ABO blood group alleles and the risk of pancreatic cancer in a Japanese population

Abstract

Several studies have investigated a possible association between the ABO blood group and the risk of pancreatic cancer (PC), but this association has not been fully evaluated in Asian populations. The present study aimed to assess the impact of genotype‐derived ABO blood types, particularly ABO alleles, on the risk of PC in a Japanese population. We conducted a case–control study using 185 PC and 1465 control patients who visited Aichi Cancer Center in Nagoya, Japan. Using rs8176719 as a marker for the O allele, and rs8176746 and rs8176747 for the B allele, all participants’ two ABO alleles were inferred. The impact of ABO blood type on PC risk was examined by multivariate analysis, with adjustment for potential confounders to estimate odds ratios (OR) and 95% confidence intervals (CI). An increased risk of PC was observed with the addition of any non‐O allele (trend P = 0.012). Compared with subjects with the OO genotype, those with AO and BB genotypes had significantly increased OR of 1.67 (CI, 1.08–2.57) and 3.28 (CI, 1.38–7.80), respectively. Consistent with earlier reports showing a higher risk of PC for individuals with the non‐O blood type, the previously reported protective allele (T) for rs505922 was found to be strongly correlated (r ²=0.96) with the O allele. In conclusion, this case–control study showed a statistically significant association between ABO blood group and PC risk in a Japanese population. Further studies are necessary to define the mechanisms by which the ABO gene or closely linked genetic variants influence PC risk. (Cancer Sci 2011; 102: 1076–1080)

The incidence of pancreatic cancer (PC) is increasing in Japan, and this cancer is now the sixth leading cause of cancer death.^{( 1} , ^{2 )} Early detection of PC in its operable stage is difficult and curative treatment such as complete surgical removal is limited, which together give PC a 5‐year relative survival rate of only 5.5%.^{( 3 )} This suggests that epidemiological approaches to predicting the risk of PC may play an important role in identifying PC high‐risk groups and ultimately decreasing the number of PC deaths.

Risk factors for PC include advancing age, smoking, obesity, diabetes mellitus, family history of PC, alcohol consumption and chronic pancreatitis.^{( 2} , ⁴ , ⁵ , ⁶ , ⁷ , ^{8 )} Recently, a large prospective cohort study also showed a statistically significant association between ABO blood type and risk of PC.^{( 9 )} Specifically, this study suggested that people with blood type O have a lower risk of PC than those with blood type A, B or AB. A genome‐wide association study (GWAS) comparing PC cases and controls (PanScan), which includes the forementioned prospective cohort study population,^{( 9 )} identified a single‐nucleotide polymorphism (SNP) within the first intron of the ABO gene (rs505922) that was associated with PC risk.^{( 10 )} Furthermore, using almost the same population as PanScan, Wolpin et al. ^{( 11 )} reported that the ABO blood type could be inferred from two SNPs (rs505922 and rs8176746), and concluded the relationship between PC risk and ABO blood type. The same study found that PC risk increased with the addition of each non‐O allele, with the highest risk noted among participants with blood genotype BB.^{( 11 )}

However, most studies of the association between PC risk and ABO blood types conducted to date have been in Caucasian populations. Interestingly, a recent GWAS of PC among Japanese showed only a weak association between PC risk and 9q34.2 (rs505922).^{( 12 )} To further evaluate the potential impact of ABO loci on PC risk among the Asian population, we conducted a case–control study to assess the impact of ABO gene polymorphisms on PC risk in a Japanese population.

Materials and Methods

Study subjects.  The study cases were 185 PC patients with no prior history of cancer who were diagnosed at Aichi Cancer Center Hospital (ACCH), Nagoya, Japan, between January 2001 and November 2005. The control subjects were 1465 randomly selected non‐cancer outpatient visitors to ACCH during the same period who had no history of any cancer. All subjects were enrolled during their first visit to ACCH in the Hospital‐based Epidemiological Research Program II at ACCH (HERPACC‐II) between January 2001 and November 2005. The framework for HERPACC‐II has been described elsewhere.^{( 13} , ^{14 )} Briefly, all first‐visit outpatients to ACCH aged 20–79 years were asked to complete a self‐administered questionnaire regarding their lifestyle before the development of current symptoms. The responses were checked by trained interviewers. Outpatients were also asked to provide a 7 mL blood sample. A total of 96.7% of contacted subjects completed the questionnaire (28 766 out of 29 736) and approximately 50% of responders provided a blood sample. All data were loaded into the HERPACC database, which is periodically synchronized with the hospital cancer registry system to update the data on cancer incidence. Approximately 35% of subjects (46% of male subjects and 28% of female subjects) were diagnosed with cancer within a year of their first visit. In this study, we defined patients who were diagnosed with PC within a year of their first visit as a case population. This approach was aimed at taking a window period for final diagnosis of PC rather than prospectively identifying cases. A total of 75.7% of PC cases have histological confirmation. Among those with histological confirmation, 92.1% of them were diagnosed as ductal adenocarcinoma. Our previous study showed that the lifestyle patterns of first‐visit outpatients corresponds with those of individuals randomly selected from Nagoya’s general population, confirming external validity for the study.^{( 15 )} The present study was approved by the Ethics Committee of Aichi Cancer Center and informed consent was obtained at the first visit from all participants.

Selection and genotyping of the ABO blood group allele.  In this study, we examined four loci on the ABO gene, namely rs8176719, rs8176746, rs8176747 and rs505922. While the rs505922 locus has previously been reported to be associated with PC risk,^{( 10} , ^{11 )} the other three are responsible for the ABO blood group phenotype in the Japanese population.^{( 16} , ¹⁷ , ¹⁸ , ^{19 )} Upon examination of the cDNA nucleotide sequence of the ABO gene on chromosome 9q34.2, the following has been found for the Japanese population: a 261G deletion in exon6 (rs8176719) results in the O allele and C796A and G803C in exon7 (rs8176746 and rs8176747, respectively) distinguish the B allele from the A allele.^{( 16} , ¹⁷ , ¹⁸ , ¹⁹ , ^{20 )} The DNA of each subject was extracted from the buffy coat fraction using a DNA Blood Mini kit (Qiagen, Tokyo, Japan). All loci were examined by TaqMan assays (Applied Biosystems, Foster City, CA, USA). The principle of the TaqMan real‐time PCR assay system using fluorogenic probes and 5′ nuclease has been described by Livak.^{( 21 )} Commercial probes and primers were used for rs505922 (Applied Biosystems), while custom probes and primers were specifically designed for the other three loci. Genotyping for the ABO 261G deletion (rs8176719) was done by PCR using 5′‐CCTGTGTGGATGTGCAGTAGGA‐3′ and 5′‐CGTTGAGGATGTCGATGTTGAA‐3′ primers, and Fam‐TCCTCGTGGTGACCCCTTGGC‐TAMRA and Vic‐ATGTCCTCGTGGTACCCCTTGGCT‐TAMRA probes.^{( 16 )} Genotyping for the ABO C796A and G803C (rs876746 and rs876747, respectively) was done by PCR using 5′‐CTGCACCTCTTGCACCGAC‐3′ and 5′‐AGGCCTTCACCTACGAGCG‐3′ primers, and Fam‐CCCGAAGAACCCCCCCAGGTAGTAGAAA‐TAMRA and Vic‐CCCGAAGAACGCCCCCATGTAGTAGAAA‐TAMRA probes.^{( 16 )} All of the assays were done in 96‐well PCR plates using a 7500 Fast Real‐time PCR System with the corresponding 7500 Fast System SDS software (Applied Biosystems). For assays using custom probes, we used positive and negative control samples as described elsewhere.^{( 16 )} Amplification reactions were done in duplicate with 30 ng (5 μL) of template DNA, 2× TaqMan Universal Master Mix buffer (Applied Biosystems), and 20× primer and probe mix (Applied Biosystems). Thermal cycling was performed under the following conditions: denaturation at 95°C for 20 s followed by 40 cycles of 95°C for 3 s and 62°C for 30 s. In our laboratory, the quality of genotyping is routinely and statistically assessed using the Hardy–Weinberg test and retyping of a random sampling of 5% of subjects.

Assessment of exposure.  Exposure to potential PC risk factors was assessed from the self‐administered questionnaire results, which were completed before diagnosis during the first visit to ACCH and reviewed by trained interviewers. Subjects were specifically questioned about their lifestyle before the onset of the symptoms, which impelled their visit to ACCH. Daily alcohol consumption in grams was determined by summing the pure alcohol amount in the average daily consumption of Japanese sake (rice wine), shochu (distilled spirit), beer, wine and whiskey, with one cup of Japanese sake (180 mL) considered equivalent to 23 g of ethanol; one drink of shochu (108 mL) to 23 g; one large bottle of beer (633 mL) to 23 g; one glass of wine (80 mL) to 10 g; and one shot of whiskey (28.5 mL) to 11.5 g. Cumulative smoking exposure was measured in pack‐years (PY), the product of the average number of packs per day and the number of years of smoking. Height and bodyweight at baseline and weight at age 20 years were self‐reported. Current body mass index (BMI) and BMI at age 20 were calculated by dividing the weight in kilograms by the height in square meter, and expressed as kilogram per square meter. The past histories of subjects were also obtained from the self‐administered questionnaire results. Family history of PC was considered positive when at least one parent or sibling had a history of PC.

Statistical analysis.  All statistical analyses were performed using Stata version 10 (Stata Corp., College Station, TX, US). A P‐value <0.05 was considered statistically significant. Differences in characteristics between cases and controls were assessed using the Chi‐squared test. Odds ratios (OR) and 95% confidence intervals (CI) were estimated using an unconditional logistic regression model adjusted for potential confounders. Potential confounders considered in the multivariate analysis were age, sex (male or female), PY of smoking (<5, <20, <40 or ≥41), drinking habit (non‐drinker, <5, <23 or ≥23 g/day), current BMI (<18.4, <22.4, <24.9, <29.9 or ≥30.0 kg/m²), BMI at age 20 years (<18.5, <22.4, <24.9, <29.9 or ≥30.0 kg/m²), history of diabetes (yes or no), and family history of PC (yes or no). Gene‐environmental interactions, including interaction terms, were assessed by the logistic model. We evaluated linkage disequilibrium (LD) by means of linkage disequilibrium coefficients (r ²). Accordance with Hardy–Weinberg Equilibrium was assessed by the Chi‐squared test.

Results

Background characteristics of all subjects are shown in Table 1. Men accounted for 68.7% of case subjects and 74.9% of control subjects. Compared with the control group, the case group had a significantly higher prevalence of heavy smokers (P = 0.005), higher prevalence of a history of diabetes mellitus (P < 0.001), lower current BMI (P = 0.021) and higher BMI at age 20 years (P = 0.018).

Table 1.

 Characteristics of case subjects and controls

	Case (%) n = 185	Control (%) n = 1465	P‐values†
Age (years)
<40	10 (5.41)	75 (5.12)	0.999
≥40 and <50	19 (10.27)	147 (10.03)
≥50 and <60	60 (32.43)	479 (32.70)
≥60 and <70	60 (32.43)	484 (33.04)
≥70	36 (19.46)	280 (19.11)
Sex
Male	127 (68.65)	1097 (74.88)	0.068
Female	58 (31.35)	368 (25.12)
Current BMI (kg/m2)
<18.5	15 (8.11)	61 (4.16)	0.021
≥18.5 and <22.5	84 (45.11)	555 (37.88)
≥22.5 and <25	47 (25.41)	478 (32.63)
≥25 and <27.5	25 (13.51)	245 (16.72)
≥27.5	14 (7.57)	111 (7.58)
Unknown	0 (0.00)	15 (1.02)
BMI at age 20 years (kg/m2)
<18.5	14 (7.57)	168 (11.47)	0.018
≥18.5 and <22.5	112 (60.54)	950 (64.85)
≥22.5 and <25	36 (19.46)	226 (15.43)
≥25 and <27.5	11 (5.95)	72 (4.91)
≥27.5	6 (3.24)	14 (0.96)
Unknown	6 (3.24)	35 (2.39)
Cigarette pack‐years
<5	69 (37.30)	638 (43.55)	0.005
≥5 and <20	22 (11.89)	206 (14.06)
≥20 and <40	34 (18.38)	308 (21.02)
≥40	60 (32.43)	304 (20.75)
Unknown	0 (0.00)	9 (0.61)
Drinking (g ethanol/day)
Non	56 (30.27)	488 (33.31)	0.568
<5	53 (28.65)	425 (29.01)
≥5 and <23	43 (23.24)	342 (23.34)
≥23	31 (16.76)	193 (13.17)
Unknown	2 (1.08)	17 (1.16)
History of diabetes mellitus
Yes	37 (20.00)	126 (8.60)	<0.001
No	148 (80.00)	1339 (91.40)
Family history of pancreatic cancer
Yes	8 (4.32)	58 (3.96)	0.811
No	177 (95.68)	1407 (96.04)

†Chi‐squared test. BMI, body mass index.

Table 2 shows the allele frequencies of the ABO gene among the control population. Each participant’s two ABO alleles were inferred using rs8176719 as a marker for the O allele, and rs8176746 and rs8176747 for the B allele. The resulting allele frequencies were consistent with information from the HapMap project and previous studies.^{( 17} , ¹⁸ , ^{22 )} Rs505922 was in strong but not perfect LD with the O allele (r ²=0.96).

Table 2.

 Frequencies of ABO alleles among control participants

ABO alleles	SNPs†		ABO allele frequencies (%)
	rs8176719	rs8176746/rs8176747
O	del	C/G	54.64
A	G	C/G	28.33
B	G	A/C	17.03

†For rs505922, the T and C allele frequencies were 55.01% and 44.98%, respectively. SNP, single‐nucleotide polymorphism.

We assessed the risk of PC according to genotype‐derived ABO blood type among all study participants (Table 3). Compared with individuals with blood type O, those with blood type A were at greater risk of PC. As shown in Table 4, an increased risk of PC was observed with the addition of any non‐O allele (trend P = 0.012). Compared with individuals with the OO genotype, those with AO and AA genotypes had OR of 1.67 (95% CI, 1.08–2.57) and 1.53 (95% CI, 0.80–2.91), respectively, whereas individuals with the BO and BB had OR of 1.24 (95% CI, 0.74–2.06) and 3.28 (95% CI, 1.38–7.80), respectively (Table 5).

Table 3.

 Multivariable‐adjusted OR (95% CI) for incident PC by genotype‐derived ABO blood type

	O	A	AB	B
No. cases (%)/controls (%)	38 (20.54)/428 (29.22)	85 (45.95)/572 (39.04)	22 (11.89)/141 (9.62)	40 (21.62)/324 (22.12)
Multivariable‐adjusted OR†	1.0	1.64 (1.08–2.48)	1.67 (0.94–2.98)	1.42 (0.88–2.31)

†Multivariable adjustment by age, sex, smoking status, history of diabetes mellitus, and BMI, drinking habit, family history of PC and BMI at age 20 years. CI, confidence interval; OR, odds ratio; PC, pancreatic cancer.

Table 4.

 Multivariable‐adjusted OR (95% CI) for incident PC with the addition of non‐O allele

	O,O	O,X	X,X	Trend P
No. cases/controls	38/428	101/745	46/292	0.012
Multivariable‐adjusted OR†	1.0	1.50 (1.00–2.25)	1.77 (1.11–2.83)

†Multivariable adjustment by age, sex, smoking status, history of diabetes mellitus, and BMI, drinking habit, family history of PC and BMI at age 20 years. X: A or B allele. CI, confidence interval; OR, odds ratio; PC, pancreatic cancer.

Table 5.

 Multivariable‐adjusted OR (95% CI) for incident PC by genotype‐derived ABO blood group allele

Second allele	First allele
	O	A	B
O
No. cases/controls	38/428	69/455	32/290
Multivariable‐adjusted OR†	1.0	1.67 (1.08–2.57)	1.24 (0.74–2.06)
A
No. cases/controls	–	16/117	22/141
Multivariable‐adjusted OR†	–	1.53 (0.80–2.91)	1.67 (0.93–2.98)
B
No. cases/controls	–	–	8/34
Multivariable‐adjusted OR†	–	–	3.28 (1.38–7.80)

†Multivariable adjustment by age, sex, smoking status, history of diabetes mellitus, and BMI, drinking habit, family history of PC and BMI at age 20 years. CI, confidence interval; OR, odds ratio; PC, pancreatic cancer.

Because smoking, obesity, history of diabetes mellitus, and alcohol consumption are well‐known modifiable risk factors for PC, we used joint models to explore the possibility that the effect of ABO blood type might be modified by potential risk factors. In combination with smoking (≥5 pack‐year), overweight at age 20 years (BMI ≥ 25 kg/m²), diabetes mellitus, or drinking alcohol (≥5 g ethanol/day), the non‐O blood type was associated with OR of 2.24 (95% CI, 1.22–4.10), 2.38 (95% CI, 1.14–4.95), 4.06 (95% CI, 2.19–7.55) and 1.74 (95% CI, 1.05–2.88), respectively, compared with subjects who had the O blood type and lack the exposure (smoking, obesity, history of diabetes mellitus, and alcohol consumption). These OR are almost consistent with a multiplicative OR model for the joint effects of these factors and non‐O blood type. These was no evidence for statistically significant interactions.

Discussion

The present study found that the risk of PC was higher among those with the non‐O blood type than those with the O blood type in a Japanese population. By analyzing polymorphisms in three loci of the ABO allele coding region (rs8176719, rs8176746 and rs8176747), we identified the ABO blood type of each patient. The rs505922 locus was in strong LD (r ²=0.96) with the O allele. To our knowledge, this is the first study to replicate the correlation between SNPs at the ABO gene loci and the development of PC observed in Western populations in an Asian population.

Although previous reports have suggested an association between ABO blood type and PC risk,^{( 23} , ²⁴ , ²⁵ , ^{26 )} these studies had notable limitations, such as a small sample size and poorly selected control populations.^{( 9} , ^{27 )} Recently, some large studies have identified more reliable and consistent associations between ABO blood type and PC risk.^{( 9} , ¹⁰ , ^{11 )} These studies showed that people with a non‐O blood type had a greater risk of PC than those with the O blood type. However, these study populations referred partially to the same samples and were primarily composed of Caucasian subjects.^{( 9} , ¹⁰ , ^{11 )} Meanwhile, an association between serotype‐derived ABO blood type and PC incidence among Han Chinese patients has been reported.^{( 27 )} Among Han people, those with blood type A or AB had a higher risk of PC than those with blood type O, but did not have a higher risk than those with blood type B. In our study, compared with subjects with the OO genotype, only those with AO and BB genotypes had significantly increased OR of 1.67 (CI, 1.08–2.57) and 3.28 (CI, 1.38–7.80), respectively. Although these inconsistencies among the Asian population with blood type B require further evaluation, as a whole, our study confirmed previous findings in Western populations.^{( 9} , ¹⁰ , ^{11 )} Genetic variants in the ABO gene therefore appear associated with PC risk not only in Western but also in Asian populations.

Nevertheless, a recent GWAS of 991 PC case subjects and 5209 controls indicated that several chromosomal loci (6p24.3, 12p11.21 and 7q36.2) are significantly associated with an increased risk of PC in Japanese.^{( 12 )} The SNPs at these loci are within the same LD blocks as genes that have been implicated in the oncogenesis of PC.^{( 12 )} However, these loci do not include the ABO gene locus (9q34.2), and only a weak association between PC risk and ABO gene locus was detected. Although the reason for this weak detection of the ABO gene locus is unclear, the results may have been affected by the heterogeneity of the control population and/or potential confounders.

The mechanism by which genetic variants in the ABO gene locus influence the risk of PC has not been fully investigated. The ABO gene encodes a glycosyltransferase that catalyzes the transfer of carbohydrates to the H antigen, forming the antigenic structures of the ABO blood groups.^{( 10 )} The A and B alleles of the ABO gene encode proteins that differ minimally in amino acid sequence but catalyze the transfer of different carbohydrates (N‐acetylgalactosamine or galactose) onto the H antigen to form the A or B antigens. In contrast, the O allele contains a single base deletion and does not lead to the production of A or B antigens.^{( 10 )} ABO or ABH antigens are expressed on the surface of red blood cells and numerous other tissues throughout the body.^{( 28 )} Glycoconjugates, such as the ABO antigen, are important mediators of intercellular adhesion and membrane signaling, both of which are critical to the progression and spread of malignant cells.^{( 11} , ^{29 )} These cell surface molecules are recognized by the host immune response and may promote immunosurveillance for malignant cells.^{( 11} , ^{30 )} One hypothesis proposes that the difference in ABO antigens in gastric mucins influences the properties of Helicobacter pylori binding, and accounts for the difference in PC risk among the ABO blood type.^{( 31 )}

Two recent GWAS have suggested that SNPs at the ABO gene locus are associated with two serum markers of inflammation‐tumor necrosis factor‐α (TNF‐α) and soluble intercellular adhesion molecule‐1 (ICAM‐1).^{( 32} , ^{33 )} TNF‐α is an inflammatory cytokine known to modulate the rate of pancreatic ductal cell apoptosis, and plasma levels of soluble ICAM‐1 are associated with the risk of developing diabetes mellitus, a known predisposing factor for PC.^{( 34} , ^{35 )} These reports suggest the possibility that the ABO blood group may correlate with the systemic inflammatory state, thereby influencing the risk of PC.^{( 11 )} We therefore suspect that the association between ABO blood type and risk of PC results from a genetic variation that is closely linked to the ABO gene locus.

Our study has several methodological issues that warrant discussion. First, because the control population was selected from non‐cancer patients at Aichi Cancer Center Hospital, it is reasonable to assume that these subjects represented the same base population as that from which the cases were selected, warranting internal validity. With regard to the external validity of our results, we previously showed that individuals selected randomly from our control population were similar to the general population of Nagoya City in terms of the exposure of interest.^{( 15 )} Second, blood type was derived from genotype rather than serological data. Although methods for determining blood type based on a subject’s DNA are well established,^{( 16} , ²⁰ , ³⁶ , ^{37 )} measurement error and exposure misclassification might have occurred. Additionally, the non‐deletion type O allele and cis‐AB allele could not be detected using our method, albeit that these alleles are extremely rare among the Japanese population.^{( 16} , ¹⁸ , ^{19 )} The ratio of Cis‐AB to total AB blood type was estimated to be approximately 0.012%, and the allele frequency of cis‐AB was 1.1 × 10⁻⁵ among the Japanese population.^{( 38} , ^{39 )} Third, a potential source of bias was the medical background of the controls. However, our previous study in women demonstrated that this had only limited impact: more than 66% of non‐cancer outpatients at ACCH have no specific medical condition, while the remaining 34% have specific diseases such as benign tumors, non‐neoplastic polyps or both (13.1%), mastitis (7.5%), gastrointestinal disease (4.1%) or benign gynecological disease (4.1%).^{( 2} , ^{40 )} The situation for men is thought to be comparable.^{( 2 )} Lack of consideration of chronic pancreatitis as a potential cause of PC in the analysis could also be a limitation in the present study; however, the consideration of drinking, a definite cause of chronic pancreatitis, might minimize this possibility.^{( 41 )} Moreover, a potential limitation was residual confounding by known and unknown risk factors; in particular, the limited number of cases, particularly when stratified by genotype, indicates the need to replicate our findings in a larger study. In addition, case–control studies have an intrinsic information bias. However, the HERPACC system is less prone to this bias than typical hospital‐based studies, because the data for all patients are collected before diagnosis. Chiefly, participants and investigators were not informed about the association between PC and factors such as the ABO genotype, mitigating information bias in the analysis. Lastly, our study was limited to a Japanese population, and the results cannot necessarily be extrapolated to other populations.

In summary, our case–control study showed that SNPs at the ABO gene locus were associated with the risk of PC in a Japanese population. This finding supports the recently reported association between non‐O blood types and elevated risk of PC. Further investigation into these findings in other ethnic groups is warranted, and the mechanism by which these SNPs influence PC risk should be fully elucidated.

Disclosure Statement

None of the authors have any financial or other interest that could be construed as a conflict of interest with regard to the submitted manuscript.