Germline Analysis from Tumor–Germline Sequencing Dyads to Identify Clinically Actionable Secondary Findings

With the technological advancement and continued reduction in cost associated with massively parallel sequencing, application of this technology to tumor sequencing has enabled clinicians and scientists to elucidate molecular mechanisms in cancer (1) and recognize the potential of personalized oncology (2, 3). When tumor–germline dyad sequencing is performed on entire genomes, exomes, or selected genes, germline variants are “subtracted out” from those found in the tumor to identify somatic mutations (4). The primary focus then is the analysis of somatic mutations to identify driver mutations with existing targeted therapies (2, 3, 5). While the process of germline variant subtraction enhances the specificity of detecting somatic mutations (3), in some instances, a pathogenic germline mutation may be overlooked that predisposes a patient to increased cancer susceptibility (6–8). Ignoring the germline data after “subtraction” will likely miss these critical variants and identifying such germline variants via tumor-only sequence analysis would be challenging and imprecise (9).

In the course of tumor–germline sequencing, the incidental detection of germline mutations of potentially diagnostic clinical significance can and does occur as previously described by other groups (10, 11). As these concomitant and potentially unexpected findings could have significant implications for the patient and their family members, we conducted an exploratory study within patients undergoing tumor–germline sequencing to explore the frequency of “opportunistically identified” pathogenic germline variants within cancer predisposing genes.

Participants were enrolled in the LCCC1108 study (UNC clinical sequencing study, referred to hereafter as UNCseq™). Informed consent and whole blood DNA (or buccal as appropriate) were obtained from all patients through an institutional review board (IRB)-approved protocol at the Lineberger Comprehensive Cancer Center and the University of North Carolina, Chapel Hill (UNC-Chapel Hill, Chapel Hill, NC; ref. 5; NCT01457196). The UNCseq™ study aims to associate known molecular alterations with clinical outcomes in oncology and uses this information to support treatment decisions through reporting of genetic profiling to clinicians. The overarching study consent describes the collection and analysis of both tumor and germline tissue including the explicit possibility for identification of an underlying hereditary cancer predisposition. Participants consent to the reporting of all results deemed clinically significant. Consent was obtained by UNCseq™ study staff for the primary study at enrollment. Patients were referred into the UNCseq™ study team by their clinic physician and enrolled according to their treated cancer (Table 1) and thus the tumor tissue to be analyzed. All patients enrolled between November 2011 and June 2014 for the cancer types listed in Table 1 were included in our data capture for exploratory germline analysis.

Library preparation and gene capture methods have been described previously (5). Briefly, DNA was extracted from blood using a Puregene DNA Purification kit (Gentra Systems), DNeasy Blood and Tissue Kit (Qiagen), or a Maxwell MDx16 (Promega, Inc.). In each methodology, DNA was extracted according to the manufacturer's protocol. DNA was fragmented to approximately 180–225 base pairs using a Covaris E220 focused ultrasonicator instrument (Covaris, Inc.). After fragmentation, the sample was enriched for appropriately sized fragments using an automated separation step employing AMPure beads (Beckman Coulter). Fragment size enrichment and subsequent library preparation steps involving precise liquid handling steps were performed using the Agilent basic Bravo A and/or the Bravo B robot(s) (Agilent Technologies). Gene capture was performed using a SureSelect custom capture kit according to the manufacturer's protocal (Agilent Technologies). All exons of the 247 genes on the UNCseq™ panel were sufficiently captured with average coverage depth of 750× [see Supplementary Table S1; capture V6 within Jeck and colleagues (5) listing all 247 genes].

Library quality was assessed with a Bioanalyzer or Tapestation 2200 (Agilent Technologies) using either D1K Screentapes or High Sensitivity D1K Screentapes (Agilent Technologies). Completed libraries were normalized and pooled using Bravo robots guided by vWorks automation control software (Agilent Technologies), and sequenced at the UNC High Throughput Sequencing Facility using a HiSeq2500 (Illumina). Alignment and variant calling of the sequencing reads have been described previously, with the addition of Isaac and FreeBayes for variant calling as well as ABRA for read realignment (5, 12, 13, 16, 18). In brief, germline sequencing reads were mapped to the hg18 reference genome using the Burrows Wheeler Aligner (10) and ABRA (16). ABRA is a bioinformatics platform designed to improve indel detection and accuracy for estimation of variant allele frequency (16). The germline variants were then called using Varscan (5), FreeBayes haplotype-based variant detector (13), and Isaac to improve calling near indels by local realignment (18). Finally, variants were annotated using ANNOVAR (5). Generally, mean target coverage for all patients ranged from 100 to 2,000×, with the average being approximately 750×. Germline variants and variant annotations were stored in a local PostgreSQL database (14).

While the next-generation sequencing (NGS) methods used here may detect copy number variation (CNV), we did not use it for this purpose. If we had, any CNV would have been verified through a Clinical Laboratory Improvement Amendments (CLIA)-certified laboratory at UNC-Chapel Hill (Chapel Hill, NC). Validation of the assay including assessment of sensitivity and specificity to detect germline variants was not performed because this is an exploratory research study. Any variants deemed clinically significant, and thus warranting return to the patient, are confirmed on a new sample through an orthogonal method within the CLIA-certified Molecular Genetics Laboratory at the UNC-Chapel Hill (Chapel Hill, NC).

Variants were first filtered through a gene list of 36 known hereditary cancer genes and then prioritized for analysis based on minor allele frequencies, protein effect, and existence in databases of previously reported pathogenic variants (see Table 1 for analyzed genes). Allele frequency data were obtained from The 1000 Genomes Browser (http://browser.1000genomes.org/index.html), National Heart, Lung, and Blood Institute Exome Variant Server ESP6500 Data Set (http://evs.gs.washington.edu/EVS/), and/or The Exome Aggregation Consortium (ExAC, http://exac.broadinstitute.org/).

Online resources for variant classification included The National Center for Biotechnology Information ClinVar database http://www.ncbi.nlm.nih.gov/clinvar/), the Leiden Open-Source Variation Database (LOVD, http://www.lovd.nl/2.0/index_list.php), and the Catalogue of Somatic Mutations in Cancer (COSMIC, http://cancer.sanger.ac.uk/cosmic). COSMIC was used to determine whether a variant existed in tumors from similar tissues of origin. After a preliminary computational classification, variant counts were generated using an in-house python script and validated manually.

Variants underwent tiered review by trained molecular analysts in conjunction with discussion in a multidisciplinary group. Evidence curation and variant classification was performed in a manner similar to the more recently published guidelines from the American College of Medical Genetics and Genomics and the Association for Molecular Pathology (15). As the patients were not selected for clinical or family histories suggestive of a hereditary cancer predisposition, this phenotype information was not available during the variant review process. Therefore, the molecular analysts utilized an incidental or secondary variant analysis approach such that a high threshold for pathogenicity must be met for variant result. The medical and family history presented in Table 2 was obtained from medical record review after variant analysis. Following stringent review, variants classified as Likely Pathogenic or Known Pathogenic were identified as eligible for return to patients. Before results return, these variants will be confirmed through analysis of a new sample via an orthogonal method (e.g., Sanger sequencing) and verified by an American Board of Medical Genetics and Genomics (ABMGG)-certified molecular pathologist. The confirmation step was ongoing at the time of submission. Once confirmed, the hereditary cancer predisposing variants will be returned to the patients through a board-certified genetic counselor experienced in hereditary cancer. When medical record review documented a clinically known hereditary cancer predisposing variant, no additional steps for confirmation and results return were performed.

To assess the frequency of pathogenic variants opportunistically identified in a group of unselected cancer patients undergoing tumor sequencing, we analyzed germline variants from 439 patients ascertained through the UNCseq™ study (5). Although all 247 genes of the UNCseq™ panel were sequenced, we specifically investigated germline variants only in 36 genes strongly associated with hereditary cancer syndromes that were present on the somatic sequencing panel. On the basis of current knowledge about the spectrum of cancers associated with these hereditary cancer syndromes, 24 were considered concordant with the cancer types of the patients being analyzed (Table 1). These cancers included colorectal, ovarian, breast, musculoskeletal, lung, kidney, brain/CNS, melanoma, hematologic, and pancreatic cancers (17, 19, 20, 21–25). Of all cases examined, 19 of 439 (4.3%) had germline variants classified as pathogenic/likely pathogenic (P/LP) in a hereditary cancer-predisposing gene. Of these, 12 were in genes concordant with the presenting cancer at enrollment and 7 were in other hereditary cancer genes (Fig. 1; Table 2). The discrepancy in P/LP variants listed in Table 2 and Figure 1 is due to a number of cases having pathogenic variants in discordant genes [e.g., BRCA1 NM_007294.3:c.594-2A>C in patient 11 with colorectal cancer, BRCA2 NM_000059.3:c.5233_5233delA, p.(Met1745fs) in patient 15 with AML, and others]. The majority of these findings occurred in patients with colorectal, ovarian, breast, and pancreatic cancers; very few such findings occurred in patients with musculoskeletal, lung, kidney, brain, skin, or hematologic malignancies.

Overall, BRCA1 and BRCA2 harbored 11 of 19 (57.9%) of the pathogenic variants, the majority of which were classified LP because they were novel variants expected to result in an early truncation or for which existing evidence suggested a pathogenic role based on classification guidelines (ref. 15; Fig. 2). As might be expected, we discovered that a portion of the P/LP variants had been previously identified through prior clinical genetic assessment for hereditary cancer predisposition. Medical record review following variant classification revealed that the BRCA1/2 variants identified in breast and ovarian cancer patients in this study had all been previously identified through routine clinical genetic testing, indicated on the basis of medical and family history (26). However, we also discovered P/LP variants in patients whose prior clinical genetic testing was negative because it was focused on only BRCA1/2 genes. None of the ATM and CDKN2A pathogenic variants identified in breast cancer patients were previously known (Table 2), reinforcing the idea that additional P/LP variants may exist in breast cancer patients that would be missed in individuals whose clinical testing was restricted to BRCA1 and BRCA2 (27, 28).

Some patients had a prior history of cancers consistent with the variants identified in the germline analysis, but had been enrolled for cancers that were presumably unrelated (Table 2). For example, patient 11 was previously diagnosed with breast cancer at age 41 and was enrolled in UNCseq™ when diagnosed with colorectal cancer at the age of 49 years. She was found to have a pathogenic canonical splice site variant in BRCA1 (NM_007294.3:c.594-2A>C) that provides a very good explanation for her breast cancer, but there is debate about whether BRCA1 mutations predispose individuals to colon cancer (29). Similarly, patient 17 was previously diagnosed with breast cancer at the age of 52 years, but was enrolled in the UNCseq™ study for non–small cell lung cancer. She was found to have a pathogenic nonsense variant in ATM [NM_000051.3:c.352C>T, (p.Gln118Ter)] that provides a plausible explanation for her breast cancer. There is limited evidence to support the contribution of ATM to lung cancer (30).

Although relatively few patients had clearly pathogenic variants, 178 of 439 (40.5%) had a germline Variant of Uncertain Significance (VUS; Figs. 1 and 3). In 24 patients, a VUS was found in a gene relevant to the presenting cancer type, while 143 patients had a VUS in hereditary cancer genes unrelated to their cancer type. Not surprisingly, 11 patients had a VUS in both pertinent and nonpertinent genes (Fig. 3).

This article explores the yield of clinically relevant findings from germline analysis in patients undergoing tumor–germline dyad sequencing. Consistent with the range reported in previous hereditary genetic disease studies (3, 10, 31), we found that 4.3% of patients had incidental pathogenic germline variants. This frequency is lower than the approximately 12% recently reported by Schrader and colleagues in 2016 (10). While 4.3% of our patients harbored pathogenic germline variants, analysis of the same 36 genes in the study by Schrader and colleagues (10) indicated 116 out of 1,566 (7.4%) of their cases to have pathogenic variants in these genes [Table 1, this study; see eTable 7 in Schrader and colleagues (2016; ref. 10)]. Furthermore, if the data by Schrader and colleagues are limited to the same cancer types included in our current analysis, only 61 of the 116 cases remain. Thus, when given the same restrictions, Schrader and colleagues report a pathogenic variant in 3.9% of cases, which is consistent with our findings (10). Therefore, it could be presumed that in a larger population of unselected cancers run on a larger gene capture, the UNCseq™ data may have revealed the presence of additional pathogenic germline variants. Regardless, both studies suggest that a small but clinically meaningful number of patients undergoing tumor–germline sequencing will harbor germline pathogenic findings in hereditary cancer susceptibility genes.

Interestingly, the majority of P or LP variants identified in this study were in patients with breast or ovarian cancer, even though these diagnoses made up only approximately one-third of the total cohort studied. It is somewhat surprising that no patients were identified with Lynch syndrome, although this may simply be due to the relatively small number of colorectal cancer cases analyzed (N = 53). The lack of findings in patients with other tumor types is not unexpected, given the relatively small contribution of monogenic cancer predisposition in those conditions.

Another explanation for the small proportion of patients with pathogenic variants in our study could be the utilization of a tumor–germline sequencing panel created for therapeutic rather than genetic diagnosis of hereditary cancer predisposition. For instance, the NGS capture panel in our study lacks several important hereditary cancer predisposition genes such as PALB2, BARD1, BRIP1, and PMS2. While these genes are implicated in hereditary cancer predisposition, they may not necessarily be used to guide cancer treatment decisions and therefore were not on the therapeutic-focused UNCseq™ panel. Furthermore, our germline analysis did not include CNV detection. Therefore, we cannot exclude the possibility that additional pathogenic germline variants may exist within this cohort.

In our series, half of the pathogenic variants had previously been identified through clinical genetic evaluation. The other half, representing approximately about 2% of our patients, were not associated with any prior clinical hereditary cancer evaluation. For this subset of patients, the opportunistic germline analysis provides critical information for both the individual and their family, enabling potentially lifesaving interventions (9, 32). Identifying pathogenic germline variants could also provide important prognostic information, guiding surgical procedures or targeted therapeutic options for the individual cancer patient, thereby providing immediate treatment applications (2, 33). Further long-term follow-up is needed to assess whether this information ultimately benefits patients and/or their family members. We recognize that unexpected germline susceptibility information may be unwelcome to some patients depending on their personal situation or preference for information. Providers who obtain tumor–germline sequencing on their patients should thus be aware of the potential for germline findings and prepare their patients for this possibility.

That being said, providers should also recognize that analysis of the germline as an ancillary part of a tumor sequencing assay does not substitute for clinical genetic evaluation and testing for cancer predisposition, when indicated. In clinical genetic testing for hereditary cancer susceptibility, the patient has a personal and/or family history suggestive of a hereditary cancer risk, and thus an elevated prior probability of having a causative genetic lesion. Germline testing in this setting is focused on the identification of variants in relevant genes and thus findings must be assessed for their potential causal/diagnostic role and communicated within the personal and familial context of the cancer history (34). Importantly, variants of uncertain significance (VUS) are frequently returned within the diagnostic evaluation of hereditary cancer risk (35) where clinical follow up could include testing additional family members to determine, for example, if the VUS segregates with the cancer risk in the family. Interestingly, when patients with features suggestive of a hereditary cancer predisposition undergo tumor–germline sequencing, the tumor data may aid the interpretation of germline VUS variants through assessment of loss of heterozygosity (LOH) in the tumor. Current efforts within the Clinical Genome Resource are focused on determining how LOH in the tumor can support pathogenicity of a germline variant (36).

On the contrary, when clinical laboratory assessment of tumor–germline pairs is performed for prognostic or therapeutic indications, the identification of germline variation would be considered an incidental or secondary finding. As such, only P or LP findings should be reported to patients per the recommendations of the American College of Medical Genetics and Genomics (31). This notion is supported by the finding of VUS results in almost half (40.5%) of patients in the current study and almost all patients in the Schrader and colleagues study (10). This proportion was expected, given the large proportion of VUS results in other studies (10, 37–39). The clinical relevance of these variants, by definition, remains to be determined. It is important to note that if this testing had been performed for suspicion of an underlying hereditary cancer predisposition, these VUS results would likely have been reported back to the ordering clinician. Further investigation into the VUS results, including segregation analysis and functional studies, might be necessary to provide additional evidence of pathogenicity (35, 36). Although reporting and clinical correlation of VUS are appropriate through clinical genetics evaluation based on suspicion of a cancer predisposition, the sheer quantity of such findings in a germline analysis as demonstrated here and elsewhere (10) makes their routine clinical follow-up in all patients undergoing tumor–germline sequencing untenable. Because the identification of germline variants in hereditary cancer genes is not the primary goal of tumor–germline dyad sequencing, these results are most consistent with the definition of incidental or secondary findings (40); given the low prior probability of clinical relevance, the majority of variants identified are likely to be inconsequential. In accordance with guidelines for evaluating incidental findings (31), VUS results were not considered appropriate to return to patients in this study.

While some studies report less or more VUS than our results here, these differences may stem from different thresholds of variant classification or the number of genes examined within each study. Ultimately, in a setting such as tumor–germline sequencing returning VUS results to patients would produce a significant clinical burden (32), may cause undue stress, and may result in potentially unnecessary surveillance, testing or procedures for the patient and family members erroneously presumed to be “at-risk.” Given the more stringent threshold for reporting of variants considered to be incidental or secondary to the testing indication, it is essential to recognize that the absence of a reported germline hereditary cancer variant on a tumor–germline test does not rule out the possibility that a pathogenic variant does, in fact, exist. This important distinction may need to be communicated to patients who incorrectly assume that their tumor genetic evaluation included comprehensive testing for hereditary cancer predisposition. It remains imperative for oncologists to ascertain whether their patients should be evaluated specifically for germline mutations due to personal, medical, and family history indications including age at presentation, tumor phenotype, and ancestral background.

Whether NGS should be applied on a routine basis for tumor mutation profiling remains to be determined (41) although it is a major focus of precision medicine efforts. We demonstrate here that utilizing an incidental/secondary variant analysis approach for germline sequence data in unselected patients undergoing somatic sequencing may provide a small but important benefit with regard to the detection of clinically relevant, highly penetrant variants in hereditary cancer predisposition genes. Most of these findings can be ascertained through cancer genetics evaluation recommended on the basis of family history, age at presentation, ancestry, or tumor phenotype. However, some of these patients may not be referred to a cancer genetic service (42–45) and a minority will be missed due to lack of typical clinical and/or family history indications (46, 47).

Potentially unsuspected pathogenic variants have now been reported in a small, but not insignificant, proportion of cancer patients undergoing therapeutically indicated tumor–germline testing (10, 11), and our data provide further support to this scenario. Disclosing the identification of a hereditary cancer predisposition would be highly relevant to the clinical care of these cancer patients and have important implications for their relatives' medical guidance. Providers who obtain tumor sequencing will need to be cognizant of the implications of tumor–germline analysis with respect to potential incidental findings (48), understand the differences between tumor sequencing and clinical genetic testing for hereditary cancer susceptibility, and be able to effectively communicate these issues to their patients.

A.P. Hoyle holds ownership interest (including patents) in Pfizer. No potential conflicts of interest were disclosed by the other authors.

The views expressed in the submitted article are exclusively of the authors and not an official position of the University of North Carolina at Chapel Hill School of Medicine. B.A. Seifert, J.M. O′Daniel, and J.S. Berg had full access to all data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Conception and design: B.A. Seifert, J.M. O'Daniel, J.P. Evans, H.S. Earp, N. Sharpless, D.N. Hayes, J.S. Berg

Development of methodology: J.M. O'Daniel, K.C. Wilhelmsen, D.N. Hayes, J.S. Berg

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.M. O'Daniel, N.M. Patel, M.C. Hayward, D.N. Hayes

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.A. Seifert, J.M. O'Daniel, K. Amin, D.S. Marchuk, J.S. Parker, L.E. Mose, A. Marron, C. Bizon, K.C. Wilhelmsen, N. Sharpless, D.N. Hayes, J.S. Berg

Writing, review, and/or revision of the manuscript: B.A. Seifert, J.M. O'Daniel, N.M. Patel, J.S. Parker, M.C. Hayward, J.P. Evans, N. Sharpless, D.N. Hayes, J.S. Berg

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.M. O'Daniel, D.S. Marchuk, N.M. Patel, A.P. Hoyle, M.C. Hayward, K.C. Wilhelmsen, D.N. Hayes, J.S. Berg

Study supervision: J.M. O'Daniel, K.C. Wilhelmsen, J.P. Evans, H.S. Earp, D.N. Hayes, J.S. Berg

Other (review of the manuscript): H.S. Earp

This work was supported by grants from the University Cancer Research Fund and NIH grant U01 HG006487.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.