Article Text

Original article
Genome-wide association study of gastric adenocarcinoma in Asia: a comparison of associations between cardia and non-cardia tumours
  1. Nan Hu1,
  2. Zhaoming Wang1,
  3. Xin Song2,
  4. Lixuan Wei3,
  5. Byung Sik Kim4,
  6. Neal D Freedman1,
  7. Jiwon Baek5,
  8. Laurie Burdette1,6,
  9. Jiang Chang3,
  10. Charles Chung1,6,
  11. Sanford M Dawsey1,
  12. Ti Ding7,
  13. Yu-Tang Gao8,
  14. Carol Giffen9,
  15. Yaling Han3,
  16. Myunghee Hong5,
  17. Jia Huang2,
  18. Hee Sung Kim4,
  19. Woon-Puay Koh10,
  20. Linda M Liao1,
  21. Yi Min Mao2,
  22. You-Lin Qiao11,
  23. Xiao-Ou Shu12,
  24. Wen Tan3,
  25. Chaoyu Wang1,
  26. Chen Wu3,
  27. Min-Jie Wu13,
  28. Yong-Bing Xiang8,
  29. Meredith Yeager1,6,
  30. Jeong Hwan Yook4,
  31. Jian-Min Yuan14,
  32. Peng Zhang2,
  33. Xue-Ke Zhao2,
  34. Wei Zheng12,
  35. Kyuyoung Song5,
  36. Li-Dong Wang2,
  37. Dongxin Lin3,
  38. Stephen J Chanock1,
  39. Alisa M Goldstein1,
  40. Philip R Taylor1,
  41. Christian C Abnet1
  1. 1Division of Cancer Epidemiology and Genetics, National Cancer Institute, National Institutes of Health, Bethesda, Maryland, USA
  2. 2Henan Key Laboratory for Esophageal Cancer Research of The First Affiliated Hospital, Zhengzhou University, Zhengzhou, Henan, People's Republic of China
  3. 3Department of Etiology & Carcinogenesis, Cancer Institute and Hospital Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, People's Republic of China
  4. 4Department of Surgery, Asan Medical Center, University of Ulsan College of Medicine, Seoul, South Korea
  5. 5Department of Biochemistry and Molecular Biology, University of Ulsan College of Medicine, Seoul, South Korea
  6. 6Cancer Genome Research Laboratory, Division of Cancer Epidemiology and Genetics, Advanced Technology Program, SAIC-Frederick Inc., NCI-Frederick, Frederick, Maryland, USA
  7. 7Shanxi Cancer Hospital, Taiyuan, Shanxi, People's Republic of China
  8. 8Shanghai Cancer Institute, Shanghai, People's Republic of China
  9. 9Information Management Services Inc., Silver Spring, Maryland, USA
  10. 10Duke-NUS Graduate Medical School Singapore, Singapore, Singapore
  11. 11Department of Epidemiology, Cancer Institute and Hospital Chinese Academy of Medical Sciences, Beijing, People's Republic of China
  12. 12Department of Medicine, Vanderbilt-Ingram Cancer Center, Vanderbilt University, Nashville, Tennessee, USA
  13. 13Cancer Research Center, Xinxiang Medical University, Xinxiang, Henan, People's Republic of China
  14. 14University of Pittsburgh Cancer Institute, Pittsburgh, Pennsylvania, USA
  1. Correspondence to Dr Christian Abnet, Division of Cancer Epidemiology and Genetics, Nutritional Epidemiology Branch, National Cancer Institute, 9609 Medical Center Dr Rm 6e430 MSC 9768, Bethesda, MD 20892, USA; abnet{at}


Objective Genome-wide association studies (GWAS) of gastric cancer have reported differences in single-nucleotide polymorphism (SNP) associations for tumour subtypes, particularly when divided by location into the gastric cardia versus the non-cardia.

Design Here we present results for a GWAS using 2350 East Asian gastric cancer cases divided as 1189 gastric cardia and 1027 gastric non-cardia cases and 2708 controls. We also included up to 3042 cardia cases, 4359 non-cardia cases and 7548 controls for replication from two Chinese studies and one Korean study. From the GWAS, we selected 12 top SNPs for each gastric cancer subtype, 4 top SNPs for total gastric cancer and 1 SNP in MUC1 for replication testing.

Results We observed genome-wide significant associations for rs10074991 in PRKAA1 at 5p13.1 for cardia (p=7.36×10−12) and non-cardia cancers (p=2.42×10−23) with per allele OR (95% CI) for the combined endpoint of 0.80 (0.77 to 0.83). At 6p21.1, rs2294693 near UNC5CL was significantly associated with gastric non-cardia cancer risk (p=2.50×10−8), with OR (95% CI) of 1.18 (1.12 to 1.26), but there was only a nominal association for cardia cancer (p=1.47×10−2). We also confirmed a previously reported association for rs4072037 in MUC1 with p=6.59×10−8 for total gastric cancer and similar estimates for cardia and non-cardia cancers. Three SNPs in PSCA previously reported to be associated with gastric non-cardia cancer showed no apparent association for cardia cancer.

Conclusions Our results suggest that associations for SNPs with gastric cancer show some different results by tumour location in the stomach.

Statistics from

Significance of this study

What is already known on this subject?

  • There are almost one million new cases of gastric cancer each year, and Eastern Asia has the highest rates of any geographic region.

  • Several previous genome-wide association studies (GWAS) studies of gastric cancer in subjects of East Asian ethnicities have reported a modest number of associations for common single-nucleotide polymorphisms.

  • These studies have reported that many single-nucleotide polymorphisms (SNPs) have different results when analysed by location of the tumour within the stomach—cardia versus non-cardia.

What are the new findings?

  • Using a GWAS with subjects of Asian ethnicity and >1000 cases from each sublocation within the stomach, we found that rs10074991 in PRKAA1 at 5p13.1 showed genome-wide significant associations for cardia and non-cardia cancers, with the finding for cardia being novel. Furthermore, we observed that rs2294693 near UNC5CL at 6p21.1 was genome-wide significant for gastric non-cardia cancer.

  • An SNP in MUC1 (rs4072037) showed near genome-wide significance, with similar associations for cardia and non-cardia tumours.

  • Three SNPs in prostate stem cell antigen previously reported to be associated with gastric non-cardia cancer showed subsite-specific findings with some evidence for an association in the non-cardia, but no association for cardia cancer.

How might it impact on clinical practice in the foreseeable future?

  • These GWAS findings highlight the importance of studying gastric cardia and non-cardia cancer in parallel to assess differences in aetiology by location within the stomach.

  • The results from this GWAS increase our understanding of the pathophysiology of gastric cancer and support better understanding of the aetiology of this common cancer.


Despite reductions in gastric cancer incidence rates coinciding with economic development, gastric cancer remains the third leading cause of cancer death worldwide and continues to be a major public health problem.1 As a consequence of ageing populations, the International Agency for Research on Cancer projects that the number of incident cases of gastric cancer will continue to increase for several decades. About 90% of all gastric tumours present as adenocarcinomas, which can be further classified by location. Etiological studies have found numerous differences between gastric cardia adenocarcinomas occurring in the top few centimetres of the stomach and non-cardia adenocarcinomas occurring elsewhere in the stomach, including distinct results for Helicobacter pylori infection,2 ,3 body mass index,4 and other risk factors. Risk factors for gastric cardia adenocarcinomas also differ between Eastern and Western populations, including H. pylori.5

Recently, we published a genome-wide association study (GWAS) that examined the associations between common genetic variants and risk of gastric adenocarcinoma and oesophageal squamous cell carcinoma (ESCC) risk in ethnic Chinese subjects.6 We reported that the strongest association for gastric cardia adenocarcinoma and ESCC were single-nucleotide polymorphisms (SNPs) in a locus on chromosome 10q23 in the PLCE1 gene, but these SNPs showed no association with gastric non-cardia adenocarcinoma. This study demonstrated that genetic susceptibility loci differ between the two main sublocations of gastric cancer among ethnic Chinese, and therefore, each type should be investigated separately when possible.

Two other groups have published GWAS studies of gastric carcinoma in East Asians, each focusing on particular subtypes of gastric cancer. A GWAS of gastric adenocarcinoma from Japan reported significant associations for SNPs in prostate stem cell antigen (PSCA)7 and for SNPs at 1q22, which were associated with gastric adenocarcinomas of diffuse but not intestinal histological type, which are subtypes of adenocarcinoma defined in the Lauren histological classification system.8 This study7 did not report tumour location within the stomach, but other studies report that most gastric cancers in Japan are located in the non-cardia, especially those of the intestinal subtype. A GWAS of non-cardia gastric adenocarcinoma from China reported significant associations with SNPs at 5p13.1 and 3q13.31 with the strongest associations at rs13361707 and rs9841504, respectively. When the authors tested these SNPs using an independent set of 905 subjects with gastric cardia cancer, these SNPs were not significantly associated with risk.9 This study, like most studies conducted in China, did not report associations by Lauren subtype because that histological system is not widely implemented clinically.

Here we extend our previous work by examining additional potential associations for gastric cardia, gastric non-cardia and total gastric adenocarcinoma using our gastric adenocarcinoma GWAS data and three large replication sets from China and Korea.

Materials and methods

Gastric GWAS

For the National Cancer Institute GWAS, subjects were drawn from four prospective cohort studies and one large case–control study as reported in Abnet et al,6 and all subjects used in replication in the original paper were subsequently genotyped using the Illumina 660W-Quad microarray, which included scanning 725 additional gastric cancer cases and 608 additional controls. Similar genotype filtering was applied to those additional genotyped samples. The final analytic data set included 2350 gastric cancer cases and 2708 controls with genotypes for a total of 556 896 SNPs. Demographic characteristics are presented in the online supplementary table S1.

GWAS analysis and replication

We used the final analytic data set to complete full GWAS analysis for three endpoints, cardia cancer, non-cardia cancer and total gastric cancer (the two prior endpoints combined plus a modest number of overlapping subsite and not otherwise specified cases) using the models described in the statistical analysis section. We ordered all results by p value and selected a total of 29 SNPs for replication testing after excluding SNPs tagged to PLCE1 because these were studied in a prior publication.6 All 29 SNPs had a stage one p value<5×10−5, and we distributed the 29 tests among the three endpoints as the 12 top SNPs based on the cardia gastric cancer model, the 12 top SNPs based on the non-cardia gastric cancer model and the 4 top SNPs from the total gastric model. The last SNP was selected as the lowest p value SNP tagged to MUC1 because of the strong prior probability given other recent publications.10 TaqMan assays were designed and optimised at the Cancer Genomics Research Laboratory. For the assays that failed to design, linkage disequilibrium (LD) surrogates rs6717108, rs6768588, rs61364777 and rs10881372 were used for the originally selected SNPs rs11884368, rs2035265, rs3935714 and rs12403232, respectively. A total of 25 SNPs were genotyped on 3716 gastric cancer cases (1877 cardia and 1839 non-cardia) and 3912 controls in the Beijing study. A total of 24 SNPs were successfully genotyped using Sequenom platform on 2896 gastric cancer cases (1796 cardia and 1100 non-cardia) and 2826 controls in the Henan study. All 29 SNPs (23 were designed as Sequenom assays plus six were supplemented with TaqMan) were genotyped on 796 gastric cancer cases (65 cardia, 724 non-cardia and 7 NOS) and 810 controls in the Korean study. The modest proportion of gastric cardia cancer cases in Korea limits power for independent testing in this study.

Statistical analysis

For stage 1, logistic regression was performed to analyse the association between each SNP (genetic trend effect) and the case–control status with adjustments for study, gender, age and significant eigenvectors, if any. To adjust for population stratification, we used the first eigenvector from a principal component analysis of ancestry informative SNPs for total gastric and cardia cancer models only because this vector was nominally significant (p<0.05) in both baseline risk models. No eigenvector adjustment was needed for the non-cardia-only model to control for population stratification because no vectors were significantly associated with case status. Quantile–quantile plots for each outcome show no evidence of inflation with a lambda of 1.01 for each outcome (see online supplementary figure S1). In each replication study, we adjusted for gender and age (not available in the Korean study) and analysed the total gastric as well as cardia or non-cardia-only models. Fixed-effect meta-analysis was subsequently applied to combine the association results for stage 1 and three stage 2 studies together. Data analysis and management were performed with GLU (, a suite of tools available as an open-source application for management, storage and analysis of GWAS data.


Search for novel associations

Using GWAS data on 2350 subjects with gastric adenocarcinoma, which included 1189 cardia, 1027 non-cardia and 134 gastric cancer cases not otherwise specified, and 2708 controls (table 1), we selected 29 SNPs with the lowest p values after LD pruning (r2>0.6) and removal of previously identified PLCE1 SNPs because of their known association with risk for gastric cardia cancer. The 29 SNPs (see online supplementary table S2) included 12 SNPs with the lowest p values each for cardia and non-cardia, and four additional SNPs with the lowest p values for total gastric cancer, plus rs4072037 near MUC1 because it appeared to be strongly associated with gastric cancer risk in many previous candidate gene studies. Replication testing used three independent case–control sets from Henan Province, China, Beijing, China, and Korea that provided a total of 7408 cases and 7548 controls (table 1). Because of past differences in associations between cardia and non-cardia gastric cancers, we tested all 29 SNPs for both anatomic subtypes in the replication phase.

Table 1

GWAS and replication subjects by type and study

Online supplementary tables S3–5 provide complete results from the GWAS and the replication testing for cardia, non-cardia and total gastric cancer for all 29 SNPs. Table 2 highlights three SNPs with associations at or near genome-wide significance (p<5×10−8).

Table 2

Three SNPs associated with gastric cancer risk at or near genome-wide significance

At 5p13.1, rs10074991 showed genome-wide significant associations for both cardia and non-cardia cancers with p values of 7.36×10−12 and 2.42×10−23, respectively. The ORs were similar in magnitude and the overall per allele OR (95% CI) for total gastric cancer was 0.80 (0.77 to 0.83), p=4.83×10−26. This SNP localises to an intron in PRKAA1 (protein kinase AMP-activated alpha 1 catalytic subunit) (figure 1A).

Figure 1

Regional plots of association results, recombination hotspots and linkage disequilibrium for the (A) 5p13.1:40,557,405–41,130,166, (B) 6p21.1:40 821194–41413053 and (C) 1q22:15294037 9–15389737 6 gastric cancer (GC) susceptibility loci. Association results from a trend test in –log10 p values (y axis, left; red diamonds, GC association result; blue diamonds, non-cardia GC association result; light green diamonds, cardia GC results) of the single-nucleotide polymorphisms are shown according to their chromosomal positions (x axis). Linkage disequilibrium structure based on control data from the genome-wide association studies (n=1660) was visualised by snp.plotter software. The line graph shows likelihood ratio statistics (y axis, right) for recombination hotspots by SequenceLDhot software based on the background recombination rates inferred by PHASE V.2.1 using five resamplings of 100 controls. Physical locations are based on hg18. Gene annotation was based on the National Center for Biotechnology Information RefSeq genes from the UCSC Genome Browser.

At 6p21.1, rs2294693, we observed nominal associations with each gastric cancer subsite, but the association was genome-wide significant for only non-cardia cancer risk (p=2.50 x10−8), with OR (95% CI) of 1.18 (1.12 to 1.26). For cardia cancer, the OR (95% CI) was 1.08 (1.01 to 1.15) (p=1.47×10−2), suggesting either a modest or null association with risk at that subsite. rs2294693 is an intronic SNP in UNC5CL (unc-5 homolog C (C. elegans)-like) (figure 1B).

One other SNP showed an association with gastric cancer risk, but did not reach genome-wide significance. The exonic SNP rs4072037, in MUC1 at 1q22, showed a similar OR (95% CI) for cardia and non-cardia gastric cancer and a combined per allele OR (95% CI) of 0.76 (0.69 to 0.84), p=6.59×10−8 (figure 1C).

Finally, two other SNPs showed some evidence for an association with gastric cardia cancer, but had inconsistencies between studies. The point estimates for rs3935714 at 16p11.2 were consistent in data from the Beijing and Korean studies, but was not assayable in the Henan replication set and the SNP did not reach genome-wide significance in the combined estimate (p=1.27×10−6). Similarly, rs12922317 at 16p13.13 replicated in the Beijing set, but not in the other two studies (combined p=2.62×10−6, see online supplemental table S3).

Previously reported variants from the literature

In table 3, we present results from our stage 1 GWAS data, without attempted replication, for six SNPs at four loci that were gleaned from previously published gastric cancer GWAS and two SNPs from a GWAS for H. pylori seropositivity. Notably, we confirm prior independent GWAS reports7 of an association between multiple SNPs in PSCA at 8q24.3 and risk of non-cardia gastric cancer, but find no evidence for an association with cardia cancer. Shi et al reported a novel association for rs9841504 in ZBTB20 with non-cardia gastric cancer, but we saw no evidence of that association in our data. Furthermore, results were mixed for two SNPs reported in a previous pleiotropy analysis that included gastric, lung and oesophageal cancer. A nominally significant association was found for rs2285947, but no association for rs2494938. We also looked up the association for two SNPs reported to be associated with H. pylori seropositivity in Europeans,11 without regard to development of gastric cancer, because H. pylori is an important cause of gastric adenocarcinoma. We found one locus (rs368433) to be nearly monomorphic in our population and no evidence for an association with gastric cancer risk at the other locus (rs10004195).

Table 3

Associations from the stage 1 GWAS data for SNPs from previously published GWAS studies


From our GWAS analysis of gastric cancer in Chinese subjects, we selected and tested associations between 29 SNPs and risk of cardia and non-cardia gastric cancer using replication data from up to 9758 total cases and 10256 controls. One SNP, rs10074991 in PRKAA1 at 5p13.1, reached genome-wide significance for both cardia and non-cardia gastric cancers, while rs2294693 at 6p21.1 showed genome-wide significance for only non-cardia gastric cancer.

An intronic SNP in PRKAA1, rs13361707, was recently reported to be associated with risk of non-cardia but not cardia gastric cancer.9 Here we found significant associations with ORs of similar magnitude for both gastric cancer subsites for rs10074991, an SNP in perfect LD with rs13361707. Shi et al9 reported a genome-wide significant association for this SNP in a scan of non-cardia cancer and also tested the association using DNA samples from 905 cardia cancer patients. Other reports relating this SNP to gastric cancer from Korea have also confirmed an association,12 with one study reporting that the association did not differ by location in the stomach.13 Overall, there appears to be consistent evidence for an association with gastric non-cardia cancer, but the results for cardia are mixed. PRKAA1 is a catalytic subunit of the 5′-AMP-activated protein kinase, which is an energy sensor that is highly conserved across eukaryotic species. Online supplementary table S6 provides additional biological information on rs10074991 and SNPs in LD with it showing numerous potential changes in altered binding motifs.

An association between rs2294693 at 6p21.1 and risk of non-cardia gastric cancer reached genome-wide significance, but not for gastric cardia cancer alone or the combined endpoint. This intronic SNP in UNC5CL falls in a genomic region with two prior GWAS hits for upper GI cancers. Two SNPs, rs10484761 and rs2494938, are 200  and 500 kb telomeric to rs2294693, respectively, and have been associated with ESCC14 and non-cardia gastric cancer in Chinese populations.15 However, the pairwise r2 values for these three SNPs were low (≤0.015), and rs2494938 is not associated with gastric cancer risk in our GWAS data (table 3). rs2294693 and SNPs in high LD with it (see online supplementary table S6), including a synonymous change in the coding sequence, lead to alterations in numerous protein binding and other motifs, but there is only limited evidence for alterations predicted to alter gene regulation.

We previously reported a borderline significant association for rs4072037 (MUC1) for total gastric cancer using a subset of the current data.6 Our original published findings for rs4072037 included about 2240 gastric cancer cases, while the total here includes about 3146 cases. Although the current p value falls just short of genome-wide significance (p=6.59×10−8), Saeki et al16 reported that rs4072037 was associated with both diffuse-type gastric cancer (OR 1.66, 95% CI 1.44 to 1.93) and intestinal-type gastric cancer (95% CI 1.23, 1.02 to 1.48). We do not have information on diffuse versus intestinal-type gastric cancer in our study, so we cannot directly investigate the specificity of the association by Lauren classification. Further, a recent meta-analysis of 6580 cases and 10324 controls including data from both the Saeki et al study and our previous publication6 found a summary estimate for rs4072037 of 0.72 (0.68 to 0.77), p=7.82×10−25.10 The choice of referent allele has differed in prior publications because the G allele predominates in populations of European descent, while the A allele predominates in populations of East Asian descent, but studies consistently find that the A allele elevates risk compared with the G allele. MUC1 is a member of the mucin family that collectively forms the protective mucous barrier on epithelial surfaces. rs4072037 seems to be functionally important because it altered transcriptional regulation and determined splice variants in MUC1.16

Using our GWAS data, we observed significant associations between variants at 8q24.3 (PSCA) and risk of non-cardia gastric cancer, which has been reported in several prior publications (table 3).7 Jin et al also showed evidence for an association between rs2285947 at 7p15.3 and non-cardia gastric cancer risk.15 This SNP does not appear on our array so we examined a proxy, rs976516 (r2=0.85), and found further evidence for an association with non-cardia and cardia cancer, with p value (9.42×10−3) for total gastric cancer. We saw no association with either of two loci reported to be associated with H. pylori seropositivity in Europeans,11 but note that H. pylori seropositivity is highly prevalent in our populations, and we did not directly reassess the previously reported association.

Regardless of which sublocation of gastric cancer provided the initial signal from the GWAS analysis, we completed replication testing in both cardia and non-cardia gastric adenocarcinoma cases to more deeply explore how associations for the selected SNPs differed between gastric cancer locations. Our overall results showed that rs10074991 was associated with risk of gastric cardia adenocarcinoma (p=7.36×10−12) despite little evidence for an association in our GWAS subjects (p=1.27×10−2). Finally, when added to the known consistent differences between cardia and non-cardia gastric cancer for the associations in common variants for PSCA and PLCE1, our results suggest that the clearest picture of the association of common genetic variants with gastric cancer in the future will come from studies that include both gastric cancer sublocations studied independently whenever possible. Studies in Western populations have noted major differences in the risk factors for cardia and non-cardia gastric cancers, many of which suggest that cardia adenocarcinomas in the West are more similar to oesophageal adenocarcinomas. The current study coupled with previous reports on other risk factors that showed either similar5 or dissimilar results6 in China highlights the apparent differences between these tumour types in Asian versus Western countries.

In summary, our replication study found associations with variants in PRKAA1 and MUC1 for cardia and non-cardia gastric cancers. We also report a novel association with variants in UNC5CL at 6p21.1. There was some heterogeneity between our studies and those in the literature, resolution of which will require additional data.


Supplementary materials


  • Contributors NH, ZW, XS, LW, BSK and NDF are co-first authors. Study concept and design: NH, ZW, XS, LW, BSK, NDF, KS, L-DW, DL, SJC, AMG, PRT and CCA. Acquisition of data, critical revision for important intellectual and final approval of the manuscript: all authors. Analysis and interpretation of data: NH, ZW, XS, LW, BSK, NDF, SJC, AMG, PRT and CCA. Drafting of the manuscript: NH, ZW, XS, LW, BSK, NDF, AMG, PRT and CCA. Study supervision: KS, LDW, DL, SJC, AMG, PRT and CCA.

  • Funding This work was supported by the intramural research programme of the National Institutes of Health, National Cancer Institute. The Singapore Chinese Health Study was supported by National Cancer Institute grants R01CA80205 and R01CA144034. The Shanghai Women's Health Study was supported by National Cancer Institute grants R37:CA070867 and the Shanghai Men's Health Study was supported by National Cancer Institute grants R01:CA082729 and UM:CA173640. The Henan province study was supported by the National Natural Science Foundation of China, grant number 81472323; the Top Talent Support Project of Zhengzhou University, grant number ZDGD13001; and the Innovation Scientists and Technicians Troop Construction Projects of Henan Province, grant number 3047 to LDW. KS was supported by a Mid-career Researcher Program grant (2014R1A2A1A09005824) through National Research Foundation from the Ministry of Science, Information and Communication Technology (ICT) and Future Planning, Republic of Korea.

  • Competing interests None declared.

  • Ethics approval NCI Special Studies IRB.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement Data from the NCI GWAS will be available through dbGap.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.