Cancers with chromosomal instability (CIN) are held to be aneuploid/polyploid with multiple large-scale gains/deletions, but the processes underlying CIN are unclear and different types of CIN might exist. We investigated colorectal cancer cell lines using array-comparative genomic hybridization (CGH) for copy number changes and single-copy number polymorphism (SNP) microarrays for allelic loss (LOH). Many array-based CGH changes were not found by LOH because they did not cause true reduction-to-homozygosity. Conversely, many regions of SNP-LOH occurred in the absence of copy number change, comprising an average per cell line of 2 chromosomes with complete LOH; 1-2 terminal regions of LOH (mitotic recombination); and 1 interstitial region of LOH. SNP-LOH detected many novel changes, representing possible locations of uncharacterized tumor suppressor loci. Microsatellite unstable (MSI+) lines infrequently showed gains/deletions or whole-chromosome LOH, but their near-diploid karyotypes concealed mitotic recombination frequencies similar to those of MSI− lines. We analyzed p53 and chromosome 18q (SMAD4) in detail, including mutation screening. Almost all MSI− lines showed LOH and/or deletion of p53 and 18q; some near-triploid lines had acquired three independent changes at these loci. We found consistent results in primary colorectal cancers. Overall, the distributions of mitotic recombination and whole-chromosome LOH in the MSI− cell lines differed significantly from random, with some lines having much higher than expected levels of these changes. Moreover, lines with more LOH changes had significantly fewer copy number changes. These data suggest that CIN is not synonymous with copy number change and some cancers have a specific tendency to whole-chromosome deletion and regain or to mitotic recombination. (Cancer Res 2006; 66(7): 3471-9)
- Gastrointestinal cancers: colorectal
- Molecular cytogenetics
- LOH and marker studies
- Genetic instability: multistep progression
Genomic instability is often thought to be necessary for carcinogenesis. Some cancers, notably those of the colorectum, have so-called microsatellite instability (MSI). This relatively well-defined phenomenon results from defective DNA mismatch repair. MSI+ cancers seldom deviate from a near-diploid karyotype ( 1). A much more poorly defined, although probably more common, phenomenon is chromosomal instability (CIN; ref. 2). Sometimes, CIN is used to describe cancers that are found by cytogenetics or flow cytometry to have a karyotype which is aneuploid and polyploid; ∼50-65% of colorectal cancers probably fall into this category. However, CIN can also be used to describe cancers that have multiple gains or deletions of chromosome arms, or multiple translocations. The causes of CIN in human cancers and the nature of any underlying defect are unclear. One model proposes that CIN is caused by the initial acquisition of a polyploid, yet euploid, karyotype, followed by subsequent chromosomal losses and gains ( 3). Others have suggested that CIN is the direct or indirect result of specific mutations or gene silencing ( 4). In the latter case, CIN could result from a variety of underlying defects, including microtubule, centromere or centrosome dysfunction; telomere erosion; chromosome breakage; and failure of cell cycle checkpoints. In vitro experiments have shown that specific mutations or changes in gene expression can lead to aneuploidy/polyploidy. In colorectal cancer, genes proposed to cause CIN include BUB1, BUBR1 ( 5), adenomatous polyposis coli (APC; ref. 6), FBXW7/CDC4 ( 7), and aurora kinases ( 8). In vivo, however, such associations have been more difficult to show. APC mutations, for example, are found in colorectal cancers of all karyotypes ( 9).
The inconsistent or imprecise use of the term CIN partly reflects the fact that different experimental methods can be used to detect changes at the level of the chromosome or chromosome arm. In addition to cytogenetic methods, at the molecular (DNA-based) level, the two main techniques for identifying CIN-related changes are comparative genomic hybridization (CGH) and loss of heterozygosity (LOH) analysis. CGH detects chromosomal or subchromosomal gains and deletions (copy number changes) relative to total genome dosage. Thus, array-based CGH (aCGH) will find gains or deletions of chromosomal regions down to levels of 1 Mb or less (and hence will detect aneuploidy) but it will not detect polyploidy. Despite this limitation, CIN+ cancers usually show more aCGH changes than CIN− lesions because aneuploidy and polyploidy tend to co-occur.
LOH analysis relies on the assessment of polymorphic markers, such as microsatellites. In practice, the measurement is of the relative dosage of the maternal allele to the paternal allele at the site of the polymorphism. In a completely pure, homogeneous tumor, heterozygosity is truly lost when all copies of a particular allele are derived from the same parental homologue. This phenomenon also used to be described, accurately, as reduction to homozygosity. In principle, this can occur either by deletion of all copies of one homologue or by a mechanism which renders the homologues identical without a net change in gene dosage. Based on existing data and theoretical considerations, at least three types of change might cause LOH without an accompanying copy number change: (a) deletion of one chromosomal homologue and regain of the other homologue by duplication (or gain of one homologue followed by loss of the other), resulting in whole-chromosome LOH; (b) LOH extending from a telomere to involve whole or part of a chromosome arm, resulting from break-induced replication and classically termed “mitotic recombination” in cancer genetics; and (c) an interstitial region of LOH, probably also resulting from a mitotic recombination event, but for which the term “mitotic gene conversion” is sometimes used.
In cancers derived from fresh-frozen tissue, a pure tumor cell population very rarely exists. LOH (or allelic imbalance, as it is sometimes accurately called in this context) is therefore observed as a decreased dosage of one allele, rather than absence. The threshold for scoring decreased dosage is ad hoc but 50% is often used. It follows that LOH may not result from true reduction to homozygosity in these cases. Instead, copy number changes (amplification, gain, deletion) without true homozygosity can be scored as LOH, depending on the threshold used. Even in pure tumors, however, changes in relative allelic dosage often occur. For example, in a perfectly triploid cancer, all polymorphic loci would show an allelic dosage of 2:1. These changes may all be scored as LOH if the threshold is set sufficiently low despite the fact that their functional importance is highly doubtful in most cases.
It follows from the above that the relationships between CIN, copy number change, and LOH are complex. Although CIN+ cancers probably have more copy number changes and more LOH than CIN− lesions, the underlying mutational processes may often be different. We have sought to clarify the frequencies of, and relationships among, different large-scale mutations in a panel of colorectal cancer cell lines. We have used aCGH to assess dosage changes, supplemented by karyotyping and multicolor fluorescence in situ hybridization (FISH). We used single-nucleotide polymorphism (SNP) microarrays to assess LOH (SNP-LOH); in practice, our LOH assessment almost exclusively detected only true reduction to homozygosity. The results of our work show that LOH without copy number change is common and that multiple forms of CIN may exist within cancers.
Materials and Methods
Cell lines. Forty-five unselected colorectal cancer cell lines were studied ( Table 1 ). DNA from all of these was taken at the same passage from which metaphases were derived for cytogenetic analysis. The great majority of lines had a near-diploid or near-triploid karyotype.
Array CGH. In brief, the array comprised spotted DNA from 3,452 large insert genomic clones at an average spacing of about 1 Mb throughout the genome ( 10). Cy3- and Cy5-labeled test (cancer) and normal DNAs were hybridized to the arrays together with herring sperm and CotI DNA. After scanning, rejecting poorly hybridized arrays and correcting for background, the log 2 ratios of the fluorescence intensities of test (T) to control (N) were calculated, following normalization to the remainder of the genome in that tumor. The determination of significant copy number changes was done as described ( 10). The sex chromosomes were not analyzed. We used the terms “gain” and “deletion” for aCGH (copy number) changes. For consistency, we also referred to absence of a whole chromosome as deletion. All changes were mapped according to the May 2004 Human Genome freeze (http://genome.ucsc.edu/). aCGH data for some of the cell lines have previously been reported ( 10) and the remainder are available from the authors on request.
SNP microarray LOH. Cancer cell line DNAs were processed, labeled, and hybridized to the Affymetrix GeneChip Human Mapping 10K Xba131 SNP arrays according to the instructions of the manufacturer. Arrays were scanned using the Affymetrix 3000 Scanner. The MGCOS and GDAS (Affymetrix) software were used to process the array output data, according to the protocols of the manufacturer, and thresholds for scoring homozygotes and heterozygotes. Median SNP call rates of 95% (IQR = 92-98%) were achieved. Because paired constitutional DNA is not generally available for the cancer cell lines, assessment of LOH was done as follows. First, apparently homozygous autosomal regions of ≥10 Mb were identified. Then, the number of successfully scored SNPs within the region was counted and the reported heterozygosity of each SNP was determined. The probability of all the scored SNPs in the region being truly homozygous was then calculated using a Bonferroni correction to allow for a total of about 350 regions of 10 Mb in the genome. By inspection of large regions of known hemizygosity ( 11), we found that up to 2.5% of SNPs erroneously reported heterozygosity. We therefore permitted a maximum of 2.5% apparently heterozygous SNPs within a single significant region of LOH and we incorporated SNPs reporting heterozygosity into the probability calculation. Conservatively, we then classed all regions with P < 0.001 of being truly homozygous as having LOH. Each of these regions was then compared with the LOH output from the Affymetrix CNAT tool. The great majority of our LOH calls agreed with CNAT, but where there was discordance, we reached a decision by taking into account factors such as chromosomal location (e.g., relative to our calculation, CNAT over-called LOH close to centromeres and under-called at telomeres). CNAT was also tested for its ability to score copy number changes. Whereas large regions of gain or loss were successfully identified, the CNAT output was consistently far more variable and, hence, less reliable than that from aCGH (examples available from authors; also suggested by ref. 12). CNAT was therefore not used further for assessment of copy number. The SNP arrays were therefore used solely to report LOH (or simply “loss”) in its strict sense (i.e., a reduction to homozygosity). Where aCGH and SNP-LOH both found changes, we assumed precedence of the former when scoring the change.
MSI, mutation detection, and microsatellite genotyping. Standard methods were used (details available from authors). Cancer cell lines were screened for p53 and SMAD4 mutations by single-strand conformational polymorphism or denaturing high-performance liquid chromatography analyses. MSI was assessed using the near-monomorphic mononucleotide repeats BAT26 and BAT25. Microsatellites used to validate the SNP-LOH were typed using standard methods and, as for the SNP-LOH data, scored for homozygosity (one allele present) or heterozygosity (two alleles present), irrespective of relative allelic dosages.
Overall findings. In the 45 colorectal cancer cell lines, SNP-LOH analysis detected multiple genomic changes (mean, 4; median, 3; range, 0-11), which were not found by either aCGH (Supplementary Fig. S1) or cytogenetic methods (details not shown). All cell lines except SNUC2B harbored regions of LOH which were previously unsuspected on the basis of aCGH analysis. Our focus below is on these LOH changes which did not result in copy number change because we believe aCGH to be efficient at detecting the latter. Conversely, copy number changes were not consistently detected by SNP-LOH, principally because most copy number changes did not result in true reduction to homozygosity because most involved gain or loss of a single copy on a near-triploid background; the SNP algorithm generally called such asymmetrical allelic ratios as heterozygotes. When aCGH and SNP-LOH both detected changes, it was usual for the boundaries of these regions to match. However, in 45 cases in which regions of LOH overlapped with deletion or gain by aCGH, the boundaries of the changes were not coincident; it seemed that LOH without copy number change and a dosage change involving part of the same region had occurred separately. A small number of such changes were simply too complex to explain other than by multiple different events (Supplementary Fig. S2). These complex changes fitted no consistent pattern and involved varying combinations of gains, losses, and LOH events, involving all or part of chromosomes. The complexity of these changes helps to explain why attempts to map genes mutated in cancers on the basis of LOH have been largely unsuccessful.
Whole-chromosome LOH without whole-chromosome change on aCGH was found in 30 of the 45 (67%) cancer cell lines ( Table 1). There was a total of 80 such changes and a mean of two per cancer (median, 1; range, 0-10). Chromosome 21 showed no whole-chromosome LOH in any line, most chromosomes were lost in 2 to 4 lines, and chromosome 17 showed complete LOH in 11 of 45 (24%) lines. Overall, in about half of the 80 chromosomes which had undergone complete LOH without corresponding whole-chromosome gain or deletion, copy number changes involving a smaller part of the chromosome were added. The smaller gains could have occurred before or after the presumed whole-chromosome deletion and regain. The smaller deletions might have resulted either from loss without full regain or from deletion subsequent to the loss/reduplication.
LOH by mitotic recombination was found in 31 of 42 (74%) cancer cell lines ( Table 2 ). The total of 68 mitotic recombination events represented a mean of 1.5 and median of 1 such changes per line (range, 0-6). The chromosome arms most frequently involved were 5q (25% of lines), 9p (13%), 9q (9%), and 17q (10%; Table 2). Added gains and/or deletions involving part of the region with mitotic recombination were found in only 12 of the 68 (18%) changes. The sizes of the regions of mitotic recombination ranged from our minimum threshold of 10 to 102 Mb on chromosome 1q in line C32. Most mitotic recombination changes occurred too infrequently to allow mapping of consistent boundaries although there was a general tendency for the boundaries to occur within peri-centromeric repeats, leading to LOH of the whole chromosome arm. However, for the relatively high-frequency mitotic recombination change on chromosome 5q, which presumably targeted APC in most cases, noncentromeric breakpoints were found in 7 of 12 (58%) cell lines close to 69 or 99 Mb. These two sites are the locations of the only low copy number repeats between the centromere and the APC locus (110 Mb). The repeats have a complex structure (http://genome.ucsc.edu), with several tandemly repeated subregions of homology >98% and multiple regions which are >95% identical between the sites near 69 and 99 Mb. There is further similarity to a region on chromosome 5p, close to 20 Mb. The preferential location of mitotic recombination breakpoints at these repeats may be caused by some uncharacterized tendency to intrachromosomal or interchromosomal recombination, or to intrinsic chromosomal fragility.
In total, 45 regions of putative mitotic gene conversion (MGC), interstitial LOH of ≥10 Mb, were found ( Table 3 ). Fourteen (32%) of these MGCs were found together with partially overlapping aCGH changes. The remaining 31 (68%) MGCs lay in regions with no copy number change by aCGH. Chromosome 1q was the region most frequently affected by MGC, occurring in 4 (9%) cancers with a recurrent proximal boundary in a gene-poor region close to 192 to 193 Mb and with a putative minimal region between 196 and 207 Mb. Most cancer cell lines harbored 0 to 3 MGCs, but the MSI+, near-diploid cancer LS180 was a striking outlier, with 8 such events.
For four cell lines (C10, C32, HCA7, and LS411), paired constitutional DNA was available and we verified our results by typing these samples. In no case was our LOH region shown to be a false-positive finding resulting from constitutional homozygosity or hemizygosity. In addition, no regions of LOH smaller than our working threshold of 10 Mb were detected.
LOH in the MSI+ (CIN−) cancer cell lines. The MSI+ cancer cell lines in the study ( Table 1) were all near-diploid. They showed very few changes by aCGH (mean and median per line: 2 regions of gain or deletion; ref. 10). Whole-chromosome LOH (mean, 0; median, 0; range, 0-1; Table 1) was significantly less frequent in MSI+ lines than in the MSI− cancers (mean, 2; median, 1; range, 0-10; P = 0.0002, Wilcoxon test). The number of mitotic recombination events ( Table 2) was, however, similar in the MSI+ (mean, 1.5; median, 1; range, 1-4) and the MSI− groups (mean, 1.5; median, 1; range, 0-6; P = 0.73, Wilcoxon test). Gene conversions ( Table 3) were absent in the MSI+ group, apart from the strikingly high frequency in LS180 which rendered the occurrence of MGCs not significantly different between the MSI+ and MSI− groups; however, if LS180 was deemed to be an outlier and this excluded from the analysis, the frequency of gene conversions was higher in the MSI− group (mean, 1; median, 1; range, 0-4) than in the MSI+ group (mean, 0; median, 0; range, 0-3; P = 0.017, Wilcoxon test).
LOH, deletion, and gain near p53 and on chromosome 18q. We used two relatively well characterized sites of LOH and copy number change in colorectal cancer to show the power of the combined genome screens to indicate LOH and copy number change. SNP-LOH analysis of the p53 locus and chromosome 17p revealed both losses which had not been not detected by aCGH and several changes which were of greater complexity than previously suspected ( Table 4 ). Of 32 cancer cell lines with any change at p53, aCGH and SNP-LOH were concordant in 16 cases, with deletion (chromosome arm, whole-chromosome or interstitial) matching the region of LOH. One line (CACO2) had gain of the terminal part of 17p and this change was also found by SNP-LOH. In 11 other cancer lines, LOH involved the whole of chromosome 17 ( Table 4), but the same change was not present on aCGH. In 6 of these 11 lines, no change at all had been found on 18q by aCGH; whole-chromosome deletion and regain had presumably occurred. In the five other cancers with whole-chromosome LOH, part or all of chromosome 17 was additionally gained or lost ( Table 4). In two cancers (COLO320 and VACO5) with no change on aCGH, mitotic recombination was found to be responsible for LOH at p53. Finally, in LIM1863 and SW1116, mitotic recombination at p53 partially overlapped with deletion of whole or part of the chromosome, presumably owing to independent events. Of 34 MSI− lines, only three (C10, SKCO1, and COLO678) had no copy number change or LOH around p53; in addition, one MSI+ line, VACO5, had a mutation at p53 accompanied by LOH.
Some chromosome 18q loss is believed to target the SMAD4/MADH4/DPC4 tumor suppressor, but only a minority of cancers have mutations in SMAD4 ( 13). Haploinsufficiency of SMAD4 has been proposed as an explanation ( 11, 14). In this analysis, all MSI− cell lines (and no MSI+ lines except SW48) showed changes on 18q ( Table 5 ). The SNP-LOH analysis revealed a number of changes, mostly whole-chromosome LOH, which were not found by aCGH ( Table 5). Overall, 17 of 35 (49%) cancer cell lines with 18q changes had deleted the entire chromosome, as detected by both aCGH and SNP-LOH. A further five lines had interstitial or terminal deletions detected using both techniques. Four lines had whole-chromosome LOH but no aCGH change. Three lines (C70, CX1, and CACO2) had whole-chromosome LOH and a further deletion of part of the chromosome including the SMAD4 locus. Three other lines (LS123, VACO10MS, and NCI-H716) had deletions which were not present on SNP-LOH, presumably because they involved only one copy in a polyploid cell (data from this study and ref. 11). Only two lines (C32 and SW48) showed evidence of mitotic recombination ( Table 5). Finally, SKCO1 had complex changes on 18q, comprising multiple gains and deletions and consistent with previous findings using microsatellites ( 11). Deletions of 18q tended to involve a breakpoint close to the centromere (∼20 Mb) and to extend to the telomere ( Table 5).
Verification of LOH without copy number change in primary colorectal cancers. To show that LOH without copy number change occurred in primary cancers as well as cell lines, we reexamined our existing data derived from a panel of unselected colorectal cancers. Although genome-wide assessment of LOH in these cancers was not practicable, we focussed on p53 and chromosome 18q near SMAD4 by comparing copy number change assessed by aCGH and LOH assessed by a panel of three microsatellite markers. To minimize false-positive LOH, for the purposes of this demonstration analysis only, we required the following stringent criteria to be met to score LOH: (a) all informative microsatellites to show allelic imbalance (defined as a relative dosage change of >3.0 or <0.33); (b) at least two markers to be informative; (c) LOH to involve the shorter allele for at least one microsatellite per tumor (because false-positive LOH is more likely to result from unequal amplification involving the lower microsatellite allele). At p53, of 54 informative cancers, 7 (13%) showed LOH without copy number change by aCGH; by comparison, 8 of 45 (18%) cell lines showed p53 LOH without copy number change. At SMAD4, 4 of 64 (6.3%) cancers showed LOH without copy number change compared with 6 of 45 (13%) cell lines. No clinicopathologic or molecular features distinguished the lines showing LOH and no aCGH change (details not shown). These results are consistent as regards the similar frequencies of changes in the cancers and cell lines (P > 0.15 in each case, Fisher's exact test), the slightly higher frequencies in the latter perhaps reflecting the relatively stringent thresholds for scoring LOH that we have used. It is notable that in both cell lines and primary cancers, the frequency of LOH without copy number change was slightly higher at p53 than at SMAD4 ( Tables 4 and 5).
New forms of genomic instability. Given that some of the colorectal cancer cell lines had acquired very complex changes (Supplementary Fig. S2), we wondered whether combined aCGH and SNP-LOH analysis might suggest the existence of specific forms of CIN in different tumors. After excluding MSI+ cancer cell lines, all of which have evidence of defective DNA mismatch repair, we tested whether the frequencies of gains and deletions (median = 17 changes per cell line) detected by aCGH fitted the Normal distribution, which would be expected were they mostly random events, as opposed to skewed or bimodal distributions which would be found if some cell lines had a particular tendency to gains or deletions. Good fits were found (P > 0.25 for both, Shapiro-Wilk test), suggesting that copy number changes were principally derived from random events, even if they had occurred on a background of genomic instability and allowing for the fact that some changes targeted known genes such as p53. By contrast, the frequencies of none of the three types of LOH event fitted a normal distribution well (details not shown), but their less common occurrence meant that it was also appropriate to attempt to fit them to a Poisson or negative binomial distribution. The frequency of MGCs in the MSI− cancer cell lines fitted Poisson and negative binomial distributions well (χ12 = 0.00, P = 1.00). However, the frequency of mitotic recombination differed significantly from Poisson, normal, and negative binomial distributions (P < 0.04 in all cases; Supplementary Fig. S3). The frequency of whole-chromosome LOH in the MSI− lines also failed to fit a Poisson, normal, or negative binomial distribution (P < 0.001 in all cases; Supplementary Fig. S3). There was no correlation between the frequency of whole-chromosome LOH and mitotic recombination in each line (t = 0.87, P = 0.39). COLO320 and SW620 had exceptionally frequent whole-chromosome LOH, with 8 and 10 such events, respectively ( Table 1); C32, LIM1863, SW837, and the MSI+ line RKO harbored unusually high numbers of mitotic recombination events, with 6, 5, 4, and 4 respectively ( Table 2). Notably, MSI− cell lines which had higher numbers of LOH changes without copy number change tended to have fewer copy number changes (regression coefficient, −0.24; 95% confidence interval, −0.43 to −0.051; t = 2.63, P = 0.015). Thus, in addition to CIN manifest as chromosome copy number changes, we have found evidence of two alternative types of CIN in MSI− colorectal cancers, the chromosome deletion and regain phenotype and the mitotic recombination phenotype. If the cell line LS180 is considered, there may also exist a rare MGC phenotype.
Evidently, LOH without copy number change at APC and p53 and on chromosome 18q are alternatives to deletion as a method of inactivating a tumor suppressor gene at these locations. However, outside these three sites, there was no tendency for deletions and LOH without copy number change, including, specifically, mitotic recombination, to target the same genomic regions (details not shown; P > 0.5 in all cases, linear regression analysis). Given that both mitotic recombination and deletion are expected to result from the same initiating lesions, double-strand breaks in DNA, it would be expected that the two types of change would affect the same genomic locations were they simply to be background events. The different chromosomal distributions of mitotic recombination and deletions may represent selective constraints on gene dosage in some cancers (perhaps unlikely in polyploid lesions) or different genetic pathways in cancers which tend to have different relative proportions of LOH and deletion.
The use of high-density SNP microarrays has revealed that colorectal cancer cell lines harbor multiple regions of LOH which are not detected by methods, such as array CGH, which screen for copy number changes. In many cases, our use of both SNP-LOH and aCGH analyses has revealed very complex large-scale mutations. Chromosome deletion and reduplication, mitotic recombination, and gene conversion occur in carcinomas at frequencies which, when combined, are about one third as frequent as gains and deletions. This frequency estimate is a minimum one because small regions or mitotic recombination or MGC might have fallen below our detection threshold. In some cancers, mitotic recombination or whole-chromosome LOH is the predominant mode of large-scale change. MSI+ cancers, which are generally near-diploid and have very few copy number changes, have frequencies of LOH by mitotic recombination (although not of LOH by chromosomal deletion/regain or MGC) which are as high as those seen in MSI− tumors, most of which are aneuploid/polyploid. The MSI+ cancer LS180 had an exceptionally high frequency of MGC. We suggest that unknown constraints generally prevent MSI+ cancers from becoming aneuploid/polyploid but that those constraints do not apply to large-scale mutations, such as mitotic recombination or MGC, which involve no change in copy number. MSI+ cancers do, however, seem to be constrained not to undergo whole-chromosome LOH, presumably because this requires a temporary aneuploid phase before euploidy is restored. It is even possible that MSI cannot be selected on an aneuploid background.
Very few other studies to date have used independent methods to analyze gains, deletions, and LOH in tumors on a genome-wide scale. Bignell et al. ( 12) compared aCGH data from two cell lines, HCC1937 and NCI-H209, with LOH and copy number estimates derived using a forerunner of the Affymetrix GeneChip 10K XbaI SNP array. A positive association was found between the two methods but consistency was variable, with the SNP arrays being relatively good at finding large changes, in terms of physical size or copy number, but of poorer performance when analyzing smaller changes and chromosomes 17, 19, 20, and 22. Slightly more studies have used SNP-LOH to study cancers, sometimes using the CNAT or a similar tool. Primdahl et al. ( 15) analyzed 11 patients' microdissected primary and secondary bladder cancers using low-density SNP arrays. Bignell et al. ( 12) used SNP arrays to compare regions of copy number change and LOH in 20 cancer cell lines of various tissue origins. They found about five regions per cancer to have LOH in the absence of copy number change. Similarly, Huang et al. ( 16) screened the Hs578 breast cancer cell line with paired constitutional DNA using the 10K XbaI SNP array. They found some discordance between the copy number changes and LOH and concluded that SNP-LOH could be used in some cancers in the absence of constitutional DNA. Most recently, Raghavan et al. ( 17) screened 64 acute myeloid leukemia samples with near-diploid karyotypes using SNP arrays and found only single, large regions of LOH, probably resulting from mitotic recombination, in 12 of 64 lesions.
All studies of cancer cell lines must be hedged with a few caveats despite the manifest advantages of using a system free from normal tissue. Constitutional DNA is not available from most lines, raising the possibility that small regions of LOH might be missed, although we found no evidence of this in the lines with paired constitutional samples. Cancer cell lines probably undergo a small number of changes in culture although most probably provide a good representation of the original primary cancer (e.g., MSI+ lines remain near-diploid in culture). There is some genetic heterogeneity within each culture although it is also unlikely that primary cancers are homogeneous. Finally, not all types of colorectal cancer grow in culture, including many MSI− tumors with a near-diploid karyotype ( 18). Despite these potential problems, we confirmed the principal findings of our cell line analysis in primary carcinomas by showing that LOH without copy number occurs commonly at two regions analyzed in detail, near p53 and SMAD4.
For p53 and chromosome 18q, SNP-LOH found changes wholly or partially missed by aCGH in about one third of the cancers. Almost all MSI− cancers had LOH and/or deletion at p53 and on 18q. For both loci, there was some evidence of three independent genetic changes ( Tables 4 and 5). Given that loss of 18q and p53 mutations probably occurs at the late adenoma or early carcinoma stage of colorectal tumorigenesis ( 19), it is possible that these changes occurred as one of three independent events in an already near-triploid cell. The identification of LOH and/or deletion at p53 in nearly all MSI− cancers suggests that p53 is a true tumor suppressor, rather than dominant-negative or oncogene ( 20). Furthermore, our finding that chromosome 18q was commonly affected by LOH without copy number change provides good evidence against the hypothesis that simple haploinsufficiency for this region is selected in tumorigenesis ( 11).
We conclude that the use of SNP-LOH to study the common carcinomas detects many changes not found by other methods, such as aCGH and lower-resolution techniques such as multicolor FISH. Mitotic recombination (and MGC) is not restricted by ploidy or MSI status, and near-diploid MSI+ cancers can tolerate mitotic recombination, perhaps because no temporary or permanent aneuploidy is involved. In the MSI− cancer cell lines, the frequency distributions of whole-chromosome LOH and mitotic recombination were significantly skewed. It is intriguing, moreover, that cancer cell lines with a higher frequency of LOH changes (without corresponding gains and deletions) had significantly fewer copy number changes. Although the cause remains to be found, these data strongly suggest that certain cancers have underlying tendencies to mitotic recombination or whole-chromosome LOH. These two forms of CIN must therefore be added to the version of CIN which is commonly proposed to result in gains and deletions in cancers. Genomic instability in cancers may therefore comprise at least three types of CIN plus two other forms, MSI and multiple translocations ( 21). CIN seems to be a heterogeneous phenomenon and we suggest that this is taken into account when the term is used.
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.
We thank Cordelia Langford and colleagues at the Sanger Institute for printing the arrays for CGH.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
M. Gaasenbeek and K. Howarth contributed equally to this work.
- Received September 13, 2005.
- Revision received January 7, 2006.
- Accepted January 24, 2006.
- ©2006 American Association for Cancer Research.