Validation of a Blood Biomarker for Identification of Individuals at High Risk for Gastric Cancer

Infection with the microaerophilic, spiral bacterium Helicobacter pylori is the leading cause of gastric cancer, the fifth most common cancer worldwide (1), and is overall responsible for more total incident cancers each year than any other single infectious agent (2). Although a vaccine against this bacterium has not yet been successfully developed, there exists effective eradication therapy in the form of 2 weeks of triple or quadruple therapy, involving treatment with two to three antibiotics plus a proton pump inhibitor and/or bismuth (3, 4). However, mass eradication is neither feasible nor recommended, as half of the global population harbors this bacterium, but the vast majority of these individuals will not develop neoplasia (5). Moreover, population-based H. pylori eradication could increase antibiotic resistance, and in addition some benefits have been observed with carriage of the bacteria, including reduced incidence of esophageal disease (6).

Thus, there remains a pressing need to identify those individuals at highest risk for gastric cancer for targeted cancer prevention through H. pylori eradication treatment, which has been shown to reduce risk for this malignancy (7). This is particularly important in the region of East Asia, where over half of all incident gastric cancers occur in the world each year (1). In our efforts to achieve this aim, we developed a serologic H. pylori biomarker panel for gastric cancer risk in a cohort of urban men in Shanghai, China, using a fluorescent bead-based multiplex serology assay developed at the German Cancer Research Center (8). We then replicated this initial finding in a consortium of eight prospective cohorts in China, Japan, and Korea, among 1,608 incident noncardia gastric cancers and 1,958 matched controls. In this consortium, we found that seropositivity to two, Omp and HP0305, of the initial six identified H. pylori proteins (Omp, HP0305, HyuA, HpaA, CagA, and VacA) was strongly and consistently associated with cancer risk among all cohorts, so that prior to cancer diagnosis, seropositivity to both, compared with seropositivity to neither, was associated with an over 4-fold increase in the odds of gastric cancer incidence (9).

In the present study, we sought to validate these H. pylori blood biomarkers for precancerous gastric lesions in an independent East Asian population, that of the high-risk population in Linqu County, Shandong Province, China. We assessed whether our previously identified risk markers could identify individuals with prevalent gastric precursor lesions, specifically those who are on the cascade of events leading to gastric cancer.

In 2002, an intervention trial was established in Linqu County, Shandong Province, China, to compare the effect of H. pylori treatment and selective COX-2 inhibitors on precancerous gastric lesions. At baseline, study subjects completed a standard structured questionnaire, provided a blood sample, and were screened by upper endoscopy. Details of these methods have been published previously (10); briefly, 3,161 residents ages 35 to 64 from 12 randomly selected villages in Linqu were assessed for eligibility, and 2,813 (89%) individuals agreed to participate in the initial screening. Four experienced gastroenterologists conducted the endoscopies, and five biopsy samples were taken from the standard sites in the stomach according to the updated Sydney system (11). A global diagnosis was then made for each participant based on the biopsy specimen with the most severe diagnosis. A panel of three pathologists then reviewed each slide and graded as normal, superficial gastritis (SG), chronic atrophic gastritis (CAG), intestinal metaplasia (IM), indefinite dysplasia (Ind DYS), dysplasia (DYS), and cancer, following the criteria of the updated Sydney system (11) and the Padovo international classification (12). At baseline, a 5-mL sample of blood was also collected from each study participant, allowed to clot for 30 to 40 minutes at room temperature, and then centrifuged at 965 × g for 15 minutes. Serum was then aliquoted into vials and frozen immediately at −20°C and stored in a −70°C freezer.

For the present study, a total of 1,402 individuals screened by upper endoscopy at baseline were included. Because there were so few participants with normal gastric mucosa, 512 participants with SG (138) or mild CAG (374) were randomly selected as the control group. Furthermore, all participants with IM (n = 412) and DYS (n = 145) were included, and 333 participants with Ind DYS were randomly selected as the precancerous gastric lesions group. A written informed consent was obtained from each participant and the study was approved by the Institutional Review Board of Peking University Cancer Hospital.

Serum samples from all study participants were assayed for antibodies to 13 H. pylori recombinantly expressed fusion proteins (UreA, Catalase, GroEL, NapA, CagA, HP0231, VacA, HpaA, Cad, HyuA, Omp, HcpC, and HP0305; refs. 13, 14). As previously described, H. pylori multiplex serology is based on a glutathione S-transferase capture immunosorbent assay combined with fluorescent bead technology (Luminex) to simultaneously detect human IgA, IgM, and IgG antibodies to selected H. pylori proteins. Calculation of antigen-specific cutoff points (mean of the median reporter fluorescence intensity [MFI] plus three times SD, excluding positive outliers) was done using 17 H. pylori–negative sera previously classified for H. pylori status run within the same experiment. Defining H. pylori seropositivity as reactivity with ≥4 proteins has shown good agreement (k = 0.70) with commercial serologic assay, resulting in 89% sensitivity and 82% specificity (13).

Pepsinogen I and II levels in serum were determined by pepsinogen I and II ELISA assay kits (Eagle Biosciences) according to the manufacturer's instructions. Briefly, 25 μL of serum for measurement of pepsinogen I and 50 μL of serum for pepsinogen II, respectively, were applied in duplicates to a streptavidin-coated microplate. After incubation with the respective capture and tracer antibody HRP substrate was added for signal detection. The reaction was stopped with stop solution, and the absorbance was measured at 450 nm in a microplate reader. Provided assay standards were run on each plate to obtain a plate-specific standard curve for determination of the concentration (ng/mL) of pepsinogens I and II in each sample. Two control samples with a given pepsinogen I and II concentration were provided by the manufacturer and applied on each plate to ensure reliability of the assay result.

Initially, we sought to validate the analyses previously performed among the prospective cohort studies from East Asia included in the H. pylori Biomarker Cohort Consortium (HpBCC; ref. 9). Accordingly, we first assessed the individual associations of seropositivity to each of the 13 H. pylori antigens included in the multiplex serology panel, using logistic regression to produce odds ratios (ORs) and 95% confidence intervals (CIs) for each type of prevalent precancerous gastric lesion (IM, Ind DYS, and DYS) after adjusting for age (continuously) and smoking status (current vs. not current), factors that were associated both with H. pylori status and disease outcome in this population. However, when comparing ORs for H. pylori overall and for each individual antigen, there were no statistically significant differences in the results by type of precancerous lesion, so we combined them all into one outcome. The Bonferroni correction was applied to recognize P values at ≤0.0038 (0.05/13 markers).

We then examined the association of combined Omp and HP0305 seropositivity with odds of prevalent precancerous lesion, as that was the strongest finding in the HpBCC. As before, we created three categories: seronegativity to both (reference), seropositivity to only one, and seropositivity to both. To compare these markers with the established virulence factor for gastric cancer risk, CagA, we also examined the association of prevalent precancerous lesion with dual H. pylori–positive (seropositive to ≥4 proteins) and CagA-positive status. Finally, we repeated the panel of six antigens found in the Shanghai Men's Health Study (8) and replicated in the HpBCC for association with risk. For all, we used logistic regression adjusting for age and smoking.

We also examined the data for potential differences in the association by sex and smoking status through stratified analyses, and the use of a multiplicative interaction term to assess effect modification, but no differences were found.

To create a classification model for prevalent precancerous lesion, we considered two populations: all study participants, and only those who were H. pylori positive. The rationale is that our primary motivation was to determine a biomarker for high-risk of gastric cancer, so that those individuals could be targeted for H. pylori eradication, and thus limiting the population to H. pylori–positive individuals would make sense in this instance. However, beyond eradication therapy for H. pylori–positive individuals, identification of high-risk individuals in the population at large (regardless of H. pylori status) could also be beneficial in terms of discovering precursor lesions earlier, and in changing screening schedules so as to diagnose gastric cancer at earlier stages. Thus, we performed receiver operating characteristic (ROC) curve analyses among both populations to calculate the area under the curve (AUC) and to compare models with and without H. pylori antibody biomarkers.

As a secondary analysis, we performed ROC curve analyses among the subset of participants for whom a valid pepsinogen result was obtained, to determine if this measure of gastric atrophy strengthens the model. Among these 546 individuals (226 controls and 320 cases), the pepsinogen I:II ratio was independently associated with prevalence of precancerous lesion (OR for individuals with a ratio <4, compared with those ≥4 = 2.37; 95% CI, 1.15–4.90). However, inclusion of the pepsinogen I:II ratio did not change the main results with Omp and HP0305 and did not significantly improve the AUC. Moreover, the pepsinogen I:II ratio was highly inversely associated with Omp and HP0305 status (which are themselves highly correlated, Pearson correlation coefficient of 0.52, P < 0.0001), with Pearson correlation coefficients of −0.34 and −0.24, respectively (both with P values of <0.0001). Thus, as this measure was available on less than half of our study population, was highly correlated with our serologic antigen biomarkers, and did not improve the classification ability beyond our validated risk markers, we did not include it in our final models.

Individuals in Linqu County with prevalent precancerous lesions (IM, Ind DYS, or DYS), as compared with controls (those with SG or mild CAG), were more likely to be of older age and a current smoker (Table 1).

Overall, 54% of controls were identified as H. pylori seropositive as compared with 83% of individuals with prevalent precancerous lesions, leading to an age- and smoking status–adjusted OR of 4.51 (95% CI, 3.50–5.81; P < 0.0001). Controls were also less likely than the cases to be seropositive to each of the 10 H. pylori proteins previously identified as potential risk biomarkers in the HpBCC, the consortium of prospective studies in East Asia that this study sought to validate (all P < 0.0001), as well as an 11th antigen, HP0231 (P = 0.008; Table 2).

The strongest association among the individual antigens was for Omp, seropositivity to which was associated with an over 5-fold odds for the prevalence of precancerous lesions (OR, 5.37; 95% CI, 4.20–6.89). These results did not significantly differ when separating out cases by individual diagnosis (IM, Ind DYS, or DYS; Supplementary Table S1).

In replicating the panel of six antigens (Omp, HP0305, HyuA, HpaA, CagA, and VacA) originally created in the preliminary work for this study among the participants of the Shanghai Men's Health Study (8), seropositivity to 4 to 5 or all 6 of these specific H. pylori antigens, compared with seropositivity to three or fewer, resulted in a significant over 4-fold increase in the odds of prevalent precancerous lesion (OR, 4.69; 95% CI, 3.63–6.07; and OR, 4.41; 95% CI, 2.92–6.66; Table 3). When focused on the two antigens found to be the strongest markers of risk in the HpBCC, compared with individuals seronegative to both Omp and HP0305, individuals seropositive to one or to both were at increased odds of a prevalent precancerous lesion (OR, 3.29; 95% CI, 2.43–4.46; and OR, 7.43; 95% CI, 5.59–9.88, respectively). Among H. pylori–positive participants alone, the strength of these associations was slightly reduced, but all remained statistically significant, with the strongest association still for those seropositive to both Omp and HP0305 compared with those seronegative to both (OR, 6.20; 95% CI, 3.97–9.66; Supplementary Table S2).

Finally, a classification model for prevalent precancerous lesion in Linqu County, including age (continuous), smoking status (current vs. not current), H. pylori status (defined as seropositivity to 4 or more of the 13 H. pylori antigens assessed), and Omp and HP0305 antibody status (separately), resulted in an AUC of 0.7510 (95% CI, 0.7245–0.7774). This model was significantly better than the one that included age, smoking, H. pylori status, and seropositivity to the established H. pylori virulence factor CagA (AUC, 0.7184; 95% CI, 0.6908–0.7461, P for difference in the AUC with the Omp and HP0305 model = 0.0002), as well as to the same model but with H. pylori status included only (AUC, 0.7143; 95% CI, 0.6864–0.7422, P for difference in the AUC with the Omp and HP0305 model < 0.0001; Fig. 1 and Table 4). When limiting the population for the model to H. pylori–positive individuals only, the stronger predictive ability of Omp and HP0305 as compared with CagA remained (AUC, 0.7139; 95% CI, 0.6773–0.7505, compared with AUC, 0.6623; 95% CI, 0.6248–0.6998, respectively, P value for difference = 0.0004; see Supplementary Table S3). For a range of probabilities, the positive predictive value of the combined status of Hp⁺ Omp⁺ HP0305⁺ in our population is fairly high, at >80% for the majority, but these biomarkers did not achieve a similarly high negative predictive value (ranging from 70.5% down to 46.2%).

In this cross-sectional study of gastric cancer precursor lesions among a high-risk population in China, we validated two H. pylori biomarkers we originally identified in a pilot study of urban men in Shanghai, China, and then replicated in a consortium of prospective cohort studies of men and women in China, Japan, and Korea. The consistency and strength of the associations with antibody seropositivity to the H. pylori proteins Omp and HP0305 suggest that these biomarkers could substantially add to a screening program that seeks to identify individuals at highest risk for gastric cancer for closer surveillance and to be targeted for H. pylori eradication, an established method for reducing gastric cancer risk (15, 16). Adding motivation for this plan is the data from China showing that H. pylori eradication, even among individuals who have already developed precancerous lesions (particularly IM or DYS, the outcomes in the present study) is as, if not more, effective in reducing gastric cancer incidence than among those with normal or CAG histopathology (17).

Examining antibodies to multiple H. pylori–specific proteins to uncover potential biomarkers of gastric cancer risk has been performed in other populations, the original of which was a German case–control study that found significant associations between antibodies to eight individual H. pylori proteins, including HP0305, with an OR of 2.34 (95% CI, 1.46–3.74), but not Omp (OR, 1.41; 95% CI, 0.89–2.26; ref. 18). Since our original publication of our pilot study in 2012 (8), second to be performed only to the study above, there have been numerous additional studies, with findings generally strongest for the known virulence factors VacA and CagA. Specifically, recently in the MCC-Spain case–control study, only seropositivity to CagA and VacA were found to be individual predictors of noncardia gastric cancer risk, with no associations found for Omp or HP0305 (19). In a case–control study in northeastern Iran, a population with a high risk of gastric cancer and a high prevalence of H. pylori, again only antibodies to CagA and VacA were found to be associated with risk, with no associations found for Omp or HP0305 (20). In a Swedish case–control study, all antigens were significantly associated with gastric cancer risk, and after performing principal component analysis, the authors derived two factors associated with increased risk of noncardia gastric cancer: the first, and most strongly associated with risk, included CagA, VacA, and Omp; the second included NapA and Catalase (21). The only other study to explore results from the German Cancer Research Center H. pylori multiplex serology assay in East Asian populations—the geographic region in which the greatest number of gastric cancers occur each year—was in the Linxian Nutrition Intervention Trial cohort (9). In this study of one population in China, and the only other prospective investigation, only two H. pylori antigens passed the Bonferroni correction for multiple testing—Omp and HP0305 (OR, 2.30; 95% CI, 2.36–3.88; and OR, 2.16; 95% CI, 1.40–3.33; ref. 22). Also, previously, a subset of Linqu County samples were assayed by recomLine analysis (Mikrogen; ref. 23) to determine seropositivity to six H. pylori antigens (CagA, VacA, GroEL, UreA, HcpC, and gGT) and found that CagA was an independent predictor of advanced gastric lesions (24). Longitudinally, both CagA and GroEL were also seen to predict progression of gastric lesions, although neither Omp nor HP0305 were included in these analyses.

In all of our analyses of H. pylori antigen-specific association with gastric cancer risk and within each individual East Asian study population—in the Shanghai Men's Health Study (8); the individual cohorts that comprise the HpBCC (Japan Public Health Center-based Prospective Study I and II, Korean Cancer Prevention Study II, Korean Multicenter Cancer Cohort I, Linxian Nutrition Intervention Trial, Shanghai Men's Health Study newly identified cases, and the Shanghai Women's Health Study; ref. 9); and now the Linqu County trial, reported on in the present manuscript—antibodies to Omp and HP0305 have significantly been associated with gastric cancer risk. This consistency highly suggests that these are replicable markers of risk for East Asian populations. These results also support more research into the mechanisms of Omp (an outer membrane protein, known as HP1564) and HP0305 (a hypothetical protein, also shown to be expressed in outer membrane vesicles). Previously, studies have been conducted to characterize these proteins, suggesting their roles in bacterial colonization and as proinflammatory agents (25–29), but examination of the role of these proteins in carcinogenesis on the molecular level has not yet been performed.

Other groups have also looked to develop risk prediction models for gastric cancer that include biomarkers, such as two from Japan: in Fukuoka, for which the model included a dual measure incorporating H. pylori IgG antibodies and pepsinogen levels, and hemoglobin A1c levels, in addition to age, sex, and smoking status (30); and from the National Cancer Center in Tokyo, whose model included again H. pylori titers and pepsinogen levels, along with age, sex, smoking status, family history of gastric cancer, and consumption of highly salted food (31). Both of these models produced c-statistics above 0.70 but below 0.80, as in our current study. In a rural county of Northern China with high gastric cancer mortality, a risk prediction model for gastric precancerous lesions of five circulating biomarkers (pepsinogen I, pepsinogen II, pepsinogen I/II ratio, H. pylori IgG status, and gastrin-17 levels) produced a c-statistic of 0.803 (32). In the present study, pepsinogen levels did not significantly improve the predictive capability of our classification model for prevalent precancerous lesion: among individuals with a valid pepsinogen assay, the model including age, smoking, H. pylori status, and Omp and HP0305 seropositivity, the AUC was 0.7454; adding pepsinogen moved it only slightly to 0.7477. Among H. pylori–positive individuals, there was also no appreciable difference (AUCs of 0.7043 and 0.7070, respectively), and pepsinogen was no longer significantly associated with prevalence (OR, 1.92; 95% CI, 0.87–4.20). Furthermore, pepsinogen was highly correlated with Omp and HP0305 in this study, with pepsinogen I:II ratios decreasing (indicating greater gastric atrophy) with increasing Omp/HP0305 category, so that for individuals seronegative to both Omp and HP0305, the median pepsinogen ratio was 15, compared with 12 for individuals seropositive to only one, and 7 for individuals seropositive to both (all P values comparing pepsinogen ratios by the Wilcoxon rank sum test <0.01).

Although it is a limitation of the present study that we do not have pepsinogen successfully measured on all participants, our results on the subset of individuals with valid pepsinogen measurements suggests adding pepsinogen does not improve the classification ability of the model. In Japan and Korea, where pepsinogen is assayed regularly, different technologies are used (including latex agglutination and immunoradiometric assay; ref. 33), and as part of their large screening programs can take place soon after blood draw, thus avoiding the degradation of samples. Furthermore, a recent study to explore differences in results comparing three different pepsinogen tests did find some significant differences comparing the Biohit ELISA assay (similar to our Eagle BioSciences ELISA) with the Japanese Eiken latex agglutination system, although in general concluded that the assays have “good relative agreement” (34). Finally, it is possible that pepsinogen does not add value to our classification model in that unlike the models presented above that sought to predict future gastric cancer, ours was developed to determine prevalent precancerous lesion, which may be a state when atrophy is no longer the strongest signal.

Our finding of a validated biomarker that includes antibodies to just two specific H. pylori antigens, that can be determined in one assay, versus the previous models above all requiring at least 3 assays (2 for pepsinogen I and II and 1 for H. pylori, plus HbA1c), as well as other lifestyle characteristics not always available in electronic medical records, suggests the Omp and HP0305 markers could potentially be part of a feasible test for determining risk among a large population base. Furthermore, none of these previous risk prediction models considered the heterogeneity of H. pylori, which is particularly important in East Asia, where the majority of the population is H. pylori positive. In fact, when we compared our precancerous lesion classification model including the Omp and HP0305 biomarker to the same model with H. pylori dichotomous status alone, the Omp/HP0305 model performs significantly better (AUC, 0.7510; 95% CI, 0.7245–0.7774 compared with AUC, 0.7143; 95% CI, 0.6864–0.7422, P for difference <0.0001; Supplementary Fig. S1). Another limitation is that we did not additionally have a conventional measure of H. pylori such as an ELISA or immunoblot assay for comparison with our serological measures. However, in our previous work in the Shanghai Women's Health Study, we found a similar if slightly higher prevalence of H. pylori among controls by multiplex serology (94.6%) than using conventional methods (92.2%), which is what might be expected as it is possible that multiplex serology is more sensitive than standard ELISA (9, 35). Additionally, in the study in the Linxian Nutrition Intervention Trial alone, adjustment for conventional ELISA H. pylori seropositivity did not change the significant results found using multiplex serology with Omp and HP0305 (22). Finally, we must also note that a limitation of the present study is that we did not produce a prediction model for risk of incidence gastric cancer, but rather one for prevalent gastric precancerous lesion. However, the findings all validate what we found previously in prospective cohort studies, and it is well established that individuals with precancerous lesions are at highest risk for gastric cancer. The present study was also performed among a population at high risk in East Asia, as with the previous studies in our group. The ability to generalize our results to non–East Asian populations is not known.

In conclusion, in populations with a high prevalence of the known carcinogen H. pylori and a high incidence of gastric cancer, it has been established that biomarkers are needed to identify individuals at highest risk of developing cancer for targeted eradication. Our identification, replication, and now validation of two novel H. pylori biomarkers for gastric cancer risk in East Asia, Omp and HP0305, have specific importance for contributing to targeted H. pylori eradication schemes, as the treatment for these cancer-causing bacteria is a relatively straightforward course of 10 to 14 days of two to three antibiotics and a proton pump inhibitor, which has already been shown in clinical trials to reduce gastric cancer risk by 50%. Furthermore, the evidence presented here suggests that these two markers can contribute to a high-risk classification model in East Asia, as they predict prevalence of precancerous gastric lesions beyond the established, and highly prevalent, known virulence marker of CagA. Moreover, they can be easily measured to then result in AUCs at a similar level to risk prediction models that include larger panels of biomarkers that are also more cost- and time intensive to assay.

Y.-L. Qiao is a consultant/advisory board member for MSD. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M. Epplein, X.-O. Shu, S.H. Jee, M. Pawlita, W.-c. You, K.-f. Pan

Development of methodology: M. Epplein, S.H. Jee, M. Pawlita

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Epplein, J. Butt, Y. Zhang, C.C. Abnet, G. Murphy, W. Zheng, X.-O. Shu, S. Tsugane, Y.-l. Qiao, P.R. Taylor, T. Shimazu, K.-Y. Yoo, S.K. Park, J. Kim, S.H. Jee, W.-c. You, K.-f. Pan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Epplein, L.H. Hendrix, C.C. Abnet, G. Murphy, S. Tsugane, T. Shimazu, S.K. Park, M. Pawlita

Writing, review, and/or revision of the manuscript: M. Epplein, J. Butt, C.C. Abnet, W. Zheng, X.-O. Shu, S. Tsugane, P.R. Taylor, T. Shimazu, S.H. Jee, T. Waterboer, M. Pawlita, W.-c. You, K.-f. Pan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Epplein, X.-O. Shu, Y.-l. Qiao, T. Waterboer, W.-c. You, K.-f. Pan

Study supervision: M. Epplein, M. Pawlita, W.-c. You, K.-f. Pan

Other (PI of a parent study): W. Zheng

Other (led one of the parent studies and provided samples for the study): X.-O. Shu

This work was supported by the NCI at the NIH (R01 CA174853 and K07 CA151782 to M. Epplein).

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.