Array comparative genomic hybridization, with a genome-wide resolution of ∼1 Mb, has been used to investigate copy number changes in 48 colorectal cancer (CRC) cell lines and 37 primary CRCs. The samples were divided for analysis according to the type of genomic instability that they exhibit, microsatellite instability (MSI) or chromosomal instability (CIN). Consistent copy number changes were identified, including gain of chromosomes 20, 13, and 8q and smaller regions of amplification such as chromosome 17q11.2-q12. Loss of chromosome 18q was a recurrent finding along with deletion of discrete regions such as chromosome 4q34-q35. The overall pattern of copy number change was strikingly similar between cell lines and primary cancers with a few obvious exceptions such as loss of chromosome 6 and gain of chromosomes 15 and 12p in the former. A greater number of aberrations were detected in CIN+ than MSI+ samples as well as differences in the type and extent of change reported. For example, loss of chromosome 8p was a common event in CIN+ cell lines and cancers but was often found to be gained in MSI+ cancers. In addition, the target of amplification on chromosome 8q appeared to differ, with 8q24.21 amplified frequently in CIN+ samples but 8q24.3 amplification a common finding in MSI+ samples. A number of genes of interest are located within the frequently aberrated regions, which are likely to be of importance in the development and progression of CRC.
Colorectal cancer (CRC) is the second most common malignancy in the Western world and is responsible for ∼20,000 deaths in the United Kingdom per annum. The lifetime risk in the general population is ∼5% (1) . The majority of CRC cases are sporadic, but the autosomal, dominantly inherited syndromes familial adenomatous polyposis and hereditary nonpolyposis CRC account for up to 5% of cases. Thus far, a handful of genes have been identified in which somatic mutations contribute to the pathogenesis of CRC, including APC, SMAD4, p53, KRAS, and β-catenin (2 , 3) . The earliest aberration detected in CRC is mutation of APC (found as germ line mutations in familial adenomatous polyposis patients), which generally appears to be necessary for cancer initiation (4) . Mutant KRAS is found in ∼50% of CRCs, mutant p53 in ∼60%, and mutant SMAD4 in 10–20% (5 , 6) . CRCs can be categorized according to the type of genomic instability that they are thought to exhibit, namely chromosomal instability (CIN), characterized by an aneuploid/polyploid karyotype (6 , 7) or microsatellite instability (MSI), characterized by defective mismatch repair and a near-diploid karyotype (8 , 9) . Other CRCs are near-diploid and MSI−. Genetic pathways defined at the chromosomal level are mirrored by genetic changes at the level of the gene. For example, p53 and SMAD4 mutations are more common in CIN+ CRCs (accompanied by loss of 17p13.1 and 18q21.1, respectively), whereas inactivation of the BAX and TGFBIIR genes, by frameshift mutations in tandem repeats, occurs in MSI+ lesions (10) .
Genomic copy number changes are found frequently in cancers and are believed to contribute to their development and progression through inactivation of tumor suppressor genes, amplification of oncogenes, or more subtle gene dosage changes. Comparative genomic hybridization (CGH; Ref. 11 ) was developed to allow genome-wide screening for such copy number changes, and CGH investigations into CRC have revealed consistent gains and losses (12, 13, 14, 15, 16) . In particular, gain of chromosome 20q is a widespread finding in primary CRCs (67%) as is loss of 18q (49%; Ref. 16 ). Other consistent regions of copy number gain are 7p, 8q, 13q, and 12p along with deletions of 8p and 4p. Overall, a smaller number of aberrations have been detected in diploid compared with aneuploid cancers (16) . CGH has also been used to investigate any differences in chromosomal gains and losses between MSI+ and CIN+ cancers (17 , 18) . It was found that chromosomal imbalances were more frequent in CIN+ than MSI+ cancers, and cell lines and differences in the particular chromosomes involved were also detected. Gains of 4q (15%) and 8q (8%) and losses of 9q (21%), 1p (18%), and 11q (18%) were the most common findings in MSI+ cancers. However, gain of 8q (50%), 13q (35%), and 20q (25%) and loss of 18q (55%), 15q (35%), and 17p (30%) were detected most frequently in CIN+ cancers (17) . In addition, different types of aberration were detected depending on the particular mismatch repair defect identified (18) .
Conventional CGH has a limited resolution and can only detect losses of ∼10 Mb or greater (19 , 20) . High-level amplifications have a maximum resolution of 3 Mb (21) . The resolution of CGH has been improved by replacing the metaphase chromosomes as the hybridization target with mapped and sequenced clones (bacterial artificial chromosomes, P1-derived artificial chromosome, and cosmids) arrayed onto glass slides (22 , 23) . The resolution of this matrix (22) or array (23) CGH is limited only by the insert size and density of the mapped sequences used. The capability of the technique to provide high-resolution mapping of variation in copy number has been demonstrated in the analysis of breast tumors, where the fine structure of amplicons was resolved and potential candidate genes identified (24) . In this study, we describe the use of array CGH with a resolution of ∼1 Mb (25) to investigate copy number changes occurring in CRC by the analysis of 48 cell lines and 37 primary cancers.
MATERIALS AND METHODS
Cell Lines and Primary Cancers.
DNA was extracted from 48 CRC cell lines derived from local collaborators or public sources. In almost all cases the MSI status and ploidy were already known, and several authors have reported molecular changes, such as mutation of APC, KRAS, and p53, in some of the cell lines (10 , 26 , 27) . Fresh frozen samples of primary sporadic CRCs, derived from patients at St. Mark’s Hospital, were obtained. All of the cancers contained >70% neoplastic tissue on histological review. MSI analysis (28) was undertaken, and ploidy was assessed using flow cytometry (data not shown). DNA was extracted from 7 MSI+ and 30 CIN+ cancers using standard methods; no MSI−CIN− (or MSI+CIN+) cancers were included in the study because these are very rarely represented in the set of known CRC cell lines. Cell line and cancer details can be downloaded from Supplementary Table 1 ⇓ (Excel spreadsheet). Cell line DNAs were hybridized against a normal female lymphoblastoid cell line C0009-SAH (HRC160; European Collection of Cell Cultures no. 93010702) obtained from the European Collection of Cell Cultures (Salisbury, United Kingdom). Primary cancers were hybridized against a sex-matched control DNA pool created from 20 normal blood DNAs obtained from Human Random Collection of the European Collection of Cell Cultures. Clones that had been shown previously to report copy number changes in hybridizations of individual normal DNAs against the pool of normal DNAs (data not shown) were considered to be polymorphic in the normal population and were excluded from analysis.
Test and control DNAs were differentially labeled using a Bioprime labeling kit (Invitrogen, Carlsbad, CA), as described previously (25) , with minor modification. Briefly, a 130.5-μl reaction was set up containing 450 ng of DNA and 60 μl of 2.5× random primer solution. The DNA was denatured for 10 min at 100°C, and 1.5 μl of 1 mM Cy5-dCTP or Cy3-dCTP (NEN Life Science Products, Boston, MA) and 3 μl of Klenow fragment were added on ice to a final reaction volume of 150 μl. The reaction was incubated overnight at 37°C and stopped by adding 15 μl of stop buffer. Unincorporated nucleotides were removed using G50 spin columns (Pharmacia Biotech, Piscataway, NJ) according to the manufacturer’s instructions.
Microarray Hybridization and Image Analysis.
Hybridization to the array was carried out as described previously (25) . Briefly, test and control-labeled DNAs were combined, precipitated together with 135 μl of human Cot1 DNA (Roche, Mannheim, Germany), and resuspended in 60 μl of hybridization buffer [50% formamide, 10% dextran sulfate, 0.1% Tween 20, 2× SSC, 10 mm Tris-HCl (ph7.4)], and 6 μl of yeast tRNA (100 μg/μl; Invitrogen, Carlsberg, CA). This hybridization solution was incubated for 1 h at 37°C before application to the array. A prehybridization solution of 80 μl of herring sperm DNA (10 mg/ml; Sigma-Aldrich Company Ltd., Dorset, United Kingdom) and 135 μl of human Cot1 DNA (Roche), precipitated together and resuspended in 160 μl of hybridization buffer, was applied to the array and incubated at 37°C for 1 h. The arrays were scanned on an Axon 4000B scanner (Axon Instruments, Burlingame, CA), and images were quantified using the software program “Spot” (29) . The fluorescence intensity ratio for each clone (after local background subtraction) was calculated and normalized by dividing each ratio by the median ratio of all of the autosomal clones and plotted against the mapped genomic position (NCBI31; freeze date, November 2002).
The determination of significant copy number changes detected in array CGH for cancer and cancer cell lines, where much of the genome has been altered, is not straightforward. Measurement variation will vary from hybridization to hybridization, so it is important to set statistically defined thresholds for copy number change by analysis of the variation for each hybridization. We have approached this by identifying regions of modal copy number (normalized linear ratio of 1.0) in each hybridization, using an arbitrary threshold determined from independent normal versus normal hybridizations and then using the variation of the defined modal region to set thresholds specific to that hybridization. Specifically, the largest contiguous chromosomal region reporting a modal copy number was identified. A region was considered to be modal if 95% of the clones fell within 99% confidence intervals calculated from the linear ratios of autosomal clones in a normal male versus normal female hybridization (data not shown). The size of the region used varied from 100 Mb to 246 Mb. The SD of measurements in these modal regions ranged from 0.03 to 0.14. The linear ratios reported by the clones within the defined modal region were used to calculate the coefficient of variation and 99% confidence intervals for each hybridization. Clones that fell outside the 99% confidence intervals were identified as reporting significant copy number changes. We have shown previously that our arrays demonstrate close to theoretical copy number changes in response to autosomal single copy gains and losses (25) . For the cell lines, gains were considered to be single copy or greater if the reported ratio was greater or equal to the theoretical value for a single copy gain (1.5) minus the 99% confidence interval at this value for each hybridization and double copy number or greater if the reported ratio was greater or equal to the theoretical value for a double copy amplification (2.0) minus the 99% confidence interval at this value for each hybridization. Deletions were considered to be single copy or greater if the reported ratio was greater or equal to the theoretical value for a single copy deletion (0.5) plus the 99% confidence interval at this value for each hybridization. In no hybridization did these defined thresholds overlap the 99% confidence intervals for modal clones. The theoretical values for single and double copy number changes were calculated relative to a diploid karyotype; therefore, classification of copy number changes, which occurred in samples with an increased ploidy, would be underestimates of the true level of copy number gain or loss. For the primary cancers, the observed changes were not classified as single or double copy number changes because the cancer samples are inevitably contaminated with DNA from surrounding normal tissue. Contaminating normal tissue results in suppression of the reported ratios in regions of amplification or deletion. The exact amount of contamination for each sample could not be readily determined; hence, precise thresholds could not be set. Copy number change of the sex chromosomes was not analyzed because the cell lines were all hybridized against female control DNA. To identify trends in copy number gain, the “mean percentage of gain by clones” (MPG) for each particular chromosome or region of interest was used. The MPG was determined by calculating the percentage of samples with a significant copy number gain for each clone and calculating the mean percentage gain for all of the clones in the chromosome or region of interest. An equivalent calculation, “mean percentage of loss by clones” (MPL) was applied for copy number losses. The MPL was determined by calculating the percentage of samples with a significant copy number loss for each clone and calculating the mean percentage loss for all of the clones in the chromosome or region of interest. The genomic locations of all of the clones named specifically in the text were confirmed by bacterial artificial chromosome end sequencing and fluorescence in situ hybridization. Raw, normalized log2 ratios for all of the cell lines and cancers can be downloaded from Supplementary Table 1 ⇓ (Excel spreadsheet).
Metaphase Fluorescence in Situ Hybridization.
Degenerate oligonucleotide primed amplified clone DNA (amplified using primers degenerate oligonucleotide primed 1, 2, and 3) created for array construction (25) was labeled with biotin-16-dUTP (Roche) or digoxigenin-11-dUTP (Roche) by nick translation. Hybridizations were carried out using conventional methods to chromosomes prepared from a normal male lymphoblastoid cell line (HRC575; European Collection of Cell Cultures no. 94060845). Biotin-labeled probes were detected using Avidin TexasRed (Molecular Probes Inc., Eugene, OR), and digoxigenin-labeled probes were detected with a combination of mouse antidigoxigenin (Vector Laboratories Ltd., Peterborough, United Kingdom) and goat antimouse-FITC (Sigma-Aldrich Company Ltd.) antibodies.
The most common finding in both cell lines and cancers was gain of chromosome 20. The MPG was 49% for cell lines and 50% for cancers (for definition, see “Materials and Methods”). The region of chromosome 20 gained most frequently was 20q13.3 (RP4-563E14 at 62.3Mb) in 65% of cell lines and 81% of cancers. Gain of chromosome 13 was also a common finding in both cell lines (MPG = 43%) and cancers (MPG = 46%), along with gain of chromosome 7 (MPG = 22% of cell lines; 46% of cancers) and the long arm of chromosome 8 (MPG = 31% of cell lines; 27% of cancers). The most common region of copy number loss was the long arm of chromosome 18 (MPL = 52% of cell lines; 32% of cancers). In particular, three clones detected the most frequent copy number loss: two at 18q21.1 (RP11-313C14 at 43.9Mb in 60% of cell lines and cancers and RP11-25O3 at 49.7Mb in 60% of cell lines and 49% of cancers) and one at 18q12.2 (RP11-19F9 at 35Mb in 56% of cell lines and 59% of cancers). In addition, chromosome 8p was also a consistent region of copy number loss (MPL = 33% of cell lines; 29% of cancers).
The overall pattern of copy number change was very similar in the cell lines and cancers (Fig. 1) ⇓ . However, loss of chromosome 6 was a significantly more common event (P < 0.005, χ2 test) in the cell lines (MPL = 21%) than the cancers (MPL = 1%). In addition, gain of chromosome 15 occurred more frequently in the cell lines (MPG = 21%) than the cancers (MPG = 1%; P < 0.005, χ2 test). Furthermore, gain of chromosome 12, in particular, the short arm, was a common finding in the cell lines (MPG = 32%) but not the cancers (MPG = 3%; P < 0.005, χ2 test). Although chromosome 12p was not generally gained in the cancers, one particular clone, RP11-59H1 at 12.9 Mb, was gained in 22% of cancers and 31% of cell lines. The most frequently gained clone in the cell lines (38%) was RP11-4N23 at 13.6 Mb. The total number of clones reporting significant gains and losses was calculated, and the median number per sample was compared between the cell lines and the cancers. The cell lines showed a median number of aberrant clones of 975 (range = 598-2556), and the cancers showed a median number of aberrant clones of 536 (range = 48–1152). This difference is statistically significant (P < 0.005, Mann-Whitney U test).
The frequency with which each clone on the array reported a high-level amplification (double copy number or greater) in the cell lines was calculated and is displayed in Fig. 1 ⇓ as yellow bars. Clones reporting amplifications in ≥2 cell lines are listed in Table 1 ⇓ . The most common region of amplification of chromosome 20 was again at 62.3 Mb (reported by clone RP4-563E14; amplified in 4 cell lines). The most common region of amplification of chromosome 13 was 13q12.13-q12.2 (RP11-44J9 at 26.2 Mb; amplified in 9 cell lines), and the most frequently amplified region of chromosome 8 was 8q24.21 (126-128 Mb; reported in 7 cell lines). The most frequently amplified region of chromosome 17 was 35.8 Mb (RP5-986F12), and the most frequent amplification of 7p was 29.3 Mb (RP11-550A18). The frequency with which each clone reported a deletion (single copy or greater) in the cell lines is displayed as black bars in Fig. 1 ⇓ . Clones reporting deletions in ≥2 cell lines are listed in Table 1 ⇓ . Consistent regions of deletion included chromosome 4q34.3 (RP11-226A18; deleted in 4 cell lines) and, again, chromosome 18q21.1.
Small regions of consistent copy number change were also observed (Tables 2 ⇓ and 3 ⇓ ). For example, chromosome 17q11.2-q12 (33–38 Mb) was a consistent region of gain in both cell lines (MPG = 21%) and cancers (MPG = 22%). RP5-986F12 at 35.8 Mb was the most frequently amplified clone (double copy number or greater) occurring in 6% of cell lines. Gain of this clone occurred as part of both broad and narrow regions of copy number change (Figs. 2, A and B) ⇓ . In addition, chromosome 4q34-q35 was a common region of copy number loss in cell lines (MPL = 35%) and cancers (MPL = 38%) as shown in Fig. 2, C and D ⇓ . The most frequently deleted clone defining the peak of the copy number change in the cell lines (8%) was RP11-226A18 at 182.4 Mb. Examples of 4q34-q35 loss are shown in Fig. 2, E and F ⇓ .
The genome-wide pattern of copy number changes in the MSI+ and CIN+ cell lines and cancers was compared (Supplementary Fig. 1). CIN+ samples reported significantly more aberrations than MSI+ samples (cell lines P < 0.005; cancers P < 0.05, Mann-Whitney U test). Gain of chromosome 20 was the most common overall finding, but when the analysis was split by the microsatellite status of the samples, a marked difference emerged. Gain of chromosome 20 occurred more often in CIN+ cell lines (MPG = 51%) and CIN+ cancers (MPG = 59%) than in MSI+ cell lines (MPG = 6%) and MSI+ cancers (MPG = 12%; P < 0.005 for both cell lines and cancers, χ2 test). However, clone RP11-563E14 was gained in 14% of MSI+ cell lines and 43% of MSI+ cancers and gained (single copy or greater) in 7% of MSI+ cell lines. Loss of 18q was also a more frequent event in CIN+ cell lines (MPL = 71%) and CIN+ cancers (MPL = 42%) than MSI+ cell lines (MPL = 12%) and MSI+ cancers (MPL = 1%). However, loss of chromosome 18q21.1-21.2 (44–47 Mb, including clones RP11-313C14 and RP11-46D1) was detected in MSI+ cell lines (MPL = 21%) and MSI+ cancers (MPL = 29%). Loss of the short arm of chromosome 17 occurred more frequently in CIN+ cell lines (MPL = 38%) and cancers (MPL = 29%) than MSI+ cell lines (MPL = 3%) and cancers (MPL = 1%). In addition, loss of chromosome 8p was a frequent event in CIN+ cell lines (MPL = 47%) and cancers (MPL = 37%) but occurred much less frequently in MSI+ cell lines (MPL = 3.4%). Loss of 8p was never found in MSI+ cancers but was frequently gained (MPG = 28%; Fig. 3 ⇓ ). The overall peak of copy number gain on chromosome 8q was band 8q24.21 from 126–128Mb. However, in MSI+ cancers a different region 8q24.3 (144–145Mb) was the most frequently gained (RP11-472K18 at 144.3Mb in 36% of MSI+ cell lines and 57% of MSI+ cancers and RP11-349C2 at 145.3 Mb in 36% of MSI+ cell lines and 71% of MSI+ cancers; see Fig. 3 ⇓ ).
The frequency with which the clones reported amplifications (double copy number or greater) and deletions (single copy or greater) in the CIN+ and MSI+ cell lines is displayed in Supplementary Fig. 1 as yellow and black bars. Clones reporting amplifications in ≥2 CIN+ cell lines and 1 MSI+ cell line and clones reporting deletions (single copy or greater) in ≥2 cell lines (CIN+ and MSI+) are listed in Table 4 ⇓ . There were no clones that were amplified in >1 MSI+ cell line. Only one clone reported a deletion in >1 MSI+ cell line. This clone, RP11-241C9, is mapped to chromosome 1q41 at 214Mb.
Copy number losses reported by clones representing three genes, APC, p53, and SMAD4, were compared with previously published loss of heterozygosity (LOH) data in the cell lines (10 , 27 , 30) . In the case of APC, it was found that LOH occurs most often in cell lines with a normal copy number at this locus (17 of 22 samples), probably by mitotic recombination, as suggested previously (30) . Conversely, LOH at SMAD4 is generally associated with copy number loss (25 of 26 samples), and in cases where LOH was not detected, copy number is normal (11 of 12 cases). LOH at p53 is not associated with copy number change, suggesting a mechanism of mitotic recombination or gene conversion.
A genomic microarray with a resolution of ∼1 Mb was used to investigate chromosomal imbalances in a set of 48 CRC cell lines and 37 primary CRCs. The collection included both MSI+ and CIN+ samples. Analysis of the array hybridizations revealed consistent regions of copy number change. Many of the findings were in agreement with those observed previously in conventional CGH investigations (12, 13, 14, 15, 16) , for example, gain of chromosomes 20, 13, and 8q and loss of chromosomes 18q and 8p. However, the increased resolution of array CGH allowed the identification of the most frequently altered regions within these large scale gains and losses and also the detection of previously unreported changes, for example, amplification of 17q11.2-q12 and deletion of chromosome 1q41. The most common region of gain was chromosome 20, especially 20q13.3 (RP4-563E14 at 62.3Mb). Candidate genes that map to this region include the following: (a) LIVIN, an inhibitor of apoptosis that has been associated with the progression of bladder cancer (31) and detected at high levels in a CRC cell line (32) ; (b) PTK6, a protein kinase, which when overexpressed in mammary cells leads to sensitization to epidermal growth factor and partial transformation (33) and also shows increased expression in primary colon tumors (34) ; (c) HD54, which is involved in calcium-mediated signal transduction and cellular proliferation (35) ; and (d) EEF1A2, a transcription elongation factor found to be amplified and overexpressed in ovarian cancer (36) . Gain of chromosome 13 was also a common finding, and the most frequently amplified region was 13q12.13-q12.2 from 26–28 Mb. Genes of interest that map to this region include FLT3, a tyrosine kinase receptor in which activating mutations have been found in acute myeloid leukemia (37) and FLT1, a vascular endothelial growth factor receptor found to be expressed in gastric and breast carcinoma cells (38 , 39) . The most frequent region of copy number loss was the long arm of chromosome 18, particularly 18q21.1, including the known tumor suppressor genes SMAD2 (40) and SMAD4 (41) , which function in the transforming growth factor signaling pathway to mediate growth inhibition. Allelic loss at 18q21.1 has been detected in up to 60% of CRCs (42) , and mutation of SMAD4 has been identified in 50–60% of CRCs with loss of 18q21 (10) .
In addition to large-scale rearrangements, smaller regions of copy number change were also detected. For example, amplification of 17q11.2-q12 (33–38 Mb) was a common finding, and the peak of this amplification was reported by clone RP5-986F12 at 35.8 Mb. Genes of interest that are located in this region include AATF, which functions as an antagonist to apoptosis (43) , and TBC1D3, a Rab GTPase amplified in 15% of prostate cancers (44) . Consistent regions of deletion (single copy or greater) were also identified, for example, RP11-226A18 at 4q34–35 (182.4 Mb). The proapoptotic gene caspase-3, which is down-regulated in gastric cancer (45) , maps to this region. In addition, RP11-241C9 at 1q41 (213.9 Mb) was the only clone reporting a deletion in >1 MSI+ cell line. Transforming growth factor β2 is located in this deleted region.
Within the limitations of our analysis, imposed by both the unknown ploidy of the cancer samples and the variable degree of contamination by normal cells, the overall pattern of copy number change in the cell lines and cancers was markedly similar; however, some differences were detected. The number of aberrations (all of the gains and losses) was significantly greater in the cell lines than the cancers, and loss of chromosome 6 and gain of chromosome 15 were far more common events in the cell lines. It has been observed previously that loss of chromosome 6, especially 6q, occurs more frequently in cell lines than primary tumors (26) . The additional changes seen in the cell lines could have arisen if they were advantageous to cells growing in cell culture conditions and may not always be relevant to cancer formation. Gain of the short arm of chromosome 12 was another more frequent occurrence in the cell lines than the cancers. However, one clone, RP11-59H1 at 12.9 Mb, was gained in the cancers, perhaps suggesting that this region of 12p, in particular, could be the target of the general increase in copy number seen in the cell lines. Genes of interest that map to this region include MGP, a matrix GLA protein that has been shown previously to be overexpressed in breast cancer (46) .
The analysis was split by the type of genomic instability that the samples exhibited, chromosomal (CIN+) or microsatellite (MSI+) instability. Overall, MSI+ samples had significantly fewer aberrations. This finding follows expectation; MSI+ cancers tend to maintain a diploid karyotype, whereas aneuploidy is a common feature of CIN+ cancers (8 , 9) and agrees with findings reported previously (17 , 18) . The copy number changes that were detected in the MSI+ cancers are more likely to be causative in the initiation and progression of the cancer rather than a consequence of a generally unstable karyotype and likely to contain target genes of the rearrangement. For example, gain of the whole of chromosome 20 was a common finding in CIN+ cancers but a much less frequent event in the MSI+ samples; yet, gain of a particular clone, RP11-563E14 at 62.3 Mb, was frequently detected in the MSI+ cancers and cell lines, again suggesting that this region could be an important target of the gain of chromosome 20. Dividing the analysis by microsatellite stability status has also been informative in focusing the attention on a particular region of chromosome 18q. Although loss of the entire long arm of chromosome 18 is not a common event in MSI+ samples, loss of 18q21.1–21.2 (44–47 Mb) is a frequent occurrence, suggesting that this area is likely to contain candidate genes that may be the target of the aberration. The pattern of copy number changes reported by the chromosome 8 clones differs distinctly between the MSI+ and CIN+ samples. Loss of 8p is a common event in CIN+ cell lines and cancers; however, this is not the case in the MSI+ samples. Not only is loss of 8p not frequently detected, but in the MSI+ cancers, 8p is actually a common region of copy number gain, suggesting that loss of 8p is an important event, specifically in the development of CIN+ CRCs. Furthermore, gain of the long arm of chromosome 8 was a common finding in all of the samples. However, although gain of 8q24.21 was the most prominent finding in the CIN+ samples, a different region, 8q24.3, stands out in the MSI+ samples, and RP11-472K18 and RP11-349C2 are the most frequently gained clones. Candidate genes that map to this region include RIG-E, a receptor that shares homology to epidermal growth factor receptors and is highly expressed in leukemic cells (47) , and GLI4, a member of the Kruppel family of transcription factors found to be overexpressed in brain tumors (48) .
A number of genes located within the reported regions of copy number change are associated with transforming growth factor signal transduction, suggesting that the abrogation of this pathway may be important in the development of CRC. For example, the only clone to report a deletion in >1 MSI+ cell line was RP11-241C19 at 1q41; transforming growth factor β2, which has been shown to induce growth inhibition and apoptosis (49 , 50) , maps to this region. In addition, caspase-3, which maps to the 4q34-q45 frequent region of deletion, has been implicated in transforming growth factor β-mediated apoptosis (51) , and SMAD2 and SMAD4, key mediators of transforming growth factor β signaling (52) , are located within the most common region of loss of chromosome 18. Furthermore, BAD, a member of the BCL-2 family of proapoptotic genes, maps to 11q13.1, a region of frequent copy number loss. This protein is inhibited by the 14-3-3ε protein but promotes cell death on its release, which occurs via the cleavage of 14-3-3ε by caspase-3 (53) . The tumor suppressor CDKN2A (P16-INK4A), an activator of caspase-3-mediated apoptosis (54) , is located within the region of deletion on chromosome 9p21-p22. Additional genes that act in this pathway are located within common regions of gain. For example, FLT1, a receptor for the vascular endothelial growth factor that promotes angiogenesis, cancer invasion, and metastasis in breast cancer, is located within the most frequently amplified region of chromosome 13. Stimulation of this receptor in epithelial cells results in decreased caspase-3 activity and prevention of apoptosis (55) . In addition, RAF-1, which functions downstream of the Ras family of GTP-ases in the promotion of cell cycle progression (56) , is located within the 3p25.2 region of copy number gain. It has been shown that RAF-1 is activated by vascular endothelial growth factor (57) and that RAF-1 signaling can lead to the inhibition of apoptosis via the phosphorylation of BAD (58) .
Overall, the analysis of CRC cell lines and cancers by array CGH has revealed consistent regions of copy number change ranging from gain of whole chromosomes to the loss of a single clone. The genome-wide pattern of copy number change was strikingly similar between cell lines and cancers, although a few obvious differences, loss of chromosome 6 and gain of chromosomes 15 and 12p, were reported. CIN+ samples had a significantly greater number of aberrations than MSI+ samples, particularly gain of chromosome 20 and loss of chromosomes 18q and 8p. In addition, the target of chromosome 8q gain appeared to differ depending on microsatellite status with 8q24.21 frequently gained in CIN+ samples and 8q24.3 gain associated with MSI+ samples. A number of genes of interest are located within frequently aberrated regions, which, on additional investigation, may prove to be of importance in the pathogenesis of CRC.
We thank the Wellcome Trust Sanger Institute Mapping Core Group, Carol Scott, and Elizabeth Huckle for clone selection and verification, the Wellcome Trust Sanger Institute Micorarray Facility for printing the arrays, Judy Fantes for fluorescence in situ hybridization analysis of clones RP11-59H1 and RP11-4N23, Sarah Edkins for the cell line Colo320 HSR, and Ying Liu for some karyotype data.
Grant support: This work was supported by the Wellcome Trust and Cancer Research United Kingdom.
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.
Note: Supplementary data for this article online at http://www.sanger.ac.uk/Teams/Team70/supplemental-data/.
Requests for reprints: Nigel P. Carter, The Wellcome Trust Sanger Institute, Wellcome Trust Genome Campus, Hinxton, Cambridge CB10 1SA, United Kingdom. Phone: 44-0-1223-494860; Fax: 44-0-1223-494919; E-mail:
- Received February 2, 2004.
- Revision received April 22, 2004.
- Accepted May 14, 2004.
- ©2004 American Association for Cancer Research.