Multigene Hereditary Cancer Panels Reveal High-Risk Pancreatic Cancer Susceptibility Genes

Purpose The relevance of inherited pathogenic mutations in cancer predisposition genes in pancreatic cancer is not well understood. We aimed to assess the characteristics of patients with pancreatic cancer referred for hereditary cancer genetic testing and to estimate the risk of pancreatic cancer associated with mutations in panel-based cancer predisposition genes in this high-risk population. Methods Patients with pancreatic cancer (N = 1,652) were identified from a 140,000-patient cohort undergoing multigene panel testing of predisposition genes between March 2012 and June 2016. Gene-level mutation frequencies relative to Exome Aggregation Consortium and Genome Aggregation Database reference controls were assessed. Results The frequency of germline cancer predisposition gene mutations among patients with pancreatic cancer was 20.73%. Mutations in ATM, BRCA2, CDKN2A, MSH2, MSH6, PALB2, and TP53 were associated with high pancreatic cancer risk (odds ratio, > 5), and mutations in BRCA1 were associated with moderate risk (odds ratio, > 2). In a logistic regression model adjusted for age at diagnosis and family history of cancer, ATM and BRCA2 mutations were associated with personal history of breast or pancreatic cancer, whereas PALB2 mutations were associated with family history of breast or pancreatic cancer. Conclusion These findings provide insight into the spectrum of mutations expected in patients with pancreatic cancer referred for cancer predisposition testing. Mutations in eight genes confer high or moderate risk of pancreatic cancer and may prove useful for risk assessment for pancreatic and other cancers. Family and personal histories of breast cancer are strong predictors of germline mutations.


INTRODUCTION
Pancreatic cancer (PC) is the fourth most common cause of death resulting from cancer in the United States. 1 Epidemiologic studies have suggested that 10% to 20% of PCs are associated with an inherited component, with familial PC, defined as kindreds containing at least two affected first-degree relatives, as an established entity of inherited disease. 2 PC is a component of hereditary breast-ovarian cancer syndrome, 3,4 Lynch syndrome, 5,6 familial adenomatous polyposis, 7 familial atypical multiple mole melanoma syndrome, 8 hereditary pancreatitis, 9 Peutz-Jeghers syndrome, 10 and Li-Fraumeni syndrome. 11 Recent studies involving familial PC kindreds have further characterized the role of BRCA1/2, CDKN2A, ATM, and PALB2 in PC susceptibility. [12][13][14] Until recently, germline studies of PCs have focused on single cancer predisposition genes. 15,16 The first panel-based study of 13 cancer predisposition genes among patients with PC identified 11 mutations (3.8%) in ATM, BRCA1/2, MLH1, MSH2, MSH6, and TP53, 17 whereas a 22-gene panel-based study identified ATM, BRCA1/2, CHEK2, and PALB2 mutations in 13% of 96 sequentially collected PCs. 18 A majority of these mutations were identified in PCs with a family history of pancreatic, breast, ovarian, or colorectal cancer, suggesting that a better understanding of PC risk will depend on evaluation of families with broad constellations of tumors. 18 More recently, panel-based approaches identified germline Purpose The relevance of inherited pathogenic mutations in cancer predisposition genes in pancreatic cancer is not well understood. We aimed to assess the characteristics of patients with pancreatic cancer referred for hereditary cancer genetic testing and to estimate the risk of pancreatic cancer associated with mutations in panel-based cancer predisposition genes in this high-risk population.
Methods Patients with pancreatic cancer (N = 1,652) were identified from a 140,000patient cohort undergoing multigene panel testing of predisposition genes between March 2012 and June 2016. Gene-level mutation frequencies relative to Exome Aggregation Consortium and Genome Aggregation Database reference controls were assessed.

Results
The frequency of germline cancer predisposition gene mutations among patients with pancreatic cancer was 20.73%. Mutations in ATM, BRCA2, CDKN2A, MSH2, MSH6, PALB2, and TP53 were associated with high pancreatic cancer risk (odds ratio, > 5), and mutations in BRCA1 were associated with moderate risk (odds ratio, > 2). In a logistic regression model adjusted for age at diagnosis and family history of cancer, ATM and BRCA2 mutations were associated with personal history of breast or pancreatic cancer, whereas PALB2 mutations were associated with family history of breast or pancreatic cancer.
Conclusion These findings provide insight into the spectrum of mutations expected in patients with pancreatic cancer referred for cancer predisposition testing. Mutations in eight genes confer high or moderate risk of pancreatic cancer and may prove useful for risk assessment for pancreatic and other cancers. Family and personal histories of breast cancer are strong predictors of germline mutations. mutations in 4% (33 of 854) of patients with apparently sporadic PC 19 and in 25% (44 of 176) of patients with advanced PC. 20 Here, we report results from panel-based clinical testing of 1,652 patients with PC from a large cohort of > 140,000 patients evaluated by a single diagnostic laboratory and calculate gene-specific risks of PC by comparison with Exome Aggregation Consortium (ExAC) and Genome Aggregation Database (gnomAD) reference controls. 21,22

Study Population
Patients with PC (N = 1,819) were identified from a large cohort of > 140,000 patients undergoing multigene panel testing of seven to 49 cancer predisposition genes between March 2012 and June 2016 at Ambry Genetics 23 (Aliso Viejo, CA; Appendix Table A1). Demographic and personal and family cancer history information was provided by the ordering clinician using test requisition forms, clinic notes, and pedigrees. Clinical histories and molecular results were reviewed and summarized. Exclusion criteria, including the presence of neuroendocrine tumors or intraductal papillary mucinous neoplasms, reduced the number of patients for analysis (N = 1,652; Appendix). The study was approved by the Western Institutional Review Board.

Multigene Panel Testing
Mutation testing was performed by sequencing of targeted custom capture products from several multigene panels and targeted chromosomal microarray analysis, as previously described. 24 Genomic DNA was isolated from each patient's blood or saliva specimen using a standardized methodology (Qiagen, Valencia, CA). Sequence enrichment was performed by incorporating the genomic DNA into microfluidics chip or microdroplets along with primer pairs or by a bait-capture methodology using long biotinylated oligonucleotide probes (RainDance Technologies, Billerica, MA; Integrated DNA Technologies, San Diego, CA), followed by polymerase chain reaction and then next-generation sequencing analysis (Illumina, San Diego, CA) of all coding exons plus at least five bases into the 5′ and 3′ ends of all the introns and untranslated regions. A targeted chromosomal microarray was used for the detection of gross deletions and duplications for all genes except PMS2 (Agilent, Santa Clara, CA). Gross deletion and duplication analysis of PMS2 was performed using MLPA kit #P008-B1 (MRC-Holland, Amsterdam, the Netherlands) and Sanger sequencing. Initial data processing and base calling were performed using RTA 1.12.4 (HiSeq Control software [version 1.4.5]; Illumina). Sequence quality filtering at Q20 was executed with CASAVA software (version 1.8.2; Illumina, Hayward, CA). Sequence fragments were aligned to the reference human genome (GRCh37), and variant calls were generated using CASAVA. Mutations were annotated with the Ambry Variant Analyzer, a proprietary alignment and variant annotation software (Ambry Genetics). All mutations identified by Ambry Genetics are submitted to the ClinVar public database.

Statistical Methods
The observed frequency of all pathogenic mutations within each gene in white patients with PC was compared with the frequency of pathogenic mutations in the ExAC non-Finnish European (NFE) non-The Cancer Genome Atlas (TCGA) reference control after data cleaning and filtering (Appendix) as previously described. 23 Copy number variants in all genes and mutations in pseudogene homology regions (PMS2 exons 9 and 11 to 15) were excluded from cases and controls for risk estimation, because these alterations were not individually validated in ExAC or gnomAD controls. Established low-penetrance mutations (eg, APC p.Ile1307Lys) were excluded. Associations between combined mutations in each gene and PC were estimated by odds ratios (ORs) and corresponding 95% CIs based on Fisher's exact test. P values < .05 were considered statistically significant. Genes were categorized as high risk (OR, > 5.0), moderate risk (OR, 2.0 to 5.0), or of no clinical relevance (OR, < 2.0). Similar studies were conducted using a combined gnomAD NFE and gnomAD Ashkenazi Jewish reference control data set, henceforth referred to as gnomAD. Although these gno-mAD controls partially overlap with ExAC NFE non-TCGA controls, the substantially increased number along with updated variant calling algorithms identified gnomAD as an independent reference control data set. Sensitivity analyses for associations were performed for associations between genes and age at diagnosis; cases of PC tested with a targeted PC panel; all races and ethnicities combined; personal history of breast cancer or melanoma; family history of PC, breast cancer, ovarian cancer, uterine or endometrial cancer, melanoma, or colorectal cancer; and mutations meeting strict PASS criteria in ExAC. 25 Associations between mutations and age at PC diagnosis were evaluated using the Kolmogorov-Smirnov test. Associations with personal and family histories of other cancers were also evaluated by logistic regression, with adjustment for family history and age at diagnosis.

Characteristics of Study Population
The phenotypic characteristics of 1,652 patients with PC of all races and ethnicities and those of 1,256 white patients are listed in Table 1

Associations Between Pathogenic Mutations and PC
Mutations in ATM, BRCA2, CDKN2A, MSH2, MSH6, PALB2, and TP53 were significantly associated with high risk of PC (OR, > 5), whereas deleterious mutations in CHEK2 and BRCA1 were associated with moderate risk (OR, > 2; Table 2). Results for all panel genes are listed in Appendix Table A5. Association analyses using gnomAD reference controls confirmed all significant associations, and gene-specific risk estimates were highly similar, except for slightly attenuated risk for PALB2 mutations and increased risk for TP53 (Appendix Table A6).
The same genes were associated with increased PC risk when considering patients of all races and ethnicities compared with ExAC all race and ethnicity controls (Appendix Table A7) and after excluding those who had previously tested negative for BRCA1/2 mutations before panel testing (Appendix Table A8). Risk estimates for most genes were slightly diminished when including only those patients with PC for whom PC was the first cancer diagnosis, although MSH2 and TP53 mutations were no longer significantly associated with moderate risk of PC because of the decreased number of mutations in patients with PC, and the modest OR associated with CHEK2 was marginally significant (Appendix Table A9). In contrast, analyses using only ExAC NFE non-TCGA variants in the high-quality PASS category marginally increased the ORs for each gene (Appendix Table A10). Sensitivity analyses were also performed after excluding patients with PC with a family history of breast, ovarian, endometrial, colorectal, melanoma, or pancreatic cancer (Appendix Tables A11 to A16, respectively).

Characteristics of PCs With Mutations in PC Predisposition Genes
The frequency of mutations in the high-and moderate-risk PC predisposition genes was increased in patients with PC with a personal history of breast cancer (Table 3), with almost two-fold more mutations observed in ATM (6.80%), BRCA2 (6.50%), PALB2 (3.38%), BRCA1 (2.00%), and TP53 (0.91%). Results from logistic regression analysis confirmed these findings for ATM (P = .0065) and BRCA2 (P = .0092; Table 4). In contrast, mutations in the mismatch repair genes CHEK2 and CDKN2A collectively decreased from 4.89% to 2.52% in the context of personal history of breast cancer (Table 3). Mutations in ATM, BRCA2, and PALB2 were also more frequent in patients with PC with a family history of breast cancer (firstor second-degree relative; Table 3). In contrast, only PALB2 and MSH2 displayed a substantial increase in mutation frequency among patients with a family history of PC, and only CHEK2, MSH2, and TP53 had increased frequencies of mutation among patients with PC with a family history of colorectal cancer (Table 3). Results from logistic regression analysis confirmed the association of PALB2 mutations with 4 ascopubs.org/journal/po JCO™ Precision Oncology family history of PC (P = .029) or breast cancer (P = .0056) and the association of CHEK2 mutations with family history of colorectal cancer (P = .014; Table 4).

Performance of Genetic Testing Criteria Among Mutation Carriers
Consensus clinical genetic testing guidelines include PC as a component tumor for seven of the confirmed PC genes in this study (BRCA1/2, MSH2, MSH6, ATM, PALB2, and CDKN2A). [27][28][29] Clinical histories of patients with mutations in these genes were evaluated to determine whether the respective genetic testing criteria were met ( Table 5). Although a majority of BRCA1/2 and all MSH2 mutation carriers displayed histories consistent with testing criteria, ≤ 50.0% of ATM, CDKN2A, PALB2, and MSH6 carriers met criteria. In addition, no CDKN2A families met diagnostic criteria for familial atypical multiple mole melanoma syndrome, 30 and 38.9% (seven of 18) did not report any personal or family history of melanoma.

DISCUSSION
Here we report a study of cancer predisposition gene mutations among patients with PC on the basis of a cohort of individuals undergoing hereditary cancer multigene panel testing from a single clinical laboratory. Results from casecontrol studies of the PC cases and ExAC reference controls identified six genes associated with high risk (OR, > 5) of PC (ATM, BRCA2, CDKN2A, MSH6, PALB2, and TP53), consistent with previous smaller studies and segregation studies from PC families. MSH2 was also associated with a high risk of PC; however, additional studies are needed to confirm these findings, because this association was based on a limited number of mutations detected among PC cases. There has been some debate regarding the contribution of BRCA1 mutations to PC risk, because early studies were enriched for founder mutations from Ashkenazi Jewish patients with PC.
Here we show that BRCA1 mutations are associated with a moderate risk (OR, > 2) of PC, even in a series of sensitivity analyses accounting for potential modifying effects of other cancers. CHEK2 mutations were also associated with a moderate risk of PC; however, this association was either diminished (OR, < 2) or nonsignificant in several sensitivity analyses. In addition, the association of CHEK2 with PC was attenuated (OR, 1.64; 95% CI, 1.02 to 2.62; P = .046) when including the common p.I157T variant in the analyses, consistent with the lower penetrance of this alteration. Given the instability of the risk estimates, additional studies are needed to establish the influence of CHEK2 mutations on PC risk. Despite the association of STK11 with high risk of PC, no mutations were detected in this cohort. One likely explanation is that STK11 mutations are unlikely to occur in the absence of pathognomonic clinical characteristics associated with Peutz-Jeghers syndrome, and therefore, patients with suspected Peutz-Jeghers syndrome may be referred for single-gene testing more often than multigene testing. Pathogenic mutations in other panel genes were still sufficiently uncommon to allow assessment of associations with risk (eg, APC, MLH1).
The risk estimates for PC associated with each of these established predisposition genes will help improve clinical PC risk assessment. For some genes, these results offer more precise estimates than previously reported, whereas for others, such as PALB2 and ATM, we are the first to characterize the level of risk, to our knowledge. It should be noted that the interpretation of the risks reported here is specific to patients referred for hereditary cancer genetic testing based on a personal or family history of cancer (at least one diagnosis of PC in the family), and thus, these data may not be applicable to the general population or unselected PC cohorts. Despite the enrichment for cases with personal or family history of cancer, these risks are derived from a broader clinical cancer testing cohort compared with previous studies selected for classic syndromic phenotypes such as FAMMM and therefore demonstrate that PC risk from syndromic genes remains high across a range of clinical histories. Furthermore, this enrichment presented an opportunity to explore predictors of germline mutations.
In total, 13% of patients had mutations in genes significantly associated with increased risk for PC across a range of sensitivity analyses (ATM, BRCA1, BRCA2, CDKN2A, MSH6, PALB2, and TP53). Consistent with results from a previous study of 96 sequentially recruited patients from the Mayo Clinic, 18 90% (158 of 173) of the mutations in the risk-associated genes in this study  ExAC NFE non-TCGA controls were used in this study because of the lack of a large series of matched controls. Although the use of large reference data sets is not ideal, the large sample size allows precise estimation of the frequency of mutations in individuals without cancer and is likely reflective of the general population. In addition, we applied many data cleaning steps and used consistent criteria for selection of mutations in the clinical cohort of patients with PC and the ExAC controls to ensure that the data sets were adequately normalized for case-control association analyses. Another potential limitation of this study is the quality of the clinical history information available for patients with PC. In a recent assessment of the quality of clinical history information for patients undergoing hereditary cancer panel testing, pedigrees and/ or clinic notes were provided for 46% of randomly selected patient cases (unpublished data). When compared with pedigrees and clinic notes, a vast majority of proband cancers were reported completely (95%) and accurately (> 99%) on test requisition forms. Completeness and accuracy remained high (97%) for PCs reported on test requisition forms. Among family members, 76% of melanomas and > 80% of breast, ovarian, colorectal, endometrial, and pancreatic cancers were reported with ≥ 98% accuracy on test requisition forms. Therefore, the variant frequencies and PC risk estimates presented in this analysis were derived from a laboratory-based 8 ascopubs.org/journal/po JCO™ Precision Oncology

Exome Aggregation Consortium Reference Controls
The Exome Aggregation Consortium (ExAC) contains exome sequence data from 60,706 unrelated individuals sequenced as part of various disease-specific and population genetic studies. All of the raw data from these projects were reprocessed through a common pipeline. Principal component analysis was performed to identify population clusters corresponding to individuals of European, African, South Asian, East Asian, and admixed American ancestry. Europeans were separated into individuals of Finnish and non-Finnish European (NFE) ancestry. ExAC also contained patient cases of cancer from The Cancer Genome Atlas (TCGA). Exclusion of sequence data from these patient cases yielded ExAC non-NFE non-TCGA reference controls.

Genome Aggregation Database Reference Controls
The Genome Aggregation Database (gnomAD) contains sequencing data of 123,136 exomes and 15,496 genomes from unrelated individuals sequenced as part of various disease-specific and population genetic studies. The raw sequence data were reprocessed through the same pipeline and jointly variant called to increase consistency across projects. The gnomAD data set contains individuals sequenced using multiple exome capture methods and sequencing chemistries. The resulting variation in coverage was incorporated into the variant frequency calculations for each variant. gnomAD was quality controlled and analyzed using the Hail open-source framework for scalable genetic analysis. gnomAD provides allele frequencies separately for several races and ethnic groups, including non-NFE and Ashkenazi Jewish individuals.

gnomAD Data Cleaning and Filtering
• Restricted to gnomAD NFE exome data combined with gnomAD Ashkenazi Jewish exome data.
• Pathogenic variant classification rules: Same as in ExAC rules 1 to 8.

Review variants with allele count ≥ 15 by Integrative
Genomics Viewer and by frequency in control data from dbSNP.
• Allele number was calculated as average of all variants within the coding region of a gene of interest. This is important for ExAC, gnomAD, and Ambry patient cases, because different numbers of individuals were tested for each variant.
*ExAC controls were restricted to non-Finnish Europeans and also excluded The Cancer Genome Atlas patient cases. †gnomAD was restricted to non-Finnish European and Ashkenazi Jewish controls.       Abbreviations: ExAC, Exome Aggregation Consortium; gnomAD, Genome Aggregation Database; ND, not determined; OR, odds ratio; PC, pancreatic cancer.
*ExAC controls were restricted to non-Finnish Europeans and also excluded The Cancer Genome Atlas patient cases. †gnomAD was restricted to non-Finnish European and Ashkenazi Jewish controls.  *ExAC controls were restricted to non-Finnish Europeans and also excluded The Cancer Genome Atlas patient cases. †gnomAD was restricted to non-Finnish European and Ashkenazi Jewish controls.