Identification and Management of Pathogenic Variants in BRCA1, BRCA2, and PALB2 in a Tumor-Only Genomic Testing Program

In recent years, the use of tumor genomic testing has dramatically increased. Though not performed specifically to identify germline alterations, tumor-only testing can uncover pathogenic or likely pathogenic (P/LP) variants in cancer predisposition genes that may prompt clinical germline testing (CGT). Systemic use of tumor-only genomic testing may also provide insights into the prevalence of pathogenic variants in cancer predisposition genes across diverse neoplasms. Finally, both germline and somatic alterations may impact treatment choices and outcomes (1–7).

Germline P/LP variants in BRCA1/2 confer an increased lifetime risk of breast, ovarian, pancreatic, and high-grade prostate cancers (8–11). Germline P/LP variants in PALB2 increase the lifetime risks of breast, pancreatic, and ovarian cancers (9). Historically, clinical guidelines for germline testing have been relatively restrictive, requiring early ages at cancer diagnosis or specific patterns of family history (12, 13). More recently, the 2021 National Comprehensive Cancer Network (NCCN) guidelines state that germline genetic testing is clearly indicated in patients with cancers associated with these genes to inform treatment options, independent of family cancer history (8). Since 2019, the NCCN has also recommended germline testing for patients with cancer who have a P/LP variant on tumor genomic testing that has clinical implications if identified in the germline (14).

Since 2013, our institutions have utilized a targeted next-generation sequencing assay, known as OncoPanel (15), to characterize tumor genomics among patients with cancer. In August 2014, OncoPanel Version 2.0 was introduced and detects single nucleotide variants, indels, copy-number alterations, and structural variants across over 300 genes, including BRCA1, BRCA2, and PALB2.

Among 7,575 consecutively tested patients with diverse neoplasms, we sought to characterize the frequency of P/LP variants in BRCA1/2 and PALB2 detected by somatic tumor testing according to underlying tumor type and to assess the factors associated with CGT among patients whose tumor genomic testing revealed P/LP variants in BRCA1/2 and PALB2. We were especially interested in the frequency of tumor-detected P/LP variants and utilization of germline testing in patients presenting with BRCA-associated tumors (breast, pancreatic, prostate, and ovarian) compared with non-BRCA cancers. As a secondary aim, we characterized whether the identified P/LP was somatic or germline. We focused on these genes since germline pathogenic variants in them influence tumor biology and responsiveness to systemic therapies (1–6, 16), and have potential familial implications.

The Dana-Farber/Harvard Cancer Center (DF/HCC; Boston, MA) institutional review board (IRB) approved protocols #17-000, #11-104, and #18-122 for this study. Eligible cancer patients prospectively provided informed written consent for tumor genomic sequencing (OncoPanel) under protocols #17-000 or #11-104 (PROFILE) and this consent allows for results to be linked to clinical information (17). The current project received IRB approval (DF/HCC IRB Protocol #18-122) and was conducted in accordance with the ethical guidelines outlined by the Belmont Report.

Patients were eligible if age ≥18 at the time of OncoPanel testing and had been seen at Dana-Farber Cancer Institute (DFCI) for at least one visit. For patients who had multiple OncoPanel tests performed, only the first assay was included, regardless of tumor type or biopsy site. Our cohort included patients who underwent OncoPanel testing that resulted between October 1, 2016, and June 6, 2018, since during this period, pathogenic variants in BRCA1/2 and PALB2 were reported without filtering of potential germline variants, and because this preceded the 2019 NCCN guidelines that recommended genetic testing for identified pathogenic variants on tumor genomic testing. This period also permitted adequate follow-up time to assess subsequent CGT.

Starting in August 2014 and through the duration of our study (2016–2018), OncoPanel Version 2.0 was utilized, and included the full coding regions of 300 genes plus selected intronic regions across 35 genes. Mean target coverage for the cohort was 339X, with 50X minimum required to pass.

Genetic alterations were matched to ClinVar (http://www.ncbi.nlm.nih.gov/clinvar) and published data using canonical nomenclature or genomic coordinates to determine their classification as P/LP, variants of uncertain significance (VUS), benign/likely benign (B/LB) or conflicting interpretations for germline classification data according to the American College of Medical Genetics and Genomics (ACMG) guidelines (18, 19). The “Best Effect” annotation (20) was not used as it can alter the “c dot p dot” format and prohibit appropriate matching with the paired known alteration in a germline database (21). Reported BRCA1/2 and PALB2 alterations in ClinVar and their associated classification were extracted from the database on August 23, 2018.

Unmatched variants and variants with a “conflicting interpretation of pathogenicity” entry in ClinVar were individually reviewed by genetics experts (J. Sotelo, A. Chittenden, S. Kane, J. Mercado, and I. Gomy) in order to classify all alterations as P/LP, VUS, or B/LB. Due to genetic aberrations in the somatic/tumor setting leading to transcript sequence misalignment, there were at times discrepancies between nucleotide and amino acid location, which required manual review (22). For all variants without ClinVar entries, genomic coordinates were requested from the OncoPanel team to account for possible nucleotide numbering misalignment due to the destabilized nature of tumor DNA. These genomic coordinates were again matched against ClinVar entries and the procedure outlined above was applied.

Any variant with “Conflicting Interpretations of Pathogenicity” entry in ClinVar was reviewed by our genetic experts. If it was determined that the classification of conflicting interpretations arose from the expected reclassification over time by multiple sources from a VUS to a B/LB, the variant was classified as B/LB in our data set. For any variant that appeared to be reclassified over time as P/LP, a separate analysis was performed by an expert reviewer using ACMG criteria to determine final classification (18).

For all remaining variants, an expert curation was performed according to ACMG criteria. Data for curation were obtained from Alamut Focus version 1.0 (Interactive Biosoftware, Rouen, France, www.interactive-biosoftware.com), Genome Aggregation Database (gNOMAD; ref. 23), published literature (23), and ENIGMA consortium Database (ENIGMA, https://enigmaconsortium.org/wp-content/uploads/2018/10/ENIGMA_Rules_2017-06-29-v2.5.1.pdf).

A secondary curation was applied by another blinded expert to assure concordance in classification using research germline knowledge bases (Pathoman, Varsome, etc.). No variants remained unclassified after this process. Variants determined to be P/LPs by expert review were labeled as such in contrast to variants that matched to ClinVar or were identified in ClinVar by expert review, which were labelled as classified by “ClinVar.”

Many missense variants were ultimately classified as VUS as they represented alterations not otherwise reported in germline databases and literature. Some of these missense variants had such high population frequencies they likely represented population polymorphisms and were ultimately reclassified as LB. As loss of function is the known mechanism of disease for all genes analyzed, truncating and frameshift alterations were often classified as LP, due to the type of alteration already providing a PVS1 criteria (18) and quickly meeting threshold criteria for LP.

Tumor BRCA1/2 and/or PALB2 alterations identified by OncoPanel were compared with germline results obtained as standard-of-care commercial testing to determine whether a P/LP variant was germline or somatic. If the canonical nucleotide and/or amino acid change was not available for the germline data but clinical notes documented a germline P/LP variant in the gene identified on OncoPanel report, then the tumor-detected genetic alteration was also considered germline. Patients with germline BRCA1/2 and/or PALB2 testing were classified as having: (i) germline testing prior to the OncoPanel result date, (ii) germline testing within 12 months of the OncoPanel result date, or (iii) no record of germline testing.

Primary cancer diagnosis was assigned using the DFCI Oncology Data Retrieval System (OncDRS) and categorized by disease type. The “Other cancer” category included tumors listed in Supplementary Table S1 and these were non-BRCA tumors. The “Unknown” category included: adenocarcinomas, not otherwise specified (NOS); poorly differentiated carcinomas, NOS; squamous cell carcinomas, NOS; and cancers of unknown primary. For cases where the tumor had a P/LP variant and were classified as an “Other cancer” or “Unknown,” we manually reclassified tumors based on available information in the medical record (Supplementary Table S2).

From OncDRS, we also collected somatic allelic fraction, coverage, tumor purity, and tumor mutational burden (TMB) that was provided from the OncoPanel next-generation sequencing platforms as previously described (15). TMB was defined as the number of somatic, coding, base substitution, and indel mutations per megabase (Mb) of genome examined. From OncDRS, we obtained biopsy type (metastatic recurrence, local recurrence, primary vs. unspecified), biopsy site (tissue type), and reporting date.

For breast cancer cases, tumor subtype including: (i) hormone receptor–positive, HER2-negative (HR⁺HER2^–), (ii) HER2⁺, any HR, and (iii) triple-negative was obtained by review of medical records. Patients with ER-negative, PR⁺ low, and HER2^– tumors were classified as triple-negative breast cancer (TNBC). For patients with metastatic breast cancer, subtype is the receptor status at the time of metastatic diagnosis if a biopsy was performed. If no biopsy was performed for patients with metastatic disease, the subtype is the pathology at the time of initial breast cancer diagnosis. For patients with nonmetastatic breast cancer, subtype was determined by the pathology findings at initial diagnosis. For patients with multiple primary breast cancers or recurrent breast cancer, subtype of the tumor sample tested by OncoPanel was used. Patients with Phyllodes tumors did not have estrogen receptor or HER2 testing. Five cases classified as breast tumors were identified on this review to be non-breast malignancies and therefore were reclassified as such for our data analyses (four were pancreatic cancer and one was a salivary gland tumor).

For patients with P/LP variants in BRCA1/2 and/or PALB2, chart abstractions were performed for diagnosis date, prior and subsequent cancer diagnoses, family history of cancer, family history of known BRCA1/2 or PALB2 alteration, and Ashkenazi Jewish ancestry.

Descriptive statistics were used to summarize the proportion of patients with P/LPs in BRCA1, BRCA2, and/or PALB2, and the proportion of patients with P/LP variants who had germline testing before or after OncoPanel results. Patients’ demographic and tumor characteristics were compared between P/LP status versus non-P/LP status in BRCA1, BRCA2, and PALB2 genes respectively, using Fisher exact test or Wilcoxon signed-rank test where applicable.

Logistic regression was used to assess the association between patients’ personal and family characteristics and likelihood to undergo germline testing. For univariate analyses, personal history of breast cancer at age ≤45, TNBC at age ≤60, ovarian cancer, non-neuroendocrine pancreatic cancer or metastatic prostate cancer; or a family history of breast, ovarian, melanoma, pancreatic, or prostate cancer, family history of a cancer predisposition gene, or Ashkenazi Jewish ancestry were included in the model. These characteristics were chosen considering NCCN guidelines for germline testing from 2016 to 2018. Multivariable logistic regression was performed using all variables listed above.

The associations between TMB, both as continuous variable and as binary variable (TMB ≥10 mut/mb vs. Other), and BRCA1/2 and PALB2 status were evaluated via the Wilcoxon signed-rank test or Fisher exact test. All analyses were performed using SAS9.4.

All data generated in this study are available within the article and its supplementary data files.

The final analytic cohort included 7,575 patients (Fig. 1). Median age was 62 years (range 18–99), 4,085 (53.9%) were female, and 6,825 (90.1%) were White (Table 1). OncoPanel was performed on the primary tumor in 4,808 (63.5%) and metastatic tumor in 2,045 (27.0%). Among 1,005 (13.4%) patients who had at least one tumor-detected variant in BRCA1/2 or PALB2, 272 patients had P/LP variants in BRCA1 (n = 90), BRCA2 (n = 173), and/or PALB2 (n = 29), of whom 16 had more than one (Fig. 1).

For P/LP variants, 179 of 292 (61.3%) were classified by ClinVar; expert review was required to classify the other 113 (38.7%). Frameshift variants in BRCA1/2 and PALB2 were more likely to be classified as P/LP (170/176, 96.6%) compared with missense alterations, which tended to be classified as B/LB or VUS (901/919, 98.0%).

Age, race, ethnicity, and type of sample tested were not significantly different in the overall cohort compared with the subset of patients with P/LP variants in BRCA1/2 or PALB2 (Table 1). Female gender was associated with slightly higher prevalence of P/LP variants in BRCA1 (1.5% vs. 0.8%; P = 0.004), but not BRCA2 or PALB2.

OncoPanel parameters including median tumor purity and mean sample coverage were not significantly different between tumors with P/LP variants in BRCA1/2 or PALB2 compared with tumors without. Median single nucleotide variant count and median TMB were significantly higher for tumors with a P/LP variant in BRCA1, BRCA2 and PALB2 (median TMB 8.4, 9.9, and 12.9, respectively) compared with tumors without P/LP variants in these genes (median TMB 6.1; Table 1).

Overall, the prevalence of P/LP variants was 6.7% (102/1,514) in patients with BRCA-associated tumors (e.g., breast cancer, prostate cancer, ovarian cancer, and pancreatic cancer) and 2.9% (169/5,786) in patients with non-BRCA–associated tumors (all others), P < 0.001. Among 716 patients with breast cancer, 5.3% (38) had a P/LP variant in BRCA1/2 or PALB2, including 5.9% of 442 HR⁺/HER2^– tumors, 1.5% of 132 HER2⁺ tumors, and 7.3% of 137 triple-negative tumors (Table 2). P/LP variants in BRCA1/2 or PALB2 were detected in 11.8% (34/288) of ovarian cancers, 6.5% (18/275) of pancreatic cancers, and 5.1% (12/235) of prostate cancers. A notable proportion of patients with nonmelanoma skin cancer (8/63, 12.7%), endometrial cancer (20/264, 7.6%), colorectal cancer (43/858, 5.0%), head and neck cancer (10/217, 4.6%), and lung cancer (31/1,099, 2.8%) were found to have P/LP variants in BRCA1/2 or PALB2 on tumor-only testing (Table 2).

Among the 272 patients with a P/LP in BRCA1/2 and/or PALB2 on tumor genomic testing, 59% (n = 161) had metastatic disease and 38% (n = 102) had nonmetastatic disease at the time of OncoPanel testing. The remainder (n = 9) had either a hematologic malignancy or a brain or spinal cord cancer. Among the patients with BRCA-related tumors, 72.5% (74/102) had metastatic disease at the time of their OncoPanel testing while 50.9% (86/169) of patients with a non-BRCA tumor had a metastatic diagnosis. One with Unknown cancer type had metastatic disease at the time of OncoPanel testing.

For BRCA1/2 and/or PALB2, 103 of 272 (37.9%) patients had germline testing prior to OncoPanel testing, 29 (10.7%) were tested within 12 months after OncoPanel testing, and 140 (51.5%) had no germline testing (Table 3). Among patients with a BRCA-related tumor, 81.4% (83/102) had germline genetic testing performed and among patients with non-BRCA tumors, 29.0% (49/169) had germline genetic testing, (Supplementary Table S3; Supplementary Fig. S1). There was 1 patient with Unknown cancer type that did not have germline genetic testing. Among patients with BRCA-related tumors (n = 102), 65.7% underwent germline testing prior to OncoPanel, 15.7% within 12 months of the OncoPanel result, and 18.6% remained untested. In contrast among patients with non-BRCA tumors (n = 169), 21.3% underwent germline testing prior to OncoPanel, 7.7% within 12 months of the OncoPanel result, and 71.0% did not have germline testing (Supplementary Table S3; Supplementary Fig. S1).

Among patients with an OncoPanel-detected P/LP variant in BRCA1/2 or PALB2 who also underwent germline genetic testing, the P/LP variant was confirmed germline in 65.4% (BRCA1), 73.5% (BRCA2), and 64.3% (PALB2) of patients (Table 3). By tumor type, BRCA-related tumors with P/LPs were more likely than non-BRCA tumors to have a confirmed germline P/LP in BRCA1/2 and PALB2 (78.3% versus 57.1%, P = 0.02 (Supplementary Table S4). Notably, among patients with colorectal cancer and endometrial cancers who underwent germline testing, only 31.6% (6/19) and 25.0% (2/8), respectively, had confirmed germline P/LPs (Supplementary Table S4).

Among patients with germline testing, 74.4% (67/91) of ClinVar classified P/LPs were identified to be germline compared with 61.9% (26/41) of P/LPs determined by Expert Review (P = 0.23; Supplementary Table S5). 38.1% of Expert review P/LPs were somatic compared with 25.6% of ClinVar P/LPs and this difference was not statistically significant (P = 0.11).

For the 140 patients who did not have germline testing prior or within 12 months after OncoPanel testing, there was no documentation of discussion of genetic testing referral in the medical record for 113 (80.7%), 14 (10.0%) were deceased prior to or within 3 months of OncoPanel report issuing, 5 (3.6%) were offered but declined testing, and for 8 (5.7%) patients, referral was made, but there no documentation of testing by 12 months (Table 3).

By current NCCN criteria, 85.4% of patients who had germline testing completed prior to OncoPanel testing had at least one clinical indication for testing (Table 4). All patients who underwent testing within 12 months of receiving OncoPanel results also had at least one clinical indication for germline testing (Table 4). There was at least one clinical factor for germline testing among 32.8% of patients with a BRCA1 P/LP, 59.6% of patients with a BRCA2 P/LP, and 8.7% of patients with a PALB2 P/LP (Supplementary Table S6). Among the 140 patients who did not have germline testing, less than half (47.1%) had personal or family characteristics that could have prompted referral for germline testing according to current NCCN criteria (Table 4). Among the 19 patients with a BRCA-related tumor who were untested, 78.9% (n = 15) had a clinical indication for germline testing and among the 120 patients with non-BRCA tumors, 40.8% (n = 49) had a clinical indication for testing. One patient with Unknown cancer type who did not have CGT did have a clinical indication for germline testing. Overall, patients with tumors classically associated with BRCA1/2 and PALB2 were significantly more likely to have germline testing before or within 12 months after an OncoPanel-detected P/LP variant in BRCA1/2, or PALB2 than those presenting with non-BRCA1/2–associated tumors (81.4% vs. 29.0%; P < 0.0001; Supplementary Table S3).

On multivariate analyses, early-onset breast cancer (OR, 10.1; P = 0.006); ovarian cancer (OR, 14.9; P < 0.0001); metastatic prostate cancer (OR, 11.3; P = 0.03); family history of breast or ovarian cancer in close relative (OR, 3.9; P = 0.0003); family history of melanoma, pancreatic or prostate cancer (OR, 2.1; P = 0.04); and known cancer predisposition syndrome in the family (OR, 19.8; P = 0.006) were associated with an increased odds of germline testing (Table 5). Among patients without germline testing prior to OncoPanel results (n = 169), the following factors were associated with an increased odds of germline testing within 12 months after OncoPanel: ovarian cancer (OR, 12.1; P = 0.004); metastatic prostate cancer (OR, 43.4; P = 0.002); family history of breast or ovarian cancer in a close relative (OR, 6; P = 0.003); and known cancer predisposition syndrome in the family (OR, 12.6; P = 0.04; Supplementary Table S7).

Among 7,575 patients with diverse cancers, unselected for age, sex, or family history, 272 (3.6%) had P/LP variants in BRCA1/2, and/or PALB2 identified on tumor-only genomic testing. As expected, P/LP variants were found in tumors with which germline mutations in these genes have been associated: ovarian, pancreatic, breast, and prostate cancers. We also observed notable frequencies of P/LP variants in patients with tumors not typically associated with mutations in these genes including ≥5% of nonmelanoma skin cancers, endometrial, colorectal, and small cell lung cancers. The prevalence of P/LP variants did not vary by age, race, ethnicity, or type of sample (primary tumor, local recurrence, metastasis).

Our findings of a clinically meaningful prevalence of P/LP variants in cancer predisposition genes uncovered on tumor genomic testing confirm and extend results of others. In one study (24), P/LP variants in BRCA1/2 or PALB2 were identified in 2.8%, 3.0%, and 0.6% of 17,566 of tumors, though this was not supplemented by germline testing. The higher prevalence of deleterious alterations may be explained by an enrichment for ovarian (14.1%) and endometrial cancers (8.4%) compared with our cohort. From TCGA, pathogenic variants on tumor-only testing in the three genes of interest were identified in 1.9% (201/10,389) of diverse tumors (25). However, the methodology used in these two studies to identify and curate pathogenic variants can miss germline pathogenic variants (26).

There are ongoing research efforts to pair tumor genomic testing with normal DNA testing. In one study, tumor genomic testing (MSK-Impact) was matched with normal DNA, and 4.1% (488/11,947) of tumors that underwent sequencing were identified to harbor a germline P/LP variant in BRCA1/2 or PALB2 (27). The manuscript did not report the frequency of somatic P/LPS in these genes by tumor type. Only 29.3% (143/488) of germline P/LPs in BRCA1/2 or PALB2 were identified among patients with BRCA1/2-associated tumors (breast, ovarian, pancreatic, and prostate). In another effort where 1,015 patients with metastatic solid tumors underwent targeted sequencing of their tumors and normal tissue, germline pathogenic variants in BRCA1/2 and PALB2 (3.2%, 32/1,015) were identified in diverse neoplasms (28), but the proportion of patients with somatic pathogenic/actionable BRCA1/2 and PALB2 was not reported.

A recent report investigated the prevalence of pathogenic germline variants in an unselected population of patients with cancer, regardless of whether tumor genomic testing was performed (29). Among 2,984 patients with cancer unselected for family history and tumor type, 76 (2.5%) patients had a P/LP germline variant in BRCA1/2 and PALB2 (29). In comparison with this finding, our cohort and the MSK-Impact cohort were enriched for patients with metastatic disease, and this could explain the higher prevalence of BRCA1/2 and PALB2 in these studies.

In contrast to the above studies, we also conducted manual chart review to assess the real-world utilization of CGT before and after tumor genomic testing, as well as to ascertain family history and other factors included in NCCN guidelines for germline testing. This allowed us to determine whether tumor genomic testing might identify additional carriers of deleterious alterations in BRCA1/2 and PALB2 who would not otherwise have had germline testing. We indeed find that 62.1% (169/272) of patients with P/LPs in BRCA1/2 and or PALB2 on tumor genomic testing had not had germline testing prior to tumor genomic testing; only an additional 10.7% of patients underwent germline testing within 12 months after the tumor-only genomic result was available. Of note, one third (89/272, 32.7%) of patients with P/LP variants in BRCA1/2, and PALB2 did not have another clinical indication for germline testing by traditional criteria. Among tumor-identified P/LP patients who did not have germline testing before or within 12 months after the OncoPanel result was available, 52.9% did not meet any personal or family history criteria that would have prompted CGT. Moreover, patients presenting with non-BRCA related tumors were significantly less likely to have germline testing, even after OncoPanel results were available. Our findings confirm and further extend the work by Vlessis and colleagues (30) where patients with non-breast and non-ovarian cancer diagnoses were less likely to have germline genetic testing despite tumor-genomic findings showing pathogenic/actionable variants in BRCA1/2.

Among the subset of patients who had both CGT and tumor-only OncoPanel testing, 70.5% (93/132) of P/LP variants were confirmed germline, and the remainder were presumed somatic. Among patients with colorectal and endometrial cancers who underwent germline testing, the majority of P/LPs in BRCA1/2 and PALB2 were somatic, which may reflect the high TMB that is reported in these tumors, and not a driver of pathogenesis (31). Overall, the findings are noteworthy in light of the implications of deleterious germline alterations in BRCA1/2 and PALB2 on personal and familial cancer risk, and the potential responsiveness of tumors in germline carriers to PARP inhibitors (1–6, 16). Breast cancer with somatic BRCA1/2 pathogenic variants have sensitivity to PARP inhibitors as well (1). There is a need, however, to determine the role of PARP inhibitor therapy in non-BRCA tumors with germline and somatic P/LPs. There are ongoing efforts investigating BRCA-biology in non-BRCA tumors and case reports suggest that tumors respond to PARP inhibitors in this setting, however, more work is needed in this area (32–37).

In our cohort, the main barrier to germline genetic testing was the lack of referral by the oncologist to genetic counselors (80.7%), despite a well-established genetics program at our institution and availability of genetic counselors. This is a challenge at other institutions where 31% of patients with detected somatic BRCA1/2 variants were not recommended to undergo germline testing (30). At other institutions even when automatic referrals are placed for germline testing, 62% of referred patients do not undergo testing (38). It is important to note that during the timeframe of this study, OncoPanel was considered a research test and NCCN guideline criteria did not yet recommend germline testing for patients with a P/LP result in BRCA1/2 and/or PALB2 found on tumor testing. Annotation of BRCA1/2 and PALB2 in the OncoPanel report provided to clinicians was performed by individual pathologists, and the report format underwent several changes over time to make potentially germline findings more apparent to ordering providers. In contrast, for the current analysis, we manually rereviewed and curated all variants with a central expert panel including clinical geneticists. Since the timeframe of our study, OncoPanel has been expanded to a clinical test across a wider range of indications and the report format has been updated to better highlight potentially actionable alterations, including findings for which germline testing should be considered. Finally, in the near future, our institutions plan to offer paired tumor and germline testing to patients with cancer, with processes for genetic counselors to automatically contact all patients who have germline P/LP variants in cancer predisposition genes, without the need for manual physician referral. However, it is not unlikely that other users of tumor-only testing could miss either the notification or the significance of potential germline findings, and not appreciate the importance of reporting the findings to their patients and family members, despite concerns from experts that there is a need to develop workflows to appropriately alert providers and patients of the findings (39).

Of note, of 1,271 total variants in BRCA1/2, and/or PALB2 identified in our study, only 292 variants (in 272 patients) were classified as P/LP. Although 179 of 292 (61.3%) were classified by ClinVar, 113 (38.7%) variants required classification by expert review. Nomenclature disparities further complicate the reporting of germline variants on tumor genomic testing reports (21). Experts in the field have developed a framework for assessing somatic results that warrant a referral for germline testing (40–42). However, many clinicians are not equipped to apply this framework in practice. Our results highlight the importance of high-quality variant annotation in tumor genomics reports and clear communication with ordering clinicians.

The current study has several strengths. First, our analytic cohort included a large number of patients across diverse malignancies, unselected for family history or germline testing for inherited cancer predisposition, and consecutively tested using a single platform. Second, all variants were curated by an expert team. Third, we performed detailed chart abstractions to identify other clinical indications for germline testing and to quantify the proportion of patients who underwent germline testing, before or after tumor genomic testing.

Our study had several limitations. The majority of patients were non-Hispanic White. We relied on chart abstractions for family cancer history, known cancer predisposition gene status and ancestry. We were not able retrospectively to determine directly whether the OncoPanel result influenced the decision to refer for germline testing, only whether such testing was or was not performed. In patients with P/LPs in BRCA1/2 and/or PALB2 who had germline testing for hereditary cancer genes and the variant was not detected (deemed somatic), we were unable to verify with the commercial testing company if the variant was detectable considering their technology. Next-generation sequencing platforms can miss deletion and duplication abnormalities since somatic sequencing assays often focus on exonic regions. However, prior work has demonstrated that germline P/LPs are more likely be detected on germline testing than on tumor genomic testing and not vice versa (26). Finally, we limited our analysis to only 3 genes.

In conclusion, we report a low (3.6%), but clinically meaningful rate of deleterious alterations in BRCA1, BRCA2, and PALB2 detected by tumor-only genomic testing across a diverse range of malignancies, including in patients with cancers not classically linked to these genetic syndromes. A high proportion of patients with P/LP variants in BRCA1/2 and PALB2 detected on tumor-only testing did not have personal or family history that would otherwise prompt CGT. Finally, the uptake of CGT after a positive tumor-only genomic test for these genes was low, especially among patients with tumors not classically associated with germline BRCA1/2 alterations. Given the increasing implications of BRCA1/2 and PALB2 alterations on familial cancer risk and treatment of the index patient, our data suggest consideration of more liberal criteria for germline testing in patients with cancer, and highlight the need for systems-based approaches to meticulously annotate variants in cancer predisposition genes and drive additional management including referral for germline testing.

J. Sotelo reports other support from Thermo Fisher Scientific outside the submitted work. I. Gomy reports personal fees from Fleury Group S.A. outside the submitted work. J.S. Kim reports personal fees from Dewpoint Therapeutics outside the submitted work. M.M. Awad reports grants and personal fees from Genentech, Bristol-Myers Squibb, and AstraZeneca; personal fees from Merck, Maverick Therapeutics, Blueprint Medicines, Syndax Pharmaceuticals, Ariad, Nektar, Gritstone, ArcherDx, Mirati Therapeutics, NextCure, Novartis, EMD Serono, NovaRx, and Iovance Biotherapeutics; and grants from Lilly outside the submitted work. P.A. Konstantinopoulos reports personal fees from Alkermes, Kadmon, Mersana, Artios, and Repare Therapeutics; grants from Pfizer, Merck, Merck KGaA, Eli Lilly, and Bristol-Myers Squibb; and grants and personal fees from Bayer, GlaxoSmithKline, and AstraZeneca outside the submitted work. M.B. Yurgelun reports grants from Janssen Pharmaceuticals; and personal fees from UpToDate outside the submitted work. B.M. Wolpin reports grants from Celgene and Eli Lilly and personal fees from BioLineRx, Celgene, and GRAIL outside the submitted work. M.E. Taplin reports personal fees from AstraZeneca, Janssen, Arcus Biosciences, Blue Earth, Pfizer, Roivant Sciences, Clovis, Epizyme, Constellation Pharmaceuticals, Bayer, Myovant Sciences, AbbVie, UpToDate, OncLive, Research to Practice, UTSW, and Astellas during the conduct of the study; and personal fees from Roswell Park outside the submitted work. B.E. Johnson reports personal fees from Novartis, Boston Pharmaceuticals, Checkpoint Therapeutics, Chugai, Daiichi Sankyo, AstraZeneca, Foundation Medicine, G1 Therapeutics, Genentech, GlaxoSmithKline, Hengrui Therapeutics, Janssen, Jazz Pharmaceuticals, and Eli Lilly outside the submitted work; in addition, B.E. Johnson has a patent for EGFR Mutation Testing licensed and with royalties paid from DFCI and MGH. J.E. Garber reports other support from Ambry Genetics, Invitae Genetics, and AstraZeneca, and personal fees from Helix Genetics outside the submitted work. N.U. Lin reports personal fees from AstraZeneca, Denali Therapeutics, Prelude Therapeutics, Affinia Therapeutics, Voyager Therapeutics, Aleta BioPharma, Daiichi Sankyo, and Puma; grants from Merck, Genentech, and Zion Pharmaceuticals; and grants and personal fees from Seagen, Pfizer, and Olema Pharma outside the submitted work. No disclosures were reported by the other authors.

B.L. Bychkovsky: Conceptualization, data curation, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, writing–review and editing. T. Li: Software, formal analysis, investigation, writing–original draft, writing–review and editing. J. Sotelo: Data curation, investigation, visualization, writing–original draft, writing–review and editing. N. Tayob: Software, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. J. Mercado: Data curation, methodology, writing–original draft. I. Gomy: Data curation, investigation, methodology, writing–original draft, writing–review and editing. A. Chittenden: Data curation, investigation, writing–review and editing. S. Kane: Data curation, investigation, writing–review and editing. S. Stokes: Data curation, investigation, writing–review and editing. M.E. Hughes: Data curation, investigation, writing–review and editing. J.S. Kim: Data curation, investigation, writing–review and editing. R. Umeton: Data curation, investigation, writing–review and editing. M.M. Awad: Investigation, writing–review and editing. P.A. Konstantinopoulos: Investigation, writing–review and editing. M.B. Yurgelun: Investigation, writing–review and editing. B.M. Wolpin: Investigation, writing–review and editing. M.E. Taplin: Investigation, writing–review and editing. R.E. Newmark: Investigation, writing–review and editing. B.E. Johnson: Resources, supervision, investigation, writing–review and editing. N.I. Lindeman: Resources, supervision, investigation, writing–review and editing. L.E. MacConaill: Resources, supervision, investigation, writing–review and editing. J.E. Garber: Conceptualization, resources, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, writing–review and editing. N.U. Lin: Conceptualization, resources, formal analysis, supervision, investigation, visualization, methodology, writing–original draft, writing–review and editing.

The authors would like to acknowledge Kaitlyn T. Bifolck and Valerie Hope Goldstein for their editorial support. The authors would also like to acknowledge the DFCI OncDRS for the aggregation, management, and delivery of a portion of the clinical and operational research data used in this project. The content is solely the responsibility of the authors. Research support for this study has been provided by: The DiNovi Family Profile Fund at Dana-Farber, The Legere Family Fund for Profile, Breast Cancer Research Foundation (to N.U. Lin and J.E. Garber), Pan-Mass Challenge (to Dana-Farber Cancer Institute Breast Oncology Center), Fashion Footwear Association of New York (to Dana-Farber Cancer Institute Breast Oncology Center), de Beaumont Foundation (to N.U. Lin), NCI Specialized Program of Research Excellence (SPORE) Grant 1P50CA168504 (to Dana-Farber/Harvard Cancer Center), and National Comprehensive Cancer Network Oncology Research Program-Pfizer Independent Grants for Learning and Change (to N.U. Lin).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.