The study reveals genomics diversity in EOC. It is conceivable that some of these alterations are involved in the tumorigenesis of EOC but the pathogenesis is likely regulated by aberrations of histotype specific alterations. By stratifying based on histotypes, we were able to identify alterations in the lower prevalent clear cell, endometrioid, and mucinous samples. An example is ERBB2. Several groups have investigated alterations of ERBB2 in EOC with mixed results
[29, 30]. However, when stratified into histotypes, the high prevalence of ERBB2 amplification in mucinous is clearly evident in our results and other studies
[26–28]. In addition, our results support ERBB2 as a potential copy number driver gene in EOC. The differences of ERBB2 copy number alterations amongst the histotypes could be due to the origin of histotypes in EOC. Our study demonstrates the importance of histotype-specific analyses where the differing copy number landscape amongst the histotypes adds to the mounting evidence that EOC should not be treated as one disease.
76 cancer genes listed in Table
1 were found to be copy number altered in EOC. They include ERBB2, TPM3, BRCA1, BRAF, KRAS, and PIK3CA; some of which are potential copy number drivers e.g. PIK3CA and BRAF in serous histotypes. Another interesting observation was KRAS, a gene reported to be mutated in mucinous tumors. In our study, KRAS was not found significantly altered in mucinous tumors where mutations are common. However, KRAS was significantly amplified in serous tumors where mutations are rare in high grade tumors. The reciprocal relationship between KRAS mutation and copy number alterations is also observed in gastric cancer
. The 8q24.21 region harboring MYC, on the other hand, was altered in most histotypes other than mucinous. TCGA study also indicates MYC is highly amplified in high grade serous tumors. This suggests that MYC inhibitors may be applicable for these histotypes. For cancer genes which have been reported to have somatic or germline mutations in EOC, 3 were found to harbor copy deletions in serous histotype: PIK3R1, BRCA1, and STK11. For BRCA1, our finding was concordant with previous report that BRCA1 locus could be lost via either deletion or epigenetic silencing other than mutation in sporadic EOC
A number of the candidate drivers in Table
1 are also implicated in translocation aberrations, e.g. TPM3, BCL9, GMPS, ZNF384 and SS18L1. It’s interesting that these genes were amplified in endometrioid and/or serous tumors. TPM3 and BCL9 reside in 1q21, a frequent site for chromosomal rearrangements. TPM3 was specifically amplified in endometrioid tumors and the gene has been shown to constitute a fusion gene with NTRK1 which belongs to the group of TRK oncogenes reported for papillary thyroid carcinoma
. Interestingly, NTRK1 is also significantly amplified only in endometrioid tumors and further investigation is required to ascertain if this is due to gene fusion. BCL9 is a novel oncogene in Wnt signaling pathway, playing a critical role in epithelial-mesenchymal transition in colon epithelium and adenocarcinomas
[33, 34]. Translocation of BCL9 has been reported with 14q32
 and the gene was amplified in both endometrioid and serous tumors. Translocations for GMPS, ZNF384, and SS18L1 were also found in leukemia and synovial sarcomas
[36–40] and all were amplified in serous tumors.
There are several drugs targeting the genes, e.g. for ERBB2, inhibitors include Trastuzumab, Lapatinib, and Pertuzumab. Lately, a clinical trial on combination of Pertuzumab, Trastuzumb and Docetaxel improved outcome of patients with HER2 positive metastatic breast cancer
. BRAF mutations are more common in low grade serous while BRAF amplification is more common in high grade serous. Our data showed that it is a potential copy number driver and hence may be targetable by BRAF-inhibitors in serous tumors. Most BRAF inhibitors target various mutations and its efficacy on amplified BRAF is not yet well understood. A study has shown that BRAF amplified colorectal cancer cells acquired resistance to the MEK1/2 inhibitors selumetinib
. PIK3CA is significantly amplified in serous histotypes and could be a potential target for PI3K inhibitors. In a study of PI3K inhibitor on breast and gynecologic malignancies harboring PIK3CA mutations, patients with the mutations treated with the inhibitor showed higher response rate than patients without the mutations
In combining the 3 datasets, there was concern with regards to the genetic diversity amongst the Chinese Japanese, and Caucasian samples. The Hapmap
 and Human Genome Diversity projects
 have showed that these ethnic groups are different, though Chinese (CHB) and Japanese (JPT) tend to have high similarity in population structure. As genetic differences can be evaluated via principal component analysis (PCA)
, we used PCA to assess the copy number data of the 3 cohorts. No distinct clustering between the groups (Additional file
7: Figure S3) was observed, suggesting that in this particular copy number landscape, the genetic effect is not evident and therefore has minimal effect in the analyses. We also used ANOVA test to assess ~200 housekeeping genes between the 3 datasets; none of the genes showed significance (Additional file
8: Table S5). Note that ERBB2 also did not show any significance. Nevertheless, genetic differences were taken into consideration in the preprocessing protocol. Individual dataset was normalized with respect to the relevant ethnic group from Hapmap data, i.e. Dataset1 with JPT, and Dataset2 with CHB. Ethnic-specific common structural polymorphism was also filtered out (see Data analysis) to ensure the copy number alterations identified in this study are de novo alterations in tumors.
We recognize that the regions identified could still be limited by the individual sample size of the histotypes. The larger number of copy number altered genes in serous tumors could be attributed to the larger sample size in this collection. We performed sub-sampling analyses to ascertain the effects and in addition, to ensure robustness of results, we used stringent criteria to filter the regions as well as criteria to consider CNA genes if it were supported by at least 2 dataset (Additional file
1: Figure S1). The flip side of this filtering was that true regions of alterations could be filtered out (as shown in the green area in Figure
1), leading to probably more false negatives. Nevertheless, we observed that despite the filtering and limited sample size of some histotypes, significant regions were still observed in the less prevalent histotypes; e.g. the 1p36.33, 2p11.1, 19q13.31, and 20q13.33 amplification and 9q32 deletion in clear cell tumors (n=29); 1q21.2-3 amplification in endometrioid tumors (n=20); 17q12 amplification and p24.1 deletion in mucinous tumors (n=19). Note that endometrioid tumors were not available in Dataset 2 although the total number of tumors was comparable with clear cell and mucinous. The concordance criteria of agreement on 2 datasets in the analytical workflow would thus bias the identification of regions for this histotype. Despite this, significant alterations were still observed for endometroid (e.g. TPM3) and given the stringent criteria; these are likely high confidence alterations. It should be noted that the samples were stratified according to the 4 main histotypes, including some borderline cases. Although borderline cases are presented clinically as a different subtype, they were included to simplify the stratification of histotypes and analyses. The significance of this approach can be seen in ERBB2, where both mucinous and mucinous borderline cases harbor amplification and corresponding up regulation of expression as well. This was similarly observed in other studies
[26–28]. To assess if copy number alterations differ between borderline and non-borderline tumors and would thus cause bias in our analyses, we evaluated PCA of these samples (Additional file
9: Figure S4). No distinct clustering was observed between the borderline and non-borderline groups.