Introduction

Gastric cancer is the fifth most common cancer and third leading cause of cancer‐related death worldwide [1]. The prevalence of early gastric cancer is increasing in East Asian countries. However, advanced gastric cancer is regarded as a refractory disease. Age, sex, histopathological type, and stage have been reported as prognostic factors [2]. Studies have attempted to elucidate the molecular biology of gastric cancer by biomarkers (e.g., p53 or E-cadherin) [3, 4]; however, prognostic markers are difficult to establish due to histopathological diversity and heterogeneity [5, 6].

Next-generation sequencing has contributed to the establishment of new molecular classifications of gastric cancer that focus on the biological characteristics. The Cancer Genome Atlas (TCGA) identified four genomic gastric cancer subtypes [7]. The Asian Cancer Research Group (ACRG) focused on gene expression profiles and classified gastric cancer into four subtypes with distinct clinical outcomes [8]. However, whether the molecular classification predicts recurrence and survival in gastric cancer patients undergoing radical gastrectomy remains unclear because previous publications included patients who received non-curative surgery.

Shizuoka Cancer Center has conducted a comprehensive genetic analysis project (High-tech Omics-based Patient Evaluation [Project HOPE]) since 2014 [9]. We have reported genetic alterations according to cancer type in a large Japanese cohort (5,143 samples from 23 cancer types) using whole-exome sequencing and a transcriptome analysis of fresh-frozen tissue specimens (Japanese version of The Cancer Genome Atlas: JCGA) [10]. Although this study comprehensively reported genetic profiles in various cancers in Japan, the clinicopathological features (especially survival) of each cancer type were not shown. The present study aimed to establish a new molecular classification of gastric cancer that predicts survival in patients undergoing radical surgery using data from Project HOPE.

Methods

Study design and participants

This study included advanced gastric cancer patients who underwent radical gastrectomy for gastric cancer from January 2014 to March 2019. The following exclusion criteria were applied: (1) special histological tumor type (except lymphoid stroma); (2) gastric stump cancer; (3) deficient multiple omics data; (4) any chemotherapy before gastrectomy; and (5) palliative (R2) gastrectomy. Clinicopathological data were collected from a prospectively recorded database or electronic medical records. The pathological stage was determined according to the Japanese Classification of Gastric Carcinoma [5]. Treatment, including surgery, adjuvant chemotherapy, and follow-up, was performed according to the Japanese Gastric Cancer Treatment Guidelines [11].

Clinical samples

Gastric cancer and surrounding non-cancerous tissue (≥ 0.1 g) were dissected from surgical specimens. Small or unclear tumors were excluded from the cohort. Samples were treated as previously described [10].

Data for the analysis of somatic alterations

The DNA extraction and gene mutation analysis protocols have been previously described [10]. All samples were analyzed by whole-exome sequencing (WES) and 409-gene panel sequencing (including genes considered to be important for gastric cancer) with a high read depth (mean read depth of 1169). The results from WES and panel sequencing were combined.

Gene expression signature analysis

RNA extraction and the gene expression analysis were performed as previously reported [10]. WES and part of the gene expression profile data were submitted to the National Bioscience Database Center (NBDC) Human Database as ‘Controlled-Access Data’ (Research ID, hum0127.v1; https://humandbs.biosciencedbc.jp/en/).

Molecular classification based on The Cancer Genome Atlas classification

The tumors were classified into four TCGA subgroups based on the previous TCGA report [7]. Tumors that tested positive in the Epstein–Barr virus (EBV) in situ hybridization (EBER-ISH) test were classified as the EBV subgroup. EBER-ISH was performed when the pathologist suspected EBV-related gastric cancer by tumor-infiltrating lymphocytes. The microsatellite instability subgroup was defined from our in-house criterion: tumor mutational burden (TMB) ≥ 10/Mb and MSI sensor score ≥ 1 or COSMIC (http://cancer.sanger.ac.uk/cosmic) MSI mutational signature (signature 6 in version 2) contribution ratio ≥ 0.5 [13]. Thus, the different proportions may due to racial differences. In contrast, the clinicopathological and genomic features of our Japanese cohort were similar to the TCGA cohort, suggesting that the TCGA classification is applicable to other races. On the other hand, a feature that was not described in the TCGA study was represented in our cohort. In particular, DLC1 mutation, which was frequently observed in the EBV subgroup of our cohort, was not listed as a significant feature in the TCGA study. However, DLC1 tended to be more frequently mutated in the EBV subgroup (2/19; 10.5%) in comparison to the GS (3/58; 5.2%) or CIN (8/147; 5.4%) subgroups in the TCGA cohort. DLC1 mutation may be associated with EBV-related gastric cancer; however, it should be noted that the numbers of tumors in the EBV subgroup were limited in both the TCGA study and our cohort. The TCGA classification was established from the viewpoint of genomic alterations in molecular biology, not survival. Actually, OS did not differ among the four subtypes in the TCGA cohort [7]. Survival analyses according to TCGA classification in other cohorts showed no consistent trends, presumably due to background differences [7, 8, 21]. DSS in each of our subgroups was better in the order of EBV, MSI, GS/CIN; however, this was not statistically significant. Patients with GS and CIN showed almost equivalent survival.

In the ACRG classification, in our Japanese cohort, the proportion of MSS/EMT was higher, and that of MSS/TP53− was lower in comparison to the ACRG cohort. Since the ACRG cohort consisted of similar East Asian population (Koreans), the difference may be due to differences in the expression profile analysis, computer processing, or the definitions of each subgroup (e.g., we used the expression ratio of the tumor and non-cancerous tissue, which differed from the ACRG study, and the methods of averaging each expression value were different). In our cohort, patients with MSI showed the best DSS, while patients with MSS/EMT showed the worst DSS, which corresponded to the ACRG report; however, the results were not statistically significant, possibly due to the small number of events [8].

When constructing our new classification, TMB-high tumors were first extracted, because TMB-high has been regarded as a distinctive feature. TMB-high cancer is believed to create numerous neo-antigens, which promote the infiltration of cytotoxic (CD8+) T-cells and activated Th1 cells in the tumor microenvironment [22]. Though MSI is a leading cause of TMB-high, various mechanisms promote a TMB-high status [7, 22]. TMB was reported as a stronger prognostic factor than MSI-high by Doming et al. and Lee et al. [23, 24]. In this study, patients with HMT showed favorable DSS, while in patients with TCGA MSI, DSS was not significantly better in comparison to the other TCGA subgroups, suggesting that TMB may be a stronger prognostic factor than MSI. The T-cell-inflamed expression signature in HMT was more distinct than in the other non-T-cell-inflamed (EMTH + EMTL) subgroups, but it was widely dispersed. No clear association was observed between HMT and the anti-tumor immune response in this study.

The association between immunity and the development or progression of cancer has long been reported, and recently molecules involved in cell-mediated immunity are reported to play a crucial role [25, 26]. T-cells are reported to have an important role in the progression of gastric cancer and other solid cancers [27, 28]. We previously reported that the upregulation of immune-related genes was associated with favorable survival in gastric cancer patients who received adjuvant chemotherapy [29]. Then, TCI was extracted from non-TMB-high tumors. Both HMT and TCI have a common biological feature of an activated immune response in the tumor microenvironment, although the biological backgrounds are not the exactly same. Moreover, these subgroups showed comparable DSS. Thus, in a further survival analysis, HMT and TCI were integrated in a single subgroup.

After extracting HMT and TCI, the remaining tumors with an inactivated immune response were divided by the EMT expression signature. The relationship between EMT and tumor progression, metastasis, and survival is well known, and the mesenchymal subgroup showed poor survival in recently reported molecular classifications [6, 30]. Consistent with these reports, EMTH showed the poorest survival. In a pathological analysis, undifferentiated tumor with dense desmoplastic stroma (so-called scirrhous tumor) was often observed in patients with EMTH. EMT was reported to be associated with chemoresistance to 5-fluorouracil [31, 32]. Patients with EMTH also appeared to show a poor response to adjuvant chemotherapy with S-1 in this study. Conventional chemotherapy may be suggested for non-EMTH tumors, which are relatively sensitive to chemotherapy, while a new treatment strategy (e.g., drugs circumventing immunosuppression [EMT reportedly plays a role in suppressing anti-tumor immunity] [33] or drugs targeting mesenchymal-cell-specific proteins [34]) should be developed for EMTH tumors.

The present study was associated with some limitations. First, this was a single-institutional study. Although external validation with the TCGA cohort showed the reproducibility of our classification, further external validation using cohorts with different races and backgrounds is needed. Second, the tumor and stroma were analyzed without microdissection. Thus, in the tumor microenvironment of TCI, the locations at which immune-related molecules are actually expressed and the interaction between tumor and stroma remain unclear. Further evaluation of the localization of immune-related molecules by immunohistochemical staining is needed. Third, analyses to detect gene mutations in low-purity tumors may have yielded false-negative results, since tumors were not selected by purity. However, all samples were dissected from the tumor-rich area designated by a pathologist, and sampling from either small or uncertain tumors was avoided. Genomic data were obtained from fresh-frozen tissue and peripheral blood. Thus, the somatic mutation data are considered superior to data from FFPE tissue. Moreover, the probability of detecting mutations was increased, even in low-purity tumors, by combining the results from WES and 409-gene panel sequencing (including genes considered to be important for gastric cancer) with a high read depth (mean read depth of 1169). Thus, the possibility of false-negative results appears quite low. Finally, the TCGA classification methods in our cohort are not the exactly same as the original methods. Several similar studies have demonstrated that these classifications well reflect the characteristics of TCGA subgroups [8, 13, 21]. Accordingly, the results of the present study may reproduce the results of the original TCGA study.

In conclusion, we established a new molecular classification that predicted survival in gastric cancer patients undergoing radical gastrectomy.