Decomposing the genetic risk of autism spectrum disorder into discrete molecular subtypes underlie clinical heterogeneity based on transcriptome profile.

backgroundAutism spectrum disorder (ASD) is a highly heterogeneous neurodevelopmental condition with a complex genetic architecture. While over a thousand risk genes have been cataloged, a fundamental challenge remains how this vast genetic landscape translates into diverse clinical manifestations. To address this, we propose a "many-to-few" framework, shifting from traditional "many-to-one" convergence model toward an intrinsic organizational architecture where disparate ASD risk genes funnel into discrete molecular dimensions.

methodsBy leveraging similarity network fusion (SNF) to integrate bulk and single-nucleus RNA sequencing data, we decomposed the 311 ASD reliable risk genes into three stable, spatiotemporally distinct molecular subtypes. These subtypes represent coordinated expression programs that integrate functionally diverse genes, such as those involved in synaptic signaling, mRNA stabilization, and histone modification. Mapping de novo variants from the SPARK cohort onto these subtypes enabled stratification of probands into three genetically defined subgroups (S1-S3) with divergent clinical profiles.

resultsTranscriptomic decomposition partitioned the 311 reliable risk genes into three stable molecular subtypes, Synaptic Signaling (C1), mRNA Stabilization (C2), and Histone Modification (C3), exhibiting divergent spatiotemporal trajectories and cell-type-specific enrichment. Patient subgroups stratified based on these molecular subtypes displayed significant differences in adaptive functioning, core symptoms, and psychiatric comorbidities, while a reference group (S4) lacking de novo variants of these highly reliable ASD risk genes exhibited the most preserved functions. Furthermore, diverging rare and common genetic liability profiles across subgroups, particularly between S1 and S3, provide empirical support for a molecular subtype-based liability threshold model.

conclusionsOur study establishes a biologically informed framework that links intrinsic molecular subtypes to multidimensional phenotypic constellation, advancing mechanistic insight and offering translational potential for precision stratification and intervention.