Breakpoint profiling of 64 cancer genomes reveals numerous complex rearrangements spawned by homology-independent mechanisms

Abstract

Tumor genomes are generally thought to evolve through a gradual accumulation of mutations, but the observation that extraordinarily complex rearrangements can arise through single mutational events suggests that evolution may be accelerated by punctuated changes in genome architecture. To assess the prevalence and origins of complex genomic rearrangements (CGRs), we mapped 6179 somatic structural variation breakpoints in 64 cancer genomes from seven tumor types and screened for clusters of three or more interconnected breakpoints. We find that complex breakpoint clusters are extremely common: 154 clusters comprise 25% of all somatic breakpoints, and 75% of tumors exhibit at least one complex cluster. Based on copy number state profiling, 63% of breakpoint clusters are consistent with being CGRs that arose through a single mutational event. CGRs have diverse architectures including focal breakpoint clusters, largescale rearrangements joining clusters from one or more chromosomes, and staggeringly complex chromothripsis events. Notably, chromothripsis has a significantly higher incidence in glioblastoma samples (39%) relative to other tumor types (9%). Chromothripsis breakpoints also show significantly elevated intra-tumor allele frequencies relative to simple SVs, which indicates that they arise early during tumorigenesis or confer selective advantage. Finally, assembly and analysis of 4002 somatic and 6982 germline breakpoint sequences reveal that somatic breakpoints show significantly less microhomology and fewer templated insertions than germline breakpoints, and this effect is stronger at CGRs than at simple variants. These results are inconsistent with replication-based models of CGR genesis and strongly argue that nonhomologous repair of concurrently arising DNA double-strand breaks is the predominant mechanism underlying complex cancer genome rearrangements.

Identification of complex genomic rearrangements in cancers using CouGaR

“…CouGaR identified at least one contig (≥3 copies) in 443 of the 467 samples analyzed, and on average we found 6.25 contigs per sample. Interestingly, the 6.25 contigs per sample is nearly twice as many as previously reported (Malhotra et al 2013). The lower number reported earlier could be attributed to both lack of methods that were capable of identifying these CGRs accurately and to the smaller number of samples analyzed.…”

“…It has been previously reported that CGRs are found in 5%-9% of all cancers (Malhotra et al 2013;Zack et al 2013) and ∼25% of bone tumors (Stephens et al 2011). However, these estimates depend on three factors: first, the type of data utilized (array or sequencing); second, the computational method implemented; and third, the specific types of cancer samples analyzed.…”

“…We also compared CouGaR with two published methods, BAMBAM (Sanborn et al 2013) and nFuse (Malhotra et al 2013), on DIPG29. On this sample, CouGaR reported 16 breakpoints, while BAMBAM reported 157 and nFuse reported 25 breakpoints.…”

“…More recently, certain types of large-scale complex genomic rearrangements (CGRs), such as chromothripsis, breakage-fusion-bridges, and double minutes, have been discovered within tumor genomes and implicated in tumorigenesis ). CGRs involve three or more distant regions of the genome abnormally joining together and have been implicated in 5%-9% of all cancers (Malhotra et al 2013;Zack et al 2013) and ∼25% of bone tumors (Stephens et al 2011). These rearrangements can form distinct highly amplified contigs that harbor oncogenes, resulting in 10-to 100-fold increases in oncogene copy count, which may potentially drive tumorigenesis (Korbel and Campbell 2013).…”

Discovery of recurrent structural variants in nasopharyngeal carcinoma

“…Except for small insertions, SMASH consistently detected breakpoints with allelic frequencies >1.7%. To see if SMASH produces results comparable to other SV algorithms, we compared its breakpoint detection to HYDRA (Quinlan et al 2010;Malhotra et al 2013) on a set of simulated breakpoints (Supplemental Material; Supplemental Table T3). Overall, SMASH breakpoint detection rates were similar to HYDRA, with SMASH more accurately resolving breakpoint coordinates and HYDRA more consistently detecting short insertions.…”