Gene expression prediction

PrediXcan, a recently published software that imputes tissue-specific gene expression levels from subject’s genetic profile²⁴, was used to assess gene expression differences between patients from AS and Non_AS trios. Briefly, PrediXcan makes use of available expression quantitative trait loci (eQTLs) data from GTEx project³⁵ to predict expression levels in a tissue-dependent manner and compares expression levels across different phenotypes. Genotype files from ASD trios with available diagnose information (N = 241) were converted from.bed to.dosage format. Predixcan was run with --predict and –assoc --logistic arguments to predict expression levels per individual and perform statistical comparisons per gene between AS and Non_AS subjects. Given the extensive literature connecting cortical dysfunctions and autism, either from imaging³⁶, gene expression^37–39 or rare disrupting variation⁴⁰ studies, GTEx brain frontal cortex was used as reference transcriptome. Differences in predicted gene expression in frontal cortex between AS and Non_AS patients were assessed.

S-PrediXcan²⁵, an extension of PrediXcan software that infers its results using only summary statistics from GWAS, was then used on GWAS summary data from ASD and comorbid disorders with significant polygenic transmission (SCZ, MDD, ADHD) to estimate gene expression differences between cases and controls using imputed expression data from GTEx (brain frontal cortex).

Correlation between gene expression differences between AS and Non_AS (measured by per gene Z-values derived from PrediXcan test) and gene expression differences between cases and controls from comorbid disorders (measured by per gene Z-values derived from S-PrediXcan test) was analyzed. This relation was studied under different P thresholds from S-PrediXcan.

GTEx expression data from lung and cerebellum, other brain tissue with less consistent evidences in psychiatry that frontal cortex⁴¹ were used as comparative controls. Body mass index (BMI) summary statistics³² were also used in S-PrediXcan analyses in same tissues as a negative control, as it has not been associated with ASD across literature and repeatedly used as control. Finally, Alzheimer disease (ALZ) summary statistics⁴² were used in S-PrediXcan analyses in same tissues as a comparison between neurodevelopment and a neurodegenerative disorder.