A common copy-number breakpoint of ERBB2 amplification in breast cancer colocalizes with a complex block of segmental duplications

Gene amplification is a cellular process characterized by a selective increase of a particular genomic region without a proportional increase of the entire genome [1‐4]. The selective increase accompanies the overexpression of a particular gene within the genomic region that confers a growth advantage to the cell. The growth advantage derived from gene amplification has long been recognized as an important problem for cancer patients. Increased copy numbers of proto-oncogenes, such as MYC, MYCN, and ERBB2, leads to the overexpression of oncogene products that drives abnormal cell proliferation [5‐9]. Abnormal cell proliferation results in cancer progression and poor patient survival [10, 11]. In addition, gene amplification is an underlying mechanism for acquired therapy resistance, as cancer cells counteract therapeutic agents by overactivating either therapy-target genes (for example, BCR-ABL amplification) or alternative survival pathways (for example, MET amplification) [12‐17]. Despite these adverse effects on survival of cancer patients, little is known about amplification mechanisms, and in particular, about the initiating processes of gene amplification.

During the processes of gene amplification, extra copies of large genomic segments accumulate in a cell. The accumulation could be initiated either (a) by aberrant recombination that results in the unequal distribution of chromosomal materials between daughter cells [18‐22] or (b) by the loss of DNA-replication control that leads to the extra round of segmental DNA replication within a single cell cycle [23‐25]. In normal cells, these processes are tightly regulated and are less likely to initiate gene amplification [26, 27]. In contrast, cancer cells often lack these controls and could initiate the processes. Furthermore, cellular surveillance systems (checkpoints) that ensure genome integrity at several stages of the cell cycle are impaired in cancer cells [28, 29] and could fail to eliminate cells with extra copies. Once the accumulation is initiated, it could lead to further accumulation by the growth advantage conferred by the amplified gene(s). Therefore, defining initiating processes is the key for the better understanding of the amplification mechanism. However, defining initiation processes in tumors in vivo is not an easy task, as current methods for evaluating gene amplification may not be feasible for capturing the amplification mechanism. Gene amplification has been measured as the increase of copy-numbers of particular genomic regions by array-comparative genomic hybridization (array-CGH) [30, 31]. Although array-CGH covers the entire genome and identifies amplified regions that are important for tumor phenotypes with high confidence, such highly amplified regions may not be the initiating regions but rather the end products of adaptive evolution of cancer genomes. Next-generation sequencing could provide both copy-number profiles and somatic breakpoint sequences in cancer genomes [32, 33]. Because of the copy-number increases, breakpoint sequences tend to be biased toward amplified regions and may represent late events during amplicon formation.

The difficulty in identifying initiation processes in tumors in vivo is typified by the ERBB2 amplification in breast cancer [34, 35]. ERBB2 (v-erb-b2 erythroblastic leukemia viral oncogene homolog 2) encodes an epidermal growth-factor receptor HER2 (human epidermal growth factor receptor 2) and is amplified in 10% to 20% of invasive breast tumors [5, 11]. As increased HER2 protein stimulates growth-factor signaling pathway and drives cell proliferation, ERBB2-amplified (HER2-positive) tumors are associated with advanced stages, recurrence, and poor patient survival [36, 37]. Although the clinically significant phenotype has been known for more than two decades, the amplification mechanism remains elusive. Such information could be important for the better understanding of the etiology of ERBB2-amplified tumors and may have implications in future clinical practice. ERBB2-amplified tumors have been treated with the monoclonal antibody trastuzumab [38]. Trastuzumab binds to HER2 and downregulates growth signaling and thus has significantly improved treatment outcomes for patients with HER2-positive tumors [39‐41]. An accurate diagnosis of ERBB2 amplification is critical, because trastuzumab is solely designed (and effective) only for tumors with ERBB2 amplification. Not only the mechanism of action, but also fatal cardiac side effects [42, 43] and high costs (more than $100,000/year per patient) [44‐46] indicate the necessity of accurate diagnosis. Currently, fluorescence in situ hybridization (FISH) and immunohistochemistry (IHC) are two major diagnostic tests for identifying responders and nonresponders to trastuzumab [47]. However, these current diagnostic tests have some issues, including variable results between institutions and ambiguous diagnoses, such as "equivocal" in IHC [48, 49]. Preanalytic factors, such as the processing of specimens, the fixation method, and the choice of antibodies also introduce variability [50].

Amplification mechanisms could provide new information that may be useful to clarify issues associated with current tests. ERBB2 amplification occurs as the amplification of a genomic region surrounding ERBB2. A particular haplotype within the region may be more susceptible to ERBB2 amplification than other haplotypes. In this scenario, defining haplotypes by using patients' normal DNA could help to clarify ambiguous cases. From the tumor-biology point of view, it is not known why a subset of tumors develops ERBB2 amplification. For example, according to the cell-of-origin model [51], only a subset of breast tumors derived from luminal progenitor cells is HER2 positive. A better understanding of the amplification mechanism could tell us whether the lineage determination is random or has any genetic basis.

To understand the initiating mechanisms of ERBB2 amplification, we took integrated genomic, molecular, and bioinformatic approaches. Array-CGH data indicated that ERBB2-amplified tumors showed a unique pattern of copy-number transitions [52] that could result from a specific amplification mechanism (breakage-fusion- bridge (BFB) cycles). By using the BFB cycles as a guide, we identified a genomic region that could initiate ERBB2 amplification. The region displays a large (300-kb), complex block of duplicated segments (sequence similarity ≥ 90%) and several deletion polymorphisms. Such repeated sequences could be important in the initiation of ERBB2 amplification, as it has been observed in model systems that the frequency of gene amplification is shown to be determined by the presence of repeated sequences at the recombination sites [53‐55]. Deletion polymorphisms of such repeated sequences may affect the initiation, and thus the frequency of ERBB2 amplification. Our results indicate an important role of a complex genomic region in the etiology of primary breast tumors.

The ERBB2 gene is located at chromosome 17q11.2-12. Previous studies have shown that the amplified regions (ERBB2 amplicon) reside within chromosomes as homogeneously staining regions (HSRs) [61‐65], but not in extrachromosomal, double-minute chromosomes (DMs). Deletions of the telomeric side of ERBB2 are common [66, 67], indicating the involvement of DNA breaks in the ERBB2 amplification. A large genomic region surrounding the ERBB2 gene is amplified, and within the amplified region, ERBB2 is located in the most highly amplified segment [52, 68, 69]. Copy number decreases gradually as it goes farther from ERBB2, and ends as copy-number loss (a gradient of copy-number increase). Therefore, elucidating underlying mechanisms (a) for the intrachromosomal amplification and (b) for the gradient of copy-number increase could lead to the better understanding of the mechanism of ERBB2 amplification.

One mechanism underlying intrachromosomal amplification is a well-established amplification mechanism called the breakage-fusion-bridge (BFB) cycle. The BFB cycle consists of a series of recombination events and is initiated by a chromosome break (Figure 1A) [18, 19, 22, 70]. The replication of a broken chromosome would lead to a chromosome structure called sister chromatid fusion, in which sister chromatids are fused at a broken end. The resulting chromosome with two centromeres will have another chromosome break when two centromeres segregate into different daughter nuclei. Such a break could be resolved into sister-chromatid fusion and would initiate another round of a break and fusion. Therefore, the BFB cycles could result in the accumulation of genomic segments within the chromosome.

The accumulation of genomic segments by the BFB cycles could result in the gradient of copy-number increase (Figure 1B). An initial break could occur at the telomeric side (a blue segment) and lead to the formation of a dicentric chromosome. In the following cycle, a chromosome break at the centromeric side (a yellow segment) would be resolved into another dicentric chromosome. Further duplications and breaks would create a chromosome that accumulates segments within the chromosome. A chromosome having a segment harboring ERBB2 (a red segment) at very high copy number (Figure 1A, marked by a star) could be favored because of the growth advantage from ERBB2 overexpression. In such a chromosome, genomic segments flanking the ERBB2-harboring segment would also accumulate; however, because the flanking segments do not confer a growth advantage, their copy number would not be as high as that of the ERBB2-harboring segment. As a result, copy-number analysis for such a chromosome would show the different degree of copy-number increases between segments, and the highest increase would be seen for the segment harboring ERBB2 (Figure 1B). Importantly, such a scenario could predict a copy number transition for the initiating region of the BFB cycles. The initiating region (next to the blue segment) is marked by the transition from the copy-number loss to the low-level copy-number gain (red arrow) and is situated on the telomeric side of the ERBB2 gene.

In this study, we described a common copy-number breakpoint that potentially initiates ERBB2 amplification in primary breast tumors. The region is complex and consists of a large number of duplicated segments that form direct and inverted repeats. The sequence identities between duplicates are very high (> 90%), and some of them are more than 99% identical to each other. These duplicated segments are associated with the KRTAP gene family members, but not with high-copy repeats, such as SINE elements. Duplications appear to have occurred recurrently and predominantly within the region during primate evolution. These results suggest that the complex region could be more fragile than other unique loci and could play a mechanistic role in ERBB2 amplification.

Several lines of evidence support the unstable nature of complex genomic regions in the human genome. First, genomic regions with duplicated segments are preferred sites of non-disease-causing structural (copy number) variants [56, 100]. The increased frequency of structural variants is due to the recombination between duplicated segments (non-allelic homologous recombination, NAHR [93‐96]) in the germline. NAHR between segmental duplications leads to deletions, duplications, and inversions. Second, NAHR between duplicated segments also causes clinical phenotypes called genomic disorders [93‐96, 101]. NAHR between duplicated segments occurs recurrently and generates either duplications or deletions that determine the phenotypes of diseases. Recurrent NAHR for genomic disorders further supports the unstable nature of complex regions. Furthermore, the blocks of duplicated segments have been shown to be the most dynamic regions of the genome during primate evolution [102‐104].

These facts would strongly argue for the unstable nature of complex genomic regions. Indeed, the important role of segmental duplications in creating somatic mutations in cancers is emerging. The breakpoints of isochromosome 17q, the most common isochromosome in human malignancy, was located within a large (> 30) inverted segmental duplication on 17p (isochromosome 17q, i17q) [105‐107]. Translocation between chromosome 9 and 22, t(9;22)(q34;q11) causes the BCR/ABL gene fusion that is the underlying etiology of chronic myeloid leukemia (CML) [108]. From 10% to 20% of the translocation occurred between the 76-kb interchromosomal segmental duplications that are located either at the centromere proximal to ABL on chr 9 or the centromere distal to BCR on chr 22. The involvement of segmental duplications was also described for the microdeletion of PTEN tumor-suppressor gene in aggressive prostate cancers [109].

At the chromosome level, breakage-fusion-bridge (BFB) cycles are likely an underlying mechanism of ERBB2 amplification for at least a subset of breast tumors, as (a) the ERBB2 amplicons predominantly reside within a chromosome [61‐65], and (b) copy-number loss at the telomeric side of the complex genomic regions (Figure 2) indicates chromosome breaks resulting in the loss of genetic materials. The BFB cycles have been shown to establish intrachromosomal amplicons for other oncogenes, such as CCND1 [110, 111]. CCDN1 resides at chromosome 11q13 and is frequently amplified in head and neck tumors. CCND1 is surrounded by three clusters of segmental duplications. These clusters have been shown to colocalize with the boundaries of amplified regions [112], suggesting that a series of rearrangements could occur within these clusters during BFB cycles. In this regard, it is noteworthy that, in addition to the complex region described in this study, additional complex regions exist within ERBB2 amplicons. At the centromeric side, two large (a few hundred kb) euchromatic gaps of human genome assembly (hg18) are noted: one at 1.5 mega-base (Mb) and another at 3.3 Mb centromeric side of ERBB2 (see Additional file 1, Figure S2) [113]. Assembly gaps represent regions with full duplicated DNAs and/or complex, unclonable regions. Similar to CCND1 amplicon, these duplicated DNAs within gaps may serve as substrates for DNA rearrangements during BFB cycles.

We further found that other commonly amplified genes are also in close proximity to complex genomic regions. Among the 13 cancer genes that are most commonly amplified and overexpressed [114], five genes (ERBB2, CCND1, MYCL1, MDM4, and MYCN) are located within 1.5 Mb from either assembly gaps or blocks of duplicated segments (see Additional file, Table S3). Additionally, chromosome 1q21, a commonly amplified region in many tumor types, has 18 gaps within 6 Mb. In contrast, neither complex genomic regions nor assembly gaps are seen within the 6-Mb region surrounding MYC oncogene, which could explain a different mechanism for MYC amplification [115].

At the DNA level, sequence homology between duplicated segments could play an initiating role in BFB cycles and gene amplification. By using model systems, we and others showed that inverted repeats preexisting in the genome can nucleate the duplication of large genomic segments [22, 53‐55, 89]. Duplicated segments could facilitate the initiation of BFB cycles in two ways.

First, inverted repeats can adopt Holliday junction-like structure by forming a cruciform. The resolution of a cruciform results in two chromosomal parts with hairpin-capped ends. The replication of a centromere-harboring part with a hairpin-capped end results in the formation of a dicentric chromosome and the initiation of BFB cycles.

Second, duplicated segments could adopt a complex secondary structure that can impose an obstacle to the progression of replication forks (Figure 8) [116, 117]. As replication fork stalling and collapse could be processed into one-ended DNA breaks [118], the complex regions may have increased DNA breaks. The 5'- to 3'-end resection of one-ended DNA breaks exposes single-stranded DNA [119]. When the end of single-stranded DNA folds back and anneals to an inverted repeat sequence (intrastrand annealing [55]), it would prime DNA synthesis (break-induced replication, BIR [120]) and fill in the single-stranded gap to create a chromosome with a hairpin-capped end. Thus, the sequence homology between duplicated segments could be mutagenic and initiate BFB cycles.

In this regard, it is noteworthy that ERBB2 amplification is absent in breast tumors from BRCA1 mutation carriers [121]. BRCA1 binds to many proteins of DNA damage response and repair and thus plays a critical role in maintaining genome integrity [122]. BRCA1 is recruited to the chromatin with damaged DNA very early [123, 124] and stimulates DNA end resection for homology-directed repair [125, 126]. As BRCA1 mutant cells could lack efficient end resection, both mutation-free (conservative) and mutagenic homology-directed repair pathways could be impaired [127]. The conservative pathway is RAD51 dependent and repairs DSBs by using sister chromatids as a template, whereas the mutagenic pathway can be RAD51 independent [128] and could use repeated segments as a template. Therefore, the fact that ERBB2 amplification is rare in tumors with BRCA1 mutation may indicate that ERBB2 amplification is dependent on mutagenic homology-directed repair. In contrast, 15% of tumors derived from BRCA2 mutation carriers have ERBB2 amplification [121]. BRCA2 also functions for homology-directed repair; however, it has a more-specific role. BRCA2 has a RAD51-binding domain and plays an important role in conservative repair [129, 130]. Indeed, in BRCA2 mutant cells, conservative repair was impaired, but mutagenic repair was not affected [127]. Therefore, the distinct ERBB2 amplification tendency between BRCA1 and BRCA2 mutant careers further suggests the involvement of recombination between repeated segments in ERBB2 amplification.

Alternatively, BIR initiated from one-ended DNA breaks at the sites of collapsed replication forks could be more processive, and repeated template switching (fork stalling and template switching, FoSTeS [131]) could result in complex genomic rearrangements and copy-number transition [132, 133]. Newly established forks from one-ended DNA breaks could invade into either sister chromatid or homologues at nonallelic loci by using duplicated sequences or microhomology [134, 135]. Invading strands can be unstable and often dissociate from template strands. The resulting free ends would repeat invasion several times at nonallelic loci to create complex genomic rearrangements. Copy-number increases from such complex rearrangements is relatively low, from twofold to threefold [132]. However, duplication and triplication of the segments could facilitate further rearrangements (for example, unequal sister chromatid exchange) and high-level amplification.

ERBB2 amplicons have been classified into two groups: a large amplicon including the TOP2A gene and a smaller, more-restricted amplicon (without TOP2A) surrounding the ERBB2 gene [66, 67]. TOP2A encodes a DNA topoisomerase II (topoII) that controls and alters the topologic state of DNA in several aspects of DNA metabolism, such as chromosome segregation, transcription, and chromatin organization [136‐138]. Because the complex region is located at the telomeric side of TOP2A gene, tumors having a breakpoint at the region belong to the TOP2A-coamplified tumors. Whether tumors without TOP2A amplification have independent common copy-number breakpoints is an important issue for future studies. It is also possible that an initiating break/recombination occurs at the complex region (or on a further telomeric side [139]) and, during the evolution of the amplicon, secondary rearrangements could delete both the region including TOP2A and the complex region from the amplicon. TOP2A deletion in ERBB2-amplified tumors is common [66, 140]. Even in coamplified tumors, TOP2A and ERBB2 resided in different chromosomal domains [64, 141], suggesting that secondary rearrangements separated the two genes from primary amplicons.

Given the established role of repeated segments in gene amplification in experimental systems, structural variants of such segments could have a significant effect in the occurrence of ERBB2 amplification [22]. Several structural variants are reported within the region, and some of them could be good candidates for the variants. However, defining the DNA sequences at breakpoints and identifying the segments responsible for ERBB2 amplification can be hampered by the complexity of the region. Therefore, as an initial step, we made an effort to define the haploblocks within the region. By combining the genotyping data from the deletion polymorphism and SNP genotypes, we were able to define two blocks, one of which showed a strong LD within the block. Our ongoing effort for further defining haploblocks and identifying genetic markers will provide a better understanding of the complex region. Such genetic markers could be useful, especially for the genomic regions where SNP markers are less well defined and genome-wide association studies [142, 143] may have a limited power.

We show here a potential initiating role of a complex genomic region in ERBB2 amplification in breast cancer. The genomic sequence of the region is still ambiguous, as Genome Reference Consortium is providing an alternative sequence assembly for the region. Furthermore, two large sequence gaps (in hg18) exist on the centromeric side of ERBB2 (see Additional file, Figure S2). These sequence gaps likely contain many repeated sequences and structural variants and could also be fragile. Therefore, ERBB2 is flanked by many complex genomic regions that may not be sufficiently investigated by current genomic technologies. Investigating such regions in detail, including the patterns of DNA rearrangements at the nucleotide level, structural variants, and haplotypes within the regions, is important for the mechanistic study of ERBB2 amplification.