A multilayered in silico analysis links UHRF1, DNA methylation and developmental chromatin memory to lineage-dependent prognosis in gastric, renal and adrenal cancers
Aberrant DNA methylation is a hallmark of cancer, but its clinical interpretation remains debated. UHRF1, a key epigenetic adaptor for DNA methylation maintenance and chromatin bivalency regulation in embryonic stem cells, is frequently overexpressed yet shows context-dependent prognostic behaviour. By integrating bulk and single-cell transcriptomics, CpG-resolution methylation, developmental chromatin states, immune profiling and clinical outcomes across gastric (STAD), clear-cell renal (KIRC) and adrenal (ACC) carcinomas, we identified a four-class UHRF1-embryonic morphogenesis (UHRF1-EM) framework resolving this paradox. This axis revealed an inverse prognostic pattern: whilst across all three tumours EM-low and EM-high states mark better or worse prognosis, respectively, UHRF1-high levels associate with favourable outcome in STAD (UH-EML), and unfavourable in KIRC and ACC (UH-EMH). The classification proved reproducible and independently prognostic after adjustment for stage and molecular subtypes, outperforming existing classifiers and exceeding pathological stage in KIRC and ACC. Multivariable models incorporating UHRF1-EM yielded uniformly positive {Delta}C-indices. Hypermethylation associated with the UHRF1-EM axis was enriched at ESC bivalent developmental loci (EM and oncofoetal genes), but not at housekeeping cell-cycle sites. In STAD, this pattern was related to oncofoetal gene downregulation and best prognosis, whereas in KIRC and ACC it matched with gene-body/enhancer methylation, higher EM expression, immunosuppressive microenvironments and worst prognosis. Together, these findings establish the UHRF1-EM axis as a clinically robust molecular classifier and support a mechanistic model in which tumour-specific epigenetic engagement of developmental loci may contribute to the prognostic inversion, providing a foundation for further mechanistic experimental validation.