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

Spontaneous genomic rearrangements are a major source of genetic diversity in cancer and the cause of numerous human disorders. While most genome structural variants (SVs) can be readily categorized into the canonical forms—deletion, duplication, inversion, and translocation—there is growing evidence that a nontrivial fraction are complex genomic rearrangements (CGRs) composed of multiple clustered breakpoints that cannot be explained by a single DNA end-joining or recombination event (Quinlan and Hall 2012). The existence of CGRs is a very old observation in both the human genetics and cancer fields. Over the years, at least 251 complex rearrangements have been cytogenetically defined in patients suffering from sporadic human disorders (Zhang et al. 2009a), and innumerable complex karyotypic configurations have been reported in human tumors (Mitelman 1994), albeit generally at very low resolution. There are also reports of complex cancer gene amplification events including multifocal clusters (for review, see Albertson 2006), highly rearranged “amplisomes” (Raphael and Pevzner 2004), and chromosome-limited “firestorms” (Hicks et al. 2006). New, however, is the apparent prevalence of CGRs as revealed by modern genome-wide methods, and the mechanisms put forth to explain them. The initial suggestion that complex SVs might be widespread came from a series of studies characterizing genomic rearrangements associated with sporadic human disorders (Lee et al. 2007; Carvalho et al. 2009; Zhang et al. 2009b). Of 61 nonrecurrent pathogenic mutations, 41% were found to be complex, generally exhibiting multiple adjacent copy number alterations (CNAs) and intra-chromosomal rearrangements. Taking into account previous (for review, see Zhang et al. 2009a) and subsequent studies (Zhang et al. 2010a,b; Choi et al. 2011; Liu et al. 2011a,b; Chiang et al. 2012), these results argue that a large fraction of spontaneous germline mutations are complex in nature. Supporting this, 5%–16% of inherited and presumably benign SVs in mouse (Quinlan et al. 2010) and human (Conrad et al. 2010; Kidd et al. 2010) exhibit multiple clustered breakpoints and/or small-scale insertions or rearrangements at the breakpoint of a larger SV. Complex germline SVs have generally been explained by replication-based models such as fork stalling and template switching (FoSTeS) (Lee et al. 2007), and microhomology-mediated break-induced replication (MMBIR) (Hastings et al. 2009a). Cancer genome sequencing experiments have revealed highly complex genomic rearrangements involving tens to hundreds of breakpoints that appear to have arisen through a single catastrophic mutational event termed “chromothripsis” (Stephens et al. 2011). The investigators proposed a mechanism involving shattering of large chromosomal regions, perhaps by ionizing radiation or one dramatic cycle of breakage-fusion-bridge, followed by double-strand break (DSB) repair. There is also evidence that chromosome missegregation can generate DSBs (Janssen et al. 2011), and formation of micronuclei at lagging chromosomes can pulverize chromosomes in a manner that might lead to chromothripsis (Crasta et al. 2012). Chromothripsis is likely the same phenomenon as “firestorms,” originally identified in breast cancer array-CGH experiments and found to correlate with patient survival (Hicks et al. 2006). The true incidence of chromothripsis in cancer, and whether or not different tumor types are more or less susceptible, remain open questions. These questions have been difficult to address because studies have used different methodologies and definitions. Microarray-based estimates of chromothripsis range from 2% to 3% in a diverse set of 746 cancers (Stephens et al. 2011), 1.3% of 764 multiple myelomas (Magrangeas et al. 2011), and 13% of 98 medulloblastomas (Rausch et al. 2012). However, identification of CGRs from microarray data is problematic, and the first two studies appear to have used subjective definitions of chromothripsis, while the latter used a relatively broad definition (10 CNAs on a single chromosome) and enriched for TP53 mutant tumors. Genome sequencing experiments suggest that the true incidence may be higher, at least in certain tumors: five of 25 bone cancers (20%) (Stephens et al. 2011) and 10 of 87 (11%) neuroblastomas showed chromothripsis (Molenaar et al. 2012). Further clouding the issue, prostate cancer genome sequencing has revealed highly complex chains of balanced rearrangements that do not fall under current definitions of chromothripsis (Berger et al. 2011). Interestingly, the incidence of chromothripsis in medulloblastomas correlates with TP53 loss (Rausch et al. 2012), indicating a potential link to DNA damage response or apoptosis and suggesting that different tumors may have a variable incidence depending on genetic background. However, the relationship is likely to be more complicated since no association between genic mutations and chromothripsis was detected in neuroblastomas despite whole-genome mutation data (Molenaar et al. 2012). The human genetics and cancer fields have converged with the description of chromothripsis events in the germline that closely resemble those reported in cancer cells (Kloosterman et al. 2011a, 2012; Chiang et al. 2012), and with the proposition that the DNA replication-based mechanisms originally proposed to explain relatively mild germline CGRs may also underlie chromothripsis in cancer (Liu et al. 2011b). This raises the important question of whether or not complex mutations in germline and somatic cells have a common origin. It has been difficult to address this question because most events characterized thus far in germline lineages are relatively mild, presumably due to ascertainment bias related to selective pressures acting during early development, and because cancer genome studies have thus far focused on the most complex subset of events. Here, we perform a systematic screen for CGRs in 64 tumor genomes, use de novo assembly to profile rearrangement breakpoints at single-base resolution, and compare mutational signatures and intra-tumor allele frequencies at both simple and complex mutational events.