Multiple rare variants in high-risk pancreatic cancer related genes may increase risk for pancreatic cancer in a subset of patients with and without germline CDKN2A mutations

Abstract

The risk of pancreatic cancer (PC) is increased in melanoma-prone families but the causal relationship between germline CDKN2A mutations and PC risk is uncertain, suggesting the existence of non-CDKN2A factors. One genetic possibility involves patients having mutations in multiple high-risk PC-related genes; however, no systematic examination has yet been conducted. We used next generation sequencing data to examine 24 putative PC-related genes in 43 PC patients with and 23 PC patients without germline CDKN2A mutations and 1001 controls. For each gene and the four pathways in which they occurred, we tested whether PC patients (overall or CDKN2A+ and CDKN2A− cases separately) had an increased number of rare nonsynonymous variants.

Overall, we identified 35 missense variants in PC patients, 14 in CDKN2A+ and 21 in CDKN2A− PC cases. We found nominally significant associations for mismatch repair genes (MLH1, MSH2, MSH6, PMS2) in all PC patients and for ATM, CPA1, and PMS2 in CDKN2A− PC patients. Further, nine CDKN2A+ and four CDKN2A− PC patients had rare potentially deleterious variants in multiple PC-related genes. Loss of function variants were only observed in CDKN2A− PC patients, with ATM having the most pathogenic variants. Also, ATM variants (n=5) were only observed in CDKN2A− PC patients with a family history that included digestive system tumors. Our results suggest that a subset of PC patients may have increased risk because of germline mutations in multiple PC-related genes.

INTRODUCTION

Germline mutations in CDKN2A, the major known high-risk melanoma susceptibility gene, have been described in 20%–40% of familial melanoma kindreds.(Goldstein et al. 2006) Several features have been shown to be associated with an increased frequency of CDKN2A mutations, most notably the occurrence of pancreatic cancer (PC) in a family.(Bergman and Gruis 1996; Borg et al. 2000; Ghiorzo et al. 2004; Goldstein et al. 2006; Goldstein et al. 1995; Lynch et al. 2002; Vasen et al. 2000) At present, though, it is unclear what additional genetic, intrinsic, or extrinsic factors predispose individuals in these families to PC. Further, although specific CDKN2A mutations have been associated with PC, many melanoma-prone families with these mutations do not have an excess of PC. These observations suggest that other factors such as mutations in other genes, or tobacco or other carcinogenic exposures may be important in the development of pancreatic cancer.(Goldstein 2004; Goldstein et al. 2006; Helgadottir et al. 2014) Further, since PC is such a rapidly fatal cancer, it is important to learn whether additional genetic alterations make individuals more susceptible. Therefore, several research groups that study familial melanoma and have multiple PC families with and without CDKN2A mutations collaborated to investigate whether these patients have mutations in multiple high-risk PC-related susceptibility genes.

Twenty-four putative high-risk susceptibility genes for familial PC have been identified (Supplemental table 1).(Grant et al. 2015; Klein 2013) Many of these genes are part of autosomal dominant (AD) hereditary cancer syndromes such as Peutz-Jeghers, breast-ovarian cancer, and Lynch syndrome (comprising the mismatch repair [MMR] genes) of which PC has been proposed to be a component cancer. In addition, several genes (e.g. ATM, PALB2, FANCA, FANCC, XRCC2) are mutated in both autosomal recessive (AR) and AD disorders; for example, diseases such as ataxia telangiectasia and Fanconi anemia require two mutations for disease manifestation (bi-allelic) yet later onset cancers such as breast and pancreatic cancer are inherited in a mono-allelic AD pattern. Pancreatic cancer risk is also increased among patients with hereditary/chronic pancreatitis. Since PC patients with CDKN2A mutations have not been assessed for mutations in these genes, we used exome sequencing data to systematically examine these 24 susceptibility genes in pancreatic cancer patients with CDKN2A mutations. In addition, we also evaluated additional available PC patients without CDKN2A mutations but with a family history of cancer, primarily melanoma or digestive system tumors, that were accrued by the same research groups.

MATERIALS AND METHODS

Study Population

The PC patients included in this study came from ongoing studies conducted in the United States, Italy, the Netherlands, and Sweden. Details of the source populations, patients, CDKN2A mutation status, and study references are presented in Supplemental table 2. All diagnoses of PC were confirmed by review of pathology reports, medical records, or death certificates. Only deceased PC patients with available blood DNA were included in this study. Each study was approved by its local Institutional Review Board and informed consent was obtained for all participants. All PC patients from a family with a CDKN2A mutation carried their respective family's CDKN2A mutation. Supplemental table 3 summarizes the sample population by geographic origin and CDKN2A mutation status.

Sequencing Methods

Whole exome sequencing

Whole exome sequencing (WES) for the PC patients was performed at the Cancer Genomics Research Laboratory, National Cancer Institute (CGR, NCI), as previously described.(Shi et al. 2014) Briefly, SeqCAP EZ Human Exome Library v3.0 (Roche NimbleGen, Madison, WI) was utilized for exome sequence capture. The captured DNA (1.1ug) was then subject to paired-end sequencing utilizing the Illumina HiSeq2000 sequencer for 2 × 100-bp sequencing of paired-ends (Illumina, San Diego, CA). For this sample set WES was performed such that 91% of coding sequence from the University of California, Santa Cruz (UCSC) human genome (hg) 19 transcripts database had >15 reads with average coverage of 71×. The exome data for this paper are archived in the CGR exome build Ensemble_New_Annotation dated 2015-10-28.

Bioinformatics Analysis

Alignment and calling of variants

Details of the bioinformatics pipeline for variant alignment and calling used in this study have been previously published.(Shi et al. 2014) Briefly, sequencing reads were first trimmed using the Trimmomatic program (v0.32).(Lohse et al. 2012) Only read pairs with both ends ≥36 bp were used. Reads were then aligned to the hg19 reference genome using Novoalign software (v3.00.05) (http://www.novocraft.com). Duplicate reads were removed using the MarkDuplicates module of Picard software (v1.126) (http://picard.sourceforge.net/). Additionally, two ends of each pair had to map to the reference genome in complementary directions and reflect a reasonable fragment length (300+/−100 bp). A local realignment around sites of insertion and deletion was performed using the RealignerTargetCreator and IndelRealigner modules from the Genome Analysis Toolkit(DePristo et al. 2011) (GATK v3.1).

Variant discovery and genotype calling of multi-allelic substitutions, insertions and deletions were performed on all individuals globally using the UnifiedGenotyper and HaplotypeCaller modules from GATK as well as the FreeBayes variant caller (v9.9.2). An Ensemble variant calling pipeline (v0.2.2 http://bcb.io/2013/02/06/an-automated-ensemble-method-for-combining-and-evaluating-genomic-variants-from-multiple-callers/) was then implemented to integrate analysis results from the three callers. Subsequently, the Ensemble pipeline applies a Support Vector Machine learning algorithm to identify an optimal decision boundary based on the variant calling results out of multiple variant callers, to produce a more balanced decision between false and true positives.

Annotation of variants

Annotation of each variant locus was made via a custom software pipeline based on public data integrated by a CGR in-house script, including Ensembl, refGene, and UCSC KnownGene databases, the dataset from University of Washington's Exome Sequencing Project (ESP6500) (http://evs.gs.washington.edu/EVS/), dbNSFP(Liu et al. 2011): database of human nonsynonymous SNPs and function predictions (https://sites.google.com/site/jpopgen/dbNSFP), the Molecular Signatures Database (MSigDB) (http://www.broadinstitute.org/gsea/msigdb/index.jsp), the National Center for Biotechnology Information (NCBI) Clinically Relevant Sequence Variations (ClinVar) and Single Nucleotide Polymorphism database (dbSNP) databases(Sherry et al. 2001) build 137, the 1000 Genomes Project(Abecasis et al. 2010), the Exome Aggregation Consortium (ExAC) database (http://exac.broadinstitute.org/), and the Human Gene Mutation Database (HGMD)(Stenson et al. 2014).

Filtering of variants

Supplementary table 1 shows the 24 PC-related genes evaluated in the current study. The genes were categorized into: MMR, (other) AD disorders, AR/AD disorders, and hereditary/chronic pancreatitis gene sets. Variants were excluded from further evaluation if they did not pass the quality control (QC) filter in the in-house bioinformatics pipeline or if the variant was reported by only one caller. Since the goal of the study was to investigate high-risk variants, the analyses focused on (very) rare exonic variants. Thus, variants were excluded from further evaluation based on the following criteria: 1. allele frequency >0.1% in the 1000 Genomes Project (overall or European sample) or ESP6500 (European sample); 2. present in >2 families from an in-house database (CGR, NCI) of >900 cancer-prone families (excluding melanoma-prone or PC families); 3. synonymous, intronic, or untranslated region (UTR) variants.

Classification/Validation of variants

Variants were classified as frameshift, stop-gain, splicing, inframe deletion/insertion, or nonsynonymous (NS) substitutions (missense). Frameshift and stop-gain variants were defined as loss of function (LOF) variants since they are expected to be protein truncating and thus deleterious. To classify NS substitutions as deleterious, we used an ensemble prediction score (Meta Likelihood ratio) [Meta LRP] that incorporates results from nine in silico algorithms (Sorting Intolerant from Tolerant (SIFT), PolyPhen-2, Genome Evolutionary Rate Profiling (GERP++), Mutation Taster, Mutation Assessor, Functional Analysis Through Hidden Markov Models (FATHMM), Likelihood Ratio Test (LRT), SiPhy, and PhyloP) and allele frequency. This ensemble score achieved the highest discriminative power compared to 18 deleterious scoring methods and also showed low false positive prediction rate for benign yet rare NS variants.(Dong et al. 2015) ExAC, HGMD, Leiden Open Variant Databases (LOVD), Align Grantham Variation and Grantham Deviation (GVGD)(Tavtigian et al. 2006), ClinVar (http://www.ncbi.nlm.nih.gov/clinvar/), and Ingenuity Variant Analysis (IVA) were examined to further categorize individual variants.

LOF variants, inframe deletions/insertions, and selected missense variants including those with the lowest 10% quality (after filtering) were technically validated using Sanger sequencing (in Genoa, Italy) or Ampliseq (at CGR). For technical validation using Ampliseq, a targeted, multiplexed PCR primer panel was designed using the Ion AmpliSeq Designer v4.4.4 (Life Technologies, Carlsbad, CA). Sample DNA (30ng) was amplified using this custom AmpliSeq primer panel (average amplicon size=244bp), and sequencing libraries were prepared following the Ion AmpliSeq Library Preparation protocol (Life Technologies), using Ion Xpress Barcode Adapters. Individual sample libraries were pooled, then templated and sequenced on the Ion Torrent Personal Genome Machine (PGM) Sequencer using Ion PGM Hi-Q Chef chemistry. Base calling and alignment were performed using Torrent Suite 4.4. Variant calling was done with GATK and Torrent Variant Caller.

Population Controls

Data from 1001 European-American/European population controls from two cohort studies (Cancer Prevention Study (CPS)-II, n=224; Prostate, Lung, Colorectal and Ovarian Screening Trial (PLCO), n=378) and one case-control study (Environment and Genetics in Lung Cancer Etiology (EAGLE), n=399) were available for inclusion in the current study to evaluate genetic burden for the PC-related genes.(Wang et al. 2012) The sequencing and bioinformatics analysis methods for the population controls followed the same processes as were used for the PC patients. However, the SeqCAP EZ Human Exome Library v3.0 + UTR (Roche NimbleGen, Madison, WI) was utilized for exome sequence capture. Variant calling for the population controls was done together with that for the entire in-house database (CGR, NCI) of >1200 cancer-prone families.

Statistical Analyses

We performed a gene- or gene-set level test of association (i.e. genetic burden) for each of the 23 PC-related genes (excluding CDKN2A) and the four gene sets (MMR, pancreatitis, AD disorder, AR/AD disorder genes). We defined a rare variant to have frequency ≤0.001. Since standard family-based methods(Chen et al. 2013; Svishcheva et al. 2014) relying on asymptotic distributions of the test statistic are inappropriate given our small sample sizes, we used the following approach. For each gene or set, the test statistic was defined as the number of cases with a rare exonic variant. We then calculated a p-value, or the probability of observing at least that many cases with a rare variant under the null hypothesis. Specifically, we created a list of the 2n_u haplotypes from the n_u controls. In the unlikely situation when a control had ≥2 rare variants in a single gene, each haplotype was assumed to carry at least one rare variant. We then used a 2-step “permutation” (i.e. random assignment) procedure. First, we randomly generated Identical-By-Descent (IBD) patterns for the familial cases using the rules of Mendelian Inheritance. Second, we assigned each of the founder chromosomes in the families to carry a haplotype randomly selected from the list of control haplotypes. After 1000 permutations, the p-value was the proportion of permutations where the number of family members carrying a rare exonic variant was at least as large as that observed in the actual data. Since the CDKN2A− cases were all unrelated, we also used Fisher's exact test for the gene-level association test for comparison.

RESULTS

After excluding seven PC patients because of insufficient DNA, sample failure, or sample mixup, 66 PC patients (43 CDKN2A+ and 23 CDKN2A−) were included in the current analyses. Table 1 shows the number of rare variants (total, LOF and missense deleterious or inframe deletion variants) in the CDKN2A+ and CDKN2A− PC patients by gene set. Supplemental table 4 shows the details for the 35 rare variants (in 15 genes) that were found in the CDKN2A+ and CDKN2A− PC patients after filtering. No variant was observed in all PC patients in a CDKN2A+ family with multiple PC patients. Individual CDKN2A+ PC patients had variants in MMR, pancreatitis and AD disorder genes but no variants in AR/AD disorder genes. In contrast, CDKN2A− PC patients had variants in each gene set with the most variants in AR/AD genes. Further, LOF variants (n=3 plus a known BRCA2 frameshift (P153)(Ghiorzo et al. 2012)) were only seen in CDKN2A− PC patients (Table 2).

Table 1.

Number of rare variants (Total, LOF, Deleterious Missense or inframe deletion) in CDKN2A+ and CDKN2A− pancreatic cancer patients categorized by gene set

Gene Set	Number of Rare Variants in CDKN2A+ PC Patients (n=14)			Number of Rare Variants in CDKN2A− PC Patients (n=21)
	Total Variants	Loss of Function Variants	Deleterious Missense Variants	Total Variants	Loss of Function Variants	Deleterious Missense or Inframe Deletion Variants
MMR	7	0	3	5	0	3
AD Disorder	4	0	1	5	1*	1
AR/AD Disorder	0	0	0	7	3	3
Hereditary/chronic pancreatitis	3	0	3	4	0	1

^*Previously reported BRCA2 frameshift (Ghiorzo et al, 2012b)

LOF, loss of function; PC, pancreatic cancer; MMR, mismatch repair; AD, autosomal dominant; AR, autosomal recessive

Table 2.

Loss of function (LOF) and deleterious variants (classified by Meta Likelihood Ratio Prediction [LRP]) in pancreatic cancer patients

Gene	Chrom	Location	rsID	Ref	Var	AA/protein change	Allele freq#	CDKN2A carrier	History of smoking: yes/no/unk	Family/Patient ID
Loss of function [stopgain, frameshift] variants:
ATM	chr11	108155142		AG		E1313fs	0	no	no	M595
ATM	chr11	108183151	rs587779852	G	T	E1978*	1.05E-04	no	unk	AJ7379
BRCA2	chr13	32914288	rs80359537	AT		H1932fs	0	no	yes	P153
PALB2	chr16	23646627	rs180177100	G	A	R414*	0	no	unk	AJ7379
Deleterious missense or inframe deletion variants classified by Meta LRP:
MSH2	chr2	47630331	rs267607911	A	G	M1V	0	yes	no^	LUMC6_019
MSH2	chr2	47637248	rs145649774	C	G	L128V	0.000349	yes	no	P147
MSH2	chr2	47641430	rs34136999,,	C	T	A272V	0.000581	no	no	P149
MSH6	chr2	48026852	rs376220212	G	A	R577H	0.000116	yes	no	P176
PMS2	chr7	6013076		G	A	P848L	0	no	unk	P1
PMS2	chr7	6022617	rs587780046	G	A	T671M	6.34E-04	no	yes	P282
CFTR	chr7	117171037	rs201958172	G	A	A120T	1.36E-04	yes	no	M402
CFTR	chr7	117171152	rs397508725	G	C	S158T	1.73E-05	yes	no	P61
CFTR	chr7	117243784	rs151048781	G	C	M952I	1.20E-04	no	yes	P115
CFTR	chr7	117250625	rs149279509	A	G	Y1014C	0.001	yes	unk	K_1001, K_1002
ATM	chr11	108122581		T	G	L542W	0	no	no	P268
ATM	chr11	108196797	rs567060474	G	A	A2274T	0.001	no	yes	P115
FANCA	chr16	89858430		CTT		E377del	0	no	yes	P282
TP53	chr17	7576911	rs145151284	G	C	T312S	8.99E-05	no	no	P268
BRCA1	chr17	41256266	rs28897673	T	C	Y105C	0.000116	yes	no, yes	LUMC19_013, LUMC19_016

^#Maximum allele frequency from public databases (see Supplemental Table 4 for details)

^{^}Extensive passive smoking exposure

Chrom, chromosome; Ref, referent allele; Var, variant allele; AA, amino acid; freq, frequency; unk, unknown

There were 14 missense variants in CDKN2A+ PC patients, seven in the MMR genes, four in different AD disorder genes (1 each in APC, PALLD, BRCA1, BRCA2), and three in the pancreatitis gene CFTR. Seven of these variants were predicted to be deleterious by summary Meta LRP score (Table 2); 3 of which were in MMR genes (MSH2, MSH6), 3 in the pancreatitis gene CFTR, and one in BRCA1. For most of these variants, however, algorithms such as CLINVAR, IVA, and Align GVGD clinically classified the variants as uncertain (Supplemental table 4).

Of 21 variants observed in CDKN2A− PC patients, 12 (Table 2) were considered deleterious [3 LOF variants (frameshift in ATM, stop-gain in ATM and PALB2), the known BRCA2 frameshift, 1 nonframe deletion in FANCA, and 7 missense variants]. One Swedish PC patient had two stop-gains, both technically validated and classified as pathogenic, one each in ATM (p.E1978*) and PALB2 (p.R414*). The third LOF variant, an ATM frameshift (p.E1313fs), was observed in an Italian patient; this variant was not seen in our in-house familial or population controls or reported in 1000 Genomes, ESP, ExAC, or Kaviar (http://db.systemsbiology.net/kaviar/). ATM had the greatest number of variants (n=5), all technically validated and most (4/5) with classification as pathogenic or likely pathogenic (Table 2, Supplemental table 4). Similar to what was observed in CDKN2A+ PC patients, most MMR variants were classified as variants of uncertain significance (VUS) (Supplemental table 4).

Table 3 shows the gene-level association test for genes with rare variants in ≥1 PC patient. For all PC patients, PMS2 showed a suggestive association (p=0.057). CDKN2A− PC patients had a significant increase for rare variants in ATM (p=0.006), CPA1 (p=0.021), and PMS2 (p=0.038). Evaluation of the MMR gene set showed a significant association for All (p=0.033) and suggestive association in CDKN2A+ PC patients (p=0.086). There was no significant gene-set association for pancreatitis, AD disorder (excluding CDKN2A), or AR/AD disorder genes for All, CDKN2A+, or CDKN2A− PC patients (data not shown), however, CDKN2A− PC patients showed a suggestive association for AR/AD disorder genes (p=0.051). Restricting the evaluation to LOF/deleterious variants (based on Meta LRP) produced significant differences only in ATM (p=0.001) and PMS2 (p=0.013) for CDKN2A− PC patients.

Table 3.

Results of gene-level association tests in All, CDKN2A+, and CDKN2A− PC patients versus 1001 population controls

		All PC Cases (n=66)			CDKN2A+ PC Cases (n=43)			CDKN2A− PC Cases (n=23)
Gene	Number of Controls with rare variant	Number of PC Cases with rare variant	No. Families with rare variant	p-value [All PC Cases vs controls]	Number of CDKN2A+ PC Cases with rare variant	No. Families with rare variant	p-value [CDKN2A+ PC cases vs controls]	Number of CDKN2A− PC Cases with rare variant	p-value [CDKN2A− PC cases vs controls]	p-value, Fisher's exact test [CDKN2A− PC cases vs controls]*
APC	43	1	1	0.931	1	1	0.823	0	--	--
ATM	51	5	5	0.237	0	0	--	5	0.006	0.005
BRCA1	26	3	2	0.24	2	1	0.317	1	0.459	0.463
BRCA2	52	4	4	0.414	1	1	0.864	3	0.108	0.116
CASR	7	1	1	0.374	0	0	--	1	0.157	0.167
CFTR	61	5	4	0.384	4	3	0.285	1	0.721	0.725
CPA1	12	2	2	0.211	0	0	--	2	0.021	0.037
FANCA	51	1	1	0.959	0	0	--	1	0.715	1.000
MLH1	19	1	1	0.687	1	1	0.499	0	--	--
MSH2	19	3	3	0.128	2	2	0.2	1	0.357	0.368
MSH6	17	3	3	0.092	2	2	0.179	1	0.303	0.338
PALB2	12	1	1	0.527	0	0	--	1	0.253	0.257
PALLD	19	1	1	0.694	1	1	0.517	0	--	--
PMS2	30	5	5	0.057	2	2	0.367	3	0.038	0.032
TP53	7	1	1	0.365	0	0	--	1	0.166	0.167

PC, pancreatic cancer

^*Gene-level association tests also computed using Fisher's exact test for comparison purposes.

Table 4 shows the number of rare variants in CDKN2A− PC patients by cancer family history for LOF/deleterious variants. ATM variants were found only in PC patients with family histories that included digestive system tumors (5/14=36%). Further, in addition to the Swedish patient with two LOF variants, three other CDKN2A− PC patients had multiple rare potentially deleterious variants two of which included ATM. Similarly, nine CDKN2A+ PC patients had multiple rare potentially deleterious variants since they had known CDKN2A mutations in addition to potentially deleterious variants in MMR, pancreatitis, or other AD disorder genes (Table 2, Supplemental table 4).

Table 4.

Number of LOF or potentially deleterious variants in CDKN2A− PC patients by cancer family history (includes cancers in studied PC patient)

Family History of Cancer#	No. Of Families by Cancer Family History	No. LOF/deleterious variants
		MMR	Pancreatitis	ATM	Other
PC (n=1) + Melanoma (n≥1)	5	1	0	0
PC (n=1) + Melanoma (n≥1) + Digestive System^ (n≥1)	8	1'	1*	3	1 [FANCA]
PC (n=1) + Digestive System (n≥1)	6	0	0	2	1* [PALB2] 1* [TP53]
PC (n=1) + Melanoma (n≥1) + Non-digestive system (n≥1)		1	0	0	1 [BRCA2 - QC]

^#Number of family members with PC, melanoma, digestive system cancer, or non-digestive system cancer

^*Patient has another rare LOF/deleterious variant in ATM

^'Patient has a rare nonframe deletion in FANCA

^{^}Includes pancreatic cancer, thus family has ≥2 members with PC

LOF, loss of function; PC, pancreatic cancer; No., number; MMR, mismatch repair; QC, quality control

DISCUSSION

The risk of pancreatic cancer is increased in some families with CDKN2A mutations. However, the reason for this increased risk has yet to be determined. We systematically evaluated 24 (including CDKN2A) purported PC-related susceptibility genes in PC patients with and without CDKN2A mutations that were ascertained by research groups investigating patients/families with CDKN2A mutations. Overall, there was a significant increase in genetic burden for carrying rare variants in MMR genes in all PC patients (CDKN2A+ and CDKN2A−) compared to population controls. Nine CDKN2A+ and four CDKN2A− PC patients had rare potentially deleterious variants in multiple genes. For the CDKN2A− PC patients, ATM showed the strongest association with ATM variants only observed in PC patients with a family history that included digestive system tumors. Most of the ATM variants were predicted to be pathogenic. Further, three of the four CDKN2A− patients with potentially deleterious variants in multiple PC-related genes had a variant that involved ATM, including one Swedish patient with previously HGMD reported pathogenic stop-gains in both ATM and PALB2. However, having two pathogenic LOF mutations in two high-risk PC-related genes did not result in a substantially earlier age at diagnosis, similar to what has been observed in several familial PC (FPC) patients(Roberts et al. 2015) as well as melanoma patients homozygous for the Dutch CDKN2A founder mutation.(de Snoo et al. 2008; Gruis et al. 1995)

Family history studies of pancreatic adenocarcinoma suggest that 5–10% of cases have a strong hereditary basis consistent with other adult onset cancers such as breast cancer and melanoma.(Goldstein and Tucker 2001; Zhen et al. 2015) However, in contrast to breast cancer with BRCA1/2 and melanoma with CDKN2A, mutations in individual PC-related predisposition genes do not account for more than a few percent of FPC patients. Genetic causes of PC show extensive genetic heterogeneity with cancer predisposition genes ATM, BRCA1, BRCA2, CDKN2A, PALB2, and the MMR genes appearing to account for the largest proportion of the known genetic causes of FPC.(Klein 2013; Zhen et al. 2015) A recent whole genome sequencing study of FPC patients confirmed the importance of these major FPC susceptibility genes and also proposed some additional candidate genes (BUB1B, CPA1, FANCC, and FANCG) for FPC.(Roberts et al. 2015) In the current study, we observed no LOF/deleterious variants in BUB1B, FANCG, or FANCC (data not shown) in either CDKN2A+ or CDKN2A− PC patients. In contrast, there was a significant gene-level association with CPA1 in CDKN2A− PC patients but neither of the observed CPA1 variants was classified as deleterious. Roberts et al(Roberts et al. 2015) also analyzed 87 predominantly hereditary cancer genes (supplementary table S3 in(Roberts et al. 2015)) in depth for protein truncating variants. Review of these hereditary cancer genes in our sample (excluding the genes systematically evaluated) revealed only one very rare (0.0083% from Kaviar version 150810-Public) LOF variant [c.2011dupA, p.I671fs] in FANCI in a CDKN2A+ PC patient.

The importance of cancer family histories in PC patients has yet to be fully investigated. A recent examination of 290 PC probands from the population-based Ontario Pancreas Cancer Study who had been selected based on family history of breast and/or ovarian cancer, PC, or no family history of either, found 11 pathogenic mutations (3.8%) in an investigation of 13 cancer predisposition genes (APC, ATM, BRCA1, BRCA2, CDKN2A, MLH1, MSH2, MSH6, PALB2, PMS2, PRSS1, STK11, TP53).(Grant et al. 2015) Of particular interest, carrier status was significantly associated with a personal or first-degree family history of breast or colorectal cancer but not a family history of PC. Our smaller non-population-based study showed a high proportion of patients with a family history of digestive system tumors carrying a deleterious variant which may be etiologically relevant or reflect ascertainment bias. Future much larger studies of PC patients with detailed family history data will be needed to better understand the cancer histories that impact PC risk.

The samples in this study came from research groups who study familial melanoma so most CDKN2A+ PC patients were from melanoma-prone families in ongoing studies in the United States, Netherlands, Italy, or Sweden. In addition, several CDKN2A+ and CDKN2A− PC patients were part of a PC case-control study conducted in Genoa, Italy. The remaining CDKN2A− PC patients included members of melanoma-prone or PC-prone families. Smoking history was not available for many patients and thus this risk factor could not be evaluated. Interestingly, though, about half of the PC patients with LOF/deleterious variants were nonsmokers (see Table 2, Supplemental table 4). Further, although of interest, small sample size precluded examination of common variants. In addition, the population controls that were sequenced on the same platform as the cases, necessary for conducting the gene-level association tests, were European (Italian) and European-American. Restriction of the gene- and gene-set-level association tests to Italian and American PC patients showed stronger evidence for an association for the MMR gene-set across all PC patient subsets: All (p=0.002), CDKN2A+ PC patients (p=0.027), and CDKN2A− PC patients (p=0.049). In addition, gene-level associations were stronger for individual MMR genes and CFTR for All [MSH6 (p=0.02), PMS2 (p=0.028), and CFTR (p=0.075)], for CDKN2A+ PC patients [CFTR (p=0.045) and MSH6 (p=0.057)], and for CDKN2A− PC patients [PMS2 (p=0.012)]. Finally, in CDKN2A− PC patients, ATM (p=0.005) showed similar evidence for association but CPA1 (p=0.198) was no longer significant.

For most PC patients with LOF/deleterious variants, it was not possible to evaluate additional relatives to determine whether they carried the variant seen in their respective families. However, one Italian PC patient proband (P268) who carried the p.L542W ATM variant recently had a brother diagnosed with cancer of the epiglottis at age 60 years. After providing a blood sample and consent for genetic testing, the brother was found to carry the same ATM variant as the proband. In addition to these two patients, the cancer family history included the siblings' father who had PC, two first cousins of the siblings with PC, and an aunt with colorectal cancer. The median age at diagnosis of the four PC patients was 58 years (range: 56 – 62 years).

In summary, five research groups from the United States, Italy, Netherlands, and Sweden that have been studying melanoma, PC, and the CDKN2A gene and have some of the largest samples of PC patients with CDKN2A mutations contributed to this study. Nevertheless after combining material from these five research groups, the sample size remained limited with sequencing data available for only 43 PC patients with and 23 PC patients without CDKN2A mutations. However, even with this limited sample size, we found a nominally significant gene-set level association for MMR genes in all PC patients with stronger evidence for this association in the Italian and American PC patients, particularly in CDKN2A+ PC patients, and for ATM, CPA1, and PMS2 in CDKN2A− PC patients. Further, numerous PC patients had rare likely deleterious variants in more than one PC-related gene suggesting that a subset of CDKN2A+ and CDKN2A− PC patients may have an increased risk of pancreatic cancer because of mutations in multiple PC-related susceptibility genes. However, many CDKN2A+ PC patients did not have deleterious variants in any of the PC-related genes and therefore other genes, exposures, and/or alternative mechanisms that involve CDKN2A likely influence risk of PC in many of these families. Additional research is needed to confirm these findings and to more fully evaluate specific variants/genes that may play a role in pancreatic cancer in patients and families with and without CDKN2A mutations.