Germline DNA Sequencing Reveals Novel Mutations Predictive of Overall Survival in a Cohort of Patients with Pancreatic Cancer

Familial pancreatic cancer (PC), defined as having two or more first-degree family members who have been diagnosed with pancreatic cancer, is thought to account for 5% to 10% of pancreatic adenocarcinoma (1). Previous studies of mutation prevalence in familial pancreatic cancer, however, have focused on a limited set of genes, mostly dedicated to those genes within the DNA damage repair (DDR) pathway (ATM, BRCA1, and BRCA2) or related to hereditary Lynch syndrome (MLH1, MSH2, MSH6, PMS2; refs. 2–7). The National Comprehensive Cancer Network (NCCN) currently endorses genetic counseling for all patients with pancreatic cancer (8). A provocative family history of cancer or Ashkenazi Jewish heritage may prompt some physicians to order further genetic testing. Recent studies have shown that pancreatic cancer family history is not predictive of germline mutations (7, 9), highlighting the value of broad sequencing in unselected groups. Specific genes are frequently tested on the basis of known cancer syndromes (10). The prevalence of mutations among patients with pancreatic cancer, who are unselected for specific risk factors such as age at diagnosis or family cancer history, is a current topic of interest amongst oncologists. Next-generation sequencing enables testing for both commonly described familial mutations, as well as rarely described variants.

We previously reported family history as a biomarker for survival in pancreatic cancer (11). We found that patients with a strong family history of BRCA-related cancers (three or more first- to third-generation relatives with breast, ovarian, or pancreas cancer) had an overall survival (OS) nearly double of those with no such family history. Family history, however, is a subjective measure and surrogate for the underlying disease biology leading to predisposition to pancreatic cancer in families.

In addition, despite evidence for a predictive benefit to platinum agents in familial PC, most practitioners use performance status as the main determinate for treatment regimen decisions. Patients with metastatic pancreatic cancer with poorer performance status (PS > 1) often receive treatment with gemcitabine plus nab-paclitaxel (gem/nab-paclitaxel), rather than the more difficult to tolerate platinum-based regimen 5-fluoruracil, irinotecan, and oxaliplatin (FOLFIRINOX; refs. 12, 13).

In this study, we assessed the prevalence of deleterious germline mutations using a 263-gene panel in a population of patients with pancreatic cancer, treated with current standard-of-care (SOC) chemotherapy, who presented to a large academic cancer institution. We also used a secondary cohort, the pancreatic cancer genome atlas (TCGA) research network germline data, to confirm the validity of some of our findings (14). Our goal was to identify additional underlying genomic alterations leading to familial pancreatic cancer in an unselected cohort including their link to family history, clinical implications, response to SOC treatment, and survival outcomes.

All patients with metastatic pancreatic adenocarcinoma who were seen at MD Anderson Cancer Center (MDACC, Houston, TX) between January 2010 and January 2016, and who received first-line SOC chemotherapy, were eligible for chart review. Cases were identified retrospectively. All patients had consented for DNA banking for clinical research. Samples were obtained from the MDACC pancreas cancer tissue bank and the Center for Translational and Public Health Genomics at Duncan Family Institute of MDACC and Patient History database Program. Clinical and pathologic information was abstracted from chart review from the MDACC electronic medical record including cancer stage, cancer histology, family history, and record of clinical genetic testing. Family history was defined as first- through third-degree (out to first cousins) relatives with breast, ovarian, or pancreatic cancer diagnosed at any age. The gastrointestinal pathologists in the MDACC Department of Pathology and Laboratory Medicine reviewed all pancreatic cancer tumors. All patients consented to DNA banking and clinical research. Of the 233 patients identified, blood samples for germline DNA testing were available from 133 cases including 95 and 38 patients treated with first-line FOLFIRINOX and gem/nab-paclitaxel, respectively. When available, outside germline mutation testing results were incorporated into our database. As specimens were used for research purpose alone, testing was not performed under Clinical Laboratory Improvement Amendments regulations. Results were not returned to the families of study participants or used to make clinical decisions. The study was conducted under the auspice of the MDACC Institutional Review Board. The study conformed to the Declaration of Helsinki. Informed written consent was obtained from each subject. A consort diagram is available in Fig. 1.

Deidentified germline genomic data was obtained from TCGA, pancreatic adenocarcinoma cohort. DNA sequencing data as well as a limited amount of clinical information was collected from all 127 patients.

Peripheral blood leukocytes were used to collect germline DNA. DNA was extracted using a QIAGEN DNA Extraction Kit (Qiagen Inc.). Targeted panel capture was performed on 500 ng of genomic DNA per sample based on KAPA library prep (KAPA Biosystems) using the T200.1 Solid Tumor Cancer gene panel (263 genes) and paired-end multiplex sequencing of samples was performed using the Illumina HiSeq 2500 sequencing platform (15). Average coverage was 450×.

Paired-end reads in FastQ format generated by the Illumina platform were aligned to the human reference genome (UCSC Genome Browser, hg19) using Burrows-Wheeler Aligner (16) by allowing three mismatches with 2 in the first 40 seeding regions and aligned reads processed using the GATK Best Practices of duplicate removal, indel realignment, and base recalibration (17). Germline single-nucleotide substitutions were detected using Platypus (18). The following filtering criteria were used: (i) total read coverage of the variant > = 20, (ii) the variants should be on-target, exonic, and nonsilent, (iii) a population frequency threshold of 1% was used to filter out common variants in the databases of dbSNP129 (PMID: 11237013), 1000 Genomes Project (PMID: 23128226), Exome Aggregation Consortium (Exac; PMID: 27535533), and ESP6500 (PMID: 23201682). Specifically, variants with a population frequency larger than 3% in any individual ethnicity group in Exac were also filtered out. A schema of variant calling and number of variants seen at each filtering step are provided in Fig. 2A.

Gene variants deemed deleterious were considered mutations as were private, nonsilent variants not filtered out by the criteria shown in Fig. 2A. Those variants suspected as deleterious without evidence of validation in the literature were classified as variants of unknown significance (VUS). Variants were classified using American College of Medical Genetics and Genomics recommendations. Supporting data was obtained from literature review using linkage, biochemical, clinical, functional, and statistical research for specific alterations (19, 20). Variants were cross-referenced to Clinvar, HGMD, and UMDBRCA genomic data banks for further variant determination (21–23).

Patient clinical characteristics of categorical variables such as race and gender are reported given frequencies and percentages. Continuous data and sequencing results were summarized with descriptive statistics. The χ² test and Fisher exact test were used to evaluate the association between two categorical variables. Wilcoxon rank sum test was used to compare the distributions of continuous variables between two different groups. Univariate and multivariable logistic regression models were used to evaluate the association between the outcomes of interest and patient demographic covariates. All tests are two-sided and P values < 0.05 are considered statistically significant. All analyses were conducted using SPSS version 24 (SPSS).

We identified 233 patients that met our study cohort selection criteria as shown in Fig. 1. The median and mean ages at diagnosis of the entire cohort were 62 and 61 years (range, 36–84 years), respectively. 136 patients were male (58.4%) and all patients were metastatic at diagnosis. 155 patients had a baseline ECOG score of 0–1, 12 with 2, and in 15 patients, the score was unknown. 178 (76.4%) patients had liver metastases at diagnosis. Nineteen patients had secondary cancers (8.2%) that included breast (n = 3), colorectal cancer (n = 3), lung (n = 3), prostate (n = 3), bladder (n = 2), ovarian (n = 1), melanoma (n = 1), and various hematologic malignancies (n = 3), as well as other cancers. Four (1.7%) of these patients had multiple cancers including two patients with lung cancer, both heavy smokers, one with bladder cancer, and one with squamous cell carcinoma of the maxilla. Blood samples for germline DNA testing were available from 133 cases. The median and mean ages of the sequenced cohort at diagnosis were 61 and 60 years (range, 36–84 years), respectively. Sequenced patients were more likely to have received first-line FOLFIRINOX than gem/nab-paclitaxel (95 vs. 38 patients, P = 0.03). Otherwise, no other significant differences were observed between the two groups. A summary of these patients is presented in Table 1. A comparison of the sequenced versus unsequenced patients is found in Supplementary Table S1.

Of the 133 sequenced patients, 29 deleterious mutations were found in 26 (19.6%) patients. Of those patients with a deleterious mutation, the median and mean ages were 56 and 55 years (range, 36–70 years), respectively. Seventy-eight (58.6%) patients were male, 111 (84%) were Caucasian, and 7 (5.3%) were of Ashkenazi Jewish descent. Of the 6 (4.5%) sequenced patients with a family history of 3 or more first- through third-degree BRCA-related relatives, 3 (50%) were found to have a deleterious mutation, and one of these patients had two inherited mutations. Conversely, of the 26 patients with a deleterious mutation and 2 (7.7%) patients had no family history of cancer. There were 21 patients with 0–1, 2 patients with 2, and 3 patients with 3 affected first-, second-, or third-degree relatives with a BRCA-associated cancer (breast, ovarian, or pancreas). Twenty-two of the 26 patients received first-line therapy with FOLFIRINOX. Clinical and tumor pathologic features for the patients with a deleterious mutation are provided in Table 2.

Of all sequenced patients, 4 (3.0%) had a BRCA1 (n = 2) or BRCA2 (n = 2) germline mutation. In addition, 6 (4.5%) patients had deleterious mutations in other pancreatic cancer predisposition genes including ATM (n = 3), CDKN2A (n = 1), PALB2 (n = 1), and TP53 (n = 1). Thus, 10 (7.5%) patients had an inherited mutation in a known pancreatic cancer predisposition gene. Six (4.5%) patients had mutations in other DDR genes including ERCC4 (n = 2) and the checkpoint gene, CHEK2 (n = 4). Nine patients had other mutations in cancer-associated genes including TERT (n = 3), AR (n = 2), CYP2C19 (n = 2), NF1 (n = 1), and SDHD (n = 1). One IL7R mutation was identified in a male patient with a deleterious frameshift deletion in CHEK2, one patient had both a HNF1A mutation and a RET mutation, and one patient with a deleterious CDKN2A frameshift insertion had an IL7R mutation as well. In total, 18 (13.5%) patients had a mutation in a gene associated with cancer risk. No patients carried definitive, pathologic, deleterious mutations in POLE, STK11, or Lynch-related genes, although VUSs were found in these genes (Supplementary Table S2). Table 3 lists the 29 mutations identified, mutation effect, and associated patient characteristics.

We identified at least one VUS in 123 (93.2%) patients, with as many as 10 variants found per patient. On average, patients had 2.6 variants per patient (Fig. 2B). The most commonly seen VUSs were in FANCA [n = 8, (6%)], NOTCH1 [n = 8, (6%)], ERCC4 [n = 7, (5%)], and FGFR4 [n = 7, (5%)], respectively. Seven patients had VUSs in known pancreatic cancer predisposition genes including 2 (1.5%) patients with PALB2 variants, 2 (1.5%) patients with PMS2 variants, and 3 (2.3%) patients with POLE variants, including two recurrent variants. All VUSs identified are listed in Supplementary Table S2.

The median OS from date of diagnosis of the 133 sequenced patients was 10.0 months [95% confidence interval (CI): 8.5–11.5 months]. Of the sequenced patients with a family history of 0–1, 2, or 3 or more first-, second-, or third-degree relatives with a BRCA-associated cancer (breast, ovarian, or pancreas), the median OS was 9.6, 8.4, and 23.7 months, respectively (P = 0.13, Fig. 3A). The median OS in patients with versus without a deleterious mutation was 10.2 versus 9.9 months (P = 0.25, Fig. 3B). The median OS of the 10 patients with a deleterious DDR (ATM, BRCA1, BRCA2, ERCC4, PALB2) gene mutation versus other sequenced patients was significantly longer than patients without a DDR mutation (17.9 vs. 9.6 months, P = 0.03, Fig. 3C). There was no difference in survival for those patients who were 60 years old or less compared with those over 60 (10.0 vs. 9.4 months, P = 0.91, data not shown). The median OS from the date of the start of chemotherapy was 9.6 versus 8.9 months (P = 0.47, Fig. 3D), in those sequenced patients who were treated with FOLFIRINOX versus gem/nab-paclitaxel, respectively.

On univariate analysis, ECOG PS > 1, family history of pancreatic cancer, family history of any cancer, and presence of 3 or more affected family members with a BRCA-associated cancer, were all significant determinants of survival (Supplementary Table S3). However, on multivariate analysis, ECOG PS >2 (HR 2.37; 95% CI, 1.28–4.37; P = 0.006) and a history of 3 or more family members with BRCA-associated cancers remained significant (HR 0.55; 95% CI, 0.32–0.93, P = 0.03; Table 4). Univariate analysis of all 233 patients is provided in Supplementary Table S4; however, the only significant predicting factor on multivariate analysis was ECOG PS (data not shown).

Of the patients who underwent sequencing, there were 0 complete responses (CR), 37 partial responses (PR), 49 patients with stable disease (SD), 33 with progressive disease (PD), and 14 that were not evaluable due to lack of follow up imaging or other reasons (Table 1). The overall response rate (ORR) was 27.8% (37 of 133 patients). The ORR of patients without deleterious mutations was 29% (31 of 107 patients). Of the 26 patients with deleterious mutations, 22 were treated with FOLFIRINOX and 4 patients were treated with gem/nab-paclitaxel (Supplementary Table S5). There were 0 CRs, 6 PRs, 8 SDs, 8 PDs, and 4 were not evaluable with an ORR of 23.1% (6 of 26 patients). However, of the patients with DDR mutations, there were 0 CRs, 5 PRs, 6 SDs, 3 PDs, and 1 was not evaluable (ORR = 33.3%, 5 of 15 patients) compared with 0 CRs, 1 PR, 1 SD, 4 PD, and 2 who were not evaluable in those without DDR mutations (ORR = 12.5%).

The prevalence of deleterious mutations decreased with increasing age at pancreatic cancer diagnosis, with frequencies of 41.7% (5/12), 20.4% (11/54), and 14.9% (10/67) for age groups younger than 45 years, 45 to 60 years, and older than 60 years, respectively. In a univariate analysis, age less than 45 nearly reached statistical significance (OR 2.94; 95% CI, 0.88–9.91; P = 0.08; Supplementary Table S6). Nearly half of all sequenced patients (56/133) had a family history of a BRCA-related cancer. There was a trend among those patients with three or more relatives with a BRCA-related cancer to have a deleterious mutation; however, this did not reach the level of statistical significance (OR = 4.53; 95% CI, 0.86–23.85; P = 0.08; Supplementary Table S5). Of all sequenced patients, 6.8% (9/133) had a personal history of an additional malignancy, including 3 patients with breast cancers. Personal history, however, did not predict for deleterious mutations (P = 0.96, Supplementary Table S6). There was a trend toward increased number of mutations in those with Ashkenazi Jewish heritage, although this did not reach statistical significance on univariate analysis. Other correlates are presented in Supplementary Table S5.

On multivariate analysis, only recurrent VUS was associated with a deleterious mutation (OR = 0.20; 95% CI, 0.06–0.62; P = 0.005; Supplementary Table S7).

We additionally performed a similar query with the limited clinical data and germline sequencing data from the pancreatic cancer patients in TCGA. Median and mean age at diagnosis for this cohort was 61 (35–88 range, years) and sixty-five (51.2%) were male. Using the same variant calling pipeline (Fig. 2A), we identified 26 deleterious mutations in 24 (18.9%) patients of the 127 patients tested. There were no BRCA1 mutations and 2 (1.6%) BRCA2 mutations. Five (3.9%) patients had mutations in ATM, three (2.4%) patients had mutations in MUTYH, two patients each (1.6%) had mutations in AR, CHEK2, DNMT3A and one (0.8%) patient each had a mutation in CDKNA, CYP2C19, ERCC2, FANCA, IL7R, MEN1, NBN, PALB2, RET and TERT. Recurrent deleterious mutations that overlapped with our own study cohort were seen in 6 (4.7%) patients (Table 3). The median OS from date of diagnosis was 19.8 months in this cohort. We see that the median survival of patients with a DDR mutation exceeds that of those without DDR mutations (24.2 vs. 19.7 months, P = 0.024, data not shown).

Our current understanding of the prevalence of familial pancreatic cancer is largely limited to twelve genes previously described by Grant and colleagues in the Journal of Gastroenterology in 2015 (2). The incidence of damaging mutations in these genes, including ATM, BRCA1, BRCA2, and TP53 is thought to occur in 5% to 10% of patients with pancreatic cancer and is mainly limited to particular genes in the DDR pathways. In this study, we have evaluated for a larger variety of mutations that may predispose to familial pancreatic cancer and found a mutation rate of almost 20% when testing 263 cancer-associated genes in this unselected cohort.

Here, we have also provided additional evidence that family history of 3 or more BRCA-related cancers is a determinate of OS; however, this was not statistically significant. The small number of patients in this group and the subjective nature of family history may contribute to the lack of significance. Interestingly, patients with mutations in genes involved in DDR or the cell-cycle checkpoint (ATM, BRCA1, BRCA2, CDKN2A, CHEK2, PALB2, and ERCC4) had a near doubling of the OS (17.9 vs. 9.6 months, P = 0.03) compared with those that did not have a mutation in one of these genes. Perhaps not all deleterious mutations are beneficial. As seen in Table 3, there is a trend toward better outcomes in those with BRCA mutations than those with ERCC4.

Familial pancreatic cancer due to mutations in mismatch repair genes may demonstrate overlapping tumor biology with those with DNA repair defects. We did not find any pathologic mutations in MMR genes other than 2 VUSs in PMS2. In previous studies of familial pancreatic cancer, the estimated incidence of MMR gene mutations was very low at 0.1%–1% (24, 25). With a larger study cohort, we may have observed mutations in these genes. Future study may also include a broader range of genes related to DNA repair such as MREIIA, NBN, and BARD1.

The frequency of each DDR mutation in the TCGA cohort differs from our own discovery cohort. This is consistent with findings from Johns Hopkins University (Baltimore, MD) that showed a large amount of heterogeneity among patients with familial pancreatic cancer (26). However, when taken as a whole, the TCGA patients with DDR mutations also demonstrated prolonged OS (numbers), confirming our own findings. In 2019, a group at Memorial Sloan Kettering Cancer Center published a study in the journal, Genetics in Medicine, where they found a 10% incidence of germline pathogenic mutations in a cohort of patients with familial pancreatic cancer (25). They tested a larger cohort of patients, but used a smaller panel of genes to test both germline and somatic mutations. Importantly, we confirm that patients with pathogenic germline defects in DNA repair genes have a better overall survival. This may argue for a biologically based survival benefit in these patients either due to improved treatment response or less aggressive tumor biology. In the future, we plan to compare germline and somatic variants in our patient cohort.

In the clinic, decisions whether to treat patients with either of the current SOC regimens, FOLFIRINOX or gem/nab-paclitaxel, is largely based on performance status. Those in better condition tend to receive the more difficult to tolerate, yet arguably more effective, FOLFIRINOX. We did find that the majority of patients identified as carrying deleterious mutations in predisposition genes were treated with first-line FOLFIRINOX (85%). Of the 4 patients with mutations that received first-line gemcitabine and nab-paclitaxel, only two were evaluable for response. This may be due to the fact that FOLFIRINOX had become SOC by 2011 for first-line metastatic pancreatic cancer and it was not until 2013 that gem/nab-paclitaxel became FDA approved. In those with BRCA-mutated cancers, previous predictive data has largely been based on treatment with cisplatin, a chemotherapeutic agent that is less frequently used with current SOC chemotherapy regimens (27). Because of this widely held perspective, patients with a stronger family history, which may be more likely to represent those with a deleterious mutation, were placed on FOLFIRINOX. If patients with deleterious mutations in DDR genes have a better prognosis, they may be in better physical condition at presentation, and given the more difficult-to-tolerate regimen by their oncologist. Unfortunately, we did not have enough patients to characterize the effectiveness of these regimens among patients with inherited mutations.

One of the limitations of this study is that the vast majority of sequenced patients were deceased. This introduces a retrospective bias, as patients who lived longer might not undergo genetic testing. Therefore, the actual overall survival of mutated patients may be longer than what we report in this study. We would ideally perform a prospective study looking at an unselected pancreatic cancer cohort where all-comers are sequenced for germline mutations. In addition, mutations, although predicted to be deleterious, may not necessarily be causative of pancreatic adenocarcinoma. Furthermore, in this study, we focused on metastatic patients. We therefore, cannot speak to the germline mutation spectrum of resectable or locally advanced patients.

Only 13.5% of sequenced patients underwent formal genetic screening. Of those that underwent screening, mutations were found in 50% (9/18) of these patients. This was due to lack of follow through in sending blood for sequencing or a determination that sequencing was not necessary. We additionally compared our profiling with commercial tests. Two commercial labs, which omit testing for ERCC4 and CHEK2, would have missed 6 patients with DDR mutant cancers. The age of these affected patients ranged from 50 to 66. As these mutations occurred in patients over the age of 45, their omission would have potentially failed to predict cancers in relatives. The prevalence of ERCC4 and CHEK2 mutations among patients 50 and older suggests that commercial platforms should be expanded to include these genes. This also demonstrates the need for a more universal approach to evaluating familial pancreatic cancer.

In conclusion, we have better characterized the relationship of moderate penetrance germline mutation to familial pancreatic cancer. We have shown that deleterious mutations in either DDR genes or cell-cycle checkpoint genes confer a better prognosis over those without these mutations. Available commercial testing omits genes that may confer prognostic information. In addition, this study highlights the need for continued family counseling, preventative imaging, and early detection in those unaffected carriers. Further prospective studies can address these critical questions.

M.M. Javle is a consultant/advisory board member for Taiho, QED, Merck, and EMD Serono. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J.B. Goldstein, M.M. Javle, F. McAllister, D.R. Fogelman

Development of methodology: J.B. Goldstein, M.M. Javle, D.R. Fogelman

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.B. Goldstein, Y. Ghelman, M.M. Javle, R.T. Shroff, G.R. Varadhachary, R.A. Wolf, D.R. Fogelman

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.B. Goldstein, L. Zhao, X. Wang, M.J. Overman, F. McAllister, A. Futreal, D.R. Fogelman

Writing, review, and/or revision of the manuscript: J.B. Goldstein, L. Zhao, X. Wang, M.J. Overman, R.T. Shroff, F. McAllister, A. Futreal, D.R. Fogelman

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.B. Goldstein, Y. Ghelman

The authors thank MDACC pancreas cancer tissue bank and the Center for Translational and Public Health Genomics at Duncan Family Institute of MDACC and Patient History database Program. This work was supported in part by the Welch Foundation's Robert A. Welch Distinguished University Chair Award (G-0040).

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.