Integrated analysis of homozygous deletions, focal amplifications, and sequence alterations in breast and colorectal cancers

Results

Optimization of Copy Number Analysis with DK.

DK was used as a standard to develop criteria for assessing amplifications and HDs with Illumina high-density SNP arrays [supporting information (SI)Fig. S1 and Table S1]. Analysis of DK libraries from 18 colorectal tumor samples identified a total of 21 amplification events, and four regions within the autosomal chromosomes where the tag density reached zero, representing HDs (Table S2). As expected, we identified low-amplitude gains and losses of chromosome arms or other large genomic regions. We did not pursue these low-amplitude copy number changes as it is difficult to reliably identify candidate cancer genes from such large regions. To ensure that the copy number changes identified by DK were bona fide amplifications or HDs, we independently examined 12 alterations by quantitative PCR and confirmed the presence of the genomic alterations in every case examined.

We then directly compared DK data to those obtained through genomic hybridization of the same DNA samples to Illumina high-density oligonucleotide array (25, 26). Using fluorescence intensity measurements, we developed an approach to detect HDs and amplifications resulting in ≥12 copies per nucleus (≥6-fold amplification compared with the diploid genome) (see SI Methods).

Using this approach, all 14 amplification events and 3 HD events identified by DK in three representative tumor samples were detected by Illumina arrays (Table S1 and Fig. S2). In all cases, the genomic boundaries identified by both approaches were similar and within the resolution expected for both methods. No additional copy number changes of the sizes expected to be detected by DK were identified by the array approach in these samples. We did identify 25 additional small HDs (<250 kb in length) that would not have been possible to detect with DK given the number of tags analyzed (24). To independently validate such smaller HDs, we used PCR and Sanger sequencing to examine genes located within small HDs and found that, in each case, multiple exons of each gene could not be amplified or sequenced. These results suggested that our approach for analysis of Illumina array data provided a sensitive and specific method for identification of amplifications and HDs, including relatively small alterations of either type.

Detection of Amplifications and HDs.

A total of 45 breast and 36 colorectal tumors were analyzed by Illumina arrays containing either ≈317,000 or ≈550,000 SNPs (Fig. S1). To determine the fraction of alterations that were likely to be somatic (i.e., tumor-derived), we analyzed these regions in 23 matched normal samples. In the normal samples, no amplifications and only four distinct HDs were detected. We removed these alterations from further analysis, as well as those corresponding to known copy number variation in normal human cells (27, 28). Finally, we removed any copy number changes where the boundaries were identical in two or more samples, because these were likely to represent germ-line variants. Based on this conservative strategy, we estimated that >95% of the 614 amplifications and 463 HDs (Table S3) represented true somatic alterations.

Breast cancers contributed to a majority of the alterations identified, comprising 68% and 80% of the total HDs and amplifications, respectively. Individual colorectal and breast tumors had on average 7 and 18 copy number alterations, respectively. Each colorectal cancer had an average of four HDs and three amplifications. Breast cancers had on average 7 HDs and 11 amplifications. Several of the tumor samples contained copy number alterations that were separated by short nonamplified or deleted sequences, presumably reflecting the complex structure of these alterations (29, 30). The copy number alterations observed encompassed on average 1.7 and 2.4 Mb of colorectal and breast tumor haploid genomic sequence, respectively. The average numbers of protein-coding genes that were affected by amplification or HD were 24 and 9 per breast and colorectal cancer, respectively.

Genes Altered in More than One Tumor.

One of the main challenges in the analysis of somatic alterations in cancers involves the distinction between those changes that are selected for during tumorigenesis (driver alterations) from those that provide no selective advantage (passenger alterations). Even in regions that have multiple copy number alterations, this distinction can be particularly difficult because regions of amplification or HD can contain multiple genes, only a subset of which are presumably the underlying targets. We reasoned that the integration of copy number analyses with sequence data would help reveal driver genes that were more likely to contain genetic alterations. To accomplish this integration, we developed a statistical approach for determining whether the observed genetic alterations of any type in any gene were likely to reflect an underlying mutation frequency that was significantly higher than the passenger rate. To analyze the probability that a given gene would be involved in a copy number alteration, we made the conservative assumption that the frequency of all amplifications and HDs observed in each tumor type represented the passenger mutation frequency (i.e., we assumed that all copy number changes were passengers). The number of actual copy number alterations affecting each gene in all tumors was then compared with the simulated number of expected passenger alterations taking into account gene size, the distribution of SNP locations, and the frequency of passenger amplifications and HDs in breast and colorectal cancers.

We integrated these copy number analyses with the sequence data of the Sjöblom et al. and Wood et al. studies (5, 31). In these studies, the protein coding sequences of 20,857 transcripts from the 18,191 genes in the RefSeq database were determined in breast and colorectal cancer samples, allowing detection of somatic sequence alterations. In the current study, the same 22 breast and colorectal tumor samples were analyzed in parallel by Illumina arrays, together with additional samples of each tumor type (Fig. S1). To integrate these different mutational data for each tumor type, we combined the probability that a gene was a driver gene based on the type and frequency of point mutations observed with the probability that the gene was a driver based on the number of observed amplifications and HDs (Fig. S1).

Table 1 lists the loci that were amplified in at least one tumor and had the highest probability of containing driver genes as determined by the combined mutation analysis (a complete list of amplifications is provided in Table S3 and amplified genes in Table S4). For genes to be considered potential targets of the amplification, the entire coding region of the gene was required to be contained within a focal amplicon. A few candidate genes in this list [e.g., CCNE1 (cyclin E) and ERBB2] were amplified in multiple tumors but were not found to be mutated by sequencing. The majority of candidate genes, however, harbored point mutations in some tumors and amplifications in others. The most striking aspect of this list of candidate genes is that only some of them had been implicated in cancer in the past: of the 19 genes indicated in Table 1, only 8 had been implicated in tumorigenesis. The known cancer genes included MYC, ERBB2 (HER2/NEU), CCNE1, CCND1, EGFR, FGFR2, and IRS2, each of which had been shown to be amplified. In addition, MRE11, which was amplified in breast cancers, has been shown to be mutated in a subset of colorectal cancers and is thought to play an essential role in maintaining chromosomal stability (32). Some genes were shown to be altered in both breast and colorectal cancers, with at least one of the tumors containing amplifications. Interestingly, among these genes, ERBB2 was found to be amplified in both breast and colorectal cancers, and FGFR2 was found to be mutated in breast cancers and amplified in colorectal cancers.

Table 2 similarly lists the loci that were homozygously deleted in at least one tumor and had the highest probability of containing drivers as determined by the combined mutation analysis (a complete list of HDs is provided in Table S3 and homozygously deleted genes are listed in Table S5). For each of these genes, at least a portion of the coding region was affected by the HD. A number of genes known to be inactivated in colorectal or breast tumorigenesis, such as CDKN2A, PTEN, and TP53 are found in this list. We also identified genes such as CHD5, MAP2K4, SMAD2, and SMAD3 that have been shown to be deleted in other tumor types but not in colorectal or breast cancers. Finally, we discovered a number of genes not known to be affected by HD in any tumor type. For example, HDs and point mutations were found in OMA1 and ZNF521 in colorectal cancers and in MANEA, PCDH8, SATL1, and ZNF674 in breast cancers. During the course of this study, we identified through independent experimentation that PCDH8 is mutated and homozygously deleted in breast cancer (33).

Pathways Enriched for Copy Number and Point Alterations.

We examined whether groups of genes belonging to certain cellular pathways were preferentially affected by genetic alterations. For this purpose, we developed a statistical approach that provided a probability that a pathway contained driver alterations, taking into account both the copy number changes and point mutations. Because the net effect of a pathway can be the same whether certain components are amplified or others deleted, all copy number alterations within a gene group were considered together. The analysis was performed by using well annotated GeneGo MetaCore databases (34). For each gene group, we considered whether the component genes were more likely to be affected by point mutations, amplifications, or HDs, as compared with all genes analyzed using a modified version of gene set enrichment analysis (GSEA) (35) rather than the total number of mutations within individual groups. This approach limits the effects of single highly mutated genes and requires the involvement of multiple genes to score a pathway as significantly affected.

These analyses identified gene groups that were enriched for genetic alterations in these tumor types (Table 3). In particular, the EGFR and ERBB gene pathways were enriched for alterations in both tumor types, involving various components of the PI3 kinase pathway (Fig. 1). One-third of genes in these combined pathways were mutated by sequence alterations, amplifications, or HDs. Enrichment of alterations in other canonical gene groups including Notch and G₁-S cell cycle transition pathways were also detected. The latter group included HDs of CDKN2A and CDKN2B genes and amplifications of cyclin D1, cyclin D3, and cyclin E3 genes in breast cancers. For all these gene groups, new genes were identified that had not been implicated by genetic alterations in these cellular processes. Finally, a variety of gene groups not known to be enriched for copy number changes in tumorigenesis were identified. These included genes implicated in cell–cell interaction and adhesion, including cadherins and metalloproteases. As an example, in colorectal cancers, a total of 33 cadherin and protocadherin genes were detected as being affected by copy number or sequence changes. In breast cancers, there was also enrichment in genes implicated in DNA topological control, including alterations in a number of topoisomerases (TOP1, TOP2A, TOP2B, and TOP3A) and helicases. Pathways showing significant enrichment for genetic alterations are listed in Table S7.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

Alterations in the combined FGF, EGFR, ERBB2, and PI3K pathways. Genes affected by copy number alterations are circled in red, whereas those altered by point mutations are circled in blue. The number of breast (B) and colorectal (C) tumors containing alterations are indicated in boxes adjacent to each gene.