Social relationships are crucial for emotional and cognitive development in social species. The need to belong is a fundamental human behavior. Social isolation, encompassing both objective lack of social connections and subjective feelings of loneliness, is a significant risk factor for morbidity, mortality, and mental disorders. While loneliness and social isolation have been linked to depression, recent research highlights their association with psychosis. Social withdrawal is a feature of early-stage schizophrenia, echoing classical descriptions of pre-schizophrenia traits. Meta-analyses show loneliness's role in the onset and maintenance of psychotic symptoms. However, the causal relationship between social isolation and schizophrenia remains unclear. Genetic factors may partly explain this relationship, as evidence supports the genetic basis of loneliness and social isolation. A multi-trait GWAS (MTAG) study revealed genetic correlation between a combined loneliness and social isolation phenotype (LNL-ISO) and schizophrenia. Schizophrenia polygenic scores have also predicted loneliness, suggesting a shared genetic etiology. However, questions remain about the direction of the association, the specific effects of shared and non-shared genetic variants, and the influence of sex. This study investigates the bidirectional genetic relationship between LNL-ISO and schizophrenia, dissecting predisposing variation, analyzing polygenic risk scores and sex effects, and evaluating causality using Mendelian randomization.

Existing research demonstrates a correlation between loneliness and social isolation (LNL-ISO) and various mental health conditions, particularly depression. However, a growing body of work points towards a significant link between LNL-ISO and schizophrenia. Studies have observed social withdrawal and isolation in the early stages of schizophrenia, aligning with historical descriptions of pre-schizophrenic traits. Meta-analyses consistently reveal a strong association between loneliness and psychotic symptoms, both positive and negative. Some research suggests loneliness may even increase subclinical paranoia in non-clinical populations. While these studies establish an association, the causal direction and underlying mechanisms remain ambiguous. Previous genetic studies indicate a heritable component to both LNL-ISO and schizophrenia, with evidence suggesting shared genetic architecture. A multi-trait GWAS (MTAG) analysis has shown a significant genetic correlation between LNL-ISO and schizophrenia, while separate studies have demonstrated that schizophrenia polygenic scores can predict loneliness levels in independent samples. However, these studies lack a comprehensive framework to address the directional nature of the association, the impact of sex differences, and the specific roles of shared versus non-shared genetic variations.

This study used a multi-stage approach to investigate the genetic relationship between LNL-ISO and schizophrenia. First, a polygenic score for LNL-ISO (PGS_LNL-ISO) was calculated using summary statistics from a UK Biobank MTAG study. This score was then applied to an independent Spanish case-control sample (CIBERSAM) to assess its contribution to schizophrenia risk. The analysis accounted for sex, age, and ancestry components. Second, schizophrenia-associated genetic variants were categorized based on their effects on LNL-ISO: variants not associated with LNL-ISO (SCZ[noLNL]), variants with concordant effects (SCZ[CONC]), and variants with discordant effects (SCZ[DISC]). Polygenic scores were then generated for each subset (PGS_SCZ[noLNL], PGS_SCZ[CONC], PGS_SCZ[DISC]) and applied to the CIBERSAM sample to determine their contribution to schizophrenia risk, stratified by sex. Heritability enrichment was estimated using LD-score regression (LDSR) for each variant subset. Partitioned heritability and LDSC-SEG analyses were performed to explore tissue-specific enrichment in brain regions. Third, to explore shared genetic etiology with other disorders, genetic covariances between schizophrenia and other phenotypes were calculated across the LNL-ISO partitions using GNOVA. Finally, two-sample Mendelian randomization (MR) analyses, including several methods to mitigate pleiotropy (IVW, WM, MR-Egger, SM, W-mode, MR-PRESSO, and CAUSE), were conducted to infer the causal relationship between LNL-ISO and schizophrenia, and between schizophrenia and the individual LNL-ISO components (perceived loneliness, social support proxies, ability to confide).

The LNL-ISO polygenic score significantly contributed to schizophrenia risk in the CIBERSAM sample (R² = 0.56%, p = 1.2 × 10⁻⁴). One standard deviation increase in PGS_LNL-ISO was associated with a 15% increased likelihood of schizophrenia. Genetic variants with concordant effects in both LNL-ISO and schizophrenia (SCZ[CONC]) showed significant SNP-based heritability enrichment (Enrichment = 3.43, p = 1.83 × 10⁻¹⁵) and a greater contribution to schizophrenia risk in females. In contrast, variants with discordant effects (SCZ[DISC]) only contributed to schizophrenia risk in males and showed negative covariance with other mental disorders. Annotation-stratified genetic covariance analyses revealed positive correlations between SCZ[CONC] and various mental disorders (depression, anxiety, ADHD, autism, alcohol dependence), while SCZ[DISC] showed negative correlations. Mendelian randomization analyses provided strong evidence for a bidirectional causal relationship between LNL-ISO and schizophrenia. The effect of LNL-ISO liability on schizophrenia risk was larger than vice versa. Analyses of individual LNL-ISO components revealed a negative causal effect of 'ability to confide' on schizophrenia and a negative causal effect of schizophrenia on 'number of people in household'.

This study provides robust evidence for a substantial genetic overlap between loneliness and social isolation (LNL-ISO) and schizophrenia, demonstrating a bidirectional causal relationship. The findings highlight the distinct contributions of concordant and discordant genetic variants, emphasizing sex-specific effects and differential correlations with other mental disorders. The greater effect of LNL-ISO on schizophrenia suggests that social isolation may be a significant risk factor, while the inverse relationship, albeit smaller, indicates that schizophrenia may also contribute to social isolation. The identified genetic variants shared between LNL-ISO and schizophrenia could potentially point to therapeutic targets for interventions aimed at reducing the risk of schizophrenia or improving social functioning in individuals with the disorder. The sex-specific effects also suggest that targeted interventions may be needed, tailoring approaches to address the unique vulnerabilities of males and females.

This research confirms a strong genetic overlap between LNL-ISO and schizophrenia, supporting the idea that social isolation plays a critical role in the development and progression of the disorder. The study's findings suggest a bidirectional causal relationship, but the impact of LNL-ISO on schizophrenia risk is stronger than the reverse. The discovery of sex-specific effects and distinct associations with other psychiatric disorders highlights the complexities of this relationship. Future research should focus on identifying specific genes and pathways implicated in this association, allowing for the development of personalized interventions that address the modifiable factors influencing social isolation and its impact on mental health.

Several limitations should be considered. The study used single-question questionnaires for LNL-ISO measures from the UK Biobank, which might not capture the full complexity of these constructs. Potential socio-economic biases in the UK Biobank sample may have affected the generalizability of genetic predictions. The partitioning of the genome into subsets of SNPs may have underpowered some analyses. The observed heterogeneity in the MR analyses, despite using multiple robust methods, warrants further investigation. Finally, the small effect sizes underscore the likely significant contribution of environmental factors, which remain to be explored in more detail.