Language-impacting genetic variants discovered in autism also influence language ability in the general population

Developmental language impairment (LI) impacts 4-7% of the population, and is associated with poor social, educational, and occupational outcomes. Despite the high heritability estimates of LI, previous genetic studies have struggled to find robust associations, largely due to small sample sizes. Studying LI in connection with psychiatric disorders, which are enriched for language issues, can provide the necessary statistical power for genetic association testing. Autism is of particular interest because of its high co-occurrence with LI (approximately 30%). We hypothesized that results from a genetic study of specific language impairments (SLI, i.e., LI that cannot be explained by broader cognitive impairment) within autism would provide novel insights into language that are relevant to the general population. We derived a specific language impairment (SLI) trait from a large autism study (SPARK, N=15,577 autistic cases) using machine learning. The SLI score represents the degree of LI that cannot be explained by age, sex, or cognitive impairment. This trait was investigated in a series of genetic analyses, including: polygenic score correlations, a heritability analysis, a genome-wide association study (GWAS), and a rare-variant exome-wide association study. Finally, we tested the generalizability of our GWAS results by computing polygenic scores for this SLI trait in a large population sample. Using the Adolescent Brain and Cognitive Development (ABCD, N=5989), we correlated SLI polygenic risk with language test scores. As expected, polygenic propensity for high cognitive performance was negatively associated with the SLI score. Surprisingly, autism polygenic risk was not associated with the language trait, either in the SPARK sample or in ABCD. The SLI trait showed significant heritability estimates, with a common variant heritability estimate of 7% (95% CI: 1-13%), and a rare variant heritability of 15% (95% CI: 11-19%). The common variant GWAS identified one genome-wide significant locus on chromosome 13 (rs11147928, p-value = 1e-8). Further analysis of our GWAS results showed significant positive genetic correlations with schizophrenia, bipolar disorder, and left-handedness. The rare-variant exome-wide association study identified 6 exome-wide associated genes (Bonferroni p-value < 0.05), including genes previously linked to educational attainment and neurogenesis (LTA and RABD8A). Finally, polygenic scores derived from our SLI GWAS in the SPARK sample were predictive of language deficits in typically developing children. These findings offer new insight into the nuanced relationship between genetic factors, language impairment, and broader psychiatric and neurodevelopmental conditions. Interestingly, although autism and LI show significant co-occurrence, autism polygenic risk was not predictive of our SLI score. Instead, PGS for schizophrenia, bipolar disorder, handedness, and cognitive performance were associated with the SLI score, suggesting intersection with risk pathways for psychosis and brain lateralization. Further, the generalization of our findings to a diagnosis-agnostic population sample suggest that the polygenic mechanisms of LI observed in autism also apply to typically developing individuals. Finally, the rare variant exome-wide study implicated a number of genes that play key roles in the brain, and the significant rare variant heritability invites further rare variant studies of language ability.