Impact of Sixteen Established Pancreatic Cancer Susceptibility Loci in American Jews

Since the 1950s, epidemiologic studies have repeatedly found pancreatic cancer to be more frequent among Jews, particularly Ashkenazi Jews, as compared with non-Jews (1–9). Most recently, Risch and colleagues (10) conducted a population-based case–control study and found an increased odds of pancreatic cancer for subjects reporting Jewish ancestry compared with subjects not reporting Jewish ancestry [odds ratio (OR), 1.81; 95% confidence interval (CI), 1.05–3.10], and Eldridge and colleagues (11) found an increased risk of pancreatic cancer mortality in Cancer Prevention Study II participants reporting Jewish religious affiliation compared with participants reporting Protestant, Catholic, Latter Day Saints, other, or no religious affiliation [risk ratio (RR), 1.43; 95% CI, 1.30–1.57].

Pancreatic cancer is one of the most lethal cancers for both men and women in the United States, with a 5-year survival of less than 2% on a population basis (12). In 2015, 48,960 diagnosed cases of and 40,560 deaths from pancreatic cancer were estimated in the United States (13). Globally, in 2012, 337,872 diagnosed cases of and 330,391 deaths from pancreatic cancer were estimated to have occurred (14). Most pancreatic cancer patients present with advanced disease; however, in early disease stages patients can undergo surgical resection, which confers significant survival advantage (15).

Established risk factors for pancreatic cancer account for about half of the disease in the general population. In the United States, cigarette smoking accounts for about 20% of the disease and non-O ABO blood group explains about 19% (16–18). Other risk factors, including uncommon hereditary factors (e.g., germline mutation in p16, BRCA1 and BRCA2, ATM, PALB2, and MMR genes), chronic pancreatitis, obesity, long-term diabetes mellitus, and Helicobacter pylori colonization, combined, account for around 15% to 20% of the disease (16, 17, 19–22). In addition, five large-scale genome-wide association studies (GWASs) have been conducted in the white European population: The Pancreatic Cancer Cohort and Case-Control Consortium I (PanScan I) study, the Pancreatic Cancer Cohort and Case–Control Consortium II (PanScan II) study (23), the Pancreatic Cancer Cohort and Case–Control Consortium III (PanScan III) study, the Pancreatic Cancer Case–Control Consortium (PanC4) study, and the combined PanScan I, II, and III (PanScan I-III), PANcreatic Disease ReseArch (PANDoRA), PanC4 study. Together these studies identified 16 common low-risk pancreatic cancer susceptibility loci that reached the genome-wide significance threshold: 10^−7.3 (i.e., P < 5 × 10⁻⁸; Supplementary Table S1; refs. 24–28).

The increased pancreatic cancer risk among Ashkenazi Jews can be partially explained by the greater prevalence of two BRCA1 mutations (185delAG and 5382insC) and one BRCA2 mutation (6174delT), which together account for about 9.1% of the disease in Jews. For pancreatic cancer, the population-attributable fraction (PAF) of disease in the Jewish population for the two BRCA1 mutations is 2.40%: 1.17% total prevalence of the two BRCA1 mutations and 3.1-fold relative risk of pancreatic cancer (19). The PAF of disease for the BRCA2 mutation in the Jewish population is 6.74%, based on 1.29% prevalence of the BRCA2 mutation and 6.6-fold relative risk of pancreatic cancer (19). The gene that determines ABO blood type (ABO) may also contribute slightly to the increased risk of pancreatic cancer among Ashkenazi Jews since the O ABO blood group frequency is lower and the B ABO blood group frequency is higher in Jews compared with white Europeans. However, the PAF of pancreatic cancer in the Jewish population has not yet been calculated on a population basis (29, 30). Other known genetic and nongenetic risk factors for pancreatic cancer are not appreciably higher in the Jewish population compared with the non-Jewish white European population (11).

To investigate further a genetic basis for pancreatic cancer in the Jewish population, we evaluated the impact of the 16 genome-wide significant pancreatic cancer susceptibility loci identified in the PanScan I-III, PanC4, and combined PanScan I-III, PANDoRA, PanC4 GWASes, in full-Jewish subjects, part-Jewish subjects, non-Jewish Southern European subjects, and the combined non-Jewish Eastern, Northern, Southern, and Western white European (non-Jewish white European) subjects. We then compared average ORs across the 16 pancreatic cancer susceptibility loci in these four populations. We chose to include non-Jewish Southern European subjects in the analysis because of all of the non-Jewish white European populations, Southern Europeans are, genetically, the closest to Jews and historically the most genetically admixed with them (31).

All subject data used in this analysis came from biosamples and data obtained from individual subjects providing written informed consent in studies that were conducted in accordance with recognized ethics guidelines (Declaration of Helsinki, International Ethical Guidelines for Biomedical Research Involving Human Subjects, Belmont Report, or U.S. Common Rule) under institutional review board approval in their respective studies.

PanScan I/II (accession phs000206.v4.p3), PanC4 (accession phs000648.v1.p1), and the Resource for Genetic Epidemiology Research on Adult Health and Aging (GERA; accession phs000674.v1.p1) datasets were obtained from the database of Genotypes and Phenotypes (dbGaP; ref. 32). Information on age, sex, case/control status, and genotype was available from all three dbGaP datasets. For the GERA dataset, only subjects genotyped on the Affymetrix Axiom Genome-Wide EUR array (white European subjects) without cancer diagnoses were considered for inclusion in our study population. PanScan I/II and PanC4 contributed cases and controls to our study population, whereas GERA contributed only controls.

Three sub-studies from PanScan I/II and PanC4 were obtained with information on genotypes and self-reported race and ethnicity. In total, 311 self-reported white European (including Jewish or half-Jewish), black, or Asian subjects from the Yale Connecticut Pancreas (Yale) study included in PanScan I/II and PanC4 (24, 25, 27), 27 self-reported Jewish or half-Jewish subjects from the Johns Hopkins University (JHU) study included in PanC4 (27), and 107 subjects with self-reported number of Jewish grandparents from the Memorial Sloan Kettering Cancer Center (MSKCC) study included in PanC4 (27), were used as reference marker subjects to help genetically identify the full-Jewish, part-Jewish, non-Jewish Southern European, and non-Jewish white European subjects on principal component (PC) plots. Information on subjects' parents' place of birth was included in the Yale study and was used to categorize non-Jewish white European subjects into Eastern, Northern, Southern, and Western Europeans (Supplementary Table S2).

The publicly available 1,000 Genomes Project Phase III (October 2014 release) dataset and the Ashkenazi Genome Consortium (TAGC) dataset (European Genome-phenome Archive Study ID: EGAS00001000664) were used as reference panels for haplotype estimations (pre-phasing) and genotype imputation (33, 34).

Subjects from PanScan I were genotyped on the Illumina HumanHap550 Infinium array, subjects from PanScan II were genotyped on the Illumina Human610-Quad array, subjects from PanC4 were genotyped on the Illumina HumanOmniExpressExome array, and subjects from GERA were genotyped on the Affymetrix Axiom Genome-wide EUR array. Quality control (QC) was performed separately using high thresholds for subject genotyping completion and variant call rate for each of the four genotyping platforms in PLINK v1.90-beta (35).

Sample replicates, failed samples, subjects with <98% genotyping completion, subjects with missing sex information, indeterminable X chromosome heterozygosity, or a discordance in reported vs. genotyped sex (reported females with >0.25 and reported males with <0.80 X chromosome heterozygosity), and related subjects (⁠|$\widehat {\pi \ }$|≥0.20) were removed from the datasets. For a pair of related subjects on the Illumina HumanHap550 Infinium array, the Illumina Human610-Quad array, and the Illumina HumanOmniExpressExome array, the control subject was removed in a case–control comparison and one subject was randomly chosen to be removed in a case–case or control–control comparison. For a pair of related subjects on the Axiom Genome-wide EUR array, the younger subject was removed. Finally, subjects across the four platforms were combined, and 55 seemingly related subjects between PanScan I/II and PanC4, 61 between PanScan I/II and GERA, and 12 between PanC4 and GERA were identified and removed (⁠|$\widehat {\pi \ }$|≥0.20). Related subjects between PanScan I/II and PanC4 were removed from the PanC4 dataset, and related subjects between PanScan I/II and GERA, and PanC4 and GERA were removed from the GERA dataset. The numbers of samples and subjects excluded from the genotyping arrays at each QC step are listed in Supplementary Table S3a.

Variants with duplicate positions, call rates <98%, extreme Hardy Weinberg equilibrium departures in controls (P < 10⁻⁷), monomorphic variants in either cases or controls, copy number variants, and variants not in autosomal chromosomes were removed. Only SNPs remained after variant QC. After QC, there were 530,771 SNPs on the Illumina HumanHap550 Infinium array, 553,743 SNPs on the Illumina Human610-Quad array, 804,262 SNPs on the Illumina HumanOmniExpressExome array, and 596,652 SNPs on the Affymetrix Axiom Genome-wide EUR array. The numbers of variants excluded from the genotyping arrays at each QC step are listed in Supplementary Table S3B.

QC was done separately for the Yale reference marker subjects in PanScan I/II genotyped on the Illumina Human610-Quad array, Yale reference marker subjects in PanC4 genotyped on the Illumina HumanOmniExpressExome array, JHU reference marker subjects in PanC4 genotyped on the Illumina HumanOmniExpressExome array, and MSKCC reference marker subjects in PanC4 genotyped on the Illumina HumanOmniExpressExome array. QC procedures discussed above in the Sample and subject QC and Variant QC section were applied to the Yale, JHU, and MSKCC reference marker datasets.

Fast principal component analysis (PCA) is an accurate approximation of exact PCA and can be used when exact PCA cannot be performed because of sample size limitations (N < 10,000 subjects). Because Jewish subjects were identified in a sample of over 50,000 subjects, exact PCA ran out of memory and FastPCA had to be used (36).

Four FastPCAs were sequentially conducted with all post-QC 80,772 genotyped SNPs common to the Illumina HumanHap550 Infinium, Illumina Human610-Quad, Illumina HumanOmniExpressExome, and Affymetrix Axiom Genome-wide EUR arrays, to identify the PanScan I/II, PanC4, and GERA (PanScan/PanC4/GERA) full- or part-Jewish subjects (Supplementary Fig. S1). FastPCA was run in EIGENSOFT v6 (36). First, FastPCA was performed on 55,930 PanScan/PanC4/GERA subjects of various races and ethnicities, including 44 self-reported black, 7 Asian, and 78 Jewish or half-Jewish reference marker subjects from the Yale study. The FastPCA PC2 vs. PC1 plot enabled visualization of the PanScan/PanC4/GERA subjects (Supplementary Fig. S2A). In total, 54,829 white European subjects to the right of the diagonal line in Supplementary Fig. S2A were retained, and a second FastPCA was run with these white European subjects, plus 78 Jewish or half-Jewish reference marker subjects from the Yale study. FastPCA PC2 vs. PC1 was again plotted to visualize the white European subjects (Supplementary Fig. S2B). From this, 47,545 white European subjects to the left of the diagonal line in Supplementary Fig. S2B were retained, and a third FastPCA was run with these white European subjects and the 78 Jewish or half-Jewish reference marker subjects, plus 29 Eastern European, 48 Northern European, 65 Southern European, and 12 Western non-Jewish white European reference marker subjects from the Yale study, 107 full- or part-Jewish reference marker subjects from the MSKCC study, and 27 Jewish or half-Jewish reference marker subjects from the JHU study. PC2 vs. PC1 from FastPCA was plotted to visualize the spread between the full-Jewish, part-Jewish, and non-Jewish white European subjects (Supplementary Fig. S2C). In total, 4,657 full- or part-Jewish subjects to the left of the two intersecting diagonal lines in Supplementary Fig. S2C were retained, and a fourth FastPCA was run on these 4,657 full- or part-Jewish subjects and 78 Jewish or half-Jewish reference marker subjects from the Yale study, 105 full- or part-Jewish reference marker subjects from the MSKCC study, and 26 Jewish or half-Jewish reference marker subjects from the JHU study. PC2 vs. PC1 from FastPCA was again plotted to visualize the spread among the full-Jewish, 3/4 Jewish, 1/2 Jewish, and 1/4 Jewish subjects (Fig. 1). All lines were placed on these four FastPCA plots based on location of the reference marker subjects and where subject clusters thinned.

Three FastPCAs were sequentially conducted with all post-QC 80,772 common genotyped SNPs to identify the PanScan/PanC4/GERA non-Jewish Southern European subjects (Supplementary Fig. S1). The first two analyses were run as described above to visualize white European subjects (Supplementary Fig. S2A and 2B). The 4,657 full- or part-Jewish subjects (Fig. 1) were removed from 54,829 white European subjects (Supplementary Fig. S2b) and a third FastPCA was run on 50,172 non-Jewish white European subjects and 29 Eastern European, 48 Northern European, 65 Southern European, 12 Western non-Jewish white European, and 2 non-Jewish Middle Eastern reference marker subjects from the Yale study. PC2 vs. PC1 from FastPCA was plotted to identify non-Jewish Southern European subjects (Supplementary Fig. S2D). Again, all lines were placed on FastPCA plots based on location of the reference marker subjects and where subject clusters thinned.

Full-Jewish, part-Jewish, non-Jewish Southern European, and non-Jewish white European subjects were visually identified on FastPCA plots (Fig. 1, Supplementary Fig. S2A–S2D). In the individual site studies, reference marker subjects reporting ancestry variously as “Jewish” or “half-Jewish” or as number of Jewish grandparents, facilitated identification of the relevant regions on the FastPCA plots. In total, 406 full-Jewish pancreatic cancer cases and 2,332 full-Jewish controls, 133 part-Jewish pancreatic cancer cases and 1,785 part-Jewish controls, 585 non-Jewish Southern European pancreatic cancer cases and 3,371 non-Jewish Southern European controls, and 6,858 non-Jewish white European pancreatic cancer cases and 43,310 non-Jewish white European controls, all with age, sex, case/control status, and genotype information, were included in the sub-group analyses.

The 16 pancreatic cancer susceptibility loci that reached the genome-wide significance threshold (P < 10^−7.3) in the PanScan I-III, PanC4, and combined PanScan I-III, PANDoRA, PanC4 GWASes, were examined in the sub-group analyses (Supplementary Table S1).

Pre-phasing, using SHAPEIT v2, and genotype imputation, using IMPUTE v2, were performed separately for white European subjects genotyped on the Illumina HumanHap550 Infinium array, the Illumina Human610-Quad array, the Illumina HumanOmniExpressExome array, and the Affymetrix Axiom Genome-wide EUR array (Supplementary Fig. S1; refs. 37, 38). Prior to imputation, variants that could not map from human genome version 18 (hg18) to human genome version 19 (hg19) were removed. The manufacturers' annotation files and the University of California Santa Cruz (UCSC) genome browser were used to align genotypes for imputation; therefore, variants not in these files were removed (N, Illumina HumanHap550 Infinium array = 1,745; N, Illumina Human610-Quad array = 2071; N, Illumina HumanOmniExpressExome array = 21,488; N, Affymetrix Axiom Genome-wide EUR array = 6,895; ref. 39). Exome labeled variants were also removed from the Illumina HumanOmniExpressExome array because of their poor representation in the 1000 Genomes Project Phase 3 and TAGC reference panels (N = 132,930). The average information metric (37) for the 9 imputed SNPs on the Illumina HumanHap550 Infinium array, the 9 imputed SNPs on the Illumina Human610-Quad array, the 8 imputed SNPs on the Illumina HumanOmniExpressExome, and the 12 imputed SNPs on the Affymetrix Axiom Genome-wide EUR array, were 0.929, 0.929, 0.937, and 0.892, respectively. Information metrics for individually imputed SNPs are listed in Supplementary Tables S4a–S4d. IMPUTE genotype probabilities for the 16 pancreatic cancer susceptibility loci were converted to genotypes using a hard call threshold of 0.49999. Each SNP genotype was coded as a count of variant alleles. Ethnicity specific European minor allele frequencies (MAF) for the 16 pancreatic cancer susceptibility loci genotyped on the Affymetrix array (control-only subjects) were confirmed to be similar to the ethnic-specific European MAFs on the Illumina arrays (case and control subjects; Supplementary Tables S4A–S4D). All association analyses were conducted in PLINK v1.90-beta using unconditional logistic regression. The full-Jewish, part-Jewish, full- plus part-Jewish, and non-Jewish Southern European populations were adjusted for age (in 10-year categories), sex, and 6 sub-group specific PCs from exact PCA. EIGENSOFT v6 ran out of memory when running exact PCA for the non-Jewish white European population, and subsequently this population was adjusted for age (in 10-year categories), sex, and 6 PCs from the non-Jewish white European FastPCA. SNP associations were considered significant at the Bonferroni correction for multiple comparisons level, P < 0.003 (0.05/16 SNPs).

An average OR in controls was estimated according to their modeled covariates (10) by calculating a weighted average,

where |$n$|=total number of controls, the |${c_j}s$| are used to select the terms of interest in the calculation, allowing adjustment of the model for potential confounders (0 or 1), the sum on |$i$| is over all |$n$| controls, |${x_{ij}}$|= the number of variant alleles per genotype (0, 1, or 2), |${\beta _j}$|=log(OR), the sum on |$j$| is over the 16 pancreatic cancer susceptibility loci, and |${w_i}$|=1, under the assumption that the controls comprise approximately representative samples of their underlying populations.

A PAF was calculated for the full-Jewish population and compared to the PAF for the non-Jewish white European population according to the prevalence of non-O ABO blood groups. Blood groups (A, B, AB, and O) were determined using the two SNPs rs8176746 and rs505922 (17, 40), and analyses were done in R v3 (41).

For 16 a priori pancreatic cancer susceptibility loci, we used a threshold of statistical significance P = 0.05/16=0.003. At this threshold, one of the 16 pancreatic cancer susceptibility loci was significantly associated with risk in the full-Jewish subjects: chr13q22.1 [rs9543325; OR, 1.36; 95% confidence interval (CI), 1.16–1.58, P = 10^−4.1). For the remaining 15 susceptibility loci, 12 of the point estimates and directions of association were consistent with the PanScan I-III, PanC4, and combined PanScan I-III, PANDoRA, PanC4 GWASes: chr9q34.2/rs505922, chr1q32.1/rs3790844, chr5p15.33/rs451360, chr7q32.3/rs6971499, chr16q23.1/rs7190458, chr22q12.1/rs16986825, chr17q24.3/rs11655237, chr2p14/rs1486134, chr7p14.1/rs17688601, chr3q28/rs9854771, chr1q32.1/rs2816938, and chr8q24.21/rs10094872; two of the point estimates and directions of association were opposite: chr13q12.2/rs9581943 and chr5p15.33/rs35226131; and one point estimate was essentially null: chr5p15.33/rs2736098 (see Table 1 for ORs and P values, Supplementary Table S4A).

We compared association results in full-Jewish subjects between the genotype threshold method (above) in PLINK v1.09-beta and the genotype dosage method (below) in SNPTEST v2 (42). We found similar ORs and P values between these two association analysis methods (OR, 1.21; 95% CI, 1.04–1.41; P = 0.016 for rs505922 and OR, 1.36; 95% CI, 1.16–1.58; P = 10^−4.2 for rs9543325, using the genotype dosage association method; Table 1).

For sensitivity purposes, when we moved the line that separates full-Jewish from part-Jewish subjects on the FastPCA plot in Fig. 1 and recalculated full-Jewish subjects' ORs and P values to include or exclude approximately 30 full-Jews, we found the results to be consistent with the original full-Jewish subjects' analysis (OR, 1.23; 95% CI, 1.05–1.43; P = 10^−2.1 and OR, 1.21; 95% CI, 1.03–1.41; P = 0.018 for rs505922 with approximately ± 30 full-Jewish subjects; and OR, 1.35; 95% CI, 1.16–1.57; P = 10^−4.0 and OR, 1.37; 95% CI, 1.17–1.59; P = 10^−4.2 for rs9543225 with approximately ± 30 full-Jewish subjects).

In part-Jewish subjects, one of the 16 pancreatic cancer susceptibility loci was significantly associated with risk: chr5p15.33 (rs451360; OR, 1.53; 95% CI, 1.17–1.99; P = 10^−2.7). For the remaining 15 susceptibility loci, 12 of the point estimates and directions of association were consistent with the PanScan I-III, PanC4, and combined PanScan I-III, PANDoRA, PanC4 GWASes: chr9q34.2/rs505922, chr13q22.1/rs9543325, chr1q32.1/rs3790844, chr5p13.33/rs2736098, chr7q32.3/rs6971499, chr16q23.1/rs7190458, chr13q12.2/rs9581943, chr22q12.1/rs16986825, chr17q24.3/rs11655237, chr2p14/rs1486134, chr1q32.1/rs2816938, and chr8q24.21/rs10094872; and three of the point estimates and directions of association were opposite: chr7p14.1/rs17688601, chr3q28/rs9854771, and chr5p15.33/rs35226131 (see Table 1 for ORs and P values, Supplementary Table S4b).

Two of the 16 pancreatic cancer susceptibility loci were significantly associated with risk in the non-Jewish Southern European subjects: chr5p15.33 (rs451360; OR,1.35; 95% CI, 1.18–1.54; P = 10^−5.1) and chr17q24.3 (rs11655237; OR, 1.34; 95% CI, 1.13–1.61; P = 10^−3.1). For the remaining 14 susceptibility loci, 13 of the point estimates and directions of association were consistent with the PanScan I-III, PanC4 and combined PanScan I-III, PANDoRA, PanC4 GWAses: chr9q34.2/rs505922, chr13q22.1/rs9543325, chr1q32.1/rs3790844, chr7q32.3/rs6971499, chr16q23.1/rs7190458, chr13q12.2/rs9581943, chr22q12.1/rs16986825, chr2p14/rs1486134, chr7p14.1/rs17688601, chr3q28/rs9854771, chr1q32.1/rs2816938, chr8q24.21/rs10094872, and chr5p13.33/rs35226131; and chr5p15.33/rs2736098 had an opposite point estimate and direction of association (see Table 1 for ORs and P values, Supplementary Table S4C).

Fifteen of the16 pancreatic cancer susceptibility loci identified in the PanScan I-III, PanC4, and combined PanScan I-III, PANDoRA, PanC4 GWASes were significantly associated with pancreatic cancer risk in the non-Jewish white European subjects (the SNP rs35226131 was not significantly associated with pancreatic cancer; Table 1; Supplementary Table S4D).

Individual ORs and MAFs were similar between Jewish and non-Jewish white European subjects. The average ORs over the 16 SNPs in the four sub-groups were similar: in the full-Jewish subjects the average OR was 1.25, in the full- plus part-Jewish subjects it was 1.30, in the non-Jewish Southern European subjects it was 1.31, and in the non-Jewish white European subjects it was 1.26.

When the published point estimates from the PanScan I-III, PanC4, and combined PanScan I-III, PANDoRA, PanC4 GWASes were compared with the point estimates from the all-white European subjects (Jewish and non-Jewish European), 15 of the 16 pancreatic cancer susceptibility loci had comparable ORs and P values (the SNP rs35226131 was not significantly associated with pancreatic cancer; Supplementary Table S1).

The expected increased risk of pancreatic cancer according to non-O ABO blood groups in the full-Jewish population compared to the non-Jewish white European population was 2%. The PAF for pancreatic cancer was 19.4% in the full-Jewish population and 17.4% in the non-Jewish white European population according to the prevalence of non-O ABO blood groups, based on a 0.65 prevalence of the non-O ABO blood groups in the full-Jewish population, a 0.57 prevalence in the non-Jewish white European population, and a 1.37-fold relative odds of pancreatic cancer.

To our knowledge, this is the first study to examine the 16 pancreatic cancer susceptibility loci found in the white European population in the higher-risk Jewish population, and the first study to use PCs to identify and distinguish part- or full-Jews from other European populations for analyses. We are confident that the overwhelming majority of full- or part-Jews in our study are of Ashkenazi descent since participants from the sub-studies included in our study were recruited in the United States, where some 90% of Jews are of Ashkenazi descent (43). Though infrequent in a United States Jewish population, there may be a few Jews of Spanish/Portuguese Jewish descent included in the full- or part-Jewish clusters identified by FastPCA plots. Other Jews, such as Yemenite, Ethiopian, and Iranian Jews are not genetically close enough to be included in these Jewish clusters (44, 45).

Our results show an association between chr13q22.1 (rs9543325) and risk of pancreatic cancer among full-Jewish subjects. While we attempted to obtain datasets with as many cases of pancreatic cancer as possible, our study was somewhat underpowered to detect associations of modest effect (OR∼1.15) and MAF (∼0.30) in the full-Jewish sub-group analysis. A power analysis with the full-Jewish sub-group sample size of 406 pancreatic cancer cases and 2,332 controls showed this sub-group analysis to have >70% power to detect an association size of 1.31 for a SNP with an MAF >0.35 (P = 0.003; ref. 46). Nevertheless, 12 of the 15 SNPs that were underpowered to detect such association showed consistent point estimates and directions of association compared with the published results in the larger original studies.

For pancreatic cancer, we calculated a slightly higher expected risk increase (2%) attributable to the higher prevalence of non-O ABO blood groups in the full-Jewish population compared to the non-Jewish white European population based on the SNP rs505922, which is in high linkage disequilibrium with the functional SNP rs8176719 on chr9q34.2. None of the other 15 white European pancreatic cancer susceptibility loci have been explored more in depth in the Jewish population or have established associations with other diseases in the Jewish population (47).

Across ethnicity, our results do not show a consistent pattern in the ORs for the 16 pancreatic cancer susceptibility loci, though we might have expected them to progress from non-Jewish white European subjects (genetically furthest away from the full-Jewish subjects) to the full-Jewish subjects, with the non-Jewish Southern European subjects (genetically closest to full-Jewish subjects of all non-Jewish white European subjects) and the part-Jewish subjects (genetically closest to the full-Jewish subjects) in between. This lack of a consistent trend may be because part-Jewish subjects are exogamously mixed more with non-Jewish Eastern, Northern, and Western Europeans than with non-Jewish Southern Europeans, as seen in Supplementary Fig. S2c, as well as the small numbers and wider confidence intervals in the part-Jewish subjects.

Finally, the average OR of the 16 pancreatic cancer susceptibility loci did not show any differences in summary odds of pancreatic cancer among the full-Jewish, full- plus part-Jewish, non-Jewish Southern European, and non-Jewish white European subjects, suggesting that other variants should be investigated for increasing the risk of pancreatic cancer in the Jewish population. Further work needs to be done to explore the association between other variants and pancreatic cancer in the Jewish population.

No potential conflicts of interest were disclosed.

Conception and design: S.A. Streicher, H.A. Risch

Development of methodology: S.A. Streicher, H.A. Risch

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.P. Klein, S.H. Olson, L.T. Amundadottir, H.A. Risch

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S.A. Streicher, A.P. Klein, L.T. Amundadottir, A.T. DeWan, H. Zhao, H.A. Risch

Writing, review, and/or revision of the manuscript: S.A. Streicher, A.P. Klein, S.H. Olson, L.T. Amundadottir, A.T. DeWan, H. Zhao, H.A. Risch

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.P. Klein, H.A. Risch

Study supervision: H.A. Risch

High-performance computing (HPC) was supported by HPC facilities operated by, and the staffs of, the Yale Center for Research Computing and Yale's W.M. Keck Biotechnology Laboratory, as well as U.S. National Institutes of Health grants RR019895 and RR029676, which helped fund the HPC cluster. The cooperation of 30 Connecticut hospitals, including Stamford Hospital, in allowing patient access is gratefully acknowledged.

This work was supported by grants from the National Cancer Institute (F31 CA 177153 to S.A. Streicher; R01 CA098870 to H.A. Risch; P50 CA062924 and R01 CA097075 to A.P. Klein; and P30 CA008748 to S.H. Olson) and by grants from the Geoffrey Beene Foundation, the Arnold and Arlene Goldstein Family Foundation, and the Society of the Memorial Sloan Kettering Cancer Center (to S.H. Olson).

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.