Divergent synaptic plasticity and pruning mechanisms underlie early- versus late-diagnosed autism: Evidence from a transcriptome-wide association study

Autism spectrum disorder is remarkably heterogeneous, both in when individuals receive a diagnosis and in how their clinical course unfolds over time. Recent work by Zhang and colleagues (2025) demonstrated that early- and late-diagnosed autism follow distinct developmental trajectories underpinned by only modestly correlated polygenic architectures, with the late-diagnosed factor showing notably stronger genetic overlap with ADHD and other mental health conditions. What has remained unclear, however, is how these genetic differences are translated—at the level of gene expression in the brain—into divergent phenotypes. Here we report parallel transcriptome-wide association studies carried out on the age-stratified autism GWAS summary statistics published by Zhang et al. (2025). Using S-PrediXcan with GTEx v8 MASHR models, we imputed genetically regulated gene expression across six brain regions and examined enrichment across thirteen biologically curated gene sets spanning synaptic plasticity, pruning, glial biology, and negative controls. Cross-tissue meta-analysis and tissue-by-pathway mapping uncovered strikingly different expression signatures for the two subtypes. Early-diagnosed autism was enriched for genes involved in mitochondrial oxidative phosphorylation, ribosomal biogenesis, cytoskeletal regulation, and cellular senescence—a profile consistent with bioenergetic and translational deficits during the critical window when synapses are first being formed and stabilized. Late-diagnosed autism, by contrast, showed its strongest enrichment among adult astrocyte markers, complement cascade components, and synapse-pruning genes, particularly within limbic circuitry, with LRRK2 emerging as the single most significant gene across the entire analysis. Together, these results point toward two mechanistically separable hypotheses. In the early-diagnosed subtype, disrupted energy-dependent processes appear to compromise the initial establishment of synaptic plasticity. In the late-diagnosed subtype, dysregulated astrocyte–complement signaling seems instead to derail the adolescent phase of synaptic pruning and circuit refinement. This framework offers a coherent explanation for observed differences in diagnostic timing, socioemotional trajectories, and patterns of psychiatric comorbidity. It also yields concrete, testable predictions regarding neuroimaging signatures, candidate biomarkers, and subtype-stratified interventions—laying groundwork for a more biologically precise approach to understanding and treating autism.