Chromosomal Breakpoints in Primary Colon Cancer Cluster at Sites of Structural Variants in the Genome

Genomic aberrations on chromosome 8 are common in colon cancer, and are associated with lymph node and distant metastases as well as with disease susceptibility. This prompted us to generate a high-resolution map of genomic imbalances of chromosome 8 in 51 primary colon carcinomas using a custom-designed genomic array consisting of a tiling path of BAC clones. This analysis confirmed the dominant role of this chromosome. Unexpectedly, the position of the breakpoints suggested colocalization with structural variants in the human genome. In order to map these sites with increased resolution and to extend the analysis to the entire genome, we analyzed a subset of these tumors (n = 32) by comparative genomic hybridization on a 185K oligonucleotide array platform. Our comprehensive map of the colon cancer genome confirmed recurrent and specific low-level copy number changes of chromosomes 7, 8, 13, 18, and 20, and unveiled additional, novel sites of genomic imbalances including amplification of a histone gene cluster on chromosome 6p21.1-21.33 and deletions on chromosome 4q34-35. The systematic comparison of segments of copy number change with gene expression profiles showed that genomic imbalances directly affect average expression levels. Strikingly, we observed a significant association of chromosomal breakpoints with structural variants in the human genome: 41% of all copy number changes occurred at sites of such copy number variants (P < 2.2e⁻¹⁶). Such an association has not been previously described and reveals a yet underappreciated plasticity of the colon cancer genome; it also points to potential mechanisms for the induction of chromosomal breakage in cancer cells. [Cancer Res 2008;68(5):1284–95]

Colorectal cancer is the second leading cause of cancer death in Europe and in the United States, with ∼300,000 new cases and 200,000 deaths each year (1). Cytogenetic and molecular cytogenetic studies clearly established that the colorectal cancer genome is defined by a specific distribution of genomic imbalances, most prominently, gains of chromosomes and chromosome arms 7, 8q, 13, and 20q as well as losses of chromosomes 4q, 8p, 17p, and 18q (2).

Within the last decade, microarray technology has been extensively applied to survey the cellular transcriptome of common solid tumors, including colorectal cancer, and for colon cancers, gene expression signatures were subsequently correlated with clinical outcome (for reviews, see refs. 3–5). However, high-resolution mapping of chromosomal copy number changes has only recently been achieved using BAC or cDNA clone-based arrays (6–10).

Chromosome 8q is one of the most frequently gained chromosomal arms in colorectal cancers (2), and it is conceivable that it contains more oncogenes than just the MYC oncogene, which maps to chromosome band 8q24.21. A potential role of chromosome 8q for the development of lymph node metastases has been previously reported (11), and overexpression of a gene, PRL-3, that maps to chromosome 8q24.3 has been implied in the development of liver metastases (12). Moreover, the 8q24 locus contains single nucleotide polymorphisms that are associated with an increased risk for the development of colon cancer (13–15).

Recently, a new class of genetic variation among humans has become recognized as a major source of genetic diversity. Termed structural variations, these polymorphisms can present themselves as copy number variants (CNV) and segmental duplications, which could be CNVs, but are not necessarily so (16–19). These polymorphisms could induce chromosomal rearrangements (20). One of our previous analyses of chromosomal aberrations in cell lines established from different carcinomas indicated that genomic copy number changes could be triggered by jumping translocations, many of which originated in the pericentromeric heterochromatin of several chromosomes (21). These regions frequently contain segmental duplications and other structural variants of the genome (22). Taken together, these data enticed us to systematically explore the genomic aberration profile and the potential involvement of structural variants of the human genome in the genesis of chromosomal aberrations in this common cancer. We therefore established a high-resolution map of genomic copy number changes in 51 primary colon carcinomas using comparative genomic hybridization (CGH) on both a BAC-based genomic tiling array for chromosome 8 and, for a subset of those, using a 185K oligonucleotide platform for whole genome coverage.

The 51 patients included in this study were diagnosed with primary adenocarcinomas of the colon, and treated at the Department of General Surgery, University Medicine Göttingen, Göttingen, Germany. All patients received standardized surgery and histopathologic workup, and tumor staging was based on WHO criteria (23). Twenty-five tumors were associated with lymph node metastases [International Union Against Cancer (UICC)-III], whereas 26 tumors were not (UICC-II). Tumor samples were obtained immediately after surgery and stored on ice for inspection by an experienced pathologist. Consistent with standard procedures, only samples with a tumor cell content of at least 70% were included in this study. Biopsies of normal adjacent mucosa were collected from some patients when possible. Table 1 summarizes the clinical data and experimental setup.

Bioptic material was in the range of 24 to 370 mg, and nucleic acids were extracted using TRIZOL (Invitrogen) following standard procedures.⁴

On average, we obtained 200 μg each of RNA and DNA. Nucleic acid quantification was determined using the Nanodrop ND-1000 UV-VIS spectrophotometer (Nanodrop). The quality of the nucleic acids after preparation was assessed using a 2100 Bioanalyzer (Agilent Technologies).

BAC array CGH platform. The 1,463 BAC clones and DNA used to construct the chromosome 8 Human-BAC microarray were a subset of the Human “32K” BAC Re-Array library from the BACPAC Resources (Children's Hospital Oakland Research Institute, Oakland, CA).⁵

The platform and details of the procedure are described in ref. 24.

Genomic DNA was digested using RsaI and AluI (Roche Applied Science), and the appropriate fragment size was confirmed on an agarose gel. After protein removal using a phenol-chloroform extraction, 600 ng of digested DNA were labeled using the Bioprime Labeling Kit (Invitrogen) to incorporate Cy5-dCTP or Cy3-dCTP (Amersham). Sex-matched tumor and reference DNA were combined and hybridized to the custom chromosome 8 BAC array in specifically designed hybridization cassettes (TeleChem International). After overnight hybridization, slides were washed and scanned on an Axon scanner using GenePixPro (3.0) software (Axon Instruments).

Oligo array CGH platform. Oligonucleotide array CGH (aCGH) was performed according to the protocol provided by the manufacturer (Agilent Oligonucleotide Array-Based CGH for Genomic DNA Analysis, protocol version 4.0, June 2006; Agilent Technologies), with minor modifications. Commercially available pooled control DNA (Promega) was used as sex-matched reference DNA in all hybridizations. Briefly, 3 μg of genomic DNA was digested for 2 h with AluI and RsaI (Promega). QIAprep Spin Miniprep Kit (Qiagen) was used for purifying the digested DNA. Tumor and reference DNA was labeled with Cy3-dUTP and Cy5-dUTP (Promega), respectively, in a random priming reaction using Bioprime Array CGH Genomic Labeling Module (Invitrogen). After 2 h of reaction, unincorporated nucleotides were removed using Microcon YM-30 columns (Millipore). Cy3 and Cy5-labeled samples were combined in equal amounts according to the incorporation of labeled nucleotides as measured using Nanodrop. Hybridization and washes were performed according to the manufacturer's protocol. Slides were scanned using a scanner (G2565BA; Agilent Technologies), and Agilent Feature Extraction software (version 9.1; Agilent Technologies) was applied for image analysis. To visualize the aCGH data, we used Agilent CGH Analytics 3.4 software (Agilent Technologies). The quality of the slides was assessed using metrics provided by CGH Analytics.

Gene expression profiles for all 51 primary colon tumors and 21 associated mucosa samples were established as previously reported (25).

BAC aCGH platform. In order to compensate for scanner distortion between the Cy3 and Cy5 channel readings, we applied a 90th interpercentile range (90IPR) normalization procedure to equalize the spread of Cy3 measurement to the spread of Cy5 measurements per array (in natural scale):

where cCy3 is the corrected Cy3 measurement, and 90IPR.Cy5 and 90IPR.Cy3 are the 95th percentile minus the 5th percentile measurements in the Cy5 and Cy3 channels, respectively. cCy3 and Cy5 measurements are then log 2–transformed, and their log 2 (ratio) are median-centralized by array using the following formula:

where MD.log 2 (Cy5) and MD.log 2 (cCy3) are the medians of log 2 (Cy5) and log 2 (cCy3) measurements, respectively. An aCGH segmentation algorithm developed under MATLAB was applied to all normalized arrays to extract segmented regions. Consensus gain or loss regions were obtained as described previously (24).

Oligo aCGH platform. The analysis of the aCGH experiments was performed with in-house developed software based on R version 2.4.1⁶

and the DNA copy package from Bioconductor.⁷ One array that did not pass the quality control criteria (derivative log ratio spread or DLRSpread > 0.3) was discarded. We also discarded features with no precise chromosomal location. The final data set was comprised of 29 arrays and 181,984 features. The data were smoothed using “smooth.CNA” function (with arguments smooth.region = 1, and smooth.SD.scale = 3), followed by the generation of chromosome segments using circular binary segmentation (CBS; ref. 26). We centralized DNA segments to the most common ploidy per array through an algorithm similar to the one offered in Agilent CGH Analytics 3.4 software. The cumulative frequency of loss score for each feature is the percentage of samples for which the segment value is below the threshold log 2 (5/6) corresponding to a loss of one DNA copy in 30% of diploid cells. Cumulative frequency is scaled to 100% = 4 (e.g., 25% = 1) in order to take advantage of the maximum range of the representation in genome, chromosome, and gene views in Agilent CGH Analytics 3.4. Likewise, the cumulative frequency of gain score for each feature is the percentage of samples for which the segment value is above the threshold log 2 (7/6).

The significance of association of chromosomal breakpoints within CNV loci was calculated as follows: the statistics for breakpoints in CNV loci is the χ² goodness of fit between the observed fraction of breakpoint in CNV loci (count of observed breakpoint in CNV loci/total observed breakpoints), and the fraction of expected breakpoints in CNV loci (total base pair of CNV areas in array/total base pair covered in array). The significance threshold for this statistical test was P < α = 0.05 (two-sided).

The correlation between average CGH copy number and average gene expression was performed using Pearson's correlation for each CBS segment with (a) ratio average values (CBS segment mean from this article), as the X-axis versus (b) average of gene expression [log 2 (ratio); from ref. 25], as the Y-axis. We excluded gene expression arrays with >30% missing data points, and to prevent distortion caused by outliers, we excluded segments containing less than six features for either gene expression or CGH prior to calculating the correlation, i.e., 10 samples and 314 of 369 segments were retained. The significance threshold for this statistical test was P < α = 0.05 (two-sided).

CGH using chromosome 8–specific tiling BAC arrays. Chromosome arm 8q is one of the most common targets of genomic amplification in colon cancer. It is also associated with the development of both lymph node and distant metastases, and contains single nucleotide polymorphisms that predispose to the development of this malignancy (2, 11, 12, 14, 15). We therefore aimed to generate a high-resolution map of genomic copy number changes by analyzing 51 primary colon tumors by CGH using a BAC clone-based genomic tiling array. Twenty-five of these tumors were associated with lymph node metastases at the time of surgery (UICC-III), whereas the remaining patients were free of lymph node metastases (UICC-II, n = 26). The clinical information is presented in Table 1.

Confirming previous results, 50% of the cases showed aberrations on chromosome 8; 37% had gains on the long arm, and 45% had losses on 8p. Two regions with the highest copy number increases mapped to genome locations 105 to 120 Mbp and 127 to 142 Mbp. This includes chromosome band 8q24.21, the genomic location of the MYC oncogene. Interestingly, in striking difference from the results suggested by conventional CGH, the short arm of chromosome 8 was not subject to loss in its entirety: in the majority of samples with 8p alterations, the loss of this arm did not include a small region close to the centromere. This region, which includes 5.5 Mbp of the short arm, was either present in normal copy number, or in fact gained to the same extent as the long arm. The summary of this analysis is presented in Fig. 1A and B. Interestingly, when we then tried to understand why chromosome 8p was prone to chromosomal breaks to such an extent, we noticed that in 9 out of 14 cases, the breakpoints coincided with sites of known structural variants identified within the human population, either CNVs or segmental duplication (Supplementary Table S1). Figure 1B summarizes the BAC array data of the 8p aberration patterns in individual cases.

High-resolution genome-wide mapping of DNA copy number changes. In order to more precisely map these breakpoints and to investigate whether the observed predilection for chromosomal breaks at sites of known structural variants applies to regions other than 8p, we profiled 31 of the 51 colon cancers analyzed with the BAC arrays by aCGH on a 185K oligonucleotide, genome-wide platform (see Table 1 for the respective cases). Regions of genomic imbalances in these tumors were determined using CBS (26). Taking the different resolution limits of the platforms into consideration, we observed an excellent congruence between the techniques, and the aberration patterns on 8p were confirmed. Our analyses also confirmed the recurrent low-level copy number changes of chromosomes 7, 8, 13, 18, and 20, which are specific for sporadic colorectal cancers (27). However, attributable to the increased resolution of this platform for aCGH, additional novel sites of chromosomal gains and deletions could be identified. Specifically, we detected 393 chromosomal breakpoints (defined as segments of copy number change) in 31 cases, for an average of 12.7 breakpoints per case (0–34). One hundred and sixty-nine breakpoint segments (including those that affected entire chromosomes) resulted in copy number increases, whereas 202 regions of copy number loss were present. Segments with copy number increase were recurrently mapped to chromosomes and chromosome arms 7, 8q, 13, and 20q, whereas losses occurred most frequently on 1p, 5q, 8p, 14, 15, 17p, 18, 21, and 22. A summary of these results is presented as cumulative gain or loss in Fig. 1C, and as a frequency distribution in Supplementary Fig. S1. Gains on chromosomes 13 and 20 were most commonly observed, and also revealed the highest level of genomic amplification, followed by copy number increases of chromosomes 7 and 8q. Chromosome arms 18q, 17p, and 8p showed the highest degree of genomic loss (both in terms of cases and actual copy number reduction).

We detected several regions whose recurrent copy number changes were not appreciated in our previous analyses of colorectal carcinomas using conventional CGH analysis (28). In addition to the above-described retention on 8p, we observed a similar pattern on chromosome 20: the breakpoint that results in copy number increase resides in the euchromatic region of 20p, and not in the centromere. In addition, we observed interstitial deletions of chromosome band 4q34.3-35.2 in three cases (CC-P19, CC-P20, and CC-P65), and a deletion that included the terminal band of the short arm of chromosome 11 (11p15.5) in two cases (CC-P23 and CC-P38). Bands 13q21.32 to 13q31.2 were deleted on this commonly gained chromosome in CC-P23, and remained in normal copy number (with the rest of the chromosome gained) in CC-P65. A few localized high-level amplifications were mapped to chromosome bands 4q13.2-13.4, 5q32-33.2, and 6p21.1 (CC-P14), and 16q12.2 in CC-P65. In CC-P23, we observed the genomic amplification of the ANKRD10 gene, which maps to distal chromosome 13.

Comparison between lymph node–negative and -positive cancers. The presence of synchronous lymph node metastases dictates the inclusion of chemotherapy in the treatment of patients with colon cancer. In order to explore whether lymph node status could be reflected by specific copy number changes on chromosome 8, as previously suggested (11), or elsewhere in the genome, we compared the distribution of genomic imbalances as determined in both groups using the oligonucleotide array platform. The percentage of chromosomal gains and losses was not different between the lymph node–positive (average, 12.9 per case) and lymph node–negative (average, 11.7 per case) carcinomas. The average number of gained or lost segments in the UICC-II tumors was 6.8 and 6.1, respectively, and for the UICC-III tumors, it was 4.8 and 7.5, respectively. In order to further analyze whether tumors associated with lymph node metastases carry distinct genomic aberration profiles, we analyzed the frequency of all CBS units in the two groups: we could not detect any CBS units that were uniquely gained or lost in either the UICC-II or UICC-III samples, nor did we detect a differential distribution of CBS units between the groups that exceeded a 30% difference threshold. The summary plots of the UICC-II and UICC-III tumors are displayed in Supplementary Fig. S2A and B.

Influence of genomic imbalances on gene expression. Genomic copy number changes are arguably one of the most recurrent features of solid tumors of epithelial origin. Consequently, numerous groups attempted to clarify the relationship between genomic copy number changes and gene expression levels; however, most of these studies focused either on the effect of whole chromosomes, or on regional amplicons (25, 29–33). We now analyzed, for the entire colon cancer genome, this correlation by plotting the average gene expression values for all CBS units against their genomic copy number (we only included those 17 cases for which we had gene expression results in both the tumor and matched normal mucosa, and those CBS segments that contained more than five genes). The analysis, shown in Fig. 2, revealed a significant correlation of genomic copy number with average gene expression levels, therefore suggesting a direct effect of gene copy on relative message levels (R = 0.66709, P = 2.2e⁻¹⁶).

CNVs. In addition to low level copy number changes, the CBS analysis revealed numerous recurrent loci of localized high-level copy number increases or decreases relative to the reference DNA. Such changes could be indicative of structural variations in the genome, either germ line or somatic. Structural variations, including CNVs, have recently emerged as a novel class of DNA segments that differ from one individual to another (20). The systematic mapping of CNVs in 270 individuals that constitute the human HapMap collection (34) suggests that ∼12% of the human genome could be subject to copy number variation (20), with as much as 3% of these regions (∼0.3% of the total genome) varying from one individual to another (35). CNVs therefore contribute significantly to human sequence variation. Applying the CBS algorithm, we could identify 120 sites that were suggestive of CNVs (i.e., sites of high-level copy number increase or decrease of no more than 200 kbp). The comparison of the variants detected in our data set derived from 31 tumors with the database on genomic variation⁸

indicated that 81 of those variants (67.5%) overlapped with known CNVs, whereas 39 (32.5%) were potentially novel sites of CNVs. A complete list is provided in Supplementary Table S2.

In order to assess whether these alterations were genomic copy number changes that emerged de novo in the tumors, i.e., somatic, or whether they would have to be considered germ line events, we hybridized tumor DNA from five patients against DNA prepared from matched normal mucosa tissue. CNVs detectable in such experiments can be considered bona fide somatic events. The initial CGH experiments revealed 54 known CNVs in these five patients (9–13). We now observe that 13 of these CNVs remained when tumor DNA was hybridized against DNA from matched normal mucosa (1–4). Based on these observations, we conclude that 24% of the CNVs are actual variants that emerged in the tumor tissue, and hence, somatic CNVs. Examples of these variant regions are shown in Fig. 3A.

Similar to fragile sites, regions of genomic copy number variation could trigger genomic rearrangements (20, 22). In order to establish to which extent genomic regions containing CNVs contribute to the emergence of chromosomal translocations (as deduced from the presence of segments of genomic copy number change by CGH), we asked how frequently chromosomal translocations coincided with the location of previously identified CNVs. Given the high resolution of our platform (16 kbp), we could manually annotate the breakpoint sequence for each segment using the Database of Genomic Variation⁸ in order to search for structural genomic variants at these genome coordinates. In the 31 cases analyzed with the oligonucleotide platform, we mapped 393 sites of genomic copy number change, 161 of which occurred at the site of known CNVs (Fig. 3C). Taking into account that ∼18% of the genome consists of segments identified as CNV, the probability that 41% of all translocations mapped to CNVs by coincidence is exceedingly low (P < 2.2e⁻¹⁶). This suggests that CNV loci (including segmental duplications) contribute significantly to the emergence of chromosomal breaks in colon cancer, and hence, to the development of genomic imbalances. CNVs that colocalized to chromosomal breakpoints in our data set are listed in Table 2. Figure 3B presents an example of a subchromosomal genomic deletion that eliminates one copy of the tumor suppressor gene APC and shows the association between the site of the chromosomal break with a known CNV. Figure 4 shows the possible emergence of genomic copy number changes in CC-P10. In this tumor, we observed chromosomal breakpoints that coincided with two segmental duplications, DC3225 on chromosome 17p and DC2472on chromosome 20p. A sequence homology of 94.51% between these two sites suggests that homologous recombination events could have contributed to a chromosomal translocation, which eventually leads to the observed pattern of DNA gain and loss.

Patterns of imbalances. Here, we present a comprehensive map of genomic imbalances in primary colon carcinomas generated using high-resolution aCGH on a genomic tiling array for chromosome 8 and a 185K oligonucleotide platform. The results are, in general, congruent with previous analyses using chromosome banding techniques (2), CGH on metaphase chromosomes (28), and aCGH with a genomic BAC platform (6–10). In fact, the summary pattern of chromosomal gains and losses in our data set and data sets reported in the literature suggest a striking conservation of genomic imbalances, and underlines the biological significance of these recurrent aneuploidies. We observed, however, a few dissimilarities between the data set presented here and previously published results. For instance, the short arm of chromosome 8 is not always lost in its entirety (as suggested by cytogenetic analyses using chromosome templates), but a minimally retained region that escapes this loss comprises chromosome band 8p11.1-11.2, which is consistent with previous aCGH analyses on genomic platforms (9, 36). A similar phenomenon on the short arm of chromosome 20 was detected. Second, we observed several regions of subtle copy number changes that were clearly below the resolution of conventional cytogenetic or CGH analyses. In patient CC-P9, we observed a localized amplification of chromosome band 6p21.1, which resulted in the significant overexpression of histone gene HIST1H2BM in this tumor. Other examples include a common deletion mapped to chromosome band 4q34-35. The most notable difference between chromosome CGH analysis, the use of an overlapping BAC array for chromosome 8, and the high-resolution oligonucleotide platform was the identification of frequent sites of small, high-level gains and losses, many of which coincided with loci of known structural variants in the human genome,⁸ which could only be mapped using the 185K oligonucleotide platform. This will be discussed separately below.

Correlation of genomic copy number and gene expression changes. The results presented here underscore the dominant role of specific and recurrent genomic imbalances, which arguably, are one of the defining features of genetic insults in colon cancer cells. We and others have therefore tried to understand the consequences of such genomic imbalances on the cancer transcriptome (25, 29–33). In general, the data are consistent with the interpretation that genomic copy number is positively correlated to transcript levels. The data set generated here now affords us the possibility to interrogate the relationship of genomic imbalances, as detected by segments of copy number change based on the CBS analysis (185K oligonucleotide arrays), with the expression levels of resident genes for the entire genome (Fig. 2). The data show that there is indeed a general, and statistically significant correlation of genomic copy number and gene expression levels and thus provide further evidence that these imbalances exert a direct effect on the cancer transcriptome, and hence, result in a massive and complex deregulation of the transcriptional equilibrium of malignant epithelial cells. This observation underlines the importance of the question as to which extent such rather global gene expression changes contribute to tumorigenesis vis-à-vis the targeted deregulation of specific genes by mutation, deletion, amplification, or epigenetic deregulation.

CNVs and potential mechanisms of induction of chromosome breakage. CNVs constitute a subset of structural variants that represent a substantial amount of interindividual genetic variation (20). The most comprehensive catalogue of structural variants in the human genome can be found at http://projects.tcag.ca/variation/. The data summarized there was generated by analyzing the genomes of 270 individuals from the human HapMap consortium using both aCGH and genome-wide single nucleotide polymorphism platforms. These variants are rather ubiquitous, comprising ∼12% of the human genome. Some of them have been shown to be associated with a particular phenotype and with disease (20). Based on a comprehensive evaluation of chromosomal breakpoints and associated genomic copy number changes in cell lines derived from solid tumors (i.e., bladder, prostate, cervix, pancreas, and breast), we could previously show that a considerable fraction of chromosomal translocations (in that case referred to as jumping translocations) originated in the pericentromeric heterochromatin of several chromosomes (21). Such heterochromatin is enriched for segmental duplications, and these show a 6:1 ratio of interchromosomal to intrachromosomal duplications. These regions can also vary in copy number between individuals, and if so, could be classified as CNVs (22). We were therefore curious as to which extent chromosomal breakpoints (as defined by sites of genomic copy number change using aCGH) colocalize with such structural variants in the genome of primary colon cancers. Surprisingly, ∼41% of all translocations resided at sites of known CNVs, including segmental duplications (Fig. 3; Table 2). Such an association is highly significant (P < 2.2e⁻¹⁶). Figure 4 suggests a possible scenario on how the observed pattern of genomic gain and loss could be explained in one of the tumors analyzed here (CC-P10). It is, however, not possible to perform cytogenetic analysis on this very sample, and therefore, one cannot formally prove that the observed pattern of imbalance is indeed caused by translocations between chromosomes 17 and 20 despite the high degree of homology (95%) between the segmental duplications that colocalize with these breakpoints. Alternatively, CNVs and segmental duplications are simply regions more prone to chromosome breakage, which can result in loss of genomic segments due to the lack of a centromere, or translocation with other regions in the genome without homology. The difference in copy number of these regions between individuals, however, is perhaps an indication that they are particularly susceptible to homology-mediated recombination, i.e., formation of chiasmata, in meiotic cells. In cells experiencing DNA damage, one could easily envision that aberrant homology-mediated repair of segmentally duplicated regions might also lead to chromosome aberrations in somatic cells, such as deletions, inversions, and translocations. Such analyses will have to be conducted using cell lines established from primary tumors. The mere fact that homologous chromosomes in an interphase nucleus rarely tend to be in the same topographical neighborhood (37) makes it more likely that a homology search will identify a duplicated region on a different chromosome. This may explain the relatively high frequency of whole chromosome arm gains and losses in aneuploid tumors. Why might these regions be more susceptible to DNA damage? First, CNVs are often found in association with gene coding regions and therefore might be expected to be in an open configuration, making them more susceptible to DNA damage. Alu sequences, satellite repeats, and regions with hallmarks of DNA fragility are found to be enriched at the boundaries of these regions, supporting the hypothesis that these areas are preferential sites of DNA double-strand breaks, making them ideal substrates for repair pathways with the potential for causing increased copy number or rearrangements. Gorgoulis et al. (38) and Bartkova et al. (39) observed an early activation of DNA damage response pathways in precancerous lesions. Serrano and colleagues showed that high expression of oncogenes triggers a permanent block in replication, termed oncogene-induced senescence (40). Oncogene-induced senescence has recently been shown to induce a DNA damage response in tissue culture models (41, 42) as well as in vivo during the development of thymocytes (43), and is able to restrict the growth of human and murine precancerous tissues (44–48). These early incidents set the stage for the events outlined above. Further progression to more advanced dysplastic lesions and to invasive carcinomas was associated with p53 inactivation and reduction of apoptosis. Interestingly, allelic loss of loci prone to DNA double-strand break formation, i.e., fragile sites was common. The authors put forward a model in which, at early stages of tumorigenesis, replicative stress triggers the formation of double-strand breaks, which in turn results in genomic instability, and through that to inhibition of apoptosis and cell cycle arrest. One could therefore reasonably speculate that CNV-induced double-strand breaks are among the earliest gross chromosomal aberrations in cancer genomes. The resulting unbalanced translocations could then, in addition to aneuploidies of entire chromosomes (which are also observed in premalignant, early dysplastic lesions), contribute to the emergence of patterns of genomic imbalances that define different tumors of epithelial origin. These speculations are potentially substantiated by our observation that ∼24% of the observed CNVs are actually de novo events, i.e., are detectable when tumor DNA was compared with DNA prepared from matched normal mucosa tissue. These data suggest that regions of copy number variation observed in the normal population continue to be subject to hypervariability and are foci of genomic instability in the tumor.

It remains to be seen whether the striking colocalization of sites of structural variants in the genome and cancer-associated chromosomal breakpoints that we observed here in colon carcinomas occurs in other epithelial neoplasms as well. It will be equally interesting to determine whether the distribution and frequency of specific CNVs is associated with population-based cancer risk.

Grant support: Intramural Research Program of the NIH, National Cancer Institute, the Deutsche Forschungsgemeinschaft (KFO 179), and a stipend from the Deutsche Krebshilfe (M. Grade).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

The authors thank Hesed M. Padilla-Nash, Buddy Chen, Joseph Cheng, and Jessica Eggert for helpful discussion, technical, and editorial assistance.