TAD boundary disruption in Xq27-q28 deletions rewires 3D chromatin architecture and may predispose to primary skewed X chromosome inactivation.

In placental females, X chromosome inactivation (XCI) establishes random monoallelic expression as XIST RNA spreads along a pre-existing three-dimensional (3D) chromatin contact network. How structural variants reshape this 3D landscape to influence XCI choice remains unclear. We previously reported two female individuals with overlapping Xq27-q28 deletions but discordant XCI, skewed in one and random in the other. By re-evaluating 17 reported cases, we found that skewing in a subset was not readily explained by classical routes of primary bias from the canonical XCI machinery or by secondary cell selection. Using the deep-learning model Orca, we predicted that the deletion in the skewed case would fuse disparate topologically associating domains (TADs) into a neo-TAD. Analysis of human preimplantation Hi-C data showed that the relevant TAD boundaries were already apparent by the 8-cell stage, before stable XCI. 3D DNA-FISH revealed reduced spatial distances between selected loci, consistent with the Orca-predicted neo-TAD architecture, in patient-derived naïve iPSCs and lymphoblastoid cells. Orca-predicted neo-TAD formation was also identified in three of six random-XCI cases, indicating that it is insufficient on its own to bias XCI. We identified 86 candidate Polycomb-associated loci with chromatin features consistent with Polycomb seed sites on the human X chromosome. The skewed case showed the most extensive disruption of these candidate loci, although candidate-locus disruption was also observed in random-XCI cases. We therefore propose a working seed convergence model in which the convergence of predicted neo-TAD formation and extensive disruption of candidate Polycomb-associated loci may predispose to skewed XCI.