Cross-cohort convergence of a microglial transcriptional state balance in idiopathic autism and Dup15q syndrome

Abstract

Autism spectrum disorder (ASD) encompasses diverse etiologies, creating a need for molecular measurements that retain biological meaning across cortical resources. We examined a predefined transcriptional balance within microglia-annotated nuclei and asked which expression component accounted for its disease-associated direction. Population A and Population B were defined as PsychENCODE/Wamsley clusters 0 and 2. A primary all-donor signature comprised 150 genes per direction derived from 44 donors with at least 10 nuclei in each population; a control-derived signature used 20 eligible controls. The same gene lists were projected separately into Velmeshev idiopathic ASD and Perez Dup15q data. Associations were evaluated using donor-level regression or donor-random-intercept models, with component and sensitivity analyses. Primary B-minus-A coefficients were positive in PsychENCODE (beta=0.058, P = 0.247, Benjamini–Hochberg q=0.247), Velmeshev (beta=0.154, P = 0.104, q=0.156), and Perez Dup15q (beta=0.151, P = 0.014, q=0.041). Dup15q provided the clearest primary statistical evidence; the idiopathic ASD estimates were directionally concordant but individually nonsignificant. External Population A coefficients were negative (Velmeshev beta=−0.157; Perez beta=−0.147), whereas Population B estimates were closer to zero. Positive B-minus-A direction was retained across alternative signature derivation, Perez cell definitions, tested technical adjustments, contamination/doublet exclusion masks, conservative gene exclusion, and all donor-removal refits within each external resource. A predefined state-balance measurement identifies a shared direction across the analyzed ASD-related resources, with the clearest evidence in Dup15q. Component analysis favors lower Population A-associated expression over a generalized increase in Population B-associated expression. This defines an interpretable molecular phenotype for mechanistic follow-up without assigning either population to a purified cell subtype.