Evidence for Widespread Associations between Broadly Expressed Genes in Blood Leukocytes and Neural Systems Response to Speech in Autism

Background: It has recently been suggested that complex phenotypes (e.g., ASD) may be understood through an omnigenic model, whereby a vast majority of genes may have some non-zero effect on the phenotype (Boyle, Li, & Pritchard, 2017, Cell). Genes broadly expressed across many tissues including the brain are numerous and may contribute significantly to the phenotype. This idea suggests broadly expressed genes measured in-vivo in non-brain tissue (e.g., blood) could be relevant for explaining variability in ASD neural phenotypes.

Objectives: To examine how widespread variation in the blood leukocyte transcriptome is associated with in-vivo neural systems response to speech in ASD subtypes and typical development. The omnigenic model predicts that association signals would be widespread across most of the transcriptome and that co-expression modules highly enriched in broadly expressed genes would harbor most of the ASD-relevant association signal.

Methods: Three age-matched groups of toddlers (mean age ~29 months) were investigated - two ASD subtypes classified as either good (n=40) or poor language outcome (n=41) (Lombardo et al., 2015, Neuron), and n=37 typically developing toddlers. Neural systems response to speech was measured using fMRI during natural sleep using a story-language paradigm identical to prior work (Lombardo et al., 2015, Neuron). RNA was measured on all individuals using Illumina microarray chips as described in previous work (Pramparo et al., 2015, JAMA Psychiatry). Weighted gene co-expression network analysis (WGCNA) was used as feature reduction step. Partial least squares (PLS) analysis was used to reveal large-scale gene expression-fMRI associations. Association signal across modules was then tested for enrichment with tissue-specific or broadly expressed gene lists directly taken from Boyle et al.’s analysis of GTEx data.

Results: Only one latent variable gene expression-fMRI pair showed a statistically significant association between module eigengene variation and neural systems response to speech (LV1: d = 83.58, p = 9.99e-5). LV1 accounts for 20.7% of the covariance between gene expression and fMRI data and is spatially constrained to primarily prefrontal and temporal cortex. Reliable non-zero association signal in at least one group was present in 64% of modules (20/31) comprising 74% of all genes investigated. Of these 20 modules, 65% (13/20) were highly enriched for broadly expressed genes (OR = 57.57, p = 0.002). Of the 13 modules where a non-zero association was present in an ASD language outcome subtype, 84% (11/13) were highly enriched in broadly expressed genes (OR = 53.16, p = 0.0001). Other tissue-specific gene lists (e.g., brain, blood, lymphocyte) were not heavily enriched in many modules or over-represented within modules containing non-zero effects.

Conclusions: Rather than finding little to no association between blood leukocyte transcriptomic signal and neural phenotypes for ASD, this work demonstrates that widespread signals in the blood leukocyte transcriptome, particularly within genes broadly expressed across many tissues including the brain, are relevant to ongoing early pathophysiology at the neural systems level in living ASD toddlers. These insights suggest much translational research potential for in-vivo monitoring of such pathophysiology in clinically important contexts (e.g., treatment) in living patients with ASD.