Main

In recent years, investigators have paid increasing attention to the sub-classification of cancer. In a detailed analysis in The Cancer Genome Atlas (TCGA), the molecular portrait of basal-like breast cancers was observed to have similar characteristics to that of high-grade serous ovarian tumours (Cancer Genome Atlas Network, 2012). Both subtypes exhibited a very high frequency of TP53 mutations, evidence of mutations in NF1 and RB1 at lower frequencies, BRCA1 and BRCA2 inactivation, and similar copy number gains and losses. Bioinformatic pattern recognition methods have also suggested that these tumour types are characterised by mutually exclusive loss of function in the same pathways (Cancer Genome Atlas Network, 2012; Ciriello et al, 2012). The authors of the TCGA report of breast cancer suggested that these results ‘indicated a related aetiology’ of these tumour subtypes (Cancer Genome Atlas Network, 2012). Our purpose was to test this hypothesis by assessing whether triple-negative breast cancers (as a surrogate for basal-like tumours) and serous ovarian cancers (most of which are high-grade) occur together more frequently in the same patient than expected. Our premise is that if risk profiles of two subtypes are correlated in the population, then these subtypes are more likely to co-occur in the same patient (Begg, 2011).

Materials and methods

Since 1999, the California Cancer Registry has routinely collected information on breast tumour markers, including oestrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2). These markers can be used to approximately classify patients into four breast cancer subtypes: luminal A (ER+ or PR+, and HER2−); luminal B (ER+ or PR+, and HER2+); HER2 enhanced (ER−, PR− and HER2+); and triple negative (ER−, PR− and HER2−). The California Cancer Registry also collected information on histologic subtypes of epithelial ovarian cancer, including the four most common subtypes: serous; endometrioid; mucinous; and clear cell. We obtained the frequencies of the subtypes of all incident cases of either cancer during the period 1999–2013, as well as the frequencies and subtypes of all cases with multiple primaries. We then used these data to compare the observed vs expected co-occurrences of all subtype combinations among all women diagnosed with both a primary breast cancer and a primary ovarian cancer.

For each of the 16 possible combinations of subtypes of the two diseases, we enumerated the observed number of cases with the specific subtype combination. This observed total was then divided by an expected value to determine which subtype combinations occurred more frequently together than expected. The expected values were calculated by multiplying the total number of observed cases with both a breast and an ovarian primary by the relative frequencies of the specific subtypes observed among all cases of cancer. These relative frequencies are provided in Supplementary Table 1. The resulting O/E ratio represents a ratio of two standardised incidence ratios. The denominator is a measure of how strongly the aetiologies of breast and ovarian cancer overall are related, while the numerator reflects the corresponding strength of the relationship between the two subtypes under consideration. Thus, if the O/E ratio is elevated it suggests that the two subtypes share a more similar aetiology than breast and ovarian cancers overall, and vice versa. The rationale for these calculations is explained in detail in the Supplementary Methods.

Results

We excluded ovarian adenocarcinomas or carcinomas with histology designated as ‘other’ and restricted the analysis to breast cancers for which tumour markers were available (74% of all breast cancer cases). Our final analytic sample included 545 women with one breast and one ovarian primary cancer that had known subtypes, but excluded 72 paired breast–ovary tumours from cases with three or more primaries and five synchronous cases. The results were not materially affected by these exclusions (data not shown). The observed and expected values for all subtype combinations are provided in Table 1. The results show a large 2.35-fold increase over expectation in the co-occurrence of triple-negative breast cancers with serous ovarian cancers. None of the other combinations of subtypes demonstrated any convincing trends, although in three combinations the co-occurrence was marginally significantly lower than expected (luminal B breast and serous ovarian; luminal A and mucinous; and luminal A and endometrioid). However, as the total observed and expected frequencies are constrained to be equivalent, the other combinations must compensate with lower observed rates to offset the large excess co-occurrence of serous ovarian and triple-negative breast cancers.

Table 1 Observed vs expected counts for combinations of subtypes in double primary cases of breast and ovarian cancers
Full size table

When we stratified by order of diagnosis (breast cancer first vs ovarian cancer first), we observed significantly elevated occurrence of the triple-negative breast/serous ovarian cancer double primary combination in both strata. When the breast cancer occurred first, there were 73 cases observed vs 25.9 expected (O/E=2.8, 95% confidence interval 2.2–3.5), and when the ovarian cancer was diagnosed first there were 30 observed vs 17.9 expected (O/E=1.7, 95% confidence interval 1.1–2.3).

Discussion

These results provide compelling support for the hypothesis that the risk profiles for basal-like breast cancer and high-grade serous ovarian cancer are strongly correlated. That is, individuals with a high risk for basal-like breast cancer will typically also experience a high risk for high-grade serous ovarian cancer, and vice versa. This knowledge should facilitate a purposeful search for the risk factors common to these subtypes. Relatively little is known about risk factors specifically for high-grade serous and basal-like breast cancers. Although reproductive and hormonal risk factors do not clearly overlap between serous and triple-negative breast cancer, what is highly consistent is that established hormonal risk factors for breast and ovarian cancers overall are much more weakly associated with these subtypes compared to other breast or ovarian subtypes (Anderson et al, 2014; Wentzensen et al, 2016).

An obvious candidate that may explain a portion of the risk association is the presence of a BRCA1 germ-line mutation. Previous studies have shown that germ-line mutations in BRCA1 are more frequent in serous ovarian cases than other ovarian subtypes, and considerably more frequent in triple-negative breast cases than other breast cancer subtypes, suggesting that BRCA1 germ-line mutations represent an important risk factor that explains a portion of the elevated concordance between triple-negative breast and serous ovarian cancers in our study (Mavaddat et al, 2012). Further, although the prevalence of BRCA1 mutations in the USA population is low, the high relative risks of breast and ovarian cancer among mutation carriers lead to much higher prevalences of germ-line BRCA1 mutations in cases, and especially in those with double primaries. While among breast cancer patients in northern California the prevalence of BRCA1 mutations is only about 2% (John et al, 2007), the prevalence in ovarian cancer patients is higher (∼13% for BRCA1 and BRCA2 combined) (Arts-de Jong et al, 2016) and the reported prevalence in cases with double breast–ovary primaries is higher still, in the region of 35% (Evans et al, 2010; Cvelbar et al, 2011; Pilarski et al, 2012). Clearly, a study involving genotyping would be necessary to resolve the extent to which these mutations may explain the increased similarity of the risk profiles of basal-like breast and high-grade serous ovarian cancers. Conversely, there are also important distinctions between these subtypes in that by definition triple-negative tumours are ER negative, whereas serous ovarian tumours are predominantly ER positive, suggesting that while there must be important commonalities in their risk profiles there must also be important differences (Shafrir et al, 2016).

It is interesting that we did not observe evidence of a shared aetiology for the other subtype combinations, even though, for example, luminal breast cancers and endometrioid ovarian cancers are characterised by hormone receptor expression (Anderson et al, 2014) and have similar reproductive risk factors (Hecht et al, 2009).

One limitation of our study is that treatment for the first primary cancer could differentially impact incidence of subtypes of the second primary tumour. For example, bilateral oophorectomy for ovarian cancer treatment may lower the incidence of hormonally sensitive luminal A breast tumours more than triple-negative tumours (Press et al, 2011; Boggs et al, 2014). Subtype misclassification is also an issue, although ovarian cancer histology is reasonably well classified and subtyping for breast cancer was relatively standardised during the period of our study (Köbel et al, 2013). Further, we used triple-negative breast cancers as a surrogate for basal-like, recognising that only about 80% of triple negatives will be truly basal-like (Rakha and Ellis, 2009; Curtis et al, 2012). While we were unable to characterise serous ovarian cancer by grade, few serous tumours are low-grade (Matsuno et al, 2013). Despite these potential sources of bias, we believe that the results are sufficiently strong that they provide substantial evidence of the common aetiology of basal-like breast cancer and high-grade serous ovarian cancer. It is critical to identify both shared and non-shared risk factors for high-grade serous and basal-like breast cancer as these are the most aggressive types of these cancers. Further research should examine risk factor profiles of tumours across anatomic sites that share molecular features.