Introduction

Globally, gastric cancer (GC) is one of the most prevalent high-mortality cancers, and is reckoned to cause 951,000 new cases and 723,000 deaths each year [1]. In China, the disease burden of GC is extremely high, and it has been indicated that the number of new GC cases and deaths for 2015 are 679,100 and 498,000, respectively [2]. The most important known cause of GC is chronic inflammation induced by Helicobacter pylori (H. pylori) infection [3, 4], but malignant transformations were found to occur in only a small percentage of infected individuals. Therefore, apart from H. pylori infection, other factors such as genetic factors, dietary factors, Epstein-Barr virus infection, smoking and obesity may also contribute to incidence of GC [5, 6].

Germline variations of genomic sequence are implicated in the predisposition to most complex traits, including gastric cancer. Genetic variants in IL1B, encoding a cytokine with a pivotal role in GC development, were among the most studied variants based on candidate gene approach, and two variants (rs1143634 and rs16944) were associated with gastric cancer in an intermediate level of summary evidence [7]. Genome-wide association study (GWAS) is an effective method to simultaneously assess a large number of single nucleotide polymorphisms (SNPs) through high-throughput genetic analysis. Basing on this method, a series of genetic loci were identified to affect GC susceptibility, including rs4072037 (MUC1) [8], rs80142782 (ASH1L) [9], rs9841504 (ZBTB20) [10], rs13361707 (PRKAA1) [10], rs7712641 (lnc-POLR3G-4) [9], rs2294693 (UNC5CL) [11], rs1679709 (BTN3A2) [12], rs2294008 (PSCA) [13], rs2274223 (PLCE1) [8]. Nevertheless, the biological mechanisms behind the association between genetic variants and GC risk still remain unclear. Besides, a study in 2016 proposed that the estimate of heritability for GC was 22%, while those established risk loci could only explain a small proportion of GC heritability [14]. Therefore, there still exist large amounts of GC associated loci in the whole genome, and they have been missed most probably due to strict significance thresholds used in GWAS. Pathway-based GWAS analysis is a new and effective strategy that can detect the associations missed by traditional GWAS and explore the biological mechanisms of diseases. Recently, this new analysis strategy has given novel insights into pathogenesis of cancers such as oesophageal squamous cell carcinoma [15], pancreatic cancer [16] and breast cancer [17].

Multi-marker analysis of genomic annotation (MAGMA) is a fast and flexible tool for gene and gene-set analysis based on GWAS data [18]. MAGMA’s gene analysis uses a multiple regression method to validly incorporate linkage disequilibrium (LD) between variants and to discover multi-variant effects. Pathway-based analysis is based on gene-set analysis that is conducted using a gene-level regression model. MAGMA is a powerful tool to detect genes and pathways associated with diseases and to help us explore the potentially biological mechanisms [19, 20].

In this study, on the basis of Chinese GC GWAS, we applied MAGMA to conduct pathway analysis to identify crucial pathways and genes that contribute to GC susceptibility.

Materials and methods

Study populations

We conducted the analysis relying on three GC GWASs from ethnic Chinese. Two GWASs from Nanjing and Beijing populations (NJ-GWAS and BJ-GWAS) were implemented by our team [10], and the other GWAS from Shanxi and Linxian (NCI-GWAS) was based on Shanxi Upper Gastrointestinal Cancer Genetics Project supported by the National Cancer Institute (NCI) [8]. Details about the above three GWASs have been described elsewhere [8, 10]. In brief, all GC cases were histopathologically confirmed, and cancer-free controls from NJ-GWAS and BJ-GWAS were matched on age, sex and geographic region, while controls were matched on age and sex in NCI-GWAS. In total, 2631 GC cases and 4373 controls were included in our analysis. Basic demographic information of the participants was shown in Supplementary table 1.

Quality control, genotype imputation and meta-analysis

We excluded SNPs with call rate < 95%, minor allele frequency (MAF) < 0.01 or Hardy–Weinberg equilibrium (HWE) P value < 1 × 10−6. Then, we performed imputation with SHAPEIT [21] and IMPUTE2 [22] for those three GWASs separately. All populations from the 1000 Genomes Project Phase III were taken as the reference set. SNPTEST [23] was performed to evaluate the relationship between each variant and GC, then we used GWAMA [24] to conduct the meta-analysis. To get more reliable results, the SNPs were selected with more stringent inclusion criteria: HWE -value > 1 × 10−3 and imputed INFO score ≥ 0.5. Finally, 6,865,316 SNPs which were shared by all three GWASs and showed no obvious heterogeneity (I2 < 75%) were included in our following analysis.

Pathway-based analysis

Pathway data from four databases were downloaded online (http://software.broadinstitute.org/gsea/msigdb/collections.jsp). Totally, 186 gene sets form Kyoto Encyclopedia of Genes and Genomes (KEGG) database, 4436 gene sets from Gene Ontology (GO) database about biological process, 217 gene sets from BioCarta database and 674 gene sets from Reactome database were included in our analysis. We performed gene-based analysis on SNP P values, and raw genotype data from Nanjing and Beijing populations were set as a reference for linkage disequilibrium (LD). SNPs located in 10 kB upstream and downstream of a coding gene were mapped to the gene. The SNP-wise mean model which is equivalent to SKAT model using inverse variance weights [25] was adopted in this analysis. The maximum and minimum number of permutations per gene were set as 1 million and ten, respectively. We got two results about the P value for a certain gene, relying on different distributions (asymptotic sampling distribution and permutation-based sampling distribution). P value relying on asymptotic sampling distribution represented the result of gene-based analysis in this study. Taking advantage of the results of gene analysis, we implemented the gene-set analysis through a linear regression model. As Bonferroni correction may be too conservative when gene sets are strongly overlapping, we adopted a permutation-based empirical multiple testing correction which is provided by MAGMA. One hundred thousand permutations were performed during the gene-set analysis.

Differential expression analysis

Differential expression analysis was performed based on the data downloaded from The Cancer Genome Atlas (TCGA) database. Differential expression among paired (32 GC tissues and 32 adjacent normal tissues) and unpaired (413 GC tissues and 32 adjacent normal tissues) samples were both calculated in our analysis.

Immunohistochemistry (IHC)

Immunohistochemistry was performed on 55 pairs of GC and matched adjacent normal tissues of Chinese patients using tissue arrays (Shanghai Outdo Biotech Co., Ltd.). The tissue sections were sequential incubations with anti-N myc interactor antibody (for NMI, ab183724, Abcam) or anti-Rac1 antibody (for RAC1, ab33186, Abcam), and EnVision™ FLEX /HRP reagent (DM842, Dako Omnis). The intensity of staining and the frequency of the stained cells were estimated by two investigators who were blinded to the patients’ information. The frequency of positive cells was scored as follows: ≤5% = 0; >5 to ≤ 25% = 1; >25 to ≤ 50% = 2; >50 to ≤ 75% = 3; and > 75% = 4. Another score was given according to the intensity of staining as follows: negative = 0; weak = 1; moderate = 2; or strong = 3. For purposes of statistical analysis, NMI and RAC1 proteins’ intensity and frequency were transformed into a staining intensity score (SIS) calculated by multiplying the staining intensity score by the frequency score [26].

Statistical analysis

Multivariate logistic regression analysis in the additive model was used to estimate the association between genetic variants and GC risk. Nanjing and Beijing GWASs took age, sex, smoking status, alcohol consumption and the top genotype principal components (PCs) as covariates. NCI-GWAS was adjusted for age (10-year categories), sex and the top PCs. EIGENSTRAT 3.0 software was used to evaluate population structure. We adopted a fixed-effect model in the meta-analysis to explore the correlation between the single genetic variant and susceptibility of GC. As for differential expression analysis among mRNA level, t test was used to measure the difference between log-2 transformed expression values of GC and normal tissues. Moreover, differential expression analysis among protein level was conducted by Wilcoxon signed rank test on SIS.

Results

Based on 2631 GC cases and 4373 controls, we first obtained the gene-based results. Among 18,449 protein-coding genes on autosomes, we identified 1465 genes with P < 0.05 (Fig. 1). After Bonferroni correction, 20 genes reached statistical significance (P < 2.7 × 10−6, 0.05/18449) (Supplementary table 2), of which 19 genes were located in known GC susceptibility regions including 1q22, 5p13.1 and 10q23.33. IRGC in 19q13.32 was a new gene, but its expression level was similar between tumor and normal tissues (P = 0.15 and 0.52 for unpaired and paired samples from TCGA, respectively, Supplementary figure 1).

Fig. 1
figure 1

Manhattan plots that show the distribution and association of genes with GC risk. The x-axis is chromosomal position and the y-axis is − log10(P). the red horizontal line represents P = 2.7 × 10−6 while the green horizontal line represents P = 0.05. Red circles showed the association at P < 2.7 × 10−6, and the known GC-associated genes (MUC1 and PRKAA1) were marked by orange circles

Full size image

Through pathway-based analysis, 244 candidate pathways were identified at the level of P < 0.05. Of the above pathways, 5 were from KEGG database, 200 from GO database, 11 from Biocarta database and 28 from Reactome database (Supplementary table 3–6). After permutation-based empirical multiple testing correction, no pathway remained. However, 3 pathways reached a less stringent threshold (corrected P < 0.5), and they were chemokine signaling pathway, potassium ion import pathway and interleukin-7 (IL7) pathway (Table 1). Stratification analysis of pathways showed no heterogeneities between the groups by age and sex (Supplementary table 7).

Table 1 Summary of significant pathways associated with GC
Full size table

To discover new potentially causal genes, we examined the relationship between genes in each of the three pathways mentioned above and GC risk. Eighteen of 183 genes in chemokine signaling pathway were identified (P < 0.05), but only RAC1 and MAP2K1 were left after false discovery rate (FDR) correction (FDR corrected P < 0.05) (Table 2, Supplementary table 8). In potassium ion import pathway, although 4 of 28 genes were found to be associated with GC risk (P < 0.05), none of them reached the FDR corrected criterion (Supplementary table 9). Among 16 genes in IL7 pathway, 4 genes reached the P level at 0.05 and NMI passed FDR correction (Table 2, Supplementary table 10).

Table 2 Summary of significant genes in significant pathways
Full size table

For the three genes passing FDR correction, we estimated the differential expression of them in paired and unpaired samples from TCGA. Results showed that NMI was overexpressed in GC for both of unpaired and paired samples, while increased expression of RAC1 in GC tissues was observed in unpaired tissues (Fig. 2). NMI and RAC1 were overexpressed in GC tissues in either early or advanced cases (Supplementary figure 2–3). However, the expression of MAP2K1 was similar between GC and normal tissues (Supplementary figure 1).

Fig. 2
figure 2

Differential expression of NMI and RAC1. a, c Exhibited results of all unpaired samples from TCGA; b, d exhibited results of all paired samples from TCGA. The x-axis showed the number of GC tissues and the normal ones used in the analysis

Full size image

NMI localized to cytoplasm, whereas RAC1 localized primarily to cytoplasm and cellular matrix (Fig. 3). In GC tissues, NMI and RAC1 expressed in cancer cells but not in tumor-infiltrating lymphocytes. Among 55 Chinese GC patients, the protein levels of NMI and RAC1 were also increased in GC tissues as compared with paired normal tissues (P = 1.48 × 10−7 and 2.35 × 10−7, respectively), for either early stage (P = 2.04 × 10−3 and 2.30 × 10−3, respectively) or advanced stage of GC (P = 1.00 × 10−5 and 2.40 × 10−5, respectively) (Fig. 3).

Fig. 3
figure 3

Results of Immunohistochemical staining of NMI and RAC1 proteins. a Immunohistochemical staining of NMI protein in two pairs of representative GC tissues and the adjacent normal tissues. b Immunohistochemical staining of RAC1 protein in two pairs of representative GC tissues and the adjacent normal tissues. c Distribution of staining intensity score of NMI protein in 55 paired GC tissues and the adjacent normal tissues. Red bars represent results of GC tissues, while blue bars represent results of the adjacent normal tissues. The first 25 paired samples came from early GC patients, and the last 30 paired samples came from advanced GC patients. d Distribution of staining intensity score of RAC1 protein in 55 paired GC tissues and the adjacent normal tissues. Red bars represent results of GC tissues, while blue bars represent results of the adjacent normal tissues. The first 25 paired samples came from early GC patients, and the last 30 paired samples came from advanced GC patients

Full size image

Discussion

In this study, on the basis of 2631 GC cases and 4373 controls, we found 3 candidate pathways and 2 candidate casual genes associated with GC in Chinese populations. In a previous study, Lee et al. [27] performed an Identify Candidate Causal SNPs and Pathways (ICSNPathway) analysis based on NCI-GWAS, and reported several hypothetical biological mechanisms of GC, including ephrin receptor binding, drug and pyrimidine metabolism, cyanoamino acid metabolism, and lipid biosynthetic process, regulation of cell growth, and cation homeostasis. Using similar dataset and ICSNP pathway analysis, Zhu et al. [28] reported similar mechanisms including ephrin receptor binding. In contrast to the ICSNPathway aiming to discover candidate causal pathways that represent the way in which the candidate causal SNPs affect GC [29], MAGMA was used to identify novel pathways associated with GC in the current study. As a result, we found three new pathway for gastric cancer, including chemokine signaling pathway, potassium ion import pathway and Interleukin-7 (IL7) pathway. We also several identified pathways such as drug metabolism—other enzymes, pyrimidine metabolism, regulation of cell growth, and cellular cation homeostasis were repeated in our analysis, apart from three new discovered candidate pathways.

Three pathways were identified in our study. They were chemokine signaling, potassium ion import, and IL7 pathway. Chemokines are a large family of small cytokines and they were initially discovered as they could recruit immune cells to a site of inflammation during an immune response [30]. Subsequent researches reported chemokines could also provide directional guidance to normal cells like neurons and germ cells during embryonic development [31], besides, they could migrate cancer cells to distant sites during metastasis [32, 33]. Nowadays, researchers have paid great attention to the relationship between chemokines and cancer, and they found that chemokines were involved in many other cancer-related processes, including facilitating growth and survival of cancer cells [34, 35] and formation of cancer blood vessels [36]. It has been reported that chemokines persisting at an inflammatory site were crucial in neoplastic progression [37], and once GC occurred, certain chemokines could be produced by tumor and might play a great role in promoting the development and progression of GC. IL-7 is an important cytokine for adaptive immune system, because IL-7 plays an essential role in the development of B and T cells such as generation of T cells in the thymus and peripheral homeostasis of T Cells [38]. A healthy immune system could help clear tumor cells effectively, and suppressed immunity might increase the risk for cancers. Additionally, cancers could suppress normal immunological surveillance function to escape being identified by immune system during the progression of cancers [39]. According to the evidence above, we inferred that both chemokine signaling pathway and IL7 pathway might participate in occurrence of cancers through immune and inflammatory processes, and facilitate the development of cancers through multiple mechanisms.

Nowadays, immunotherapy has been a new treatment for cancers and has achieved great success [40,41,42,43]. For GC, several common immunotherapies such as cancer vaccines therapy, T cell based adoptive transfer therapy and checkpoint inhibition therapy have been used to induce antitumor immunity to kill cancer cells and improve survival [44]. Although chemokines and cytokines have been commonly applied to the treatment of cancers, they were rarely used for treating GC [45,46,47]. Recent studies have found IL7 has great potential in cancer immunotherapy because of its function in immune reconstitution, enhancing the function of effector immune cells and fighting against the immunosuppressive network [48]. Therefore, in consideration of those findings of the recent studies mentioned above and our results, there are reasons to believe that new ways of immunotherapy for GC might be discovered centering on chemokines, IL7 and other cytokines in the future.

In addition to the above two pathways, potassium ion import pathway was the third one we discovered. Potassium has been reported to be associated with apoptosis, for example, loss of potassium ions is linked to shrinkage which is a basic morphological characteristic of apoptosis [49]. Besides, low level of cytoplasmic potassium is capable of activating caspases and nucleases, which is pivotal in apoptosis [50, 51]. Abnormal apoptosis process could break the equilibrium between cell proliferation and cell death, which might promote oncogenesis because “undead” cells are accumulated [52]. After oncogenesis, accumulation of potassium in T cells would suppress the activity of T cells [53]. Enhancing activity of potassium channel may enable T cells to attack cancer cells more effectively and powerfully [54]. Besides, potassium channels were reported to be correlated with multidrug resistance in gastric cancer cells [55]. Thus, process of potassium import may influence susceptibility of GC and provide new ideas for treating GC.

NMI and RAC1 were two new candidate genes discovered in our study. NMI in 2q23.3 could interact with N-myc and C-myc which are two members of the oncogene Myc family and induce transcription activity as a transcription cofactor [56]. Researchers have found that overexpression of NMI retards invasion and growth of cancer cells through inhibiting the Wnt/beta-catenin signaling [57]. Besides, NMI could suppress tumor invasion and metastasis by inhibiting NF-κB pathways in human gastric cancer cells [58]. NMI in IL7 pathway can augment the recruitment of CBP/p300 to transcription factor STAT5, which can enhance transcription of downstream target genes [59]. STAT5 proteins are required for normal T cell proliferation and NK function. Additionally, they critically regulate vital cellular functions such as proliferation, differentiation, and survival [60]. RAC1 located at 7p22.1 is a GTPase that belongs to the RAS superfamily whose members can regulate cellular events, such as cell growth, cytoskeletal reorganization, and the activation of protein kinases [61]. RAC1 in chemokine signaling pathway can interact with downstream effector PAK1. Activation of PAK1 affects mitogen activated protein kinase (MAPK), phospoinositide 3-kinase (PI3K), and Wnt/beta-catenin signaling pathways associated with inflammation and malignant transformation [62]. RAC1 is also a key activator of NF-κB and promote tumor growth by inducing the expression of inflammatory cytokines [63]. It has been proved that RAC1 was overexpressed in cancer tissues and associated with poor survival [64, 65]. In addition, researchers have indicated RAC1 can serve as therapeutic targets for cancers including GC [66, 67]. Therefore, both of the two candidate genes discovered in our study are vital in the progression of cancers and might be treated as drug targets for GC.

In conclusion, immune and inflammatory associated processes and potassium transporting might participate in the occurrence and development of GC. Besides, NMI and RAC1 were two new vital genes related to GC. Our findings might give new insight into the mechanism and treatment of GC. However, due to the genetic heterogeneity among different populations [7, 68], there might be inconsistent results according to the ethnicities as a result of differences in genetic basis as well as environmental exposures. In the future, additional studies are needed to confirm the identified pathways and genes associated with risk for GC in populations of East Asian and elsewhere, and functional experiments should be conducted to observe the effect of the two candidate genes on the development of GC. Additionally, more efforts are warranted to explore new therapies for GC based on our findings, and potential therapies may refer to immunotherapy centering on chemokines, IL7 and other cytokines, to enhance activity of potassium channel, and therapeutic targets on NMI and RAC1.