Localized Genomic Rearrangement in a Patient with Chronic Lymphocytic Leukemia
Advances in DNA sequencing have made it possible to identify the majority of somatically acquired genetic variants in cancer samples on a genome-wide basis (Ding et al., 2010; Mardis et al., 2009; Pleasance et al., 2010a, 2010b; Shah et al., 2009
). In particular, paired-end sequencing allows discovery of genomic rearrangements (Campbell et al., 2008, 2010; Stephens et al., 2009
), through sequencing both ends of 50–100 million genomic DNA fragments per sample. Alignment of the paired-end reads to the reference genome enables identification of putative genomic rearrangements.
In a rearrangement screen of 10 patients with chronic lymphocytic leukemia (CLL), we identified one patient who had 42 somatically acquired genomic rearrangements involving the long arm of chromosome 4 (A and 1B and Table S1
available online). The positions of these rearrangements relative to one another and to copy number changes on chromosome 4q reveal some striking patterns. First, the rearrangements show geographic localization within the genome. Apart from a separate 13q deletion in this patient, all rearrangements are confined to chromosome 4q and focal points on chromosomes 1, 12, and 15 (C). This is different to the patterns of genomic instability we have typically seen in breast, lung, or pancreatic cancer where rearrangements tend to be either scattered genome-wide or, if localized, are associated with substantial genomic amplification (Campbell et al., 2008; Pleasance et al., 2010b; Stephens et al., 2009
). Second, the copy number profile across the chromosome arm shows many positions at which copy number changes, but these changes alternate between just two states, namely one or two copies. Analysis of allelic ratios at germline single nucleotide polymorphism (SNP) positions on chromosome 4q demonstrated that regions of copy number 1 show loss of heterozygosity, but regions of copy number 2 retain heterozygosity (data not shown). Third, the many regions of copy number 1 are not caused by simple deletions. Instead, a series of complex rearrangements spanning the involved region generate the copy number changes, as can be seen by the distribution of rearrangements falling at change-points in copy number (A). These have both inverted and noninverted orientation, with all four orientations of intrachromosomal breakpoints represented in approximately even numbers: deletion-type (8 rearrangements), tandem duplication-type (9), head-to-head inverted (6), and tail-to-tail inverted (10). Fourth, there is pronounced clustering of breakpoints across the chromosome arm with, for example, seven rearrangements involving the 30 kb region between 77.013 Mb and 77.043 Mb, and six rearrangements in the 25 kb between 170.620 Mb and 170.645 Mb. Fifth, although the locations of DNA breaks show clustering, the two conjoined fragments of chromosome at each breakpoint are not geographically proximate. That is, there are as many rearrangements joining regions of the chromosome normally separated by tens of megabases in the germline as there are junctions between close-by regions. Sixth, there are nine rearrangements joining the long arm of chromosome 4 to other chromosomes—breakpoints on these partner chromosomes also show clustering (C).
Clustered Rearrangements on Chromosome 4q in a Patient with Chronic Lymphocytic Leukemia
The sample analyzed was collected from a 62-year-old woman with CLL who had not previously received treatment. Her subsequent clinical course showed rapid deterioration, and she was treated with alemtuzumab, but unfortunately, she relapsed quickly. To assess whether the abnormalities seen in the pretreatment sample persisted in the relapsing cells or indeed showed further evolution, we sequenced a relapse specimen collected 31 months after the initial sample. All rearrangements present in the pretreatment sample were present in the later sample (B–1D), and the striking copy number profile persisted. Furthermore, there were no new genomic rearrangements, suggesting that the process generating this complex regional remodeling had resolved before the patient was first diagnosed.
Complex Rearrangement of Single Chromosomes Is Seen in At Least 2%–3% of All Cancers
To assess whether the unusual genomic landscape observed in the patient with CLL could be seen in other cancer samples, we analyzed high-resolution copy number profiles of 746 cancer cell lines obtained using SNP arrays (Bignell et al., 2010
). Of these, 96 cell lines have at least one chromosome with >50 positions at which copy number changes (Figure S1A), many of which are caused by amplicons or other complex clusters of rearrangements. Notably, 18/746 (2.4%; 95% confidence interval, 1.5%–3.9%) cell lines have copy number profiles similar to that seen in the CLL patient, with frequent copy number changes confined to localized genomic regions rapidly alternating between one, two, or occasionally three different states (Figures S1B–S1T). Copy number changes could involve the entire chromosome (for example, SNU-C1, Figure S1G), a whole arm of a chromosome (SW982, Figure S1H), the telomeric portion of a chromosome (C32, Figure S1C), or an interstitial region of a chromosome (A172, Figure S1D). The pattern was seen in many different tumor types, including melanoma (4 cell lines), small cell lung cancer (3 cell lines), glioma (3 cell lines), hematological malignancies (2 cell lines), nonsmall cell lung cancer (1 cell line), synovial sarcoma (1 cell line), and esophageal (1 cell line), colorectal (1 cell line), renal (1 cell line), and thyroid (1 cell line) cancers. Furthermore, in segmented SNP array data from 2792 cancers, of which 80% were primary tumors (Beroukhim et al., 2010
), we find evidence for chromothripsis in a similar proportion of cases (Figure S1T).
We selected four of these cell lines for further genomic analysis with massively parallel paired-end sequencing for rearrangements and cytogenetic studies: SNU-C1, 8505C, TK10, and SCLC-21H (described later). In SNU-C1, derived from a colorectal cancer, we identified 239 rearrangements involving chromosome 15 (A and Table S2
). From 8505C, a thyroid cancer line, we mapped 77 rearrangements involving the short arm of chromosome 9 (B and Table S2
), and for TK10, a renal cancer, 55 rearrangements involving chromosome 5 (C and Table S2
Rearrangement Screens in Three Cancer Cell Lines Showing Evidence for Chromothripsis
The distinctive genomic configuration observed in the CLL patient is stamped on these three cell lines. Striking geographic localization of rearrangements is evident in these samples. Although a few rearrangements were observed elsewhere in the genome (), these are generally straightforward events such as deletions or tandem duplications and do not intersect with the regions of massive disruption shown in . The localization is especially evident in 8505C (B), in which rearrangements only involve the telomeric portion of chromosome 9p with sparing of the most centromeric bands of 9p and all 9q. As in the CLL patient, copy number oscillates rapidly between two states, with the lower copy number state showing loss of heterozygosity (LOH) and the higher copy number state retaining heterozygosity.
Genome-wide Rearrangements for Four Cell Lines with Chromothripsis, Related to
One question that arises is whether the rearrangements are all found on a single parental copy of the chromosome or whether both copies are involved. We therefore performed spectral karyotyping on the three cell lines (A and ). TK10, a hyperdiploid line, carries six copies of chromosome 5. Consistent with the observed copy number profile alternating between states of copy number 4 with LOH and copy number 6 with heterozygosity, the karyotype showed four grossly normal copies of chromosome 5 and two smaller derivative chromosomes. Similarly, in 8505C, two copies of chromosome 9 showed distinctly foreshortened p arms alongside two cytogenetically normal chromosomes. None of the three karyotypes indicated translocations involving the respective derivative chromosomes, confirming the impression from the paired-end sequencing data that the genomic remodeling of these regions was entirely intrachromosomal. Cytogenetic changes were consistently seen across all cells examined.
Spectral Karyotypes for (A) 8505C and (B) SNU-C1, Related to
The spectral karyotypes suggest that the rearrangements involve a single parental copy of the chromosome. To demonstrate this further, we designed FISH probes to five widely dispersed regions of chromosome 5 at copy number 6 in TK10 (B). From the paired-end sequencing, we predicted that the two regions at 6 Mb and 172 Mb would be joined by a head-to-head inverted rearrangement, and the three regions at 32 Mb, 66 Mb, and 150 Mb would be joined by another head-to-head inverted rearrangement and a tandem duplication-type rearrangement. These FISH probes, labeled with different dyes, were hybridized to TK10 cells (C). As expected, there were four copies of chromosome 5 per cell showing the correct genomic orientation and distribution of the five probes. In addition, each cell carried two copies of a derivative 5 chromosome in which all five probes were closely juxtaposed, as predicted by the sequencing data. These patterns were seen identically across all cells examined.
Taken together, these data suggest that at least 2%–3% of all cancers show evidence for massive remodeling of a single chromosome, involving tens to hundreds of genomic rearrangements. The consistency of cytogenetic findings across the many cells examined implies that the clustering of genomic breakpoints cannot be explained by multiple, parallel rearrangements in different subclones. In the lines studied here, the genomic remodeling occurred when there were just the two parental copies of the relevant chromosome, preceding chromosomal duplication events. This explains why copy number states alternate between heterozygous and LOH and why more than one copy of the derivative chromosome is present.
Chromothripsis Is Particularly Common in Bone Cancers and Can Involve More Than One Chromosome
Alongside the rearrangement screen in CLL, we performed rearrangement screens in primary tumor samples from 20 patients with bone cancer, including 9 with osteosarcoma and 11 with chordoma, a rare type of cancer arising in the spinal column. Strikingly, five of these patients (25%; 95% confidence interval, 10%–49%), three with osteosarcoma and two with chordoma, also show large numbers of clustered rearrangements with the hallmarks of chromothripsis.
In four of these five bone tumors, rearrangements affect localized regions of several chromosomes (, , Table S3
, and Table S4
). For example, we identified 147 somatically acquired genomic rearrangements in a chordoma sample, PD3808a, involving and linking together well-circumscribed regions of chromosomes 3q, 4q, 7q, 8p, and 9p (A). Analogous to chromothripsis involving single chromosomes, copy number in each of these chromosomal regions cycles between two different states with retention of heterozygosity in the higher copy number state. Of the 147 rearrangements, 49 are intrachromosomal and show the same back-and-forth mixture of inverted and noninverted rearrangements described above. The numerous interchromosomal rearrangements link the various disrupted regions together, implying that the resulting genomic structure is a complex medley of fragments from different chromosomes jumbled together.
Chromothripsis Involving More Than One Chromosome in Primary Samples from Patients with Bone Cancer
Chromothripsis Involving Several Chromosomes, Related to
In samples from three patients with osteosarcoma, PD3786a (B), PD3791a (A), and PD3799a (B), we identified 88, 86 and 24 rearrangements respectively with similar overall patterns of copy number change and rearrangement. PD3807a, another chordoma sample, also had 38 rearrangements interlinking well-defined regions of four chromosomes (C). Clinically, the patients ranged in age from 9 to 64 years and four of the samples were from resections of treatment-naive primary tumors, whereas one of the patients (PD3786a) had previously received neoadjuvant chemotherapy. In 1 of 13 pancreatic cancers we previously sequenced (Campbell et al., 2010
), we identified 41 rearrangements involving chromosomes 1, 4, 10, and 14 with the hallmarks of chromothripsis (H), suggesting that involvement of multiple chromosomes by this process is not restricted to bone tumors.
The Vast Majority of Chromothripsis Rearrangements Occur in a Single Catastrophic Event
There are two potential models for how such complex restructuring of a chromosome could develop. Under the progressive rearrangement model, the rearrangements occur sequentially and independently of one another over many cell cycles, leading to increasingly disordered genomic structure (A). This is the conventional view of how most complex regional clusters of rearrangements evolve, especially genomic amplification. Localization results either from rearrangement targeting a specific cancer gene or through regional abnormalities driving recurrent DNA breakage. The second model to explain the distinctive genomic structures described here is that the overwhelming majority of rearrangements occur in a single catastrophic event. In this scenario, the chromosome or chromosomal region shatters into tens to hundreds of pieces, some (but not all) of which are then stitched together by the DNA repair machinery in a mosaic patchwork of genomic fragments (B).
Genomic Features of Chromothripsis Suggest that Most Rearrangements Occur in a Single Catastrophic Event
Several characteristics of the patterns we observe here make the progressive rearrangement model difficult to sustain, and give support to the catastrophe model. The first observation is that the number of copy number states observed in the final configuration of the chromosome is restricted to two (occasionally three). With sequential, independent rearrangements, the number of different states observed would be expected to increase as the number of breakpoints rises (A). Tandem duplications increase copy number and, because many of the observed rearrangements with a tandem duplication pattern in these samples overlap with one another, we would anticipate a number of segments to have been sequentially amplified several-fold under the progressive rearrangements model. Although deletion events would tend to counteract increases in copy number, the chances of these two processes being so balanced as to generate only two copy number states fall rapidly as the number of rearrangements increases. To demonstrate this, we performed Monte Carlo simulations of the progressive rearrangement model. Rearrangements were randomly sampled from the set of breakpoints found in SNU-C1, the resulting chromosome structure calculated, and the process repeated to generate different numbers of rearrangements (C). As predicted, with increasing numbers of rearrangements, the observed number of different copy number states also rises. The observed profiles of the three cell lines and the CLL patient sit well outside the spectrum observed under simulations of the progressive rearrangement model.
In contrast, the catastrophe model predicts exactly two copy number states. Those fragments that are retained in the eventual derivative chromosome will have the higher copy number state; those that are lost to the cell will be in the lower copy number state (B).
The second problem for the progressive rearrangements model is the retention of heterozygosity in regions with higher copy number. Once lost, heterozygosity cannot generally be regained. For example, the region around 66 Mb of chromosome 15 of SNU-C1 is heterozygous, but is encompassed in the span of no fewer than 21 rearrangements with the orientation of deletions, as well as 20 tandem duplication-type and 52 inverted rearrangements (A). Under the progressive rearrangement model, a deletion that occurred early in the sequence of rearrangements would permanently remove heterozygosity between the breakpoints. Thus, deleting events can only occur late in the succession of rearrangements, once regions of retained heterozygosity have either been switched out of the region by inversion or copied by tandem duplication. When extended across all 239 rearrangements involving chromosome 15, there is major difficulty constructing a sequence of progressive rearrangements that would spare the heterozygosity found in over 20 separate segments. In contrast, alternating regions of heterozygosity and LOH is the natural consequence of the catastrophe model. With a normal parental chromosome and one shattered into many pieces, any fragment that is retained in the eventual derivative chromosome will be heterozygous; those that are lost to the cell will result in LOH in those regions (B).
A third feature arguing against the progressive rearrangement model is that breakpoints show significantly more clustering along the chromosome or chromosome arm than expected by chance (D). A clean break across double-stranded DNA (dsDNA) generates two naked ends of which none, one or two may subsequently be repaired. Some of the clustering represents erroneous repair of both sides of a dsDNA break (see B, for example). The extent of clustering observed in breakpoint locations, however, is much greater than explicable by this means alone. This presents some difficulties for the progressive rearrangements model because such nonrandom distribution of independently generated breaks would imply extensive regional variation in chromosomal fragility. Specific regions of increased propensity to rearrangement have been documented (Bignell et al., 2010
), but not to the extent observed here. Under a catastrophe model, clustering among the prolific numbers of DNA breaks would perhaps be expected, depending on the process causing the DNA damage and repair. The limited overlap between sequences at the breakpoint junction suggests that the major mechanisms of DNA repair here are microhomology-mediated break repair and/or nonhomologous end-joining rather than homologous recombination ().
Signatures of DNA Repair at the Breakpoint, Related to
In conclusion, several distinctive genomic features imply that a major catastrophic event underpins the massive, but localized, genomic rearrangement in these samples. These arguments extend to cancers where we have observed involvement of several different chromosomes. We do not argue that absolutely every rearrangement was generated in one event—indeed, a later partial duplication of the derivative chromosome is likely to explain why some samples (such as C32, C) oscillate across three copy number states rather than two. However, the majority of rearrangements seen in these examples almost certainly occurred in a single event.
Chromothripsis Can Generate Genomic Consequences that Promote Cancer Development
A cell suffering tens to hundreds of DNA breaks in a single cataclysmic event would be expected to undergo apoptosis. That a cell can survive such an insult and progress to become cancerous suggests that the extensive remodeling of the genome may confer significant selective advantage to that clone. To explore this possibility, we analyzed the genomic data for evidence of changes that might promote the development of cancer.
One small cell lung cancer cell line, SCLC-21H, demonstrates massive numbers of copy number changes on chromosome 8, mostly with the typical appearances of chromothripsis (A). Interestingly, however, the SNP array data suggest that some segments of the chromosome might be heavily amplified. We mapped 170 breakpoints, all involving chromosome 8 and showing the expected patterns of rearrangements described above (A and Table S2
). Whereas most of the chromosome oscillates among low copy number states, there are 15 discrete segments of the chromosome present at markedly increased copy number, ranging from 50 to 200 copies per cell (B). One of these segments contains the MYC
oncogene, amplified in 10%–20% of small cell lung cancers (Sher et al., 2008
). The rearrangement data demonstrate that the 15 regions are interwoven by a series of rearrangements, many of which demarcate the starts and ends of the massively amplified segments. Strikingly, we found no evidence for breakpoints linking these massively amplified regions to the other, nonamplified but rearranged, regions of chromosome 8.
Generation of a Double Minute Chromosome Containing MYC by Chromothripsis in a Small Cell Lung Cancer Cell Line, SCLC-21H
One potential mechanism for these findings is that at some stage while the cancer was evolving, chromosome 8 shattered into hundreds of pieces. Many of these were stitched together into a derivative chromosome 8, but 15 other fragments were joined to create a double minute chromosome of ~1.1Mb in size (thick lines, B). Containing MYC, it was of considerable selective advantage for daughter cells to carry extra copies of the double minute, and through further internal rearrangements (thin lines, B) and overreplication, the massive amplification evolved.
To assess this hypothesis, we performed multicolor FISH. First, we probed three nonamplified segments of chromosome 8 that the sequencing suggested were joined together through a head-to-head inverted rearrangement and a tandem duplication-type rearrangement. This revealed a single normal copy of chromosome 8 with the probes hybridized in the expected orientation and distance apart, and two derivative 8 chromosomes with the three probes closely juxtaposed (C). Thus, the cells contain a cytogenetically normal chromosome 8 and a derivative chromosome 8 generated by chromothripsis that has subsequently undergone chromosomal duplication. Second, we probed three of the chromosome 8 regions that were heavily amplified (D). This demonstrated huge numbers of extrachromosomal copies of the segments, with the probes closely abutting. In addition, there were two homogeneously staining regions identified by the probes, consistent with chromosomal integration of the double minutes. Probes for the double minute chromosomes were found in the correct orientation on the normal chromosome 8, but were absent from the two copies of the derivative chromosome 8 (A). Taken together, these findings are consistent with the model that the catastrophic shattering of chromosome 8 has facilitated the creation of a double minute chromosome, which, in this example, containing MYC, acts as a substrate for amplification, evolutionary selection and progression toward cancer.
Marker Chromosomes Created by Chromothripsis, Related to
Chromothripsis may lead to the generation of other forms of marker chromosome also. We studied the spectral karyotype of the pancreatic cancer sample with evidence for chromothripsis involving multiple chromosomes (B and S6C). Even with the low resolution of SKY, a chromosome arm with at least six cytogenetically visible stripes could be seen, indicating that the many interchromosomal rearrangements have intertwined segments from multiple different chromosomes into a distinctive marker chromosome.
A second potential mechanism by which chromothripsis could generate cancer-causing genomic changes is through loss or disruption of tumor suppressor genes. In the chordoma, PD3808a, the CDKN2A
gene is homozygously deleted (A), with one of the copies probably lost through chromothripsis. The two rearrangements demarcating the copy number change from 2 to 1 around CDKN2A
in A) appear to be part of the network of interchromosomal rearrangements interlinking regions from chromosomes 3q, 4q, 7q, 8p, and 9p seen in A. This argues that loss of this copy of the gene occurred during chromothripsis, although it is formally possible that an independent deletion of CDKN2A
might have occurred before chromothripsis. The second copy of the gene was lost through a focal deletion on the other parental chromosome, which presumably occurred as a temporally separate event (thick blue line, A).
Loss of Tumor Suppressor Genes through Chromothripsis
With so many rearrangements generated in a single genomic crisis, it is feasible that more than one cancer-causing lesion could occur in the same event. In addition to the loss of CDKN2A
described above, the chordoma sample PD3808a had a rearrangement that directly disrupted WRN
, linking the 3′ portion of this gene on chromosome 8 to an intergenic region on 9p just downstream of CDKN2A
(thick purple line, A). WRN
is a cancer gene in which germline mutation causes Werner syndrome, a condition associated with markedly increased risk of bone tumors, and in which somatic inactivating mutations have been documented in renal cancer (Dalgliesh et al., 2010
). This same patient also lost a copy of FBXW7
on chromosome 4q (A). The rearrangements around this gene link to chromosomes 3q, 7q, 8p, 9p, and elsewhere on 4q similar to those near CDKN2A
, suggesting that loss of FBXW7
occurred during the same chromothripsis event. FBXW7
is inactivated in ~6% of all cancers across many subtypes (Akhoondi et al., 2007; Kemp et al., 2005; Maser et al., 2007
). Inactivation is frequently heterozygous, supported by functional data suggesting it may be a haploinsufficient tumor suppressor gene (Kemp et al., 2005; Mao et al., 2004
). Thus, the single catastrophic event inducing chromothripsis in this patient has resulted in disruption of three tumor suppressor genes.
A number of other known cancer genes were affected by rearrangements across the samples described here (Table S5
), including ARID1A
in PD3807a (chordoma). In 8505C, the chromothripsis involving chromosome 9p has led to loss of one copy of CDKN2A
(B); the other carries a deletion of the first exon of the gene. We also identified a second patient with CLL who showed evidence for loss of two tumor suppressor genes in a cluster of rearrangements involving chromosomes 4, 9, and 13 (C). Here, single copies of both CDKN2A
, the microRNA cluster deleted in >50% of CLL patients (Cimmino et al., 2005
), were lost through interchromosomal rearrangements, whereas the other copy of the microRNA cluster was deleted in a presumably separate event (blue line, C).
Theoretically, chromothripsis rearrangements could juxtapose coding portions of two genes in the same orientation with an open reading frame, producing a potentially oncogenic fusion gene. Among chromothripsis rearrangements, we found 17 that could potentially create novel in-frame fusions (Table S5
). None generates a classic cancer-associated fusion gene, such as BCR-ABL1
, and the proportion of rearrangements generating novel in-frame fusions is similar to that observed for other types of rearrangements (Campbell et al., 2010; Stephens et al., 2009
). This suggests that most are coincidental “passenger” events, unlikely to drive cancer development.
Such dramatic restructuring of a genome will disrupt both coding sequences directly and the linkage between coding exons and regulatory elements of very many genes. We explored whether expression profiles of genes from chromosomes affected by chromothripsis differed from those of intact chromosomes. For SCLC-21H, genes from chromosomes that were not affected by chromothripsis showed an approximately normal distribution of expression levels relative to their expression in other SCLC cell lines (), as expected. On chromosome 8, however, affected by chromothripsis, expression levels were decreased in ~5% of genes in SCLC-21H relative to their expression in other SCLC cell lines (chromosome 7 versus chromosome 8, p = 0.001; chromosome 6 versus chromosome 8, p < 0.0001). Similar differences were observed for SNU-C1, in which chromothripsis affected chromosome 15 (chromosome 14 versus chromosome 15, p = 0.02; chromosome 13 versus chromosome 15, p < 0.0001).
Comparison of Gene Expression Profiles from the Chromothripsis Chromosome for SCLC-21H against Intact Chromosomes, Related to
Taken together, these data exemplify the mechanisms by which chromothripsis can promote the development of cancer. In particular, more than one cancer-causing lesion can arise from a single catastrophe, and the chaotic genomic architecture that results can inactivate or disrupt the transcription of many more genes.