Gene co-amplification and structural patterns reveal principles of extrachromosomal DNA in cancer

The amplification of oncogenes on extrachromosomal DNA (ecDNA) enables aggressive, rapidly evolving tumors. Its defiance of Mendelian segregation enables extreme copy number amplifications that escape chromosomal regulatory constraints. Our AmpliconSuite toolset is the most widely used tool for ecDNA analysis in whole genome sequencing data, now deployed on over 43,000 tumor samples. Here, we present novel insights into ecDNA biology enabled through large-scale integrative analysis with these methods. We analyzed 6,366 whole-genome sequenced tumors from combined ICGC and Hartwig Medical Foundation datasets, identifying 2,366 distinct ecDNA capturing >11,000 different genes. Our systematic gene co-amplification analysis revealed striking patterns of gene selection on ecDNA. CDK4-MDM2 co-amplification occurred predominantly via ecDNA (75% of co-amplifications). These loci, normally separated by >11Mbp on chromosome 12, preferentially assembled onto the same ecDNA molecule 93% of the time as revealed by structural analysis of the ecDNA. This suggests selective pressure for maintaining cell cycle and p53 pathway regulators on the same inheritable unit. Our analysis of frequent co-amplifications also revealed that ecDNA preferentially packages chromatin remodelers (NSD3, RSF1) alongside driver oncogenes, as well as including genes that support transcription and translation (INTS4, BRF2), creating self-contained oncogene “support” hubs. Structural analysis revealed cancer type-specific patterns to ecDNA structures, with EGFR ecDNA showing simple architectures in glioblastoma versus complex rearrangements in lung and breast cancers, suggesting distinct formation histories in different cancers. Through AmpliconRepository.org, we provide public access to these ecDNA predictions and co-amplification analysis across major cancer cohorts, currently hosting 16,000+ analyzed samples and 5,000+ characterized ecDNA amplifications. Uniquely open to community contributions, this resource enables researchers to explore patterns across datasets and validate findings. These findings reveal fundamental principles governing ecDNA formation and selection, with implications for understanding tumor heterogeneity, therapeutic resistance, and dependencies which underlie ecDNA-targeted therapies.