Pancreatic cancer as a sentinel for hereditary cancer predisposition

This study was approved by the Institutional Review Board of the University of Utah. All participants gave written consent, which included DNA sampling for molecular studies and access to medical records.

An initial set of pancreatic cancer cases (n = 66) were selected on the minimal requirements of personal history of cancer and having at least two grandparents in the genealogy data represented in the Utah Population Database (UPDB). These patients were screened with a 34-gene custom research panel. Individual family members were then linked to statewide cancer, demographic, and medical information [23]. Ages at diagnosis and family cancer history were obtained from the UPDB after sequencing and variant evaluation. Additional subjects were selected on the basis of being newly diagnosed pancreatic cancer cases ascertained at the HCH from July 2014 to April 2017 (n = 224). The pancreatic cancer cases ascertained during the interval July 2014–November 2015 (n = 151) were screened with a 59-gene custom research panel, and the cases ascertained during the interval December 2015–April 2017 (n = 73) were screened with a 14-gene custom clinical panel.

For research panel testing, blood-derived genomic DNA (100 ng) was sheared using a Covaris S2 instrument (Covaris, Woburn, MA, United States). Genomic libraries were prepared using the Ovation Ultralow Library System (NUGEN # 0329) according to the manufacturer’s instructions. Library enrichment for a 34 or 59-gene custom panel was done with the Roche SeqCap EZ Choice Library (cat# 06266339001) and the SeqCap EZ Reagent Kit Plus v2 (NimbleGen #06–953–247-001) using the manufacturer’s protocol. Individual libraries were combined into pools of 6–12 prior to hybridization, and then super-pooled for up to 96 samples per sequencing lane. Captured libraries were sequenced on an Illumina HiSeq2000 channel using the HiSeq 101 Cycle Paired-End sequencing protocol.

On the strength of preliminary data from this study, HCH began systematically offering clinical panel predisposition testing beginning December 2015, without regard to family history. From December 2015 to April 2017, clinical testing was offered to 73 sequential pancreatic cancer cases. Sixty-one pancreatic cancer patients accepted clinical testing with the 14-gene custom panel that was conducted by Invitae. The 12 individuals that declined were in poor health and/or did not see value in undergoing a genetic test. A complete list of genes captured is included in Additional file 1: Table S1.

Sequences from the Utah cohort with ≥100X mean coverage and 154 pancreatic cancer cases from the Cancer Genome Atlas (TCGA) [24] were analyzed using the USeq (useq.​sourceforge.​net) in-house pipeline, according to the Genome Analysis Toolkit (GATK v.3.3–0) best practices recommendations [25]. Variants with a mapping quality score less than 20 were excluded. ANNOVAR was used for variant functional annotation followed by conversion to Human Genome Variation Society (HGVS) nomenclature using Mutalyzer [26, 27].

Truncating variants not present in the final exon of a gene were considered pathogenic. The following filters were used to exclude variants from further analysis: minor allele frequency ≥ 0.1% in one or more populations from the Exome Aggregation Consortium (ExAC) database; [28] synonymous/intronic variants with no predicted effect on splicing via MaxEntScan; [29] variants reported as probable-non-pathogenic/non-pathogenic by more than one source with no conflicting reports in ClinVar (www.​ncbi.​nlm.​nih.​gov/​clinvar).

Variants of uncertain significance (VUS) were included if in silico predictions were suggestive of being relatively high probabilities of pathogenicity VUS (HiP-VUS). HiP-VUS had estimated prior probabilities of pathogenicity > 0.8 based on calibrated in silico predictions from publicly available databases for the mismatch repair (MMR) genes (hci-lovd.​hci.​utah.​edu), or BRCA1/2 (http://​priors.​hci.​utah.​edu/​PRIORS/​). HiP-VUS of this type were weighted according to their sequence analysis-based prior probability of pathogenicity (Prior_P) score. The VUS from the remaining genes were denoted HiP-VUS and included if at least three of the four missense analysis programs Align-GVGD, MAPP, Polyphen-2, and CADD predicted a severe score [30‐34]. This filter corresponds with an OR = 3.27 when comparing early-onset breast cancer cases with matched controls [33]. Based on the likelihood ratios identified for BRCA1/2, [35] this grouping was assigned a weight of 0.81. Canonical splice acceptor/donor variants predicted to impact splicing were given the weight of 0.97 if the effect of the variant had not been demonstrated experimentally [36]. Pathogenic variants and HiP-VUS detected by the 34−/59-gene panels were confirmed via Sanger sequencing. VUS reported by the Mayo Clinic [37] plus the non-TCGA ExAC (excluding the Finnish and undescribed populations), were graded with the same weights and severity to generate a bioinformatically equivalent set of HiP-VUS. An overview of the datasets and methods used for evaluation is shown (Fig. 1).

STATA V.13.1 (StataCorp, College Station, Texas, USA) was used to conduct meta-analyses, and calculate carrier percentages and 95% confidence intervals. The meta-analyses to compare the carrier frequencies between different pancreatic cancer cohorts were conducted using Metaprop under a random effects model, and Freeman-Tukey transformation to stabilize the variances over the studies [38]. The weighted proportions of variant carriers in the unselected pancreatic cancer cases were compared to the corresponding proportions in the non-TCGA ExAC population to estimate Standardized Incidence Ratios (SIR) [39]. Tests of significance and confidence intervals were estimated based on a Poisson distribution [40]. For the meta-analysis and SIR calculation, the genes were split into subgroups of high- and moderate-risk cancer susceptibility genes. High- and moderate-risk were defined as genes with a cumulative risk at age 80 > 32% or between 19 and 32%, respectively, for the cancer with which they are most closely associated [2]. The R package ggplot2 was used to plot the meta-analyses and SIRs [41].

In an initial set of 66 pancreatic cancer cases unselected for family history of cancer, 4 pathogenic variants were identified in BRCA2, MSH6, PALB2, and STK11. After filtering VUS, 2 HiP-VUS in ATM remained (Table 1). After weighting, 8.5% of these pancreatic cancer cases carried a variant with potential medical management impact for relatives.

Two series of cases were used for internal replication. From a set of 147 pancreatic cancer cases undergoing an in-house custom 59-gene panel, 6 pathogenic variants were identified in ATM, BRCA1, BRCA2, MRE11A, and PALB2, and 6 HiP-VUS in ATM, BRCA2, CHEK2, MSH6, and TP53. In addition to in silico predictions, CHEK2 p.(T476 M) was found to be damaging in a functional assay for CHEK2 variants, and was thus weighted more strongly towards being pathogenic [42]. Subsequently, an additional set of pancreatic cancer cases (n = 61) were tested with the 14-gene clinical panel, which identified 6 pathogenic variants and 2 HiP-VUS in 7 patients (Table 1). All pathogenic variants and HiP-VUS were identified in genes included in the original 34-gene panel. After weighting carriers, 8.6% of the pancreatic cancer cases from the latter two sets, unselected for family history, carried a variant with potential medical impact. A full list of rare variants with corresponding in silico information is available in Additional file 1: Table S2.

In order to estimate the proportion of pancreatic cancer cases with pathogenic variants or HiP-VUS that would have qualified for genetic testing, the family histories were compared to NCCN guidelines [43]. Pedigree data regarding familial cancer were available from UPDB for the carriers of pathogenic variants and HiP-VUS in 6 of the initial set of 66 pancreatic cancer patients (Additional file 2: Figure S1). Self-reported family history information was also available for the final set of 61 patients that underwent the Invitae clinical panel, including the 7 carriers of pathogenic variants or HiP-VUS. Among the 13 identified carriers, the STK11 carrier had a clinical diagnosis of Peutz-Jegher syndrome with multiple affected relatives, and 9 additional patients met criteria for HBOC or Lynch syndrome genetic testing. Thus, more than 23% of these carriers would not have qualified for testing under current guidelines.

Once a pathogenic variant is identified, cascade testing for that variant can occur on biological relatives to identify individuals who would benefit from HBOC or CRC preventive measures. For three of the five pancreatic cancer cases with positive results with the clinical panel, 17 biological relatives have undergone cascade genetic testing thus far, 10 of whom have tested positive for the family pathogenic variant. Of note, 21 (29%) pancreatic cancer cases who underwent the clinical panel passed away since the beginning of the study, 17 of whom had undergone testing. This suggests that having a family member(s) present during pre-test counseling and delegated to receive results may be beneficial in the context of utilization of pancreatic cancer as a sentinel for HBOC or Lynch syndrome.

The HCH sets of pancreatic cancer cases were combined with a published study of unselected pancreatic cancer cases from the Mayo Clinic (n = 96), [37] plus the pancreatic cancer cases from TCGA (n = 154), in a meta-analysis (Fig. 2; Additional file 1: Table S3; Additional file 3: Table S4). Among unselected pancreatic cancer cases, 3.9% (p = 2.1 × 10^− 13) carried a clearly pathogenic variant in a high-risk cancer susceptibility gene which includes pancreatic cancer in its tumor spectrum [8]. Weighed inclusion of HiP-VUS increased the proportion to 5.1% (p = 6.8 × 10^− 18). For the moderate-risk homologous recombination repair (HRR) breast cancer genes ATM, BARD1, CHEK2, and NBN, [2, 44] 3.3% (p = 7.2 × 10^− 04) of unselected pancreatic cancer cases carried a clearly pathogenic variant, and weighted inclusion of HiP-VUS increased the proportion to 5.1% (p = 2.3 × 10^− 08). Including all the high-risk genes and the moderate-risk HRR breast cancer genes, 7.9% (p = 1.4 × 10^− 12) of unselected pancreatic cancer cases carry a clearly pathogenic variant, and 10.4% (p = 9.1 × 10^− 17) carry either a clearly pathogenic variant or weighted HiP-VUS with elevated probability of pathogenicity that could enable the at-risk relatives to qualify for preventive HBOC or CRC measures.

Further, the gene burdens observed in the Utah, Mayo, and TCGA were compared to the non-TCGA ExAC (n = 49,451, excluding the Finnish and other subpopulations) as a population sample to determine SIR for subgroups of genes (Fig. 3, Additional file 3: Table S5). As a group, the high-risk susceptibility genes had a SIR = 2.6 (p = 1.6 × 10^− 05). The moderate-risk HRR genes had a slightly lower SIR = 2.3 (p = 2.0 × 10^− 05).