Transcriptomic signatures of early- versus late-diagnosed ADHD and implications for treatment heterogeneity.

Publication date: Jul 07, 2026

Attention-deficit/hyperactivity disorder (ADHD) is heterogeneous in the age at which symptoms first lead to clinical diagnosis. Recent genomic analyses of a large Danish population-based cohort showed that individuals diagnosed in childhood differ from those diagnosed in adulthood in polygenic risk scores, rare-variant burden, and patterns of genetic correlation with other psychiatric traits [1]. Childhood cases exhibited stronger overlap with autism and higher rates of rare protein-truncating variants in constrained genes, whereas late-diagnosed cases showed greater genetic overlap with depression. The functional transcriptomic mechanisms underlying these differences have remained unclear. We performed a transcriptome-wide association study (TWAS) using S-PrediXcan and GTEx v8 MASHR models from six adult brain tissues: amygdala, anterior cingulate cortex, caudate, frontal cortex, hippocampus, and nucleus accumbens. Analyses were based on GWAS summary statistics from the iPSYCH study for childhood ADHD (N = 14,878) and late-diagnosed ADHD (N = 6,961). Thirteen curated gene sets were examined, spanning synaptic plasticity, immune function, glial markers, and housekeeping controls. The transcriptomic profiles of the two subgroups showed substantial overall similarity (Spearman ρ = 0. 77 across 13,580 genes), with near-perfect directional concordance among nominally significant associations; because the two GWAS contrasted different case groups against the same 38,303 population controls, these similarity estimates should be interpreted as upper bounds rather than unbiased estimates of aetiological overlap. Despite this shared architecture, gene-set analyses suggested biologically interpretable divergences: late-diagnosed ADHD displayed stronger enrichment in long-term potentiation (enrichment ratio = 1. 34, Mann-Whitney p = 1. 42 cD7 10⁻⁸) and complement cascade pathways (1. 22, p = 5. 14 cD7 10⁻⁵), whereas childhood ADHD showed only nominally greater signals in microglia marker and MHC genes. Eighty-two genes differed in association strength between groups after FDR correction, although this delta-Z comparison should be read cautiously because the shared-control design violates the independence assumption of the test. These findings provide a functional explanation for previously reported genetic and comorbidity differences by age at diagnosis and identify an 82-gene differential list for follow-up. The enrichment patterns generate cautious, hypothesis-level questions regarding potential variation in stimulant response and tolerability, but no direct treatment, adherence, or pharmacogenomic data were analysed. This represents the first TWAS to dissect ADHD subtypes by age at first diagnosis and offers a bridge between statistical genetic associations and biologically interpretable pathways relevant to clinical heterogeneity.

Semantics

Original Article