Prevalence of Germline Pathogenic and Likely Pathogenic Variants in Patients With Second Breast Cancers

Abstract Background Few studies have examined gene-specific associations with contralateral and/or second breast cancer (SBC). Methods The frequency of pathogenic and likely pathogenic (P/LP) variants in clinically actionable genes (BRCA1, BRCA2, PTEN, TP53, CHEK2, CDH1, ATM, PALB2, NBN, and NF1) was compared between women with a primary breast cancer (PBC) and SBC who underwent multigene panel testing at a single diagnostic testing laboratory. Race- and ethnicity-specific logistic regression burden tests adjusted for age at diagnosis of first breast cancer, histology, presence of first- or second-degree relatives with breast cancer, and prior testing for BRCA1/2 genes were used to test for associations with SBC. All statistical tests were 2-sided. Results The study was comprised of 75 550 women with PBC and 7728 with SBC. Median time between breast cancers for SBC was 11 (interquartile range = 6–17) years. Restricting to women tested for all actionable genes (n = 60 310), there were 4231 (7.8%) carriers of P/LP variants in actionable genes among the controls (PBC) compared with 652 (11.1%) women with SBC (P< .001). Among Caucasians, exclusive of Ashkenazi Jewish women, those carrying a P/LP variant in a clinically actionable gene were 1.44 (95% confidence interval [CI] = 1.30 to 1.60) times as likely to have SBC than noncarriers, after accounting for potential confounders. Among African American and Hispanic women, a P/LP variant in a clinically actionable gene was 1.88 (95% CI = 1.36 to 2.56) and 1.66 (9% CI = 1.02 to 2.58) times as likely to be associated with SBC, respectively (P < .001 and P = .03). Conclusion Women with P/LP variants in breast cancer predisposition genes are more likely to have SBC than noncarriers. Prospective studies are needed confirm these findings.

Women who have undergone bilateral mastectomy for breast cancer often state that worry about developing a contralateral breast cancer (CBC) or a second breast cancer (SBC) was one of the main reasons they underwent bilateral mastectomy (1)(2)(3). Genetic testing can help clarify risk for SBCs and is frequently recommended for newly diagnosed breast cancer patients to inform surgical decisions. Studies have shown that newly diagnosed breast cancer patients who have undergone genetic testing are more likely to undergo bilateral mastectomy (4-6) especially if a pathogenic and likely pathogenic variant is identified. However, in recent years, the clinical genetic testing approach for hereditary breast cancer has shifted from single to multigene panel testing, resulting in the increased identification of patients with pathogenic variants in predisposition genes beyond BRCA1/2 (7)(8)(9). Guidelines recommend consideration of bilateral mastectomy for women newly diagnosed with breast cancer carrying pathogenic variants in BRCA1 or BRCA2 based on 20-year cumulative CBC risks of 40% and 26%, respectively (10)(11)(12)(13)(14), but recommendations for women with pathogenic or likely pathogenic (P/LP) variants in breast cancer predisposition genes other than BRCA1 or BRCA2 are not available, because cumulative long-term risks of a SBC are not well established. Although some studies have suggested a relationship between P/LP variants in genes such as CHEK2 and PALB2 and higher SBC risk (15,16), they were not sufficiently powered to reliably confirm an association.
To examine the association between P/LP variant in clinically actionable variants (according to the National Comprehensive Cancer Network [NCCN] Clinical Practice Guidelines in Oncology [NCCN Guidelines]) with SBC or CBC, we conducted a retrospective analysis of nearly 90 000 women referred for genetic testing at a single diagnostic laboratory. The objective of this study was to provide robust estimates of the prevalence of P/LP variants among patients with a CBC or SBC in lieu of a long-term longitudinal study. We compared the odds of a SBC in women who carry clinically actionable NCCN P/LP variants compared with women who do not carry any P/LP variants. Although treatment information was not available, we controlled for other potential confounders of SBC such as age at PBC diagnosis, family history, and prior BRCA1/2 testing to determine the association of P/LP variants in all "clinically actionable" genes with SBC. Our large dataset also enabled us to examine whether the prevalence of these variants differed by race and ethnicity after accounting for patient and tumor characteristics.

Patient Population
The study population consisted of female breast cancer patients who underwent multigene panel testing at Ambry Genetics from March 2012 to December 2016 (n ¼ 87 229). Demographic, family history, and clinical information (sex, age, self-reported ethnicity, personal cancer history, age at diagnosis, and breast tumor characteristics) were collected from test requisition forms provided by ordering clinicians, as well as other clinical documentation if provided. Women with potential synchronous breast cancer were identified as those who had a SBC diagnosis within 1 year of the first diagnosis (n ¼ 3951) and were excluded from primary analyses. This study did not use patient identifier information, and institutional review board approval was deemed exempt by NorthShore University Health System.

Multigene Panel Testing
Patients underwent comprehensive sequencing of BRCA1/2 and other cancer predisposition genes (Supplementary Table 1, available online), as previously described (17,18). Variants were assessed using Ambry's 5-tier classification framework based on guidelines published by the American College of Medical Genetics and Genomics and the Association for Molecular Pathology (pathogenic, likely pathogenic, variant of uncertain significance, likely benign, benign) (19,20). The classification framework incorporates multiple lines of evidence such as functional and structural impact, evolutionary conservation, allele frequency in the general population, co-segregation, casecontrol data, and phenotype (17).

Statistical Analysis
Differences between PBC and SBC patient and tumor characteristics were assessed with Fisher exact test for categorical variables and 2-sample t tests or analysis of variance for continuous variables, as appropriate. The proportion of patients who tested positive for a P/LP in a clinically actionable gene was assessed for the PBC and SBC groups. Patients were included in an analysis of a gene set (eg, RRM þ ) only if they were tested for all genes in the subset. Logistic regression tests stratified by race and ethnicity were used to test for SBC associations with P/LP in actionable gene subsets, adjusting for age at PBC diagnosis, PBC histology, presence of first-or second-degree relative with breast cancer, prior BRCA1/2 testing, and personal history of other nonbreast cancers. Logistic regression was also used to test for SBC association with CHEK2 c.1100delC specifically, adjusting for all covariates. To additionally control for confounding in the design, women with SBC were matched to those with PBC based on all covariates described above, with PBC diagnosis age categorized in 5-year intervals, and conditional logistic regression was used to estimate associations with SBC. Lastly, we also performed stratified analysis by age at diagnosis of first breast cancer (younger than 50 years vs 50 years and older) among Caucasian patients adjusting for all covariates (except residual age at PBC diagnosis). Race-and ethnicity-specific associations were reported only if carrier counts were 5 or more in both PBC and SBC groups. Sensitivity analyses were conducted to evaluate whether the inclusion of SBC cases with potentially synchronous cancers or restriction of SBC cases to those confirmed to have CBC or exclusion of ductal carcinoma in situ cases influenced the observed estimates of association. We further performed additional analyses matching women with PBC and SBC on time between age at first breast cancer diagnosis and age at genetic testing (5-year interval), as well as comparisons of SBC to PBC in which cases with more than 2 breast cancers were excluded from the SBC group (n ¼ 385). Missing values were included as a distinct category so that all observations could be included in the aforementioned analyses.
All statistical analyses were conducted with R v.3.2 (22). P values less than .05 were considered statistically significant. Tests of statistical significance were 2-sided.

Patient Demographics for PBC and SBC
Among women with PBC (n ¼ 75 550) or SBC (n ¼ 7728), after exclusion of potential synchronous cases, those with SBC were slightly more likely to be Caucasian (P <.001; Table 1). SBC cases tended to be older than PBC when referred for genetic testing (mean age at testing ¼ 63.

Tumor Characteristics Between PBC and SBC
The majority of women in both SBC and PBC groups had invasive ductal carcinoma (62.0% and 52.4%, respectively), although a substantial proportion of both groups did not report histology information (23.3% and 31.1%, respectively) (Supplementary   Table 6, available online) were also similar to those found in the restricted set of the primary analysis. In the fully adjusted model, women with P/LP variants were 1.50 times as likely (95% CI ¼ 1.37 to 1.64) to have SBC than those without P/LP variants. Odds ratios for RRM þ and RRM -P/LP variants were 1.47 and 1.41, respectively (Supplementary Table 6, available online). When the SBC group was further restricted to only those individuals with known CBC, all previously observed associations were stronger in magnitude and highly statistically significant despite the decreased sample size (Table 5). In models adjusted for all potential confounders, Caucasian women carrying P/LP variants in clinically actionable genes were 1.72 (95% CI ¼ 1.50 to 1.96) times as likely to have CBC as noncarriers. Similarly, Caucasian women with P/LP in RRMþ or RRM-categories were 1.71 (95% CI ¼ 1.43 to 2.03) or 1.52 (95% CI ¼ 1.26 to 1.81) times as likely to have CBC, respectively. As previously observed, BRCA1, CHEK2, and BRCA2 were the genes with the most prevalent P/LP variants associated with known CBC (OR ¼ 1.94, 1.81, and 1.55, respectively; all P < .001). Specifically, CHEK2 c.1100delC carriers were 1.84 (95% CI ¼ 1.34 to 2.47) times as likely to have CBC as noncarriers. Furthermore, NBN, TP53, and PALB2 were also statistically significantly associated with CBC (OR ¼ 2.79, 2.10, and 1.53, respectively; P range from .001 to .02).
When cases with more than 2 breast cancers were excluded from the SBC group, previously observed associations remain for SBC-only Caucasian women (Supplementary Table 7, available online). For clinically actionable genes, those carrying P/LP variants were 1.40 (95% CI ¼ 1.25 to 1.56) times as likely to have SBC than noncarriers. SBC-only was also associated with P/LP variants in RRMþ, RRM-, BRCA1, CHEK2, and PALB2 (OR ¼ 1.36, 1.33, 1.46, 1.52, and 1.38, respectively; all P .001).
Additionally, exclusion of stage 0 ductal carcinoma in situ cases from both the PBC and SBC groups had little impact on the association between SBC and P/LP carrier status (OR ¼ 1.43, 95% CI ¼ 1.27 to 1.60).

Discussion
In this enrichment analysis of genetic test results and detailed clinical histories from a large multi-ethnic cohort of patients tested at a single laboratory, we found that across Caucasian,  - -   --------ATM  --------BRCA1  70  3682  1.90  9  288 3.13 1.80 (0.82 to 3.50) .11 BRCA2  --------CHEK2  --------CDH1  --------NBN  --------NF1  --------PALB2  --------PTEN  --------TP53 -  --------NBN  --------NF1  --------PALB2  --------PTEN  --------TP53 - Odds ratios estimated from models adjusted for age at diagnosis of first breast cancer, histology of the first breast cancer, personal history of other cancer, presence of first-or second-degree relative with breast cancer, and prior BRCA1 and BRCA2 genetic testing. "-" indicates gene sets or specific genes for which there were less than 5 carriers in any group. AOR ¼ adjusted odds ratios; CI ¼ confidence interval; PBC ¼ primary breast cancer; RRM ¼ risk-reducing mastectomy; SBC ¼ second breast cancer. Hispanic, and African American women, those carrying a P/LP variant in a clinically actionable gene were 44%-87.0% more likely to have SBC after adjusting for multiple potential confounders such as family history, prior BRCA1/2 testing, and age at PBC diagnosis. Within each racial and ethnic group, mutated genes deemed eligible for RRM recommendation had similar effect sizes to those considered to have insufficient evidence for RRM recommendation, indicating that these findings were not driven solely by mutations in BRCA1/2. BRCA1 and CHEK2 were most prevalent among Caucasian women with SBC, whereas PALB2 and BRCA2 were most prevalent among African American women. Importantly, our sensitivity analyses comparing a subset of SBC women with clinically reported CBC to women with PBC yielded stronger associations with these genes despite reduced sample size. It is important to note that this is a retrospective analysis and does not directly link these P/LP variants with SBC. Future prospective longitudinal studies are needed to confirm a causal relationship between these non-BRCA1/2 P/LP variants and SBC and to determine the absolute risk of SBC, similar to studies on BRCA1/2 carriers. Our findings are largely consistent with the observations of smaller retrospective studies, such as the Women's Environment, Cancer, and Radiation Exposure (WECARE) study. P/LP variants in BRCA1 and BRCA2 have been reported to be associated with CBC in several studies, including WECARE (10)(11)(12)(13)(14). WECARE observed no statistically significant main effect of ATM on CBC, although such variation may enhance the deleterious effects of radiation exposure (23,24). Likewise, we did not observe an association between ATM and SBC or between ATM and confirmed CBC. In our study, CHEK2 was also strongly associated with SBC and CBC in contrast to the previous WECARE study (25). Interestingly, most CBC studies evaluating CHEK2 have focused on the c.1100delC variant (15,25,26), the results of which have been inconsistent; 1 study observed associations with CBC incidence and long-term survival (26), whereas the WECARE consortium reported no association but acknowledged small sample size and limited power (25). We confirmed that carriers of the c.1100delC CHEK2 variant were 1.5 times as likely to have SBC than noncarriers; the odds of SBC increased to approximately 1.8 in sensitivity analyses restricting to women with known CBC, in line with effect sizes reported previously (25).
Our findings also shed new light on other gene-based associations with CBC. For example, WECARE investigators previously reported a statistically significant association between PALB2 and CBC in a sample of 1124 women (93% Caucasian), with a marginal P value (P ¼ .04) based on only 5 carriers of PALB2 truncating mutations among women with CBC vs no carriers among women with only 1 primary (16). Having no control carriers further precluded WECARE investigators from estimating PALB2associated risk for CBC. Our results from a cohort of nearly 90 000 women suggest a modestly higher prevalence of PALB2 among Caucasian women with CBC compared with PBC and substantially increased prevalence among African American women (27).
Our study has limitations. Despite our large sample size, we were precluded from reporting gene-specific associations in some racial and ethnic groups because of the extremely low frequency of P/LP variants in these genes. We also lacked sufficient power to make inferences on most variant-specific effects. Hence, our analyses and interpretation of findings are based on gene-level enrichment of rare pathogenic variants, which can be confounded by allelic heterogeneity (28). Women with hormone receptor-negative PBC are at higher risk of SBC (29)(30)(31)(32), and tumors in BRCA1 carriers are more likely to be hormone negative (33); an analysis of those patients with complete ER status of their PBC showed identical findings to our analysis without adjusting for ER status. Given the high number of missing values, we could not adjust for specific tumor subtypes such as triple-negative breast cancer or HER2-positive tumors. Hormonal therapy and chemotherapy have been shown to decrease risk of CBC (34); however, our study data did not include treatment information for PBC, so we were unable to assess treatment effects and their potential interactions with P/LP variants and SBC. However, we would not expect P/LP carriers to receive different systemic treatments than noncarriers that would have impacted SBC. Lastly, we also did not have information on whether patients underwent a bilateral mastectomy at the time of PBC diagnosis, which is an important confounder because bilateral mastectomy rates have been increasing over the past decade (35)(36)(37)(38). However, studies have shown patient age at diagnosis is a larger driver of the decision to undergo bilateral mastectomy than genetic risk factors (39,40). When we examined associations in women younger than 50 years and conducted an age-matched analysis, our initial estimates did not change. Because bilateral mastectomy would reduce the risk of a subsequent breast cancer, our estimates of SBC risk for women with P/LP variants may be conservative. Caution should be exercised in using these findings to dictate clinical management. Our findings showed an association between some non-BRCA1/2 P/LP genes and SBC. These findings could be due to confounding, that is, "confounding by indication," because of ascertainment differences between cases and controls (eg, NCCN Guidelines for SBC patients allow for older diagnosis age than a patient from our PBC group). Nevertheless, we adjusted for many confounders associated with increased risk of a SBC such as family history, personal history of other cancers, and age at diagnosis based on available information using the adjusted logistic regression models, matched analysis, and stratified analysis. However, future studies are needed to confirm our findings before clinical recommendations can be made.
In conclusion, our results show that in patients carrying a P/ LP variant, the odds of having SBC are higher than in noncarriers. These findings underscore the need for future studies examining the role of non-BRCA1/2 genes in SBC and CBC to help inform the complex decision-making process that physicians and patients must navigate when results of multigene panel testing are returned for a patient with a new breast cancer.

Funding
This work was supported by the Harold and Jane Perlman Family Foundation (in memory of Joni Perlman Rosenberg).

Notes
Role of the funder: The funder had no role in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; and the decision to submit the manuscript for publication.