Polygenic Heterogeneity in Autism and in Related Phenotypes

Background: The aetiology of Autism Spectrum Disorder (ASD) is complex and largely unknown, but ASD is highly heritable with common variation estimated to explain half of the genetic risk. Shared genetics between different psychiatric disorders is well documented and genetic correlation of psychiatric disorders with phenotypes such as educational attainment and IQ has been reported. Recently, iPSYCH and Broad/MGH conducted the largest GWAS of ASD to date comprised of 13,076 cases and 22,664 controls from Denmark and subsequently meta-analyzed with 10,610 samples from the Psychiatric Genomics Consortium (PGC). As part of these analyses, we introduced a novel polygenic risk score (PRS) analysis utilizing the genetic overlap with related phenotypes to show for the first time genetic differences between the ICD10 ASD diagnostic subgroups (childhood autism, atypical autism, Asperger’s syndrome, other pervasive developmental disorder, pervasive developmental disorder, unspecified ).

Objectives: We expand these novel PRS analyses to ASD and attention deficit hyperactivity disorder (ADHD) comorbid subgroups (ASD/noADHD, ASD+ADHD, ADHD/noASD) and new and more powerful PRS.

Methods: iPSYCH is a population sample comprised of cases with one or more psychiatric diagnoses identified in the Danish Psychiatric Central Research Register and a 2% random sample from the corresponding birth cohort as controls. Archived dried blood spot samples for participants were identified in the Danish Neonatal Screening Biobank. DNA was extracted, whole-genome amplified and genotyped on the PsychChip, a customized HumanCoreExome chip. QC, imputation, PCA and polygenic risk scoring were conducted using the Ricopili pipeline of PGC. For the PRS, we used summary statistics from the latest GWAS of educational attainment (EA, n~766,000), intelligence (n~270,000) and 12 neuroticism items from the UK Biobank (n~200,000). These PRS were regressed on the ASD diagnostic and ASD and ADHD comorbid subgroups in a multivariate multiple regression model adjusted for principal components and genotyping batches. This model accounts for the correlation between different PRS and makes it possible to test an array of hypotheses.

Results: With the more powerful PRS, we reproduced our previous findings that sharing of alleles between ASD and EA is concentrated in Asperger’s syndrome and childhood autism, but not in the other ASD diagnostic subtypes. Moreover, the EA and intelligence PRS profiles of the ASD diagnostic subgroups are now distinguishable (p=0.048). With respect to PRS for neuroticism items, we found only minor differences across ASD diagnostic subgroups but marked differences across ASD and ADHD comorbid subgroups. For example, there were significant trends for mood swing PRS (p=1.0x10^-88) and feeling fed-up PRS (p=6.2x10^-74) across the comorbid subgroups, and while ASD/noADHD and ASD+ADHD had higher PRS scores for worrying too long after embarrassment, ADHD/noASD did not (p=8.1x10^-14).

Conclusions: Although current ASD GWAS are too small to identify individual genetic loci associated with ASD subgroups, using a polygenic perspective and powerful regression method we can distinguish ASD clinical subgroups . Moreover, we can begin characterizing ASD subgroups by their PRS loading for selected behavioural phenotypes, thereby, leveraging the power of GWAS to chip away at the multidimensional biology behind ASD.