that the third ventricle was associated with reduced volume with increased resilience. Evidence from multiple studies has linked enlargement of the third ventricle in first-episode psychosis and SCZ, a finding that correlates with more severe cognitive impairment.^64–68 While no evidence of moderating effects of resilience scores on these or other brain regions was found in our analysis of UK Biobank participants, it is possible that resilience loci promote compensatory effects on brain regions that are vulnerable to SCZ pathogenesis. Findings from prior research provide additional support to compensatory links between resilience loci and the brain. Resilience scores have been linked to general increases in the volume of the ventral visual pathway, a cortical area that shows altered connectivity^69,70 reduced activity in schizophrenia, possibly leading to visual perception impairments. The protective effect of resilience scores on the ventral visual pathway was strongest in the right fusiform gyrus, potentially promoting cognitive benefits. However, these results are based on single time-point, cross-sectional measures of regional brain sizes. Prospective follow-up measures of brain size are anticipated in the UK Biobank, which could offer insights into the role of resilience loci in adaptive remodeling of brain structure. Our key brain-structural findings indicated that resilience scores were associated with increases in cortical thickness and volume in multiple regions of the frontal lobe, in addition to increased volume and thickness of the ventral temporal lobe, particularly the fusiform gyrus. Notably, previous evidence shows that individuals at ultra-high risk for psychosis present with significant reductions in the fusiform gyrus, as well as regions of the frontal lobe that we found to increase in size commensurate with resilience scores.^36,71,72 The link we found between the fusiform gyrus and resilience scores mirrors an association initially reported by Hettwer et al,⁷³ who found a positive correlation between resilience scores and increased volume and area of the fusiform gyrus in a sample of 101 research participants with no personal or family history of mental illness.

It is important to acknowledge that our study did not account for individual differences in lifestyle or environmental factors, which could very well have a significant influence on the results we presented. Furthermore, the interactions that we modeled were limited to genetic proxies based on common variants, thus influences of rare variants are not reflected in our analyses. Hence, future studies should investigate the possibility of gene-by-environment interactions in the context of risk and resilience, aiming to elucidate moderating effects of genetic variants, both common and rare, in conjunction with non-genetic factors.

We have previously demonstrated significant and replicable protective effects of resilience scores against SCZ in multiple independent datasets.^10,11 The results from our current study further strengthen this evidence, serving as an additional external validation. We also found strong quantitative consistencies with respect to different polygenic scoring methods for resilience. The main effect of resilience scores and their moderating effects derived from our primary analysis of UK Biobank participants, which was based on the P + T scoring method, was highly consistent with results obtained from our analyses of resilience scores based on the Bayesian adaptive shrinkage method from PRS-CS. This finding underscores the robustness of our results across different methodologies for calculating resilience scores. However, it is important to acknowledge that potential selection biases and population admixture in the UK Biobank may have influenced our findings. The accuracy of genetic scores can be affected by differences in linkage disequilibrium (LD) structure between the base population in which the genetic scores were initially derived and the populations in which the scores were computed. While the UK Biobank is predominantly composed of individuals of European descent, there is a mix of non-European individuals, which could potentially lead to biases in genetic findings. We explicitly adjusted for ancestry with the use of genome-wide PCs as covariates to mitigate their potential confounding effects. Additional steps were taken in our study to confirm that our findings were generalizable to a subsample of UK Biobank participants who self-identified their ethnic background as white British, white Irish, or other White background. When comparing results from our analysis of the full UK Biobank dataset with the results obtained from the western European subsample, we observed strong agreement in the main effects of resilience scores and risk scores across all 299 outcomes examined. Regarding the interaction effects between resilience and risk scores, the agreement between the results obtained in the western European subsample and full UK Biobank was moderate but highly statistically significant. Taken together, the moderating effects of resilience scores that we observed in the UK Biobank showed generalizability across ancestral backgrounds. Yet, there remains the possibility of unaccounted-for confounding due to higher-order interactions between ancestry variation and the calculated genetic scores. Ancestry variation demonstrated significant associations with a multitude of phenotypic measures examined in our study, encompassing ICD-10 codes, self-reported mental health outcomes, and brain-structural profiles. This suggests that genetic diversity across ancestries can result in notable phenotypic variations, as evidenced by patterns in the UK Biobank dataset. Genetic ancestry PCs were used as covariates throughout all our analyses to account for these confounding effects. Furthermore, genetic scores for resilience and SCZ risk merely reflect the influences of heritable variation and do not capture non-genetic components of phenotypic variability. Environmental factors were not integrated into our models. We adjusted for participant age at recruitment and sex to address potential variation in phenotypic outcomes stemming from demographic differences. Additionally, we included domain-specific covariates for our analyses of the structural brain-imaging-derived phenotypes to mitigate ascertainment biases due to site-level variation and measurement noise due to head motion during image acquisition. These represent conservative steps aimed at ensuring the robustness of our primary results against likely sources of confounding. However, it remains possible that cryptic sources of ascertainment bias may have influenced our findings. As such, an important next step toward confirming the reliability of our results is conducting multiple rounds of replication in external datasets, including from other biobanks. Despite these technical considerations, population-based samples hold distinct advantages over genetic datasets that are acquired principally through clinical ascertainment. For example, the genetic relationships identified in the UK Biobank are expected to reflect signals present in the population as a whole, in contrast to relationships observed among a subgroup of patients that may be highly homogenous in terms of their clinical presentation, thus limiting statistical power to discover genetic effects relevant at the population-level. Altogether, a crucial next step in advancing this line of research is to assess the reproducibility of our findings, especially in the context of population-based studies with compositional characteristics different from the UK Biobank cohort.

Taken together, this study provides us with key insights into genetic and phenotypic links with resilience loci for SCZ. We observed a multitude of outcomes for which resilience loci moderated the influence of SCZ risk genes. Overall, our findings demonstrate first-ever evidence that resilience loci foster cross-protective effects against psychiatric and immunologic genetic risk profiles, as well as evidence that resilience scores attenuate rates of serious medical outcomes and co-morbidities, including lowering rates of suicidality measures. Utilizing phenome-wide information from the UK Biobank offers considerable value for our understanding of resilience loci. However, biobank-level datasets are currently limited in scale with lower-level molecular profiles. Embracing multi-scale ’omic approaches that encompass profiles such as transcriptome-wide profiles of transcribed gene products (including non-coding or protein-coding RNAs), proteome-wide levels of proteins, lipidomic, and metabolomic profiles, may offer crucial insights into the biological and functional relevance of resilience loci within tissues relevant to pathophysiology. Collectively, our results suggest that individuals with high genomic resilience scores have a reduced risk for psychiatric and serious medical outcomes and co-morbidities, many of which are associated with reduced life expectancy. As such, we postulate that resilience loci might facilitate adaptation to adverse outcomes, potentially enhancing longevity. Nonetheless, further research is needed to explore this hypothesis. If the results from our study are substantiated and extended to biological and environmental measures, we anticipate significant advancements in resilience genetics offering invaluable insights for future development of improved diagnostics, interventions, and prevention strategies against disease.

Supplementary Material

Supplementary material is available at https://academic.oup.com/schizophreniabulletin/.

Conflicts of Interest

In the past year, Dr Faraone received income, potential income, travel expenses continuing education support and/or research support from Aardvark, Aardwolf, AIMH, Akili, Atentiv, Axsome, Genomind, Ironshore, Johnson & Johnson/Kenvue, Kanjo, KemPharm/Corium, Noven, Otsuka, Sky Therapeutics, Sandoz, Supernus, Tris, and Vallon. With his institution, he has US patent US20130217707 A1 for the use of sodium-hydrogen exchange inhibitors in the treatment of ADHD. He also receives royalties from books published by Guilford Press: Straight Talk about Your Child’s Mental Health, Oxford University Press: Schizophrenia: The Facts, and Elsevier: ADHD: Non-Pharmacologic Interventions. He is the Program Director of www.ADHDEvidence.org and www.ADHDinAdults.com.

Data Availability

This research has been conducted using the UK Biobank Resource under Application Number 60050. Custom-written scripts used for analysis and visualizations are available on GitHub (https://github.com/hessJ/ukbb_resilience), but release of individual-level data is not permitted.