Integrative long-read sequencing, RNA-seq and machine learning framework characterizes and predicts autosomal silencing in unbalanced X-autosome translocations.

PURPOSE: X-chromosome inactivation (XCI) equalizes dosage between females (XX) and males (XY), yet its functional and clinical consequences remain challenging to interrogate in humans. Unbalanced X-autosome translocations constitute a natural model in which skewed inactivation of the derivative chromosome can spread into autosomes, contributing to phenotypic variability. Evaluating this spread and possible mediating factors is therefore essential for accurate clinical management. METHODS: Using allele-resolved long-read sequencing and RNA sequencing, we investigated DNA methylation and gene expression across autosomal regions involved in unbalanced X-autosome translocations. RESULTS: We identified gene-level variability in autosomal silencing, marked by promoter hypermethylation, gene body hypomethylation, and reduced gene expression with clinical impact in affected individuals. Using epigenetic and genomic features around silenced and active autosomal genes reported here and in the literature, we developed an interpretable machine-learning model that identified intrinsic sequence and chromatin features associated with susceptibility or resistance to silencing. This framework enables discrimination between silenced and active autosomal genes and prioritization of clinically relevant genes. CONCLUSION: Our findings reveal principles underlying the variable spreading of XCI into autosomal genes, supporting a model in which intrinsic DNA sequences and chromatin features modulate the extent of XIST-mediated silencing and its clinical impacts.