染色质网络为人类癌症预后基因表达特征的开发提供依据

Chromatin architecture is a fundamental determinant of transcriptional regulation across cancer types, yet its integration into biomarker development has been limited. Here, we leverage DNA methylation as a surrogate for large-scale 3D genome structure to reconstruct A/B chromatin compartments and derive chromatin interaction networks that capture spatially coordinated gene organization. Using these methylation-informed networks, we define gene communities that reflect shared regulatory environments and potential co-functional behavior. We apply this framework to two prototypical epithelial malignancies-prostate cancer (PCa) and breast cancer (BCa)-using publicly available DNA methylation and transcriptomic datasets to identify chromatin-constrained modules whose coordinated dysregulation marks aggressive disease biology. Across both cancer types, chromatin-driven gene communities show stronger coherence with known biological pathways than modules derived solely from expression data, underscoring the value of incorporating 3D genome context into molecular profiling. Using machine-learning approaches, we develop chromatin-informed prognostic signatures that integrate network topology with transcriptional alterations to predict metastasis and lethal outcome. These signatures demonstrate robust performance across independent datasets and consistently outperform baseline expression-only predictors, highlighting the generalizability of chromatin context as a key source of prognostic information. Functional analyses reveal that signature genes in both PCa and BCa are enriched for processes implicated in tumor progression, including lineage plasticity, hormone receptor signaling adaptation, disruption of chromatin insulation, and loss of compartmental integrity. Although the specific pathways differ in lineage-specific ways-such as androgen receptor signaling in PCa and estrogen receptor circuitry in BCa-the unifying mechanism lies in perturbations of higher-order genome organization that facilitate oncogenic rewiring. Together, these findings demonstrate that methylation-derived A/B compartment structure provides a powerful and generalizable framework for reconstructing chromatin networks and identifying biologically coherent regulatory modules. By embedding transcriptomic alterations within their 3D genome context, we advance chromatin-informed gene expression signatures with strong prognostic utility across multiple cancer types, illustrated here through prostate and breast cancer. Disclosures: AI was used to assist the preparation of this abstract.