Probing the relevance of BRCA1 and BRCA2 germline pathogenic variants beyond breast and ovarian cancer

For 30 years it has been established that germline pathogenic variants in BRCA1 and BRCA2 predispose to breast and ovarian carcinomas. Indeed, it was overt familial occurrence of these 2 cancers that allowed for the linkage analysis that proved that these 2 genes existed. Beyond breast and ovarian cancer, it has been harder to determine what other cancers are also in the “established” group of BRCA1- and BRCA2-associated tumors. In this issue of the Journal, Elze et al. (1) address this question directly by taking advantage of the excellent organization of cancer genetics services in the Netherlands and the existence of the PALGA [Pathologisch Anatomisch Landelijk Geautomatiseerd Archief (National Automated Anatomic Pathology Archive)] databank, which is a nationwide system allowing for retrieval of pathology reports and, where possible, formalin-fixed paraffin-embedded tissue blocks. The Nijmegen team conducted a historical cohort study of 2965 consecutively ascertained persons with BRCA1 or BRCA2 germline pathogenic variants from the clinic at Radboud University Medical Centre. From this large cohort, they collected data on 169 non-breast, non-ovarian cancers and looked for second hits in BRCA1 or BRCA2 and for genomic instability. They used standard incidence ratios and genomic instability scores to compare cancers with biallelic inactivation of BRCA1 or BRCA2 to cancers without biallelic inactivation of either of these two genes.

First, Elze et al. showed that, as expected, the standard incidence ratios for female breast and ovarian cancer associated with BRCA1 and BRCA2 were substantially elevated. Thus, having shown the established associated cancers gave the correct results, the standard incidence ratios associated with nonestablished cancers could be credibly reported. For BRCA1, only endometrial cancer had a statistically significantly elevated standard incidence ratio, whereas for BRCA2, male breast, prostate, pancreas, esophageal, and urinary tract cancers had elevated standard incidence ratios. Then, turning to the genomic instability scores, excluding breast cancer and using ovarian cancer as a positive control (where all 13 BRCA1- or BRCA2-related ovarian cancers had a high genomic instability score), the authors found that only for pancreas cancer with BRCA2 germline pathogenic variants did most of the tumors arising in a person with a BRCA1 or BRCA2 germline pathogenic variant have a high genomic instability score and thus could be regarded as BRCA deficient (1). Nearly all tumor types arising in those with BRCA1 or BRCA2 germline pathogenic variants, however, had some evidence for BRCA deficiency (1).

The paper by Elze et al. (1) adds to the growing literature that has used combined germline and somatic genetics to try to answer the question implied by the title of this editorial. What have others found? A recent combined study (2) of 118 BRCA1 and BRCA2 carriers with non-breast, non-ovarian primary tumors found that 32 (27%) had biallelic loss of BRCA1 or BRCA2. High homologous recombination repair defect scores were identified in 81% of tumors with biallelic BRCA1 or BRCA2 loss, whereas only 22% of tumors without biallelic BRCA1 or BRCA2 loss had a high score (P < .001). The major non-breast, non-ovarian tumor types with at least some instances of biallelic BRCA1 or BRCA2 loss were pancreas and prostate cancer (BRCA2), endometrial cancer and sarcomas (BRCA1), and stomach cancer (BRCA1 and BRCA2). But overall, most non-breast, non-ovarian tumors in germline BRCA1 or BRCA2 carriers were not associated with biallelic loss of BRCA1 or BRCA2 (respectively) and did not have homologous recombination repair defect. This is particularly true in childhood cancer, where it has been observed that despite the not infrequent presence of germline pathogenic variants in BRCA1 or BRCA2, biallelic alterations in BRCA1 or BRCA2 are very rare in pediatric cancers (3,4).

Hughley and colleagues (5) took the simple approach of comparing the prevalence of the sum of the prevalence of biallelic and monoallelic events divided by the prevalence of monoallelic events. Thus, using cases only, they derived a measure of the likely etiological contribution of complete inactivation of BRCA1 or BRCA2 to a given tumor. They called this the etiologic index (5), where the lowest possible score is 1.0, and any deviation from that number suggests the possibility of an etiological role for inactivation of BRCA1 or BRCA2 in that cancer. As expected, the established tumors—breast, ovary, pancreas, and prostate—had the highest etiologic index scores (particularly for BRCA2 for the latter 2 cancers), but high scores were seen for some nonestablished cancers (eg, endometrial [BRCA1] and gastroesophageal adenocarcinoma [BRCA1 and BRCA2]). The latter finding is notable because a study published 2 years later clearly implicated these 2 genes in stomach cancer in Japan (6), and a more recent study intriguingly suggested that it is the presence of Helicobacter pylori infection in the context of a BRCA1 or BRCA2 germline pathogenic variant that is associated with substantially increased risks for stomach cancer (7). It would have been interesting to see if the etiologic index was greater for these cases than for affected persons with BRCA1 or BRCA2 germline pathogenic variants and H. pylori infection compared with those with BRCA1 or BRCA2 germline pathogenic variants but without this infection (8). Continuing the gastrointestinal cancer theme but focusing on colorectal cancer, applying etiologic index methodology (5) to the data from the Dutch study (1) discussed here leads to an etiologic index of 2.25 for BRCA1, highly suggestive of a role for BRCA1 in the etiology of colorectal cancer.

Previous studies, relying on epidemiological comparisons between germline carriers and the populations from which they originate, have focused on families. Initial studies were published before the genes were even identified. For example, Ford et al. (9) reported an observed/expected relative risk (RR) of 3.30 (P < .01) and a maximum likelihood estimate of the relative risk of 4.11 (95% confidence interval [CI] = 2.36 to 7.15) for colon cancer by analyzing linkage data from 33 families linked to BRCA1. Surprisingly, there was no increased risk for rectal cancer, and the only other cancer with a statistically significant increased relative risk was prostate cancer (maximum likelihood estimate RR = 3.33, 95% CI = 1.78 to 6.20). After the genes were identified, the Breast Cancer Linkage Consortium revisited this and found that for BRCA1, there was a statistically significant excess number of pancreatic cancers, as well as those of the uterine body and, separately, the cervix (10). Colon cancer was also associated, but the relative risk was only 2.03 (95% CI = 1.45 to 2.85). There was no increased risk for prostate cancer. For BRCA2, the list was longer and included prostate, pancreas, gallbladder and bile duct cancer, stomach cancer, and malignant melanoma (11). However, the results of the most comprehensive study (12) have questioned many of these original associations and have identified new possible associations. Analyzing more than 5000 BRCA1 and BRCA2 families and nearly 15 000 persons with a BRCA1 or BRCA2 germline pathogenic variant, the only non-breast, non-ovarian cancers that were associated with both BRCA1 and BRCA2 germline pathogenic variants were cancers arising in the pancreas and stomach. Prostate cancer was associated with BRCA2 germline pathogenic variants. No other cancers were associated with germline pathogenic variants in either gene, most notably, neither endometrial cancer nor melanoma were associated. The result for colorectal cancer was equivocal. With all such retrospective studies with self-reported cancer history, there may be considerable misclassification, but overall the study suggests that many previous associations were false positives.

What are we to conclude from 30 years of research into this topic? For BRCA1, the established (or perhaps canonical) cancers are breast and ovarian cancer, and a weaker but statistically significant association with pancreatic cancer is almost certainly present. There is a strong but as yet inconclusive claim to include gastroesophageal adenocarcinoma in this group. Endometrial cancer, especially high-grade serous carcinoma (13,14), could well be part of the syndrome. The relationship to colorectal cancer remains uncertain (15). Notably, the case of prostate cancer remains unproven, but it seems unlikely that germline pathogenic variants in BRCA1 are a clinically significant risk factor. For BRCA2, the phenotype is broader, and here, along with breast and ovarian cancer, prostate and pancreas cancer are clearly associated with germline pathogenic variants in BRCA2. Gastroesophageal cancer is probably associated. Thus, the original 2 cancers identified as being associated with both BRCA1 and BRCA2 germline pathogenic variants remain the main players, with pancreas cancer more strongly associated with BRCA2, and gastroesophageal cancer the newest candidates for a positive association. It should be noted that the risk to those aged 80 years for gastroesophageal cancer is probably less than 4% for most BRCA1 or BRCA2 germline pathogenic variant carriers (12) [but the risk may be significantly higher in Asia (6,7)].

In terms of lifetime cancer risks, preventive and early diagnosis efforts for both BRCA1 and BRCA2 heterozygotes should stay focused on breast, ovarian, and pancreatic cancers, and on prostate cancer for men with a BRCA2 germline pathogenic variant (16-18). Surgical prevention is likely to be highly effective in reducing deaths from breast and ovarian cancer, but this will not be the case for highly fatal cancers such as BRCA2-associated prostate cancer (17,18) and pancreas and gastroesophageal cancers in BRCA1 and BRCA2 carriers. Although nonsurgical preventive efforts have stalled, new technology that can allow for detection of DNA released from dying tumor cells has offered hope that early diagnosis will change the trajectory of cancers arising in BRCA1 and BRCA2 carriers. Such approaches may be particularly valuable when the phenotypes are broad; the best extant example is in Li-Fraumeni syndrome, which is notorious for its wide phenotypic expression (19). The degree to which these techniques can accurately pinpoint the tissue of origin of the cancer will be of critical importance. Measuring all-cause mortality, rather than the proportion of down-staged cancers, will be necessary to properly assess the role of these techniques in management of persons at risk. Neverthelss, detection of circulating tumor DNA, and especially the ability to detect copy number alterations, could be an important step forward in efforts to detect homologous recombination repair defect–related tumors at an earlier, more treatable stage and ultimately to improve outcomes

The success of poly (adenosine diphosphate–ribosome) polymerase inhibitors 1 (PARP1) inhibitors in treating breast and ovarian cancers arising in those with BRCA1 or BRCA2 germline pathogenic variants (20-22) has ignited the cancer genetics world and raises the question of whether we can expect PARP1 inhibitors to be effective in non-breast, non-ovarian cancers with functional inactivation of BRCA1 or BRCA2. Two small studies have suggested that some non-breast, non-ovarian cancers arising in those with BRCA1 or BRCA2 germline pathogenic variants might respond to single-agent olaparib (23,24), and in the latter study, it is clear that biallelic inactivation of BRCA1 or BRCA2 is nearly always required for a response (24). But is this enough? A large study found that olaparib was much more effective at treating breast and ovarian cancers than it was in treating non-breast, non-ovarian cancers (25), with the possible exception of uterine sarcomas, which uniquely tend to be associated with biallelic somatic deletions in BRCA2. It is possible that, as Jonsson and colleagues posit, tumor lineage is a strong determinant of PARP1 inhibitor response (25). If this is the case, it could limit use of PARP1 inhibitors beyond the established set of BRCA1- and BRCA2-related tumors. It is, however, difficult to envision the tumor lineage link between these cancers, and perhaps the reason for the lack of response for “nonestablished” cancers lies more in the degree of homologous recombination repair defect that is present in these tumors, and the way it is measured (26).

Finally, a transformation is heading our way—artificial intelligence (AI). Nowhere could it be more relevant than in pathology. It appears that the presence of homologous recombination repair defect can be detected from whole slide images using AI with high accuracy (27-29). If this technology can be delivered cheaply and equitably to the cancer care community, it could open opportunities for therapeutic gains for many cancer patients.

Taken together, these studies suggest that testing tumors for pathogenic variants and homologous recombination repair defect–generated genomic scars (by AI or by direct testing of tumors) will be necessary when enrolling patients into studies where the mode of action of the agent is believed to depend on homologous recombination repair being inactivated. Eradicating BRCA1- and BRCA2-related cancers should be the goal for the next 30 years, and to do this we will need to focus on all 3 aspects—prevention, early diagnosis, and treatment. Detailed molecular analysis of tumors arising in persons with BRCA1 or BRCA2 germline pathogenic variants (1,2,5,25) is an important step along the way.

Data availability

No new data were generated or analyzed for this editorial.

Author contributions

William D. Foulkes, MBBS, PhD (Conceptualization; Writing—original draft) and Paz Polak, PhD (Writing—review & editing)

Funding

Not applicable.

Conflicts of interest

WDF is an associate editor and is co-author of this editorial but had no part in the editorial review of this manuscript or decision to publish the editorial. Paz Polak is an employee of Quest Diagnostics.