Relationship between ABO Blood Group Alleles and Pancreatic Cancer Is Modulated by Secretor (FUT2) Genotype, but Not Lewis Antigen (FUT3) Genotype

Pancreatic cancer, of which 95% of cases are pancreatic ductal adenocarcinoma (PDAC), is the third leading cause of cancer-related death and the second most commonly diagnosed gastrointestinal cancer in the United States with a 5-year survival rate of ∼11% (1). Established epidemiologic risk factors include tobacco smoking, obesity, diabetes, and a family history of pancreatic cancer (2). ABO blood group has been implicated in the risk of pancreatic cancer in many epidemiologic studies (3–8). Growing evidence from genome-wide association studies (GWAS) also has confirmed the association of ABO genetic polymorphisms (e.g., rs505922) with pancreatic cancer risk (9–11). In Western populations, these studies have consistently shown that individuals with A, B, or AB blood group have a greater risk of pancreatic cancer than those with O blood group.

Histo-blood group antigens are present in rich diversity in the mucosal epithelia of human respiratory, genitourinary, and digestive tracts, which serve as host immune receptor sites important for bacterial or viral attachment and infection (12). ABO antigens are also expressed on platelets. Their production and expression are encoded by gene families expressing ABO, secretor (FUT2), and Lewis-type (FUT3) antigens. The ABO gene produces either a functional glycosyltransferase that transfer specific sugar residues to a precursor substance (the H antigen) to make the A and B antigens, or a non-functional enzyme. ABO alleles (A, B, and O) in the ABO gene are tagged by single nucleotide polymorphisms (SNP) at the locus; rs505922 (as a proxy for rs8176719) marks the O allele and rs8176746 marks the B allele (4, 13). The activity of the ABO glycosyltransferase differs between two classes of A alleles, resulting in increased antigen presentation among carriers of the A¹ allele relative to carriers of the A² allele. This suggests that if the association between the A allele and a phenotype is mediated through antigen presentation, then the magnitude of the association with the A¹ allele should be greater than that for the A² allele. Indeed, differential associations between the A¹ and A² alleles have been observed with venous thromboembolism (14) and PDAC (5).

Secretor status is determined by a SNP (rs601338) at the FUT2 gene, which encodes an enzyme (galactoside 2-L-fucosyltransferase) that allows for the secretion of ABO antigens into gastrointestinal secretions and the expression of ABO antigens in mucosal epithelia (15). This suggests that secretor status may modify the association between ABO blood type and PDAC.

The FUT3 gene on chromosome 19 is polymorphic and encodes an |$\alpha$|(1,3/1,4)fucosyltransferase involved in the synthesis of Lewis antigens. These antigens are expressed in exocrine secretions and can be found on the surface of red blood cells (16). Previous studies found that some SNPs (e.g., rs812936, rs28362459, and rs3894326) at the FUT3 locus were associated with a decrease in Lewis antigen enzyme activity (Lewis negative phenotype; refs. 17–19). FUT3 and FUT2 interact to produce different forms of the Lewis antigen, resulting in the three common Lewis antigen phenotypes: Le(a-b-) (Lewis negative), Le(a+b-) (Lewis positive [Lewis a antigen] among nonsecretors), and Le(a-b+) (Lewis positive [Lewis b antigen] among secretors).

Associations between secretor status and Lewis antigen activity have been investigated individually with ABO blood groups, but interactions between the three have not been reported. Particularly for pancreatic cancer, there has been only one study (our previous work) investigating secretor status and ABO blood group in relation to pancreatic cancer risk. That study, using PanScan consortium data, did not find a statistically significant difference between the odds ratio (OR) for non-O blood group versus O blood group among secretors and the OR among nonsecretors (20), but it may have been underpowered to detect an interaction. Given current limited knowledge regarding the joint effects of secretor status, Lewis antigens and ABO blood groups on risk of PDAC, we combined data from the PanScan and PanC4 consortia to perform a population-based study including 8,027 pancreatic cancer cases and 11,362 controls with data on genetically derived ABO blood, secretor, and Lewis antigen groups. The current study more than doubles the sample size of our previous study, which only examined the interaction between ABO blood groups and FUT2.

Our study included participants in the Pancreatic Cancer Cohort Consortium (PanScan I-III) and the Pancreatic Cancer Case–Control Consortium (PanC4) including 16 cohorts and 13 case–control studies genotyped in five previous GWAS. Detailed information on cases and controls were previously described (9–11, 21–23). In brief, cases were defined as individuals with primary adenocarcinoma of the exocrine pancreas and those with non-exocrine pancreatic tumors were excluded. Each study in the consortia obtained informed consent from study participants and Institutional Review Board approvals. Of 23,064 participants, we excluded those with missing information on sex (n = 1,502), those with missing genotype data (n = 2,071), those with no genetically derived blood, secretor, or Lewis groups (n = 34), and those without covariates (n = 68). After excluding these, data from 19,389 participants remained in our analysis (8,027 cases and 11,362 controls), all of European descent. Information on covariates was collected through written questionnaires or in-person interviews (4, 20). For this study, we obtained data on age and sex as covariates for analysis.

We used SNP genotype data to define blood group alleles and genotypes for secretor status and Lewis antigens. Genotyping for participants in PanScan I-III was performed at the Cancer Genomics Research Laboratory of the NCI using Illumina HumanHap series and Illumina Omni series arrays. For PanC4 participants, genotyping was performed at the Johns Hopkins Center for Inherited Disease Research using the Illumina HumanOmniExpressExome-8v1 array. Details about genotyping, quality control, and imputation procedure can be found elsewhere (11).

To determine ABO alleles (O, A, and B), we used haplotypes of two SNPs (rs505922 and rs8176746) that were highly correlated (r²|$\approx$| 1) with the blood O and B alleles, respectively. We also used two additional SNPs, rs8176704 and rs574347, to define subtypes of the A allele (A¹ and A²; refs. 4, 20). For haplotype analysis, we used the ‘haplo.stats’ R package. To determine secretor status, we used the FUT2 SNP rs601338 (G > A), and individuals with the homozygote A/A genotype were defined as nonsecretors (20, 24). Genotypes of three other SNPs (rs812936, rs28362459, and rs3894326) on FUT3 were used to determine Lewis antigen status (positive and negative). Previous studies found that these SNPs had good genotype-phenotype correlation in Caucasians and could identify Lewis negative individuals (16, 19). In this study, we determined ABO alleles (O, A, A¹, A², and B), secretor status, and Lewis antigen groups as shown in Supplementary Table S1. (Details of all 8 SNPs are presented in Supplementary Table S2.)

We compared the distributions of age, sex, blood group alleles, secretor status, and Lewis antigens in cases and controls. To assess the relationships of ABO blood group alleles, secretor status, and Lewis antigen groups with pancreatic cancer risk, we used unconditional logistic regression models and estimated ORs and 95% confidence intervals (CI) for pancreatic cancer risk adjusting for age and sex. To further explore relationships of ABO with secretor status and Lewis antigens, we additionally performed joint analysis of ABO blood group alleles (O vs. non-O allele; number of non-O alleles; O versus A allele; number of A allele; number of A1 allele; number of A2 allele; O vs. B allele; number of B allele) with secretor status (secretor and nonsecretor) and Lewis antigens (positive and negative), separately. In the joint analyses, multiplicative interactions were also evaluated between ABO groups and secretor status and between ABO groups and Lewis antigens adjusting for age and sex. For sensitivity analyses, we also examined the interactions with further adjustment for confounding factors, for example, diabetes, obesity, smoking, and alcohol drinking. In addition, we performed a three-way joint analysis of blood group (O vs. non-O), secretor (secretor or nonsecretor), and Lewis antigen group (positive or negative) and tested the three-way interaction using likelihood ratio test. All statistical analyses were conducted with the R software program (R version 3.6.3), and we reported two-sided P values.

The genotype and phenotype data used in this study are available through dbGAP (phs000206.v5.p3 for the PanScan and phs000648.v1.p1 for the PanC4). Covariate data we used are available via data sharing agreements upon request.

The distributions of age, sex, blood group allele, secretor status and Lewis antigen groups in cases and controls is shown in Table 1. Study participants were largely in the range of age between 61 and 70 years (29% in cases and 38% in controls). The participants were a little more frequently men (54% of cases and 63% of controls), and the proportion of men was higher in controls than cases. The most common blood types were A among cases and O among controls, and the majority of both cases and controls were in the secretor group (78%) and had positive Lewis antigen (90%), i.e., normal or semi-normal function of the Lewis antigen.

We examined the associations of ABO blood alleles, secretor status, and Lewis antigen groups with the risk of pancreatic cancer (Table 2). As expected, non-O blood groups were associated with a greater risk of pancreatic cancer [adjusted OR (95% CI) = 1.39 (1.31–1.47)] compared with the O blood group. The risk was higher with increasing number of non-O alleles [adjusted OR (95% CI) = 1.36 (1.27–1.45) for one allele and 1.49 (1.36–1.63) for two alleles]. Among the non-O blood groups, individuals who were homozygotes for the A blood group had the greatest risk of pancreatic cancer [adjusted OR (95% CI) = 1.76 (1.49–2.08) for the A¹A² subtype; adjusted OR (95% CI) per A allele = 1.22 (1.16–1.28) per A¹ allele, 1.13 (1.04–1.22) per A² allele]. This is consistent with the hypothesis that individuals with the A¹ allele express more A antigen than individuals with the A² allele. When comparing risk estimates per allele between A and B, our study participants had a OR of 1.22 of pancreatic cancer risk per A allele (95% CI = 1.17–1.28) while they had a OR of 1.15 per B allele (95% CI = 1.07–1.24). Conversely, we observed no significant marginal associations of pancreatic cancer risk with secretor status and modest associations with Lewis antigen group [adjusted OR (95% CI) = 1.01 (0.95–1.09) and 1.10 (1.00–1.21), respectively].

We analyzed the joint association between ABO blood groups and secretor status in relation to pancreatic cancer risk (Table 3). We consistently observed greater risks of pancreatic cancer among individuals with non-O alleles among both secretors and nonsecretors; however, the per-non-O-allele OR was greater among secretors than nonsecretors [adjusted OR (95% CI) = 1.30 (1.24–1.36) vs. 1.10 (1.00–1.20), respectively; P_interaction = 0.001]. The difference in ABO allele ORs between secretors and nonsecretors was greatest for the A¹ allele [adjusted OR (95% CI) per A¹ allele = 1.28 (1.21–1.35) vs. 1.01 (0.91–1.13); P_interaction = 0.0002], while the differences in ORs were much smaller for A² and B alleles [adjusted OR (95% CI) = 1.13 (1.04–1.24) vs. 1.11 (0.93–1.32) for A² allele, 1.14 (1.05–1.24) vs. 1.18 (1.01–1.39) for B allele]. In our sensitivity analysis, we assessed the interaction effects of ABO blood groups and secretor status adjusting for additional pancreatic-related confounders such as diabetes, obesity, smoking, and alcohol drinking (Supplementary Table S3). All the effects of non-O alleles, in particular A¹, remained significant (all P values ≤ 0.02). We also conducted joint associations for ABO blood groups and Lewis antigen (positive or negative) (Supplementary Table S4; Supplementary Table S5). We found no compelling evidence of effect-measure modification after adjusting for the confounders as the OR associated with non-O alleles were similar in Lewis positive and Lewis negative individuals.

We also performed a three-way joint analysis of non-O blood group with secretor status and Lewis antigen group (Fig. 1; Supplementary Table S4). Individuals with non-O blood types had consistently higher risk relative to those with O blood type, regardless of secretor or Lewis status. Among individuals with O blood type who were also Lewis positive, secretors had lower risk than nonsecretors [adjusted OR (95% CI) = 0.85 (0.76–0.96); P = 0.006], while those with O blood type who were Lewis negative, secretors and nonsecretors had similar risks [adjusted OR (95% CI) = 0.94 (0.77–1.15) and 0.82 (0.58–1.14); P = 0.24 and 0.56, respectively]. In addition, we assessed the association between the risk of pancreatic cancer and different forms of the Lewis antigen predicted by combinations of FUT2 and FUT3 genotypes, regardless of their ABO blood group (Table 4). We found no difference in the risk among genetically inferred Lewis antigens marginal over ABO genotypes.

This large population-based study confirms previous epidemiologic evidence that in Western populations, non-O blood alleles are associated with greater risk of pancreatic cancer as compared with O blood alleles. In the subgroup analyses of A alleles (A¹ and A²), we found that the A¹ allele was the larger contributor to the risk of pancreatic cancer associated with non-O alleles as compared with A² allele. Although we did not observe significant marginal associations of secretor status and pancreatic cancer, we found evidence of effect measure modification between ABO and FUT2: for example, the OR for non-O blood types relative to O blood types was 18% larger among secretors compared with nonsecretors (P_interaction = 0.001). This multiplicative interaction appeared to be driven by the A¹ allele (P_interaction = 0.0002); the A² allele and B blood group did not have significant interaction effects with secretor status on pancreatic cancer risk.

The observed association between secretor status and risk of pancreatic cancer differed according to Lewis antigen status among individuals with O blood type: secretors had lower risk than nonsecretors among Lewis positive individual, while secretors and nonsecretors had similar risk among Lewis negative individuals. This is consistent with the biological interaction between Lewis antigens and secretor status: Lewis-positive secretors express a different Lewis antigen from nonsecretors among Lewis positive individuals, while Lewis negative individuals do not express any Lewis antigen, regardless of secretor status. This association, if true, would indicate that Lewis antigens confer risk independent of blood group antigen expression. However, this observation should be interpreted with caution: the three-way statistical interaction between ABO blood type, secretor status and Lewis antigen status was not significant in our analysis (P_interaction = 0.27), and there was no association between Lewis antigen status and pancreatic cancer risk marginal over ABO blood type.

Our findings on higher risk of pancreatic cancer in non-O blood groups are in line with previous studies on the ABO blood types and pancreatic cancer risk, including case–control studies from as early as the 1960s and 1970s (25–27). Two earlier studies including PanScan participants (one of which determined ABO blood types based on questionnaires and the other using genotype data) also found elevated risk of incident pancreatic cancer associated with A, B, or AB blood groups as compared with the O blood group (3, 4). Similarly, studies in other ethnic groups such as East Asians have identified a greater risk of pancreatic cancer among individuals with non-O blood types as compared with the risk among those with O blood type (e.g., OR = 1.6–1.7 for A vs. O blood type; refs. 8, 28, 29).

In our study, among the non-O alleles, the A allele was associated with greater risk of pancreatic cancer than the B allele [e.g., adjusted per-allele OR (95% CI) = 1.22 (1.17–1.28) and 1.15 (1.07–1.24), respectively]. This is consistent with a meta-analysis result of 24 studies (10,415 pancreatic cancer cases and 869,044 controls) that found a higher risk of pancreatic cancer associated with non-O blood groups (OR = 1.4, 1.2, and 1.3 for A, B, and AB, respectively; ref. 8). However, it is still not clear which of the non-O blood groups (A, B, and AB) is associated with the highest risk of pancreatic cancer as compared with the O blood group, and there are conflicting results (4, 8, 30).

In addition, we found that the high risk associated with the A allele was mainly due to the A¹ allele, rather than the A² allele [e.g., adjusted per-allele OR (95% CI) = 1.22 (1.16–1.28) and 1.13 (1.04–1.22) for A¹ and A² allele, respectively], which is consistent with previous findings from our consortia data. For example, a matched case–control study of PanScan using the same four SNPs that we used to define ABO blood groups (20) observed that individuals with genotype A¹/O had an OR of 1.48 (95% CI = 1.23–1.78), and genotype A¹/A¹ had an OR of 1.71 (95% CI = 1.18–2.47) relative to those with genotype O/O, while those with A²/O or A¹/A² had no significant associations. A previous large multicenter study within the PANcreatic Disease ReseArch (PANDoRA) consortium found that, as compared with the O allele, only A¹ carriers had an increased risk of pancreatic cancer [OR (95% CI) = 1.25 (1.02–1.51) for A¹/O, 1.57 (1.11–2.23) for A¹/A¹, and 1.71 (1.04–2.80) for A¹/A²], while carriers of A² and B allele did not (5). However, more work is needed to solve the mechanism of how different glycosyltransferase activity is biologically related to the risk of pancreatic cancer.

Because FUT2 gene encodes an enzyme (secretor) allowing the secretion of ABO antigens into body fluids (31, 32), we wanted to assess potential effects of the interplay between genetically derived ABO blood groups and secretor status on pancreatic cancer. Notably, we found significant interactions of secretor status with non-O alleles, particularly the A¹ allele, but not the A² allele and B allele. Our prior study of PanScan did not detect these interactions, despite being a matched case–control design, possibly due to the small study sample size (total n = 3,000) (20). However, the direction of the joint effects was consistent between our study and the previous study (i.e., a greater risk among individuals with non-O blood group and secretors). For other diseases and phenotypes, secretor status has modified the disease associations with ABO blood groups. For example, a recent report observed that the nonsecretor status reduced the risk of mortality among patients hospitalized with COVID-19; this reduction was larger among individuals with non-O blood groups, especially A blood group (33). Other studies reported that both ABO and FUT2 were associated with serum lipase activity and pancreatitis risk (15), and that the association between non-O blood types and recurrent urinary tract infection was restricted to nonsecretors (34). But in other studies, secretor status did not modify well-established associations between ABO blood group and disease (e.g., the inverse association between non-O blood types and venous thromboembolism; ref. 35).

The joint effects of Lewis antigens, secretor, and ABO blood groups have long been studied for various diseases from infections (36–39) to common diseases including cancers (40–42). Lewis antigens may be particularly relevant for pancreatic cancer, as Lewis antigen status is associated with levels of carbohydrate antigen 19–9 (CA19–9), a biomarker for early detection of pancreatic cancer (43–47). In this study, we found that Lewis antigen status was not directly associated with pancreatic cancer risk and also did not have risk interactions with ABO blood groups. This suggests that Lewis status is associated with CA-19–9 levels in the presence of disease, but is not itself a risk factor or effect modifier. Our finding will require confirmation by other population-based studies in the future.

This study was based on two large consortia of pancreatic cancer data, the Pancreatic Cancer Cohort Consortium and the Pancreatic Cancer Case–Control Consortium, which provided a large number of pancreatic cancer cases from 16 cohorts and 13 case–control studies. Thus, the large sample size afforded sufficient statistical power enough to detect significant interaction effects across ABO, secretor, and Lewis antigens. In addition, our results are unlikely to be confounded by population structure because all study participants were relatively genetically homogeneous (with a predominance of European ancestries).

Nevertheless, there are several limitations of our study to be noted. Although we adjusted for potential confounding effects of age and sex, there is a possibility of residual confounding effects from other risk factors of pancreatic cancer that were not adjusted for in our study. To complement it, we did sensitivity analyses to examine the effect modifications by secretor or Lewis activity adjusting for additional covariates such as diabetes, obesity, smoking, and alcohol drinking. Our findings remained the same despite some limitations on the covariates including missing data (up to 32%) and differences in data collection between case–control and cohort studies. Also, there is some chance of exposure misclassification because genetically determined exposure status might not be the same as the phenotype. For example, for Lewis antigens, a previous study reported significant divergence between genotypes and phenotypes [e.g., P = 0.002 for the difference in the frequencies of Lewis negative derived by genotypes and phenotypes (18% vs. 31%, respectively; ref. 48)], and other studies have reported heterogeneity of Lewis antigen expression depending on host conditions such as H. pylori infection (49) or cancer progression (50). However, our main findings about ABO and secretor (FUT2) genotypes would be likely to have no misclassification bias as ABO genotypes and blood groups have showed high accuracy in a number of previous studies (51, 52), and FUT2 genotype and secretor status also have been found to be highly correlated (20, 53). Finally, including only European-ancestry individuals limits the generalizability of our findings given that the ABO, secretor (FUT2), and Lewis antigen (FUT3) genotypes and their allele frequencies are different across ethnic groups. Hence, further genetic epidemiology studies including ethnically diverse populations are clearly warranted.

In conclusion, we found that the FUT2 polymorphism determining secretor status modulates the associations between non-O blood alleles and the risk of pancreatic cancer. Particularly, secretor individuals with A blood alleles had a higher risk of pancreatic cancer than nonsecretors with A alleles, and the effect modification by the secretor status was significant.

A.P. Klein reports grants from NCI during the conduct of the study; grants from NCI; and grants from Lustgarten Foundation outside the submitted work. H.A. Risch reports grants from NCI, NIH during the conduct of the study. P. Kraft reports grants from NIH during the conduct of the study. No disclosures were reported by the other authors.

J. Kim: Conceptualization, data curation, formal analysis, writing–original draft, writing–review and editing. C. Yuan: Writing–review and editing. L.T. Amundadottir: Data curation, writing–review and editing. B.M. Wolpin: Data curation, writing–review and editing. A.P. Klein: Data curation, writing–review and editing. H.A. Risch: Writing–review and editing. P. Kraft: Conceptualization, data curation, supervision, writing–original draft, writing–review and editing.

We thank to the participants and study staff of all studies in the pancreatic cancer consortia, PanScan I-III and PanC4, for their valuable contribution. This research was supported by grants from the U.S. NIH CA194393 (to P. Kraft) and CA261339 (to P. Kraft).

The publication costs of this article were defrayed in part by the payment of publication fees. Therefore, and solely to indicate this fact, this article is hereby marked “advertisement” in accordance with 18 USC section 1734.