Exome profiling of primary, metastatic and recurrent ovarian carcinomas in a BRCA1-positive patient

Ovarian carcinoma (OC) is the leading cause of death from gynecological cancer in western countries. The most important predisposing factors are germline mutations in inherited cancer susceptibility genes, most notably BRCA1, BRCA2, RAD51C, RAD51D and the mismatch repair genes [1, 2]. Recently, next generation (exome) sequencing of 316 OC revealed that over 20 percent of these cancers carried either somatic or germline inactivating mutations in either BRCA1 or BRCA2, thus emphasizing the importance of these two genes in the pathogenesis of OC [3]. Notably, about a quarter of women diagnosed with OC in their fifth decade will carry a BRCA1 or BRCA2 mutation [4]. Several studies have observed that BRCA1 and BRCA2 mutation carriers tend to have a better outcome than stage-matched non-carriers, and that this better outcome is largely attributable to the combination of BRCA mutation status and DNA-damaging chemotherapeutic drugs such as cisplatinum [5]. There have also been case reports of rare cures achieved in BRCA1/2 carriers with ovarian and other cancers following other, older treatments such as melphalan [6]. Together, these findings suggest that optimal alignment of chemotherapeutic agents with both host and tumor genetic events is possible and is in fact required to achieve improved outcomes. To further understand the interaction between treatment, host genetics and tumor-specific mutations, we extracted DNA from four sources obtained from a single patient carrying a deleterious mutation in BRCA1 (blood, primary tumor, omental metastasis and relapse (recurrence) following standard post-operative therapy with carboplatin and paclitaxel). These four DNA samples were then subjected to whole exome sequencing, thus allowing us to identify tumor-specific variants and to determine potential changes in allele frequencies and emergence of new variants in the different tumor samples.

We obtained ~100 million sequencing reads that passed quality control for each sample. The mean read coverage in the blood, the primary tumor, the omental metastasis, and the recurrence was 174X, 130X, 162X and 146X per base, respectively, allowing for confident detection of mutations across the entire frequency spectrum. We searched for de novo somatic mutations by excluding all variants present in the blood from the list of variants detected in the three tumor samples (Table 1). Based on the criteria described in the Methods section, we identified 39 somatic mutations in the primary tumor and a greater number of somatic mutations in the metastasis and recurrence (47 and 52 mutations). However, we found that all of the primary tumor/metastasis/recurrence-specific mutations were identified from poor alignments or variant callings, and on visual inspection of the data, the remaining mutations were also detected in the primary tumor with small numbers of supporting reads.

We proceeded to examine the change in frequency of the BRCA1 missense mutation (chr17, 41197766, S1841R) and observed increasing allele frequencies of this mutation: 0.48 in the blood, 0.57 in the primary tumor, 0.76 in the metastasis, and 0.72 in the recurrence. Upon validation using Sanger sequencing, this mutation showed consistent increase in frequency: 0.39 in the blood, 0.50 in the primary tumor, 0.68 in the metastasis, and 0.78 in recurrence. We note that the measurements from exome appear more accurate than from Sanger sequencing, because the allele frequency from exome sequencing of the inherited BRCA1 mutation in the blood sample was closer to the expected 0.5, representing heterozygosity. Although we observed increase in frequency of this mutation from blood to tumor samples, we did not observe complete loss of the wild-type allele in the tumors. Based on previous investigations of series of BRCA1 mutation-positive patients [3] the primary, metastatic and recurrent tumors will frequently exhibit complete loss of heterozygosity (LOH), and therefore the mutant allele frequency in the tumors should be close to 1, instead of 0.57 - 0.76, suggesting that the tumor samples may contain considerable proportion of non-malignant tissue. Allowing for sampling issues, it does appear that the frozen primary tissue (equivalent paraffin section images shown in Figure 2A) contains a considerable amount of non-malignant tissue, whereas, as shown in Figure 2B, the percentage of malignant tissue in the omental biopsy is higher (fat cells take up some of the sample, top of the figure). This is even more evident in Figure 2C, where there appears to be very little non-malignant tissue present. Further corroborating these data, CNV detection results showed that the allelic frequency of all the identified large deletions/duplications is increased from primary tumor to metastatic and recurrent tumors. Concurrently, we find no evidence for de novo alleles in the primary tumor that are absent in the subsequent tumors – which would have indicated that the primary tumor contained a mixture of different malignant clones. Thus, we hypothesize that the primary tumor sample we obtained for sequencing contained a relatively larger proportion of normal tissue than the metastases. The increased mutant allele frequencies among tumor samples are likely to reflect a more pure tumor sample, rather than a selection process. Moreover, CNV detection suggested that the region (17q11-17q21) containing BRCA1 gene was deleted in all tumors, including the primary. This result is consistent with LOH, and that in this patient, the inherited mutation and the somatic deletion in BRCA1 together initiated the tumor growth.

In order to validate the exome sequencing results, and further investigate the possibility of selection of driver mutations during the evolution of the tumor, we selected 26 variants with supporting reads increased by at least 10% in the metastatic or post-therapy tumors. Sanger re-sequencing validated 24/26 mutations as being present in all three tumor samples but not in the blood sample (Table 1). We found high concordance of the allele frequency estimates from exome and Sanger sequencing (R = 0.78, p = 7.865e-15, Figure 3). The degree of concordance between the two methods renders high confidence in the selected candidate gene list. However, as mentioned above, we believe that the increase in allele frequency of most of the mutations is a result of difference in tumor purity, as opposed to a selection process.

The above observation implies that most of the detected mutations were present in the primary tumor and that very little, if any selection has occurred thereafter. The most compelling hypothesis regarding the origin of BRCA1-related high-grade serous ovarian carcinoma is that in fact the majority of them arise in the fallopian tube [15]. Our findings suggest that most of the critical tumor-driving clonal evolution occurs very early in the life of BRCA1-related highgrade serous carcinomas. One can reasonably speculate that the three tumors we studied here were all in fact “secondary” to the primary origin of the tumor and metastases from the now obscured primary tumor, likely in the fallopian tube. Surgery and chemotherapy failed to eradicate the original clone. Furthermore, when taking into account the relatively lower purity of the primary tumor, it is highly likely that most of the somatic mutations detected in this study were already present at high allelic frequency and high level of clonality in the tumor arising in the ovary. In agreement with our data, Castellarin et al. have recently suggested that in high-grad serous carcinoma patients, most somatic mutations found in recurrent tumors during platinum-based chemotherapy were present in primary tumors [16]. Our data thus suggests that little genetic evolution of the tumor has taken place from time of diagnosis to relapse following three courses of highly-active chemotherapy. It is possible that the 2.5 fold increase in allele frequency of the NF1 mutation from the primary tumor to the metastasis (Table 1) indicates that this mutation appeared in the primary tumor later than for example, TP53 mutation but was required for the full metastatic phenotype. It is likely that the primary tumor that is detected in patients is descended from cells that already contain a significant and potentially lethal mutational load.

Another notable feature of our results is the presence of important cancer-related mutations (Table 1, Figure 4) and their corresponding structural rearrangements in all three tumors. Clear examples are the above-mentioned BRCA1 mutation, the missense mutation in TP53 resulting in R110P, the mutation in NF1 damaging the donor site for splicing, and the deletion in region 17q11-17q21 which removed one copy of each of these three genes. In the recent companion study of ovarian carcinoma, TP53 mutations were present in the primary, first recurrent and second recurrent tumors in three high-grade serous carcinoma patients [16]. Frequent somatic mutations in NF1 have been previously shown to co-occur with TP53 mutations [17]. The NF1-associated RAS pathway is also activated in many ovarian cancer cases [3, 17]. Novel mutations identified in other genes (Table 1) should also be considered as candidates for intensive investigation, since they were identified from all three samples. An interesting candidate mutation is the D891N change in TARBP1 (Polyphen score 1.00) [18]. TARBP1 encodes an RNA binding protein with a methyltransferase domain. Methyltransferases have previously been shown to be involved in cancer [19]. Two somatic mutations (A1198A, W893*) in this gene have recently been found in ovarian cancer [3]. Our results suggest that in the primary tumor, BRCA1 mutations might, in combination with TP53, NF1 and TARBP1 mutations contribute to the metastasis and relapse after chemotherapy. Analyzing the interaction between the RAS, BRCA1 and TP53-mediated pathways in ovarian cancer could be therapeutically worthwhile, especially if considered in combination [20, 21].

We also show that valuable additional information regarding structural rearrangements can be derived from exome data. The CNV landscapes in our samples are associated with known ovarian cancer mutations (Tables 2, 3). Interesting examples include the amplification of 8q, which is likely driven by the MYC oncogene, and the amplification of 11q13, which is common in breast and ovarian carcinoma [22]. In addition, we observed deletion of chromosome 4, which has been shown to house several tumor suppressor genes, and deletions in chromosome 4 are associated with BRCA related tumours [23]. These mutations are likely acting combinatorially to drive the development of ovarian cancer. It is interesting to note that all of these genomic rearrangements are already present in the primary tumor, suggesting that large scale mutations accumulate quickly in early oncogenesis of ovarian cancer.

This work used whole exome capture and massively parallel DNA sequencing to study targeted candidate mutations in selected genes, as well as performing a “hypothesis-free” analysis where we aimed to identify potential driver mutations by identifying variants with increased proportion of mutant alleles. Genetic evolution of tumors from diagnosis to relapse after chemotherapy was not observed. Instead, we suggest that most of the critical tumor-driving and chemotherapy resistant mutations were already present in the primary tumor. We show that high-throughput sequencing is effective in detecting large chromosomal rearrangements such as deletions and amplifications that occur in cancer. It is notable that the patient responded very poorly to platinum-based therapy; relapse after only 3 course of therapy usually betokens a very poor survival. This early platinum failure is somewhat less common in BRCA1-related cancer than in non-hereditary ovarian cancer [5], and it seems unlikely that this failure is related to type of mutation (i.e. missense mutation) that was present in this patient. The large number of deleterious somatic mutations present in the primary tumor likely contributed to the rapid progression of the disease. It will be important to conduct studies such as ours in large numbers of patients to establish whether specific exomic profiles at initial diagnosis are associated with subsequent resistance to standard chemotherapy. In these situations, alternative forms of first-line therapy may be chosen. As many similar studies are going to be carried out in the near future, correlation of such candidate lists across patients will provide unprecedented information regarding recurrent mutations in specific genes responsible for metastasis and resistance to therapy. In addition, pathway analysis of the mutated genes will allow definition of the functional pathways involved in the above processes.