The Prevalence of Cancer-Associated Autoantibodies in Patients with Gastric Cancer and Progressive Grades of Premalignant Lesions

Despite the decreasing incidence in most parts of the world, gastric cancer remains the fifth most common type of malignancy and the third most common cause of cancer-related death worldwide. According to GLOBOCAN, more than 70% of gastric cancer cases are occurring in developing countries. Eastern Asia and Central and Eastern Europe are the regions of the highest gastric cancer incidence (1) and are still distinguished by a high prevalence of Helicobacter pylori (H. pylori) infection and premalignant gastric lesions (2–4). The high mortality rate is mostly due to asymptomatic early stages of gastric cancer that lead to belated diagnosis. The screening of high-risk individuals and early detection remains the most promising strategy for reducing gastric cancer–associated mortality. Currently, endoscopy with a histologic examination of biopsy material is the golden standard for early diagnosis of gastric cancer and related precancerous lesions. This is an operator-dependent invasive procedure of relatively high cost with possible side effects (5). Therefore, to make gastric cancer a curable disease, the identification of biomarkers capable of stratification of high-risk patients and those with nascent tumors by noninvasive or minimally invasive means is still of paramount importance.

The level of serum pepsinogen (PG) I and II is used for gastric cancer detection in an opportunistic setting (6–8). Besides pepsinogen, several other biomarker classes have been proposed, including circulating tumor cells and cell-free DNA, mRNAs and noncoding RNAs, cancer-derived extracellular vesicles, and volatile organic compounds detectable in exhaled breath (9–11). Apart from the enlisted, serum autoantibodies against tumor-associated antigens (TAA), which predominantly represent aberrantly expressed endogenous self-antigens, have been proposed as promising biomarker candidates. They are stable in serum, possess high specificity to the cognate antigen, and are suitable for multiplex test development (12). Moreover, high-titer IgG class antibody responses can be readily induced even if the TAA expression is limited to a small number of cells in a clinically undetectable tumor mass (13), making them particularly attractive candidates for use in early cancer detection. In fact, several studies have reported anti-TAA autoantibody responses several years prior to clinically detectable malignancy including gastric cancer (14), non–small cell lung cancer (15), breast cancer (16), and colorectal cancer (17).

In general, cancer-associated autoantibodies are rare, but remarkably specific, biomarkers. Therefore, combinations of multiple autoantibodies have been proposed to reach satisfactory sensitivity for cancer detection (12). In our previous study, we identified a panel of 45 autoantibodies with a diagnostic relevance for gastric cancer and demonstrated that the autoantibody responses can be readily detected in early-stage gastric cancer patients. Moreover, we showed that the serological responses are partially shared between patients with gastric cancer and chronic gastritis, but not patients with gastric ulcer or healthy controls (18). This raised a question whether the defined autoantibody responses are reflecting immunological processes associated with chronic inflammation or, alternatively, reveal nascent transformed cells in preneoplastic lesions and thus sense the point of malignant transformation. It is important to address these questions to better understand the use of cancer-associated autoantibodies as possible markers for risk stratification in patients with premalignant lesions of the stomach, and the role of autoantibodies in gastric cancer development.

The existing data on when gastric cancer–associated autoantibodies appear in serum are rather limited and conflicting. For example, several studies have shown that the frequency of antibody detection against cancer-testis (CT) antigens NY-ESO-1 (CTAG1B) and MAGEA significantly increases with the stage of malignancy in several cancer types including gastric cancer (14, 19–22), suggesting that the autoantibody production is facilitated by increased tumor load. On the contrary, there are data supporting our observation that autoantibody responses against TAAs are detected with comparable sensitivity in all gastric cancer stages (14, 18, 23). Moreover, Zhou and colleagues showed gradually increasing prevalence of autoantibodies against seven putative TAAs in patients with histopathology confirmed chronic atrophic gastritis (CAG; 39%), dysplasia (46%), and cardia gastric cancer (64%; ref. 24), whereas Werner and colleagues did not detect significant seroreactivity against a panel of five gastric cancer–associated antigens in a patient group with CAG, defined by serum pepsinogen I and II measurements (14).

In this study, by applying protein microarray technology, we aimed to systematically assess the prevalence of serum autoantibodies against a set of 102 TAAs in large cohorts of gastric cancer patients from two different regions—Central and Eastern Europe, and Asia—as well as ascertain the gastric cancer–associated autoantibody prevalence in well-defined groups of patients corresponding to the sequential stages of gastric cancer development.

Serum sample collections from four countries—Taiwan, Latvia, Lithuania, and Germany—were used for the study; the cohort size and characteristics of the study population are summarized in Table 1. The serum sample collection representing Asian population was received from Taichung Veterans General Hospital. The Caucasian study population included multiple serum sample sets from the two Baltic states and Germany—Latvian gastric cancer patient serum samples were collected at Oncology Centre of Latvia, Riga East University Hospital, and healthy donor sera were provided by the Genome Database of Latvian Population; other sets of serum samples from gastric cancer patients and healthy donors were received from Lithuanian University of Health Sciences and German Cancer Research Center (the German gastric cancer patients were enrolled via studies DACHSplus, ESTHER II, and VERDI, the controls—via ESTHER I and BliTz; for detailed study description, see a recent article by Werner and colleagues; ref. 14). In addition, a small collection of prediagnostic serum samples from 18 gastric cancer patients from a population-based cohort study ESTHER I were analyzed (14).

The study population with premalignant lesions of the stomach included cohorts from Caucasian population. Serum samples from patients with different grades of stomach mucosal atrophy classified according to OLGIM (operative link on intestinal metaplasia assessment) system (25) into stages 0–IV were obtained in collaboration with the University of Latvia and Lithuanian University of Health Sciences; majority of the patients were enrolled within the framework of a population study GISTAR (https://www.gistar.eu) at Digestive Diseases Centre GASTRO, and all had undergone upper endoscopy and histopathologic examination of stomach biopsies. Another set of serum samples from CAG patients with no history of cancer was received from German Cancer Research Center, and patients were enrolled via ESTHER I study (14) and selected based on serum PG measurements (cutoffs of PG I < 70 ng/mL and PG I/PG II < 3). Serum samples were collected according to standard protocol by applying good laboratory practice. The blood samples were collected in venous blood collection tubes with clot activator and gel for serum separation and stored at +4°C until processing. The samples were centrifuged, aliquoted, and stored at –80°C.

The studies were conducted in accordance with the Declaration of Helsinki and performed after the approval by the local institutional review boards and obtaining an informed written consent from the study subjects.

The DNA fragments encoding 102 TAAs (Supplementary Table S1) were cloned into the pFN19A (HaloTag7) T7 SP6 Flexi bacterial expression vector (Promega). First, a His-tag coding sequence was introduced into 5′ region of Halo-tag coding region and then the selected antigen coding sequences were inserted downstream the Halo-tag coding sequence. Recombinant His-Halo-tag fusion proteins were expressed in BL21-AI Escherichia coli (E. coli) strain, total cell lysate was applied to Protino Ni-TED 150 nickel columns (Macherey-Nagel) to bind His-tagged proteins, columns were washed, and bound proteins eluted. The purity of the proteins was assessed by PAGE and Coomassie blue staining; the specificity and solubility were verified by Western blot using rabbit anti–Halo-tag antibody (G9281; Promega). The proteins were diluted to concentrations of 0.03 to 0.1 mg/mL in PBS and supplemented with 2% of glycerol to ensure proper spot morphology. Proteins were aliquoted in 384-well plates in ready-to-print format and stored at –80°C until use; throughout experiments, repeated thaw-freeze cycles were avoided.

The 102 antigens and the Halo-tag protein were printed by using solid pins onto nickel chelate glass slides (Xenopore) in duplicates by QArray Mini microarrayer (Genetix; Fig. 1A). Slides were dried at 37°C for at least 3 hours and stored at +4°C until processing. Slide processing included blocking in 7% (w/v) nonfat milk powder in TBS and 0.05% Tween-20 (TBS-Tw), and incubation for 2 hours with 1:100 diluted sera that had been preabsorbed with Halo-tag expressing E. coli lysates. Slides were washed overnight in TBS-Tw, incubated with 1:500 diluted rabbit anti–Halo-tag antibody (G9281; Promega), washed in TBS-Tw and incubated with Cy3-conjugated goat anti-rabbit IgG secondary antibody (1:2,000; 111-165-046; Jackson ImmunoResearch) and Alexa Fluor 647-conjugated goat anti-human IgG secondary antibody (1:1,500; 109-605-098; Jackson ImmunoResearch), washed in TBS-Tw, rinsed in distilled water, and dried by centrifugation. Arrays were scanned at 10 μm resolution by PowerScanner (Tecan) with 532 and 635 nm lasers, results were recorded as TIFF files, and data were extracted using GenePix Pro 4.0 software (Molecular Devices). Serum samples from different cohorts were randomized between microarray slides and experiments to diminish the impact of print-run and intra-assay introduced biases.

The obtained data were analyzed using an ad hoc program composed in R language. Low-quality spots were excluded from further analysis, and Alexa Fluor-647/Cy3 signal ratios were calculated for each antigen. A two-step data normalization strategy to diminish variations introduced by variable background intensities of different sera (interslide) and custom-made microarrays (interprint run) was used. For interslide normalization, the values in each serum were centered by the median ratio from all signals, and the distribution of data across an area was scaled by equalizing values of lower 70% signals to 1. Next, the interprint run normalization was done—median value of ratio of an individual antigen was equalized across all antigen printing batches. An individual cutoff value (T) for each antigen was calculated by using formula: |T = median( {{I_{HD}}} ) + 7SD( {80\% \;trimmed({I_{HD}})} )$|⁠, where I_HD is the signal intensities for a given antigen in healthy donors (HD; upper and lower 10% were trimmed to exclude outliers).

For the definition of the autoantibodies with relevance for gastric cancer, only the antigens with significant seroreactivity differences between gastric cancer and HD groups (Mann–Whitney–Wilcoxon test P value <0.05) were selected. Next, a rank value was calculated for each individual antigen by using the formula:

where N represents the number of all measured cases/controls. Finally, a score (S) for each serum was calculated as follows: |S = \mathop \sum \limit_{i = 1}^n {R_i}{I_i}$|⁠. For the data display perceptibility, serum score values underwent natural logarithm-based Log-Modulus transformation using the formula: L(S) = sign(S)*log(|S|+1), which allows to proportionally transform null and negative values.

The nonparametric Mann–Whitney–Wilcoxon test and Fisher exact test were used to evaluate the differences in seroreactivity between two independent groups of samples, and Kruskal–Wallis test was applied for multiple group comparisons. The set of diagnostically relevant antigens was used for ROC curve analyses, and AUC was calculated to evaluate the diagnostic performance of this antigen panel.

In total, 102 different antigens were expressed as His-Halo-tag C-terminal fusion proteins, purified, verified for their solubility, specificity, and purity, and used to produce antigen microarrays (Fig. 1A). The panel included 56 TAAs—among them, there were 17 known CT antigens (e.g., CTAG1B, CTAG2, DDX53, GAGE, MAGEA, IGF2BP3) and 39 other known (e.g., SOX2, IGF2BP2, TP53, MYC, MUC1) as well as novel antigens (18). In addition, 46 artificial peptides that were identified by applying T7 phage display-based SEREX approach (26) to gastric cancer in our previous studies (ref. 18 and unpublished data) were included. The complete list of antigens and their description are summarized in Supplementary Table S1.

Before large-scale testing of the serum sample collections, we assessed the technical characteristics of the developed assay. Linearity of the measurement ratios was determined to be within the range of Cy3 signal median fluorescence intensity of 7,000 to 15,000 at serum dilution 1:100, and all the recombinant proteins were diluted to a concentration fitting the defined range. Lower limit of antibody detection was determined to be approximately 100 pg/μL by using commercially available antibodies against two of the proteins, CTAG1B and Halo-tag. Repeatability of the assay was assessed by comparing replicates of seven antigens with different intensity of autoantibody signals—intra-assay CV was in range of 0.11–0.21 (mean, 0.14) and interassay CV—of 0.12–0.36 (mean, 0.19; Supplementary Fig. S1), which is acceptable variation for immunological assays (intra-assay and interassay CV should not exceed ∼15% and 25%, respectively; ref. 27).

To identify the autoantibodies relevant to gastric cancer, we analyzed the seroreactivity against the 102 antigens in serum samples from 829 gastric cancer cases and 939 healthy controls from two distinct populations, Caucasian and Asian (Table 1). After excluding low-quality spots and correcting for the variations in the protein quantity and differences across the print runs, an individual cutoff discriminating between seropositive and negative signals was calculated individually for each antigen. Next, the frequency of serum-positive signals above the defined cutoffs and mean signal intensity of the antigen-specific autoantibodies were compared between cases and controls independently for the two populations. Individual antigens with statistically significant differences (P < 0.05) between the two groups were chosen as relevant for discrimination between gastric cancer patients and controls.

In total, autoantibody responses against 13 and 8 antigens significantly differed between gastric cancer and healthy control groups of the Caucasian and Asian cohorts, respectively (Table 2, in bold). In Caucasian population, serological responses against CT antigens CTAG1B/CTAG2, DDX53, and MAGEA3, two epitopes of TP53, and known tumor antigen IGF2BP2 were predominantly found in gastric cancer patients (Table 2; Supplementary Fig. S2). Interestingly, the serological responses against the two TP53 linear epitopes were mutually exclusive in all TP53-positive cases. In Asian sample set, only CTAG1B/CTAG2 and DDX53 were found to be gastric cancer–associated, and the frequency of autoantibody responses against CTAG1B/CTAG2 and mean intensity of seroreactivity against DDX53 in gastric cancer patients were almost twice as high than that detected in Caucasian set. Overall, the frequencies of the identified cancer-associated autoantibodies ranged from 3.5% to 13% (Table 2; Supplementary Fig. S2). Surprisingly, there was a set of antigens, against which autoantibody responses were found to be predominantly associated with healthy controls, and the list of antigens was found to be rather different in Caucasian and Asian sample sets (Table 2; Supplementary Fig. S3)—only one antigen, an artificial peptide ID1625, was shared between the two populations. We calculated a rank value for each antigen to express their combined reactivity in the gastric cancer and the HD sample sets. The positive rank values are associated with predominant gastric cancer reactivity and negative rank values with predominant HD reactivity (see Materials and Methods; Table 2).

We elaborated an algorithm to assess the combined reactivity in each serum against the identified significant 13 antigens for the Caucasian cohort and 8 antigens for the Asian cohort. First, the signal intensity of a detected antigen was multiplied by the respective antigen rank (Table 2). Next, all the ranked intensities were summed for each serum generating a serum score (S) for each patient (see Materials and Methods). In Caucasian population, the mean values of the 13-autoantibody serum scores in gastric cancer and controls were 1,180 and -533, respectively (P = 1.1e–13; Fig. 1B). Next, we assessed the diagnostic value for the 13-autoantibody signature—the ROC curve analysis resulted in AUC of 0.60 [95% confidence interval (CI), 0.58−0.63, P = 1.1e–13; Supplementary Fig. S4A], sensitivity (Sn) of 21%, and specificity (Sp) of 91% at Youden cutoff. By validating the Caucasian 13-marker set in the Asian cohort, an AUC of 0.58 was obtained (Sn of 22%, Sp of 90%, P = 0.00018; mean gastric cancer, 44.67; mean HD, –894.5). The mean values of the 8-autoantibody serum scores of the Asian population in gastric cancer and controls were 1,180 and -533 (P = 7.7e–13), respectively; an AUC reached 0.64 (95% CI, 0.60−0.67, P = 7.7e–13; Supplementary Fig. S4B), Sn of 24% at Sp of 93%. The validation in Caucasian cohort resulted in an AUC of 0.52 (P = 0.057, Sn of 11% at Sp of 93%; mean gastric cancer, 250; mean HD, –1,011; Fig. 1B).

Next, to get an insight whether the autoantibody production is related to clinical and pathologic characteristics, we analyzed the obtained seroreactivity data in the context of the available clinicopathologic information. The 13- and 8-autoantibody signatures were analyzed in the Caucasian and Asian populations, respectively. In line with the results of our previous study (18), we found similar seroreactivity in all stages of gastric cancer (Fig. 1C; Supplementary Fig. S5A). Interestingly, serum scores were higher in the intestinal-type gastric cancer (mean 1770) in comparison with diffuse-type gastric cancer (mean, 778, P = 0.029; Fig. 1D) in the Caucasian sample set. At the level of individual antigens, seroreactivity against CTAG1B/CTAG2 was nearly twice as high and four times more frequent in the intestinal-type gastric cancer than diffuse-type gastric cancer patients (intestinal gastric cancer mean, 5.9; diffuse gastric cancer mean, 2.55; P = 0.0017). Moreover, anti-CTAG1B/CTAG2 seroreactivity was particularly associated with the intestinal-type gastric cancer of high grade (P = 0.0039; Supplementary Fig. S5B), whereas the serum scores between the low-grade and high-grade gastric cancer groups did not show significant differences. We found no differences for serum scores between patient groups stratified by age or gender (Supplementary Fig. S5C and S5D).

To characterize the seroreactivity against TAAs in patients with premalignant lesions of gastric cancer, we analyzed two cohorts from Caucasian population—100 patients with CAG (diagnosed based on serum PG I and II levels) and 775 individuals classified based on OLGIM system into stages 0–IV. Furthermore, we assessed the prevalence of the serological responses in different stages of intestinal-type gastric cancer, known to develop through the atrophy-intestinal metaplasia sequence.

Mean serum score value in serologically diagnosed CAG sample set was -593, which was similar to the HD group (-533). However, when OLGIM-defined patient groups were tested, we saw a tendency for seroreactivity to increase in line with the progression of IM—serum scores were similar in patients with endoscopically confirmed healthy stomach mucosa (OLGIM stage 0, mean -750) and mild/moderate IM (OLGIM stage I+II, mean -770), but were significantly higher in patients with advanced and severe IM (OLGIM stage III+IV, mean -140; P = 0.013, Fig. 2A). At individual antigen level, autoantibody reactivity against CT antigens CTAG1B/CTAG2 and DDX53 tended to appear in advanced stages of IM, and the autoantibody response frequencies were similar to those detected in stage I gastric cancer of intestinal-type (Fig. 2B). The opposite tendency was observed for HD-associated antigens SOX2 and an artificial peptide ID1625, showing more profound seroreactivity in gastritis patients and decreased in gastric cancer patients (Fig. 2B). Serum scores were similar in groups of H. pylori–positive and –negative individuals.

The availability of a small collection of prediagnostic serum samples from 18 patients with incident gastric cancer allowed us to directly assess the possible time points of autoantibody induction against the TAAs. For each patient, only two serum samples were available—one taken within a period of 1 to 9 years before diagnosis and the second after gastric cancer diagnosis. In total, 10 patients had detectable autoantibody responses against at least 1 of the 102 antigens in one or both samples (Fig. 3). Autoantibody responses against TAAs DDX53, SPAG8, and SSX2 were observed only at the time point of diagnosis, whereas autoantibodies against MUC1, BIRC5, IGF2BP1, SOX2, and MYC were detected up to 9 years before diagnosis. Interestingly, for antigens SOX2 (in 2 of 3 patients) and MYC, seroreactivity was detected only in the prediagnostic samples, which is consistent with the predominant seroreactivity in healthy individuals, but not cancer patients observed in this study.

In the current study, by applying recombinant antigen microarray technology, we systematically analyzed autoantibody responses against 102 TAAs in large case–control cohorts from two independent populations. We identified 13- and 8-autoantibody signatures capable of discriminating gastric cancer patients from healthy controls with 21% sensitivity and 91% specificity in Caucasian population and 24% sensitivity and 93% specificity in Asian population, respectively. By analyzing a well-defined serum collection representing consequential stages of gastric cancer development, we show that gastric cancer–associated autoantibodies appear in patients with late stages of intestinal metaplasia at a frequency comparable with early-stage gastric cancer (13% and 12%, respectively), suggesting that humoral immune response against TAAs is generated already during premalignant disease.

The identified gastric cancer–associated antigens (i.e., CTAG1B/CTAG2, DDX53, TP53, MAGEA3, and IGF2BP2) have been linked with this cancer type previously in studies of Caucasian (14, 18) and Asian (19, 24, 28–30) backgrounds. As expected, the two CT antigens CTAG1B/CTAG2 and DDX53 were the top markers in both tested populations—they are among the most reactive antigens in gastric cancer in several studies including ours (14, 18, 31) with similar frequencies. Immunogenicity against TAAs is triggered by their aberrant expression or mutations in transformed cells (19, 24, 32, 33). The observed predominant seroreactivity against CTAG1B/CTAG2 in patients with high-grade intestinal-type gastric cancer is consistent with previously published seroreactivity data in high-grade tumors of other cancer types (34–36), and predominant expression in intestinal, but not diffuse-type gastric cancer (37). Seroreactivity against IGF2BP2 and TP53 was also higher in the patient group with intestinal-type gastric cancer (although no statistical significance was reached at individual antigen level) altogether skewing serum score differences between the two histologic types of gastric cancer demonstrated in this study. Our findings reinforce the notion that intestinal-type and diffuse-type gastric cancer might be different diseases at a molecular level as suggested by recent large-scale “-omics” studies (38).

Interestingly, we did not find anti-TP53 antibodies in the Asian population (Table 2), which contrasts with several studies reporting anti-TP53 autoantibody responses in 8.1% to 32% of Asian gastric cancer patients (24, 28, 29, 39). Yet these studies were testing serology against the full-length protein, whereas the linear TP53 B-cell epitopes represented on our antigen microarray were shown to have predominant seroreactivity in colorectal cancer patients of Caucasian background (17). This implies that differential ethnicity-based genetic backgrounds could underlie the differences in immunodominant TP53 epitopes. This is in-line with existing evidence for differential distribution of TP53 polymorphisms and somatic mutations in different populations (40, 41). Noteworthy, a meta-analysis by Lin and colleagues demonstrated distinct immune response patterns against gastric adenocarcinomas of Asian and non-Asian backgrounds (42), which may explain the overall differences in the autoantibody signatures between the two populations seen in this study. However, the impact of preanalytical or environmental factors cannot be excluded. Therefore, these observations should be validated in independent cohorts from the respective ethnic backgrounds and considered for possible further clinical application for specific target groups discriminated by ethnicity.

Unexpectedly, we identified antigens that predominantly elicited autoantibody responses in healthy controls. Among them were known TAAs and proteins with role in cancerogenesis (MYC, SOX2, MTA1, IGF2BP3; Table 2). This brings up several questions. First, could such humoral responses play a protective role during the earliest stages of malignant transformation and prevent tumor formation in healthy individuals? Second, what is the reason for the lack or loss of seroreactivity in gastric cancer patients? Could the immunosuppressive environment of an established tumor (43) suppress such responses? Or does the lack of such antibodies reflect the loss or lack of cognate antigen expression? Previous studies on SOX2 offer some insight into these questions. SOX2 is a transcription factor with essential role in maintaining pluripotency of stem cells (44). In mice, SOX2 suppresses gastric cancerogenesis through the modulation of Wnt-responsive and intestinal genes (45). In humans, SOX2 expression decreases with the progression of IM and gastric cancer (46, 47) and thus the observed differences in autoantibody frequencies could be attributed to the gradual loss of SOX2 expression during gastric cancer development. Immune responses against SOX2 in gastric cancer patients have not been studied before; however, a study in small cell lung cancer showed that the intensity of SOX2 expression is crucial for eliciting anti-SOX2 autoantibody responses (48). In the current study, we saw an increase in seroreactivity against SOX2 in patients with atrophic and metaplastic gastric lesions in comparison with healthy controls (Fig. 2B). However, what serves as the trigger for autoantibody induction against this antigen, and what is the biological role of the observed immunoreactivity is currently not known.

One of the central questions addressed within the current study is the following: Does the seroreactivity appear in patients with premalignant lesions of gastric cancer? If yes, at what extent? Intestinal-type gastric cancer develops through well-characterized sequential stages, the Correa cascade—chronic gastritis, predominantly induced by H. pylori infection, followed by multifocal atrophy, IM, dysplasia, and finally intestinal-type gastric cancer (49). The autoantibody responses against the CT antigens CTAG1B/CTAG2 and DDX53 appear in late stages of IM at comparable frequency to that seen in early-stage gastric cancer patients. This argues for the presence of nascent cancer cells at the stage of advanced IM in these patients. It has been demonstrated earlier that the induction of spontaneous autoantibody responses against CT antigens is strongly dependent on the antigen expression in cancer cells (19, 24). However, the data on CT antigen expression in patients with premalignant lesions in stomach are poor. To our knowledge, only one study by Cho and colleagues has demonstrated a promoter hypomethylation of DDX53 in correlation with its aberrant mRNA expression in 19 of 55 patients with chronic gastritis, but not healthy individuals (50). However, the expression of CT antigens MAGEA, CT45, CT7, SAGE1, GAGE, NXF2, NY-ESO-1, and CT10 has been detected in squamous dysplastic lesions of esophagus in patterns similar to those detected in squamous cell carcinomas (51). Taken together, the presence of nondetected neoplastic cells expressing the cognate antigens in metaplastic tissues is highly plausible. Advanced stages of metaplastic transformation have been suggested as the cutoff for defining population at high risk of gastric cancer in Western populations (52), and our data support the existing view that these changes in gastric mucosa might represent the point of no return toward malignant transformation. Furthermore, it would be of importance to assess also the prevalence of the cancer-associated autoantibodies in patients with gastric dysplasia as well as to systematically analyze the tumor antigen expression in premalignant gastric lesions. Thus, the detection of circulating cancer-associated autoantibodies could possibly contribute to the identification of high-risk patients and therefore be worth of considering for the integration into standard-of-care diagnostic analyses.

Within this study, we also looked directly into the possible time point of autoantibody response induction in incident gastric cancer patients in a population study setting. Although autoantibodies against several antigens appeared >1 year prior the clinical diagnosis in 3 patients (Fig. 3), and anti-DDX53, SSX2, and SPAG8 autoantibodies were found only at the time point of diagnosis. Yet 3 of 4 samples were taken 6 to 9 years prior the diagnosis and we cannot exclude that the given period between the two samplings is too long—it has been reported that the median time for IM to progress to adenocarcinoma is 6.1 years, and for low-grade dysplasia, 2.6 years (53). Serum antibodies against TP53, MYC, MUC1, annexin I, 14-3-3 theta, and LAMR1 are found 1 to 5 years prior to the clinical diagnosis of colorectal, lung, and breast cancers (15–17, 54, 55). We saw cancer-associated autoantibody response induction in late stages of IM, which most likely is linked to the presence of antigen-expressing cancer cells. Unfortunately, additional samples in-between the two time points and/or information about endoscopy examinations for these patients were not available; therefore, definite conclusions from these results cannot be drawn.

Strengths of our study include the use of a large exceptional collection of serum samples including cases with gastric cancer and various well-characterized precursor lesions, and healthy controls. Besides, an inclusion of both Caucasian and Asian populations allowed the mutual validation of results. The developed microarray platform allowed to assess the seroreactivity patterns systematically in a sensitive high-throughput manner. However, there are some limitations. We applied traditional statistic methods to define antigens with predominant seroreactivity in gastric cancer patients and controls; yet, multiple testing corrections were not applied, and we did not use the adjusted P values for data interpretation because many of the identified markers would not have withstood such correction. We argue that it is not biologically meaningful in the context of cancer serology and is supported by the fact that all the identified gastric cancer–associated antigens in this study represent well-recognized TAAs in gastric cancer as shown by independent studies of Caucasian (14, 18) and Asian (19, 24, 28–30) backgrounds. Another drawback is that the antigen epitopes on the developed array most likely underrepresent the whole gastric cancer–associated TAA pattern, and addition of other relevant TAAs could increase the diagnostic value established within this study. In general, the defined AUC values were rather low and probably insufficient for risk stratification on their own.

Taken together, this is the first and largest study systematically assessing gastric cancer–associated autoantibody responses in an exceptional collection of samples from two populations including well-characterized groups of patients corresponding to the sequential stages of gastric cancer development. Study shows that gastric cancer–associated autoantibody responses are increased in patients with advanced-grade premalignant lesions, are more prevalent in the intestinal-type gastric cancer, and indicate potential population-based differences in immunodominant epitope usage that should be considered in potential clinical application. It is still necessary to establish optimal screening methods in high gastric cancer prevalence areas, and cancer-associated autoantibodies might make a valuable contribution to the stratification of patients with a high risk of developing a gastric neoplasia leading to decrease in gastric cancer–associated mortality rates.

No potential conflicts of interest were disclosed.

Conception and design: I. Meistere, U. Rulle, S. Isajevs, M. Leja, L. Kupčinskas, J. Kupčinskas, C.-Y. Wu, A. Linē, Z. Kalniņa

Development of methodology: I. Meistere, P. Zayakin, K. Siliņa, U. Rulle, S. Isajevs, M. Leja, A. Linē, Z. Kalniņa

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): I. Meistere, P. Zayakin, D. Šantare, I. Kikuste, S. Isajevs, M. Leja, L. Kupčinskas, J. Kupčinskas, L. Jonaitis, H. Brenner

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): I. Meistere, P. Zayakin, S. Isajevs, L. Jonaitis, H. Brenner, A. Linē, Z. Kalniņa

Writing, review, and/or revision of the manuscript: I. Meistere, S. Werner, K. Siliņa, S. Isajevs, M. Leja, L. Kupčinskas, J. Kupčinskas, L. Jonaitis, C.-Y. Wu, H. Brenner, A. Linē, Z. Kalniņa

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Werner, A. Pismennaja, J. Kupčinskas, C.-Y. Wu

Study supervision: L. Jonaitis, Z. Kalniņa

We are thankful to the Latvian Genome Data Base for long-term collaboration and providing the study with valuable serum samples from healthy individuals as well as the Biobank of University of Latvia and Riga East University Hospital for the offered samples. We gratefully acknowledge the excellent cooperation of German clinics and practices in patient recruitment and of Labor Limbach in sample collection. We thank the staff of the Division of Clinical Epidemiology and Aging Research for excellent work in blood sample preparation, data collection, monitoring, and documentation.

I. Meistere, P. Zayakin, K. Siliņa, U. Rulle, A. Pismennaja, D. Šantare, I. Kikuste, S. Isajevs, M. Leja, L. Kupčinskas, J. Kupčinskas, L. Jonaitis, C.-Y. Wu, A. Linē, and Z. Kalniņa were supported by Taiwan-Lithuania-Latvia mutual funds collaboration project No. MOST 103–2923-B-075A-001-MY3 (Taiwan), TAP-LLT-13–022 (Lithuania), and 11-13/IZM14-9 (Latvia). S. Werner and H. Brenner were partly funded by grants from the German Research Council (DFG, Grant No. BR1704/16-1, the BliTz study), Baden Württemberg State Ministry of Science, Research and Arts and by the German Federal Ministry of Education and Research (the ESTHER I and II studies), and German Cancer Aid (Deutsche Krebshilfe (the VERDI study).

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.