Transcriptomic and Cellular Decoding of Regional Brain Vulnerability to Neurodevelopmental Disorders

Neurodevelopmental disorders are a major public health concern but remain poorly understood in biological terms. In particular, although we know that genetically-distinct neurodevelopmental disorders induce distinct changes in regional brain anatomy, we do not understand how such specificity arises. Here, by leveraging diverse neurodevelopmental disorders of known genetic origin, we show that this specificity can be explained by cell-type dependent gradients of gene expression that are evident in the healthy brain. We first used in vivo structural neuroimaging to map altered cortical organization in neurodevelopmental disorders arising due to 6 different genomic copy number variation (CNV) disorders, including both duplications (chromosomes X, Y and 21) and deletions (X-chromosome, 22q11.2, 11p13). Comparisons with publicly-available postmortem gene expression maps from healthy adults revealed that the spatial pattern of anatomical change in each disorder was preferentially associated with the spatial expression profile for genes within the causal CNV region. Next, by gathering a comprehensive set of single-cell gene expression signatures, and using normative postmortem data to map expression gradients for each canonical cell class in the brain, we identified specific cell-classes that expressed CNV genes and closely tracked the spatial pattern of cortical disruption in each disorder (e.g. NCAM2-expressing oligodendrocyte precursors in Down syndrome, MAPK1-expressing inhibitory neurons in del22q11, PTCHD1-expressing astrocytes in X-chromosome aneuploidies). Finally, we used two orthogonal approaches to validate our imaging-transcriptomic associations against direct measures of altered gene expression in tissue from CNV carriers. First, we demonstrated that CNV genes with differential dosage sensitivity in patients show patterns of brain expression that were differentially correlated with cortical disruption. Second, we established that CNV patients with more severely disrupted gene expression in blood-derived tissue show more extreme cortical disruption in brain MRI. Thus, cell-type dependent gradients of gene expression that are intrinsic to the human brain can be used to infer disease-specific drivers of regional brain vulnerability without reference to any postmortem brain tissue in patients. Furthermore, this "transcriptional vulnerability model" for prediction of regional neuroanatomical disruption makes it possible to estimate the severity of altered brain organization in a given patient from the severity of altered gene expression in their blood.