跨>8,000例TCGA全基因组的结构变异驱动因素的系统性发现与分类

Structural variants (SVs)-large-scale genomic deletions, duplications, inversions, and translocations-can promote tumorigenesis by activating proto-oncogenes, disrupting tumor suppressors, generating oncogenic fusions, rewiring gene regulation, and mediating catastrophic events such as chromoplexy and chromothripsis. Yet, the role of SVs as cancer-driving mutations remains less comprehensively characterized than that of single-nucleotide variants or indels, largely due to the historic scarcity of tumor whole-genome sequencing data required for accurate SV detection. In this study, we analyzed whole-genome sequencing data from >8,000 tumor-normal pairs spanning 31 cancer types from The Cancer Genome Atlas (TCGA) to systematically characterize SVs at unprecedented scale. Compared with the flagship Pan-Cancer Analysis of Whole Genomes (PCAWG) project, our analysis includes roughly four times as many samples and six additional cancer types. Somatic SVs were identified using a custom pipeline combining Manta and dRanger with optimized downstream filters. Across all tumors, we detected >1 million somatic SVs. We developed two complementary frameworks to interpret these variants. First, to classify SVs by their genomic architecture, we inferred genomic segments and their associated copy number alterations driven by SVs ranging from a single to hundreds of breakpoints, enabling the generalization of distinct patterns through unsupervised clustering. Second, to identify candidate driver genes, we developed SVelfie, a statistical framework that detects genes significantly enriched in functional SVs predicted to confer gain- or loss-of-function effects. Applying SVelfie to 385 prostate and 333 ovarian cancer genomes revealed multiple novel candidate driver genes, with additional discoveries expected as analysis extends to the full >8,000-sample dataset. This work represents the most comprehensive analysis to date of SV drivers in cancer. By uniting large-scale WGS data with new computational frameworks for SV classification and driver detection, we expand the catalog of SV-driven cancer genes, illuminate mechanisms of SV-mediated oncogenesis, and advance the clinical utility of whole-genome sequencing in precision oncology.