Polygenic contribution to the relationship of loneliness and social isolation with schizophrenia

The study addresses whether and how genetic liability to loneliness and social isolation (LNL-ISO) relates to schizophrenia risk. Social relationships are fundamental in humans, and both objective social isolation and perceived loneliness have been linked to morbidity, mortality, and mental disorders. Loneliness has been associated not only with depression but also with psychosis, with social withdrawal common early in schizophrenia and meta-analytic evidence implicating loneliness in psychotic symptoms. Genetic studies show that LNL-ISO is heritable and genetically correlated with schizophrenia. However, the directionality of the association, the roles of shared versus non-shared variants, and sex-specific effects remained unclear. The study aims to: (1) test the contribution of LNL-ISO polygenic scores to schizophrenia in an independent sample; (2) dissect schizophrenia genetic architecture by concordance of effects with LNL-ISO and evaluate polygenic prediction, heritability enrichment, functional annotations, and sex effects; (3) examine annotation-stratified genetic covariances with related psychiatric and behavioral traits; and (4) assess bidirectional causality between LNL-ISO and schizophrenia using Mendelian randomization.

Prior research has consistently associated loneliness and objective social isolation with depression and major depressive disorder, and more recently with psychosis. Meta-analyses show loneliness contributes to the onset and persistence of psychotic symptoms and relates to both positive and negative psychotic-like experiences; experimental and observational work suggests loneliness can increase subclinical paranoia. Genetic evidence indicates loneliness and social isolation are heritable. A multi-trait GWAS (MTAG) of combined LNL-ISO in UK Biobank identified genome-wide significant loci and h2SNP around 4.2%, and showed significant genetic correlation with schizophrenia (rg ≈ 0.17). Schizophrenia polygenic scores predict loneliness in population samples, supporting shared genetic etiology. Sex differences exist in psychosis epidemiology and may influence loneliness perception and impact, though findings have been mixed. These gaps motivate deeper analysis of shared versus discordant genetic effects, sex moderation, and causal directions between LNL-ISO and schizophrenia.

- Samples and GWAS: Target case-control sample from CIBERSAM (Spain): 1,927 schizophrenia cases (65% male) and 1,561 healthy controls (55% male), genotyped and QC’d per PGC-SZ2 standards. Discovery summary statistics included PGC-SCZ2 GWAS (35,476 cases; 46,839 controls), UK Biobank MTAG LNL-ISO combined phenotype (effective N ≈ 487,647), and individual UKBB GWAS for constituent traits: perceived loneliness, social support proxies (frequency of family/friends visits; number of people in household), and ability to confide. No sample overlap between PGC-SCZ2 and the CIBERSAM target sample. An additional schizophrenia GWAS was used for sensitivity checks. - Polygenic scores (PGS): Computed PGS for LNL-ISO and constituent traits using UKBB/MTAG summary statistics; evaluated variance explained in schizophrenia case-control status via logistic regression with covariates (sex, age, 10 MDS ancestry components), LD-adjusted, and assessed across multiple P-value thresholds. Nagelkerke’s R2 was converted to liability scale where appropriate. - Genomic dissection: Intersected SCZ and LNL-ISO GWAS SNPs (≈5.66M SNPs). Partitioned SCZ SNPs into: SCZ[noLNL] (P_LNL-ISO > 0.05), SCZ[LNL] (P_LNL-ISO < 0.05), then split SCZ[LNL] by concordant versus discordant effect directions relative to LNL-ISO: SCZ[CONC] (β_SCZ and β_LNL-ISO same sign; N≈269k SNPs) and SCZ[DISC] (opposite sign; N≈217k SNPs). Conducted PGS_SCZ predictions using each subset. - Heritability enrichment and functional annotation: Applied LD Score Regression (LDSR) to estimate partitioned h2SNP and enrichment for SCZ[noLNL], SCZ[CONC], SCZ[DISC]. Performed LDSC-SEG analyses to test enrichment in brain tissues and neuronal cell types (GTEx and related annotations). - Sex-stratified analyses: Compared PGS_SCZ predictive performance in males versus females within each partition using bootstrap resampling (5,000 permutations of 500 cases and 500 controls) and two-sided t-tests. - Annotation-stratified genetic covariance: Used GNOVA to estimate genetic covariance between schizophrenia and multiple neuropsychiatric and behavioral traits across SCZ[noLNL], SCZ[CONC], and SCZ[DISC]. Applied FDR correction. - Mendelian randomization (MR): Conducted bidirectional two-sample MR between LNL-ISO (and constituent traits) and schizophrenia using multiple estimators: IVW, Weighted Median (WM), MR-Egger, Simple Mode, Weighted Mode, MR-PRESSO (with outlier removal), and CAUSE to address horizontal pleiotropy (both correlated and uncorrelated). Assessed heterogeneity (Q tests) and Egger intercepts for pleiotropy. Reported robust results consistent across methods.

- LNL-ISO PGS predicts schizophrenia risk: In the CIBERSAM sample (N_scz=1927; N_HC=1561), PGS_LNL-ISO explained R2=0.56% (95% CI −0.01, 1.13) at P_threshold=0.05 (p=1.2×10^−4). Each 1 SD increase in PGS_LNL-ISO increased odds of schizophrenia by 15% (OR=1.15, 95% CI 1.07–1.24). LNL-ISO explained more variance than perceived loneliness (R2=0.41%, p=1.42×10^−3), ability to confide (R2=0.28%, p=7.4×10^−3), and was comparable or greater than social support proxies (household size R2=0.54%, p=3.14×10^−4; visit frequency R2=0.42%, p=1.2×10^−3). - Genomic dissection and polygenic prediction: PGS_SCZ[CONC] (variants with concordant effects on SCZ and LNL-ISO) explained substantially more schizophrenia liability (R2=3.94% at P_threshold=0.5, p=8.36×10^−25) than PGS_SCZ[DISC] (R2=1.02% at P_threshold=0.01, p=8.43×10^−8). All partitions showed monotonic risk increases across PGS deciles. - Heritability enrichment: SCZ[CONC] comprised 3.8% of SNPs but explained ~13.1% of h2SNP (enrichment=3.43, 95% CI 2.86–4.01; p=1.83×10^−15). SCZ[DISC] showed no significant enrichment (1.08, 95% CI 0.58–1.58; p=0.748). SCZ[noLNL] (65% of SNPs) accounted for ~53.9% of h2SNP (enrichment=0.81, 95% CI 0.72–0.90; p=8.12×10^−5). - Brain tissue enrichment: SCZ[noLNL] enriched in GTEx brain cortex (p=8.5×10^−4) and anterior cingulate cortex (p=5.16×10^−3). SCZ[CONC] showed nominal hippocampal enrichment (p=0.041; not FDR-significant). - Sex differences: PGS_SCZ[CONC] predicted more variance in females (R2=2.24%, 95% CI 1.09–3.38; P_threshold=0.1; p=1.88×10^−13) than males (R2=1.41%, 95% CI 0.60–2.22; P_threshold=0.5; p=2×10^−13). Opposite trends were observed for other partitions. Bootstrap comparisons confirmed significant sex-based differences. - Annotation-stratified covariances: Within SCZ[CONC], schizophrenia showed positive genetic covariance with major depressive disorder, anxiety disorder, ADHD, ASD, alcohol dependence, cross-disorder phenotype, neuroticism, depressive symptoms, subjective well-being, and psychotic experiences; within SCZ[DISC], these covariances were negative. Bipolar disorder and OCD exhibited positive covariance with schizophrenia in both SCZ[CONC] and SCZ[DISC]. Educational attainment (EA) showed negative covariance within SCZ[CONC], but positive covariance within SCZ[DISC] and in SCZ[noLNL]. - Mendelian randomization (bidirectional causality): LNL-ISO → schizophrenia: IVW β=1.11 (SE 0.48), p=0.021; WM β=1.37 (SE 0.40), p=6.14×10^−4; MR-PRESSO outlier-corrected β=1.45 (SE 0.30), p=0.001; CAUSE effect y=0.61 (95% CI 0.34–0.89), p=0.003. Heterogeneity present; no significant Egger intercept pleiotropy (p=0.36). Schizophrenia → LNL-ISO: WM β=0.015 (SE 0.005), p=0.008; CAUSE y=0.01 (95% CI 0.01–0.01), p=0.003; heterogeneity present; no Egger intercept pleiotropy (p=0.48). - Constituent traits MR: Bidirectional causality between perceived loneliness and schizophrenia. Unidirectional negative causal effect of ability to confide on schizophrenia (WM β=−0.6, SE 0.19, p=0.002). Unidirectional negative causal effect of schizophrenia on number of people in household (WM β=−0.011, SE 0.003, p=5.86×10^−3). No evidence for causality for frequency of family/friend visits with schizophrenia.

Findings indicate substantial genetic overlap between LNL-ISO and schizophrenia, with variants showing concordant effects across both phenotypes carrying disproportionate heritability and stronger predictive power, particularly among females. Discordant variants contribute less to schizophrenia liability and show opposite-sign covariances with multiple psychiatric traits. Annotation-stratified covariances suggest that genetic liability to social isolation may underlie shared comorbidity patterns between schizophrenia and disorders such as MDD, ANX, ADHD, ASD, and alcohol dependence, whereas OCD and bipolar disorder share risk with schizophrenia independent of LNL-ISO-related annotations. The negative covariance between educational attainment and schizophrenia within concordant overlap, contrasted with positive covariance within discordant and non-overlapping variants, implies that social isolation-related genetic liability may mediate poorer educational outcomes often observed premorbidly in schizophrenia. Bidirectional MR supports a causal feedback loop: elevated LNL-ISO liability increases schizophrenia risk more strongly than the reverse, while schizophrenia liability also causally increases social isolation tendencies. These results integrate prior epidemiologic observations of loneliness in prodromal and chronic psychosis with genetic evidence, highlighting social isolation as both a contributor to and a consequence of schizophrenia liability. Clinically, this underscores the importance of addressing social isolation as a potentially modifiable target to influence onset, course, and comorbidity profiles.

The study demonstrates that polygenic liability to loneliness and social isolation contributes to schizophrenia risk and that schizophrenia genetic architecture can be meaningfully partitioned by concordance with LNL-ISO effects. Concordant variants are enriched for schizophrenia heritability, show stronger prediction in females, and align positively with multiple psychiatric traits, whereas discordant variants show opposite covariances and greater influence in males. Mendelian randomization supports a bidirectional causal relationship, with a stronger effect from LNL-ISO to schizophrenia. Given that aspects of social isolation and perceived loneliness may be modifiable, interventions targeting these domains could have preventive and therapeutic impact on schizophrenia and its comorbidities. Future research should leverage larger samples, validated multi-item loneliness measures, and multivariate genomic frameworks (e.g., Genomic SEM) to refine mechanisms, explore environmental moderators, and test sex-specific pathways.

- LNL-ISO and constituent measures derived from UK Biobank single-item questionnaires rather than validated multi-item scales (e.g., UCLA Loneliness Scale), potentially introducing measurement error, although convergent validity has been reported. - UK Biobank sample characteristics may introduce socio-economic and selection biases that could affect genetic predictions and generalizability. - Partitioning the genome into subsets reduces SNP counts and statistical power for heritability enrichment and downstream analyses; larger datasets may improve precision. - Mendelian randomization analyses exhibited substantial heterogeneity; despite using robust methods (WM, MR-PRESSO, CAUSE) and finding consistent results, residual pleiotropy or instrument weaknesses cannot be fully excluded. - Effect sizes are small, indicating that while genetic factors contribute to the link between social isolation and schizophrenia, environmental influences likely play a major role and warrant further study. - Additional analytic approaches (e.g., Genomic SEM) could complement the SNP-subsetting strategy and provide deeper multivariate insights.