Characterising processing conditions that artifactually bias human brain tissue transcriptomes

Autopsy-derived brain tissue analysis is crucial for understanding neurobiology, but post-mortem handling can introduce artifacts. We studied adult human brain transcriptomic signatures from tissue immediately extracted from brains (< 0 hours) and compared to autopsy brain tissue with short (~6 hours) and long (~36 hours) post-mortem intervals (PMIs). Significant deviations in gene signatures were observed in both short and long PMIs compared to immediately extracted tissue, which we defined as B rain A rtifact G enes (BAGs). By subjecting brain samples to processing variables that are unavoidable in autopsy programs ( post-mortem time and temperature), we characterized a set of artifact-responsive genes and mapped this signature onto matched single-nucleus RNA-seq data, revealing that it was predominantly glutamatergic neurons that exhibited the earliest induction of artifact genes followed by oligodendrocytes later. Using deep learning, we distilled this broader processing-response program into a predictive signature, called Time and Temperature Response genes Underlying Transcriptional Heterogeneity (TTRUTH) and provide an Open Science tool for assigning TTRUTH scores to brain RNA-seq data. Together, this work will help better standardize datasets, enable additional sample stratification, and enhance data interpretation.