Comparing the Phenotypic and Genetic Differences Between Autistic Regression and Autism Without Regression Using the SPARK Cohort

The biology of autistic regression (Autreg) remains poorly understood. We investigated Autreg's phenotypic and genetic landscape in 51 633 individuals from the SPARK cohort. We calculated the frequencies of parental-reported language regression and regression affecting social, play, toileting, and/or other developmental domains, and ages at skill regression and regain within SPARK. We compared developmental milestones, quantitative developmental measures, and selected other phenotypes in Autreg and autism without regression (Autnreg). Rare damaging coding variants and polygenic scores (PGS) in Autreg and Autnreg were compared using genome sequencing data. Rare variants were analyzed using the Transmission And De novo Association (TADA) framework in 1 107 Autreg and 974 Autnreg individuals, followed by pathway enrichment analysis. PGS for autism, attention-deficit/hyperactivity disorder, epilepsy, and schizophrenia were initially compared. PGS validation was conducted in 14 001 Autnreg and 11 638 Autreg individuals using larger SPARK genotyping data. We identified regression in 43% of autistic individuals. Other regression occurred later than language regression, and 60% of skill regain occurred within two years. Autreg was associated with lower Vineland-3 Adaptive Behavior Composite scores (p<0.001) and >twofold increased odds of epilepsy and intellectual disability (p<0.001). Developmental milestones and quantitative measures had modest differences. TADA analysis highlighted genes involved in nucleotide catabolism, transcription, and focal adhesion, with PTEN emerging as a strong candidate for Autreg. Schizophrenia PGS explained 0.087% of variance in liability to regression among autistic individuals. In conclusion, Autreg was associated with increased neurodevelopmental morbidity. Schizophrenia PGS and rare variants explained a minimal fraction of its etiology.