Characterizing associations between disruptive de novo rare variant burdens and phenotypic combinations in over 3,000 autistic children: towards building a public clinical genetic resource

The genetic architecture of autism spectrum disorders is scaffolded by both common variants (as described in the latest PGC-ASD freeze) and rare variants. Within the rare variant space, 373 genes are now associated with neurodevelopmental disorders (NDDs) including autism and intellectual disability (Autism Sequencing Consortium, Fu, et al, Nat Genet, 2022). Carrying a disruptive de novo variant in these genes is associated with substantial phenotypic differences starting early in development, such as delays in walking and talking (Kuo, et al, JAMA Pediatr, 2022). While de novo variant burdens have been linked to single phenotypes in autism, de novo variant burdens have rarely been evaluated for different combinations of phenotypes. By assessing these genotype-phenotype relationships across multiple autism cohorts, we aim to reliably (1) determine which phenotypic combinations are most associated with carrying a disruptive variant, and (2) predict the probabilities with which an autistic child may carry a disruptive variant based on their individual combination of phenotypes. We harmonized complete whole exome sequencing and phenotyping data in a combined sample of 3,440 autistic children aged 4 to 17 years from two independent cohorts. Disruptive variants were defined from de novo SNV/Indel/CNV calls as genomic disorders, protein truncating variants, copy number deletions, or missense variants with MPC score>2 in any NDD-associated gene. 30 phenotypes were included that were reportable by age 4, spanning demographics, co-occurring diagnoses, developmental milestones, and cognitive abilities. We built logistic regression prediction algorithms using LASSO-selected phenotypes to predict whether a child carries a disruptive variant. We defined ∼60,000 possible combinations of phenotypes and estimated average predicted probabilities for each phenotypic combination using 10,000 bootstrap predictions. The overall disruptive variant rate was ∼7.5% in autism and ∼1.3% in the general population. Phenotypes that were most associated with carrying a disruptive variant included delayed walking, lower IQ, and congenital heart anomalies. Average predicted probabilities of carrying a disruptive variant ranged from under 1% in children without intellectual disability or congenital heart anomalies who began walking by 12 months, to over 55% in children with intellectual disability and congenital heart anomalies who began walking after 18 months. Across phenotypic combinations, the ranges of predicted probabilities were sufficiently narrow to provide meaningful information about whether an autistic child has an elevated probability of carrying a disruptive variant compared to the general population. Phenotypic combinations in early development can usefully predict probabilities of carrying rare variant burdens for autistic children. By comparing these predicted probabilities to average probabilities of rare variant burdens in the general population, we can translate these genotype-phenotype associations into real-world clinical recommendations. We have therefore begun implementing our prediction algorithm in a public-facing web resource for clinicians and families of autistic children. Given the accessibility of our algorithm's input phenotypes in pediatric care settings, our novel resource can inform referrals for clinical genetic testing and support early genetic diagnosis in autism.