Marc J. Williams

High-level copy-number (CN) amplification (HLAMP) is a major mechanism of oncogene activation in human cancer. Despite progress in therapeutically targeting HLAMPs, the processes underlying HLAMP evolution are incompletely understood, leaving critical knowledge gaps in their etiology and mechanisms of therapeutic response. To address this, we analyzed using novel computational approaches, the evolutionary trajectories of HLAMP in single-cell whole-genome sequencing of 86,239 cancer cells from 94 patients and 8 experimental systems. We found that two common etiologies of oncogene amplification - intrachromosomal amplification (ICamp) and extrachromosomal circular DNA (ecDNA) - lead to distinct CN distributions across cells in amplitude, variance and clone specificity. Notably, ICamp events exhibited widespread subclonal specificity, indicating their dynamic evolution through numeric or structural modulation associated with clonal expansions. In contrast, ecDNAs exhibited higher cell-to-cell CN variation linked to structural rearrangements, whereby complex ecDNA architectures were resolved at single-nucleotide resolution in individual cell genomes. Through joint analysis of ecDNA structure and their cellular abundance, we observed both divergent and convergent modes of ecDNA evolution shaped by DNA damage, selection, and chromosomal re-integration. Finally, modeling single-cell distributions of ecDNAs substantially improved bulk genome-graph based predictions, revealing that ecDNA prevalence is tissue-type specific, rather than broadly shared across cancer types, with implications for ecDNA-directed therapy.

Per genomic bin copy number calls (gains=3, diploid=2, loss=1) for data presented in Figure 1.

Per segment raw sWGS copy number data for patients presented in Figure 2.

Per genomic bin copy number calls for single-cell data presented in Figure 3.

Cohort information per sample for the longitudinal metastasis cohort (Figure 4), including description of organ, timepoint sampled, tumour stage at diagnosis, whether the sample is treatment naïve, if the sample was treated in the prior interval, the number of days and months elapsing since the first timepoint, in addition to per sample genomic metrics such as PGA, number of segments, patient ploidy, per sample GD classification and overall patient GD classification.

Supplementary note where we describe the gland mathematical modelling and simulations

Clinical information for CRC cohort presented in Figure 1.

Per sample, per segment, raw copy number data (log2 ratios) and integer calls for the longitudinal metastasis cohort presented in Figure 4.

Per patient clinical information for the CRC cases presented in Figure 2.

Pairwise comparison metrics for data presented in Figure 1.