A Continuous Constraint-by-Timing Landscape in Autism Common-Variant Risk Genes Without Identifiable Discrete Subtypes

Autism spectrum disorder (ASD) shows substantial phenotypic and genetic heterogeneity. Whether autism-associated genes form a single continuum or discrete molecular groups remains unresolved, partly because gene-list membership is often used both as a clustering input and as a validation target, and partly because reported partitions are rarely tested against the analytic choices that produced them. We performed gene-based association analysis of the iPSYCH-PGC autism GWAS (N = 46,350) using MAGMA and characterized the selected genes on five continuous biological features: evolutionary constraint, cortical cell-type specificity, prenatal-expression bias, peak-expression age and mean brain expression, deliberately excluding curated autism gene lists so that they could serve as held-out validation. Principal component analysis identified two continuous axes capturing 63.4% of variance, a constraint-by-expression-breadth gradient and a developmental-timing gradient, along which SFARI and FMRP genes were significantly shifted (p = 3.6×10⁻⁵ and p = 4.4×10⁻³). A constraint-defined region of this landscape was recoverable across false-discovery selection thresholds and nested across them. Discrete partitions were not. Ward clustering at k = 3 exceeded its feature-shuffled null at q < 0.23, but re-running the pipeline at every threshold from q < 0.10 to q < 0.35 recovered that pattern at only 3 of 26 thresholds, and it inverted one hundredth of a q-unit away in either direction; consensus clustering selected no k; and spectral clustering produced silhouettes that rose with sample size while permutation p-values stayed flat. Autism common-variant risk genes therefore occupy a continuous, unevenly dense biological landscape in which no discrete partition survives the analytic choices used to obtain it.