Imputation of brain-regional transcriptomes from dorsolateral prefrontal cortex gene expression: the dlpfcgenie

Over the past two decades of postmortem brain transcriptomic research in neuropsychiatry, the overwhelming majority of studies have focused primarily on RNA quantification in DLPFC, despite many other brain regions holding pathophysiological significance. The development of brain-regional transcriptome imputation methods, including our Brain Gene Expression and Network Imputation Engine (BrainGENIE) algorithm, provide a resource for us to impute brain-regional gene expression profiles for regions that have either been uncollected or sampled in smaller quantities by leveraging more widely accessible DLPFC transcriptomes. We introduce our novel DLPFCGENIE algorithm, which we trained using postmortem brain data released by the Genotype Tissue Expression (GTEx) Consortium. After confirming that our DLFPCGENIE replicates its predications in four completely independent datasets, we deployed DLFPCGENIE to the PsychENCODE dataset to examine gene expression changes associated with autism spectrum disorders (ASDs) (ASD n=66, controls n=77). By combining our findings with an atlas of transcriptomic results across the cortex,2 we created a compendium of gene expression changes related to ASD in 17 brain regions. Building upon our previous work that examined molecular correlates of brain-structural differences, we analyzed the relationship between ASD-related changes in brain-regional gene expression with brain-structural differences associated with ASD. Among the ASD-related changes in gene expression, we identified 99 genes that were associated with brain-structural differences, which were found to be enriched in autophagic gene sets responsible for the breakdown and turnover of proteins. Moreover, cell-type-specific deconvolution analyses revealed that ASD-related changes in oligodendrocytes were associated with brain-structural differences. Our DLPFCGENIE algorithm provides a valuable resource to investigate transcriptional features linked with ASD, and enabled us to relate transcriptomic features and cell abundance with brain-structural differences.