Unraveling Genetic Variants underlying Schizophrenia Phenotypes: An Original GWAS in Hong Kong Chinese with Cross-Ethnic Meta-Analysis and Predictive Modeling

BackgroundSchizophrenia (SCZ) is a highly heritable and heterogeneous disorder with diverse clinical presentations and cognitive deficits. The specific genetic variants contributing to this variability remain largely unknown. This study aims to uncover the genetic bases of various clinical phenotypes such as age at onset (AAO), positive/negative symptoms, self-harm and aggression in SCZ using genome-wide association studies (GWAS). Few large-scale GWAS have explored these phenotypes, especially in non-Europeans. Understanding the genetics may reveal new therapeutic targets in the future. Additionally, it is unclear whether the genetics of SCZ phenotypes overlap with the genetic risk for SCZ itself or other psychiatric disorders. Previous studies using polygenic risk scores have provided limited insights. Aims and ObjectivesThis study aims to: (i) identify genetic variants associated with SCZ clinical phenotypes using GWAS, (ii) decipher the genetic relationship between SCZ phenotypes and other psychiatric disorders via PRS, (iii) gain biological insights into SCZ phenotypes through gene and pathway analyses of GWAS data, and (iv) predict the severity of SCZ phenotypes using a machine learning model that combines clinical and genomic factors. MethodsThis observational study is primarily based on two cohorts of SCZ patients from Hong Kong, with genotyping data collected. A third cohort from dbGaP was used to study genetics of AAO. Clinical phenotypes such as self-harm, aggression, hospitalization, and AAO were collected. PRS were calculated for psychiatric and cognitive traits. GWAS were conducted for the phenotypes within each cohort and meta-analysed. Gene-based analyses identified associated genes, which were then subjected to pathway enrichment analysis. Machine learning models were built using PRS and clinical variables to predict relevant outcomes such as hospitalization risk. ResultsGWAS identified several significant genetic variants associated with SCZ phenotypes across the three cohorts. A variant rs60648049 was associated with hospitalization count in cohort 1. In cohort 2, a variant was linked to age at onset (AAO). The dbGaP cohort revealed two variants significantly associated with AAO. Meta-analysis across cohorts identified three variants ( rs144645158, rs142498233 and rs185213255)) significantly associated with AAO. Gene-based and transcriptome-wide analyses implicated genes in pathways related to synaptic function, neurodevelopment, and immune processes. Males, lower education, and genetic factors like lower IQ PRS were associated with increased aggressive behaviours. Self-harm was linked to lower education and ADHD PRS. Earlier AAO correlated with male sex, higher education, and PRS for bipolar disorder and major depression. Machine learning models predicted hospitalization status with an area under the receiver operator characteristic curve of 63.4%, with AAO and aggressive behaviour as top predictors. Models were also built to predict the number of hospitalizations (RMSE=1.067) and total hospitalization duration (RMSE=3.063 months), with AAO, self-harm and aggressive behaviour being significant predictors of greater severity. Conclusions and ImplicationsThis GWAS examining clinical profiles in Chinese SCZ patients, combined with European samples, identified several genetic loci associated with age at onset. Nominal trends indicating genetic overlaps with psychiatric disorders, related traits, and cognitive functions suggest shared pathophysiology and potential therapeutic approaches. Machine learning models showed moderate accuracy in predicting SCZ severity. Collectively, these findings enhance our understanding of SCZs genetic architecture, and provide a foundation for future research and clinical strategies aimed at improving patient outcomes.