Relationship Between Tumor Necrosis Factor-α rs361525 Polymorphism and Gastric Cancer Risk: A Meta-Analysis

Tumor necrosis factor (TNF)-α, a major part in inflammatory, infectious and tumor processes, and is pivotal at the early stages of gastric cancer. Relationship between its risk and TNF-α rs361525 polymorphism has been demonstrated, but remains conflicting and controversial. Therefore, a meta-analysis was conducted to more precisely estimate this relationship. PubMed, Web of Science, EMBASE and CNKI were comprehensively searched to find out relevant articles through October 5, 2017. The strength of the relationship was assessed using pooled odds ratios and 95% confidence intervals. Totally 20 articles were included involving 4,084 cases and 7,010 controls. No significant relationship between TNF-α rs361525 polymorphism and increased GC risk was found in the whole populations. Subgroup analyses uncovered TNF-α rs361525 polymorphism intensified the risk of GC among Asians under five models, but decreased the risk of GC among Caucasiansin the allelic and dominant models. Subgroup analysis by genotyping methods revealed increased risk for other methods. In conclusion, this meta-analysis suggests TNF-α rs361525 polymorphism is related to the risk of GC, especially for Asians.


INTRODUCTION
Gastric cancer (GC) is the fourth major malignancy and the second dominant cause of cancer-induced death in the world (de Martel et al., 2012). In 2017, 28,000 new cases and 10,960 deaths of GC were projected to occur in the United States (Siegel et al., 2017). Helicobacter pylori infection contributes to causing the progression of chronic inflammation to GC (Fox and Wang, 2007). Some studies demonstrated the extremely low risk of developing GC in H. pylori-negative subjects (Uemura et al., 2002). However, nearly all H. pylori-positive subjects have chronic gastritis, and only 1-2% develop to GC. Therefore, other factors such as genetic factors and lifestyle may play important roles in the gastric tumorigenesis (Carcas, 2014).
Tumor necrosis factor (TNF)-α belonging to the TNF/TNF receptor cytokine superfamily can be found in plasma or serum of healthy people, as well as some cancer patients (Balkwill, 2006). TNF-α production by tumors is related with hormone irresponsiveness, poor prognosis, and cachexia/asthenia (Szlosarek and Balkwill, 2003;Tisdale, 2004). The TNF-α-blocked experimental mice were resistant to skin and colorectal cancer occurrence (Moore et al., 1999;Popivanova et al., 2008). (Oshima et al., 2014). The TNF-α/TNFR1 signaling could promote GC occurrence by inducing NADPH oxidase organizer 1 (Noxo1) and G protein subunit alpha 14 (GNA14), which are crucial in the tumorigenicity and stemness of GC cells, in tumor cells (Oshima et al., 2014). The above observations suggest TNF-α plays important roles in the etiology of GC.

Literature Search
Two investigators systematically searched PubMed, Elsevier, EMBASE, and CNKI through Oct 5, 2017 to identify relevant studies using the following search terms: "Gastric Neoplasm, " "Stomach Cancer, " "Gastric Cancer, " "Gastric Carcinoma, " "Gastric Adenocarcinoma, " "tumor necrosis factor alpha, " "TNFα," "polymorphism, " "SNP, " and "variant." In addition, all cited references were reviewed to find out studies that were not FIGURE 1 | Selection for eligible papers included in this meta-analysis. included in the above electronic databases. When two studies overlapped, we chose the latest study or the one with larger sample size. There was no restriction on language, ethnicity or region of study population. GC was diagnosed according to classification criteria.

Data Extraction and Quality Assessment
From each included study, data including name of first author, country of origin, publication year, ethnicity, age, and genotype numbers in cases and controls was extracted. When more than one ethnicity were involved, genotype data was processed separately. Data extraction and study quality assessment based on the Newcastle-Ottawa Scale (NOS) (Stang, 2010) were conducted by two investigators independently. The NOS score varies from 0 up to 9: high-quality study: >7; medium-quality study: 4-6; poorquality study: <4. All conflicting information was discussed and resolved with consensus.

Genotype and Gene Expression Correlation Analysis
Genotype data of TNF-α rs361525 polymorphism and its mRNA expression data were available from the International HapMap Project and GTex portal (https://www.gtexportal.org/home/), respectively (Gong et al., 2017).

Statistical Analysis
The strength of relationship between GC risk and TNF-α rs361525 polymorphism was investigated by using crude ORs and 95%CIs. The following comparisons for this relationship were made: the dominant model (AA+GA vs. GG), the recessive model (AA vs. GA+GG), the heterozygote model (GA vs. GG), the homozygote model (AA vs. GG), and the allele model (A vs. G). Stratification analyses were carried out by ethnicity, source of control (SOC), Hardy-Weinberg equilibrium (HWE), genotyping method and NOS score (He et al., 2014). The null hypothesis that all studies evaluated the same effect was tested by Cochran's Q-statistics. When significant heterogeneity was found (P > 0.10 or I 2 > 50%), a random-effect model was used; otherwise, a fixed-effect model was applied (Higgins and Thompson, 2002). Sensitivity analysis was conducted by omitting one study at a time to test the relative influence on the pooled estimate. The significant findings were evaluated by calculating false-positive report probability (FPRP). An FPRP threshold of 0.2 and a prior probability of 0.1 were set to detect an OR for a correlation with the tested genotype. FPRP <0.2 implied a significant relationship (He et al., 2013). χ 2 test was carried out to clarify whether the observed genotype frequencies conformed to the HWE. Potential publication bias was examined by Begger's and Egger's linear regression tests (Wassen and Jertborn, 2006), with the significant level at P < 0.05. All statistical analyses were conducted on Stata 11.0 (StataCorp, College Station, USA).

Quantitative Analysis
The results concerning the relationship between TNF-α rs361525 polymorphism and GC risk were summarized in Tables 2, 3. This relationship was insignificant in the overall population (A vs. G: OR, 1.06; 95%CI, 0.83-1.35, P = 0.646, Figure 2). Stratification analyses of ethnicity indicated TNF-α rs361525 polymorphism intensified the risk of GC among Asians in most of the comparisons (AA+GA vs. GG: OR, 1.46; 95%CI, 1.11-1.91, P = 0.007, Figure 3), but decreased the risk among Caucasians in the allele and dominant models (AA+GA vs. GG: OR, 0.73; 95% CI, 0.54-0.99, P = 0.043, Table 3). Subgroup analyses by genotyping methods revealed increased risk for other methods (A vs. G: OR, 1.54; 95%CI, 1.04-2.30, P = 0.033, Figure 4) and this relationship also held true in the HWEpositive studies. Stratified analysis by SOC did not find significant correlation in the hospital-or population-based studies ( Table 3). Similar results were observed in the subgroup analysis of NOS score.

Sensitivity and Publication Bias
In the sensitivity analysis, no overall significant change was found when any single study was removed, suggesting our results are statistically robust. Neither Egger's nor Begg's tests (GA vs. GG, Figure 6) showed any evidence of publication bias in this meta-analysis.

FPRP Analyses
The FPRPs for significant results at different p levels are shown in Table 4. At the level of 0.1, some FPRPs were all <0.20, indicating the significant associations between TNFα rs361525 polymorphism and GC risk were noteworthy (Table 4). However, the FPRPs for other significant associations were larger, suggesting some possible bias due to sample size reduction existed in some subgroups, which should be validated by larger-size studies in the future.

DISCUSSION
It is hypothesized chronic inflammation plays a crucial role in the etiology of GC and other cancers. Epplein et al. found the upregulated circulating levels of inflammationrelated cytokines such as TNF-α may intensify the risk of GC (Epplein et al., 2013). TNF-α-induced protein secretion from H. pylori is involved in the development of GC (Suganuma et al., 2012). It is assumed the stimulation of the TNFα/TNFR1 signaling in the tumor microenvironment enhances GC progression by inducing Noxo1 and GNA14 (Oshima et al., 2014). TNF-α is a promising effective target for GC therapy. The TNF-α rs361525 polymorphism could change TNFα gene transcription and adjust TNF-α generation (Kaluza et al., 2000). Many studies have reported the relationship between TNF-α rs361525 polymorphism and GC risk (Jang et al., 2001;Wu et al., 2002Wu et al., , 2003Wu et al., , 2004Glas et al., 2004;Lee et al., 2004;Lu et al., 2005;Zambon et al., 2005;Kamangar et al., 2006;Xing et al., 2006;Garcia-Gonzalez et al., 2007;Hou et al., 2007;Zeng et al., 2007;Crusius et al., 2008;Bai et al., 2009;Yang et al., 2009;Whiteman et al., 2010;Yin et al., 2012;Essadik et al., 2015;Xu et al., 2017), but with conflicting findings. Given such conflicts, several meta-analyses in this field have been conducted (Zhou et al., 2011;Yu et al., 2013;Rokkas et al., 2014;Hui et al., 2016). Zhou et al. firstly meta-analyzed the association between TNF-α rs361525 polymorphism and cancer risk (Zhou et al., 2011), but found no significant result in the general populations (Zhou et al., 2011) or in the subgroups of cancer type including GC (Zhou et al., 2011). The two subsequent meta-analyses also did not observe an association between TNF-α rs361525 polymorphism and GC risk (Rokkas et al., 2014;Hui et al., 2016). Noticeably, Hui et al. revealed TNF-α rs361525 polymorphism could significantly increase the risk of digestive system cancer, but not GC (Hui et al., 2016). However, Yu et al. uncovered a significant relationship between TNF-α rs361525 polymorphism and increased GC risk in Asians, but not in Caucasians (Yu et al., 2013). Recently, new studies in this field have emerged, which further necessitates a new comprehensive meta-analysis. Here our data are consistent with Yu et al. It is worth noting that we found a significant association of TNF-α rs361525 polymorphism with GC among both Asians and Caucasians in the stratified analysis by ethnicity, while Yu et al. did not find any association among Caucasians (Yu et al., 2013). We found TNF-α rs361525 polymorphism increased the risk of GC among Asians, while this SNP seemed to protect Caucasians from GC. There are several possible interpretations for the different findings between Asians and Caucasians. Firstly, GC may be genetically and clinically heterogeneous among different populations. Secondly, varying sample sizes of Asians and Caucasians may also account. The third reason may be the differences in genotyping methods and random errors. The fourth reason may be the varying prevalence of H. pylori among different populations. Last but not least, different geographical environments and dietary pattern may also be influential (such as Asians eat more pickled and fried food). We believe this meta-analysis has several strengths over previous meta-analyses. Firstly, our sample size was larger. Secondly, sensitivity analysis proved the reliability and stability of our data. Thirdly, we conducted subgroup analyses of ethnicity, HWE, SOC, NOS score, and genotyping methods, and explored the potential sources of heterogeneity. Finally, we calculated false-positive report probability and statistical power.  However, this study also has some limitations. Firstly, no subgroup analyses of confounding factors such as age, sex, smoking or H. pylori infection were conducted. Secondly, potential gene-gene or gene-environment interactions were not assessed. Thirdly, only one SNP of TNF-α gene was investigated. Finally, the relationship between this SNP and clinical manifestations of GC was not examined.
In conclusion, this meta-analysis indicates the TNF-α rs361525 polymorphism increases the risk of GC among Asians, but decreases the risk of GC among Caucasians. This finding should be validated by larger case-control studies in other ethnicities.

AUTHOR CONTRIBUTIONS
ZK conceived the entire study. TX and HZ analyzed the data. TX and ZK performed statistical analysis. ZK and HZ wrote the paper. All authors read and agreed with the final version of this paper.