Psychiatric polygenic risk as a predictor of COVID-19 risk and severity: insight into the genetic overlap between schizophrenia and COVID-19

COVID-19 exhibits wide inter-individual variability in susceptibility and severity. Established risk factors include older age, male sex, high BMI, and comorbid cardiometabolic and respiratory conditions. Many of these conditions and immune dysregulations (e.g., chronic inflammation, autoimmunity) are prevalent among psychiatric patients, especially those with schizophrenia (SCZ), bipolar disorder (BD), and major depressive disorder (MDD). Prior studies suggest SCZ patients may have increased COVID-19 prevalence, ICU admission, and mortality. Genetic studies of COVID-19 implicate loci such as 3p21.31 and ABO, and HLA variation has been suggested to influence susceptibility and severity. Psychiatric disorders show immune involvement and shared biological pathways with COVID-19 at transcriptomic and genomic levels. This study tests the hypothesis that common variant polygenic risk for SCZ, BD, and DEP predicts COVID-19 susceptibility and severity, and explores whether immune-related genetic variation drives any observed overlap. The purpose is to quantify genetic overlap using PRSs and to assess sex-specific effects, thereby informing biological commonalities that may underlie worse COVID-19 outcomes in psychiatric populations.

- Epidemiology: COVID-19 severity and mortality are higher among older adults, males, and those with comorbidities (CVD, respiratory disease, obesity). Some studies report higher COVID-19 prevalence, ICU admission, and mortality among SCZ patients. - Immuno-psychiatric overlap: SCZ, BD, and MDD are associated with chronic inflammation (elevated CRP, IL-1β, IL-6, TNF-α, TGF-β, IFN-γ, VEGF), impaired adaptive immunity, and autoimmunity. COVID-19 can induce systemic inflammation, neuroinflammation, oxidative stress, and microbiome disruptions, paralleling abnormalities reported in psychiatric disorders. - COVID-19 host genetics: GWASs implicate 3p21.31 (e.g., SLC6A20, LZTFL1, CCR9, FYCO1, CXCR6, XCR1) and ABO (9q34.2); candidate genes include ACE2 and TMPRSS2; HLA variation is suggested to influence susceptibility and severity. - Psychiatric genetics: Immune-related pathways, B-lymphocyte activity, and MHC loci are implicated in SCZ/BD; inflammatory pathways converge across DEP, BD, and SCZ. Cross-trait analyses indicate multiple genes shared between COVID-19 and SCZ/BD and dysregulated expression in peripheral blood. - Knowledge gaps: Degree to which shared SNP-based genetic risk explains COVID-19 susceptibility/severity in psychiatric populations remains unclear; sex differences and the contribution of immune-specific variation are insufficiently characterized.

Design: Polygenic scoring study using summary statistics from large psychiatric GWASs to compute PRSs applied to an independent COVID-19 cohort; post-hoc genetic correlation (LDSC) when PRS associations were significant. Target sample: 15,045 individuals after QC: 9371 COVID-19 positive cases (53.7% female; mean age 62.6) collected across 34 Spanish hospitals (SCOURGE consortium) and 5674 controls with unknown COVID-19 status (46.3% female; mean age 53.1) from the Spanish DNA biobank and GR@ACE consortium. Ethical approval: Galician Ethical Committee (2020/197); informed consent obtained. Genotyping and imputation: Genotyping, QC, and imputation per Cruz et al. (2022). Imputation via TOPMed r2 (GRCh38); data realigned to GRCh37 to match discovery summary statistics using UCSC Genome Browser; imputed variants with call rate <98% removed. After imputation and QC, 6,317,562 markers retained. Discovery GWAS for PRS construction: - SCZ: PGC Wave 3 Schizophrenia meta-analysis; 161,405 individuals (≈80% European; 67,390 cases, 94,015 controls). - BD: Largest PGC BD GWAS/meta-analysis; 41,917 BD cases (type I/II) and 371,549 controls (European descent) from 57 cohorts. - DEP: PGC MDD cohorts (excluding 23andMe) plus UK Biobank broad depression; total 246,363 cases and 561,190 controls. PRS computation: PRS-CS (Bayesian regression with continuous shrinkage priors) using all SNPs and 1000 Genomes European LD reference panel. Grid search over phi (1e-6, 1e-4, 1e-2, 1) to optimize shrinkage; 10,000 MCMC iterations; default remaining parameters. Global PRSs used all genome-wide variants; immune PRSs (post-hoc) restricted to variants within genes annotated to KEGG immune pathways (MAGMA v1.08b gene annotation; ±5 kb windows), selecting SNPs present in both discovery and target data. Outcomes and statistical analysis: Logistic regression to test PRS prediction of: - RISK: SARS-CoV-2 infection (case/control), - SYMP: symptomatic vs asymptomatic among cases, - HOSP: hospitalized vs not, - CRIT: critical status (severity score 4) vs non-critical. Covariates: sex, age, first 10 principal components (total sample); sex-stratified models adjusted for age and PCs. Pseudo R2 on the liability scale estimated to compare variance explained (prevalences: cases 9.9%, symptomatic 6.9%, hospitalization 0.5%, critical 0.5%). PRSs were also analyzed by deciles to estimate odds ratios (95% CI). Predictive performance assessed via AUC (ROC), comparing covariate models with vs without PRS. Multiple testing correction across 4 outcomes × 3 PRSs: α = 0.05/12 = 0.004. Genetic correlation: LDSC regression to estimate SNP-based genetic correlations (rg) between SCZ risk and COVID-19 case/control and hospitalization outcomes, using summary statistics from Cruz et al. (2022) and COVID-19 HGI (R7). SYMP not available in GWAS, so no rg estimated for that outcome.

- Sample and markers: Final target sample N=15,045 (9371 cases; 5674 controls). Post-QC/imputation markers: 6,317,562. - PRS composition: SCZ global PRS from 708,399 variants; BD from 707,450 variants; DEP from 699,966 variants. Immune PRS (SCZ) from 7,375 variants across 1,010 KEGG-immune genes. Significant SCZ PRS associations (passing α=0.004): - Total sample: - RISK (case/control): β=96,906.49, SE=26,991.08, p=3.30E-04; liability-scale PRS R2≈0.002 (covariate model R2≈0.199). AUC unchanged with PRS (73.2% vs 73.2%). - SYMP (symptomatic vs asymptomatic): β=125,266.50, SE=26,527.47, p=2.33E-06; PRS R2≈0.002 (covariate R2≈0.170). AUC 72.3% vs 72.2% (no improvement). - HOSP (hospitalization): β=100,892.50, SE=29,053.52, p=5.15E-04; PRS R2≈6×10^-4 (covariate R2≈0.417). AUC 81.4% vs 81.4%. - Females: - RISK: β=128,303.20, SE=38,597.79, p=8.87E-04; PRS R2≈0.001 (covariate R2≈0.182). AUC 71.3% vs 71.2%. - SYMP: β=154,087.10, SE=37,285.84, p=3.59E-05; PRS R2≈0.007 (covariate R2≈0.148). AUC 70.0% vs 69.8%. - HOSP: β=127,810.40, SE=42,244.92, p=2.48E-03; PRS R2≈0.001 (covariate R2≈0.397). AUC 80.8% vs 80.8%. - Males: - SYMP significant: β=115,629.30, SE=38,890.53, p=0.003; RISK and HOSP nominal p≈0.03–0.035 but not significant after correction. AUC 77.1% vs 77.1%. - PRS decile analyses showed increasing odds with higher SCZ PRS deciles for case/control, symptomatic status, and hospitalization (figure referenced), but exact ORs not provided in text. Non-significant findings: - Immune SCZ PRS: No significant associations in total or sex-stratified analyses. - BD and DEP global PRSs: No significant associations with any COVID-19 outcomes. - LDSC regression: No significant SNP-based genetic correlations between SCZ risk and COVID-19 case/control or hospitalization using either Cruz et al. (2022) or COVID-19 HGI R7 summary statistics. Overall: SCZ PRS associates with higher risk of infection, symptoms, and hospitalization, especially in women, but the added predictive value is minimal (AUC unchanged; liability-scale PRS R2 ≤ ~0.2%).

The study addressed whether common genetic liability to psychiatric disorders predicts COVID-19 susceptibility and severity. Significant associations of the SCZ PRS with infection risk, symptomatic presentation, and hospitalization suggest partial SNP-based overlap between SCZ risk and COVID-19 outcomes. The effects appear stronger or more consistently detectable in females, aligning with known sex differences in COVID-19 genetic architecture and potential pleiotropy in females. The lack of improvement in AUC and very small PRS R2 indicate limited predictive utility for clinical risk stratification at present; nevertheless, the associations highlight biological commonalities worth further exploration. The immune-restricted SCZ PRS was not significant, suggesting that non-immune genetic factors (e.g., neuronal, metabolic, vascular pathways) or broader polygenic architecture contribute to the observed overlap. Transcriptomic studies support shared dysregulation across multiple pathways in SCZ and COVID-19. The absence of significant genetic correlations by LDSC underscores methodological differences: PRS-CS uses posterior effect sizes and aggregates polygenic burden at the individual level, whereas LDSC correlates GWAS effect sizes at the SNP level; such differences can yield divergent outcomes. Additionally, SYMP had no corresponding GWAS for rg estimation. Disorder specificity is notable: despite high genetic correlation between SCZ and BD, only SCZ PRS associated with COVID-19 outcomes. This may reflect disorder-specific loci, phenotypic heterogeneity within BD (e.g., psychotic features), or power/sample definition issues. DEP PRS null results could reflect lower heritability and broad, self-reported phenotypes in discovery GWAS diluting signal. The lack of association with critical illness suggests that extreme severity may depend more on distinct disease-specific genetics and strong nongenetic factors (e.g., age, comorbidities). Findings support the concept that shared common variant risk contributes modestly to worse COVID-19 outcomes in those with higher SCZ polygenic liability. Future studies incorporating sex chromosomes, rare variants, and refined psychiatric phenotypes may clarify mechanisms and improve interpretability.

This study demonstrates that higher polygenic risk for schizophrenia modestly predicts increased risk of SARS-CoV-2 infection, symptomatic disease, and hospitalization for COVID-19, with more consistent effects in women. No associations were detected for bipolar disorder or depression PRSs, and immune gene-restricted PRSs were not significant. Genetic correlation analyses did not confirm SNP-level overlap, highlighting methodological differences and potential complexity of shared biology. Main contributions: - Evidence for modest overlap between SCZ polygenic risk and multiple COVID-19 outcomes. - Sex-stratified analyses reveal stronger or more detectable associations in females. - Immune-restricted PRS results suggest broader biological contributions beyond immune pathways. Future directions: - Incorporate sex chromosome loci and rare variants (e.g., sequencing, CNVs) to capture additional shared risk. - Develop and analyze COVID-19 GWAS for symptomatic status to enable rg estimation with SYMP. - Refine psychiatric discovery samples (e.g., MDD diagnosed cohorts; BD subtypes with psychosis) to reduce heterogeneity. - Adjust for psychiatric diagnoses and psychotropic medication use in target cohorts to disentangle direct genetic effects from treatment and comorbidity influences. - Explore pathway- and tissue-specific PRSs and cross-ancestry analyses to generalize findings.

- Predictive performance was minimal: inclusion of SCZ PRS did not change AUCs; liability-scale PRS R2 values were very small (~0.0006–0.002). - Immune-restricted PRS may have reduced power due to fewer variants, potentially missing broader biological contributions. - No data on psychiatric diagnoses or psychotropic medication use in the COVID-19 cohort, limiting adjustment for potential confounders; medications may influence COVID-19 outcomes. - SYMP outcome lacked corresponding GWAS, precluding LDSC rg for this trait; LDSC regressions for available traits were non-significant. - X chromosome variants were not included in PRSs; sex chromosome effects may be relevant to observed sex differences. - DEP discovery sample included broad, self-reported phenotypes, potentially diluting PRS signal; heterogeneity in BD subtypes may also obscure associations. - Critical illness analyses did not show PRS effects, possibly due to reliance on disease-specific genetics and strong non-genetic factors (e.g., age), and limited prevalence reducing power.