Polymorphisms of Tumour necrosis factor-α-308 (rs 1800629) and gastric cancer susceptibility: A meta- analysis of associations studies with trial sequential analysis

Background: Gastric cancer is globally the fth most common cancer. Several studies have assessed the relationship between tumour necrosis factor-alpha (TNF-a- 308) and the risk of gastric cancer. These individual genetic association studies showed inconclusive results. The objective of the present study was to synthesis evidence on the association between TNF-a-308 polymorphisms and gastric cancer risk by meta-analysis of data from eligible studies. Methods: We performed a meta-analysis of genetic association studies, according to the PLOS One checklist. We searched relevant case-control studies in health-related electronic databases. The methodological quality of included studies was assessed by the Newcastle-Ottawa quality assessment scale. The strength of association was calculated as odds ratios (ORs) with 95% condence intervals (CIs). Pooled ORs and 95 % CIs were estimated using random-effects model or xed effect model, based on between-study heterogeneity. We analysed the strength of association under four genetic models (allele, dominant, recessive and additive models). Subgroup analyses on ethnic groups, Hardy-Weinberg equilibrium (HWE) status, status of Helicobacter pylori infection and study quality were done for robustness of the estimates. Publication bias was detected by inspection of funnel plot asymmetry. To estimate the required information size, we performed trial sequential analysis (TSA) that classied the effect estimates as ‘rm evidence of effect’ or ‘potentially spurious evidence of effect’. Results: A total of 35 studies, comprising 11353 cases and 12827 controls were identied. Based on 28 studies that met HWE, there was overall signicant association between TNF-α-308 polymorphisms and gastric cancer risk under the dominant model (OR 1.19, 95%CI 1.1-1.29, I 2 :37%), as well as Asians (OR 1.2, 95%CI 1.05-1.38, I 2 :53%) and Cassian subgroups (OR 1.19, 95%CI 1.07-1.31, I 2 :28%). Based on 13 high quality studies under the dominant model, overall signicant association was also found (OR 1.38, 95%CI 1.07, 1.77). The TSA plot indicated the analyses was with the required information size. There was no publication bias. In the subgroup analysis by ethnic groups, the quality of studies impacted on the estimates. Conclusions: The ndings suggest that TNF-α-308 gene polymorphism plays an important predisposing role for gastric carcinogenesis, and can serve as a useful screening marker.


Introduction
Gastric cancer (ICD10: C16) is globally the fth most common cancer with the highest prevalence in both sexes in Asia (74.5%) and the third leading cause of cancer death. In 2018, the estimated number of deaths attributed to gastric cancer was 782,685 [1]. As such, a greater understanding of the risk factors that play a role in gastric carcinogenesis can improve preventive and therapeutic interventions [2].
Epidemiologic studies showed that gastric cancer is multifactorial in aetiology [3,4], in which dietary factors and Helicobacter pylori infection may contribute to its development [3]. Of note is that a high prevalence of these risk factors do not always correspond to a high incidence of gastric cancer, suggesting that other susceptible factors such as genetic variations and environmental differences may be involved in gastric carcinogenesis [5]. For instance, it has appeared that only 1-2% of the infected will develop gastric cancer in their lifetime, albeit with more than half of the world's population being infected with H. pylori infection [6].
Gastric neoplasms are composed of cancer cells and other "non-cancer" compartments (including immune cells), which are the major players in gastric cancer disease progression and aggressiveness [7]. Tumour necrosis factor (TNF) is a pro-in ammatory cytokine, which is produced mainly by the immune cells such as macrophages, dendritic cells, lymphocytes and mast cells [8]. The tumour microenvironment, composed mainly of in ammatory cells, is a crucial player in the neoplastic process, fostering proliferation, survival and migration [9]. Thus, TNF-α may act as a tumour promoter through in ammation. Moreover, TNF-α can induce the transcription of a wide range of other proin ammatory cytokines and chemokines, amplifying the in ammatory cascade against the infection [2]. Individual studies reported the signi cant relationship between TNF-α -308 (rs 1800629) and the risk of gastric cancer [10,11]. However, other studies reported differently [12][13][14]. The published reviews in this eld [15][16], did not report the required information size leading to concerns over con dence in estimates. We have also found more individual studies on this. Taken together, the objective of this study was to summarize the evidence of association between TNF-α-308 and the risk of gastric cancer by meta-analysis of data from eligible studies.

Materials And Methods
This is a meta-analysis of genetic association studies (GAS), following the Plos One checklist (S 1 Table).

Study search
We searched relevant studies in the health-related databases of PubMed, Ovid Medline, Embase, Google scholar and Web of science. To maximize search scope, we used a simple term combination strategy: ("gastric cancer" OR "gastric carcinoma" OR "stomach cancer" OR gastric adenocarcinoma AND "tumour necrosis factor-alpha-308" OR "rs1800629" OR "TNF-α-308 G>A" OR "TNF-α-307 G>A"). The search strategy in PubMed database is provided (S2 Table). The search was limited to the publications in English until November 2020. We also manually searched the references of included studies and relevant systematic reviews to capture any additional studies.

Inclusion criteria
Human studies that assessed gastric cancer, irrespective of site or histological type were included, if they met all the following criteria: 3. The study recruited healthy people as the controls; 4. The study assessed histopathologically con rmed gastric cancer as an outcome; 5. The study reported the genotype frequency in cases and controls; 6. The study provided su cient data to compute odds ratio (OR) and its 95% con dence interval (CI) as the outcome measurement.
Gastric cancer is as de ned in the primary studies. Studies which did not meet the inclusion criteria were excluded. Studies done on family or sibling-pairs were also excluded. Studies on gastric adenoma are not considered.

Data extraction
One investigator (WST) screened the titles and abstracts and selected the relevant full-text articles, following the inclusion criteria. Two investigators extracted the data (NHH, WST) from each study individually, by using a pretested data extraction sheet. Information collected included: rst author, publication year, country, study setting, the number of cases/controls, ethnicity (Asian or Caucasians), method of genotyping and genotype/allele frequencies in cases/controls. If an allele frequency was zero in both case and control, we added 1 to that allele, following the Laplace approximation [17]. Any discrepancy between the two investigators was discussed with the third investigator (CN) to reach consensus.

Assessment of the methodology quality
The two investigators (WST, CN) independently evaluated the methodological quality of studies, using the Newcastle-Ottawa Scale (NOS) [18]. The assessment is based on the three domains such as 'selection of the study groups' (4 points), 'comparability of the groups' (2 points) and 'ascertainment of the exposure' (3 points). A total score for each study can vary from 0 (the worst) to 9 (the best). A score achieved ≥7, 6-5 or ≤4 is regarded as good, moderate or low quality study, respectively. Any discrepancy between the two investigators was resolved by consensus.

Statistical analysis
We assessed Hardy-Weinberg equilibrium (HWE) in the control populations of the included studies using the goodness-of-t test and p>0.05 was considered to indicate consistency with HWE [19]. As described elsewhere [20], the strength of the association between TNF-α-308 G>A and the risk of gastric cancer in each study was estimated using OR and its 95% CI. Between-study heterogeneity was determined with the I 2 test, which indicates the percentage of total variation across studies attributed to the heterogeneity rather than chance. I 2 values >50% is regarded as substantial heterogeneity [21]. For pooling of the estimates, the summary ORs and its 95% CIs were calculated with the random-effects model (The Der Simonian and Laird method) in the presence of substantial heterogeneity. Otherwise, we used the xed-effect model. We calculated the summary ORs and its 95% CIs in four genetic models: the allelic contrast model (A vs G), the dominant model (AA+GA vs GG), the recessive model (AA vs GA+GG), and the additive model (AA vs GG). In order to investigate the source of heterogeneity, subgroup analyses were employed under the dominant model for ethnicity, H. pylori infection status and study quality. Due to inconsistent reporting, we were unable to analyse by location or histological type of gastric cancer. For sensitivity analysis, we reassessed the relationship between TNF-a 308 and the risk of gastric cancer in all four genetic models only with studies in conformity of HWE. The publication bias was assessed by visual inspection of funnel plots under dominant model [22,23].

Trial sequential analysis
To estimate the required information size, we performed trial sequential analysis (TSA) [24]. It is classi ed as ' rm evidence of effect' or 'potentially spurious evidence of effect', depending on whether the cumulative Z-curve cross the monitoring boundaries or not [23]. Meta-analysis was done with RevMan 5.3 (The Cochrane collaboration, Copenhagen) and rworldmap package in R ® version 3.6.1 (The R Foundation). TSA plot was done with TSA software (Copenhagen Trial Unit, Centre for Clinical Intervention Research, Copenhagen).

Study search
Fig 1 illustrates a four-phase study selection process. The initial search yielded a total of 1335 records. After removing the duplicates and screening of abstracts, 45 potentially eligible full-text articles that were retrieved. We included a nal of 35 studies (with 11353 cases and 12827 controls) in this review [2,5,[10][11][12][13][14]. Summary of the 10 excluded studies were provided (S3 Table).
Study characteristics Table 1 shows the characteristics of 35 studies identi ed. Of these, slightly more than half (54.3%, 19/35) were from the Asian region. The years of publication spanned from 2001 to 2017. The participants were adults with male predominance in all these studies. Eighty percent of the studies were consistent with HWE in genotype distribution of the controls.
The most frequent ve studies were performed in China, South Korea or Brazil. Fig 2 shows (24/35. 68.6%) included H. pylori infected gastric cancer cases, albeit with variation in distribution. For instance, all cases (100%) were infected with H. pylori in one study [32], while this was only 46% in another study [2] (Table 1).

Effect estimations
The genotype frequencies in individual studies are presented in Table 2 (Fig 3A, Fig 3B, Fig 3C and Fig 3D).

Subgroup analyses
Under dominant model, TNF-α-308 G>A has signi cantly associated with an increased risk of gastric cancer with the high or moderate quality studies, but not with poor quality studies (S1 Fig). Based on 24 studies in which patients were infected with H. pylori, a signi cant association was observed between TNF-α-308 G>A and gastric cancer risk (OR, 1.17, 95% CI, 1.05-1.29). On further strati cation, this was only with the Caucasian group (OR, 1.17, 95% CI, 1.04-1.32) ( Table 3).
A sensitivity analysis based on 28 studies that were consistent with HWE, showed TNF-α-308 G>A polymorphism was signi cantly associated with an increased risk of gastric cancer under the dominant model in overall analysis (OR, 1.19, 95%CI:1.1-1.29, I 2 :37%), regardless of ethnic groups. Moreover, this association showed a decreased statistical heterogeneity (i.e. I 2 values from 69% to 37%) ( Table 3,

TSA plot
We performed TSA of the dominant model with the use of an overall type I error of 5% and type II error of 20%. The included total participants in this meta-analysis reached the required information size (for an expected RRR 26%). Brie y, a TSA monitoring boundary crossed with Z curve, con rms the presence of robust evidence (Fig 4). In such case further studies are not needed to provide su cient information.

Summary information
The present study provides evidence on the relationship between TNF-α-308 G>A and the risk of gastric cancer, comprising 11353 cases and 12827 controls from 35 individual studies. The major observations are as follows; 1. Based on 28 studies that met HWE, TNF-α-308 G>A SNP was signi cantly associated with the gastric cancer risk under the dominant and addictive models.
2. On strati cation, the HWE status of the controls, ethnicity, pylori infection status or study quality had an impact on the effect estimates.
3. The TSA plot revealed that the required information size for evidence of effect was su cient. Any future studies in this eld will less likely change the direction of estimates.
The association was statistically signi cant only for the Caucasians in overall allele model, indicating a dominance of racial speci c factors. This difference may be explained partly due to variations in the frequency of the A allele between the different ethnic groups that could contribute to the diverse results. Moreover, it might also be related to difference in environmental factors such as smoking and diet between these two major ethnic groups. This was indirectly supported by an individual study in Poland, in which 72% of the gastric cancer cases were smokers (ex-smokers or current smokers) [39] as the effects of inflammatory polymorphisms might have been masked by smoking [35]. Due to paucity of data, we were not able to perform subgroup analysis with the smoking status of participants included in the studies identi ed. A published meta-analyses [53] reported that H. pylori infected cases had higher risk of developing gastric cancer. This was also observed in the present review. This could be explained based on immune-biological plausibility. H. pylori infection activates the cytokines production in the lining of the stomach including in ammatory-related genes such as TNF-a in the present analysis.
Our ndings were comparable with earlier reviews, in which the signi cant association was limited to the Caucasians [53,54] in the dominant models [11,54]. Although there are more studies in this analysis, the results in general, retained the evidence of association. The most commonly studied in ammatoryrelated genes in gastric diseases include TNF-α, among others. TNF-a has been shown to inhibit the gastric acid secretion which is important in inducing cell apoptosis and promoting epithelial cell damage [57].

Public health Implications
The difference in association between the ethnic groups observed in the current analysis has implications. Studies had reported that the regulation of tumor immunity factors at the genetic and gene expression level may be different in the Asian and non-Asian gastric cancer populations, and this can affect the region-speci c effects on therapy outcome and prognosis [58].

Study limitations
We acknowledge the study limitations. Only 36% of the studies in this review used TaqMan SNP genotyping assays, which is the preferred technology due to its high throughput and high accurate [59] compared to other methods. Hence, accuracy of genotype frequency is a concern. Some studies included were with small sample sizes. Hence, there might be type II statistical error. Meta-analysis is a retrospective pooling of published studies, and type II errors are, therefore, less likely than in individual studies.
There might be other confounding factors that were not included in our subgroup analyses. For example, infections with carcinogenic potentials (e.g. EBV) or smoking and alcohol drinking of the participants were not addressed due to limited data. This bias was likely to be pronounced as the calculations used unadjusted assessment of ORs. The effects of inflammatory polymorphisms might have been masked by smoking [35]. Moreover, it is likely to miss relevant studies that are available in non-English or non-indexed databases. Furthermore, there might be interactions of TNF-α and other genes such as interleukins (gene-gene interaction/synergism) or other potential confounding factors such as nutritional status and life style of the patients that might have signi cant roles in the gastric cancer risk. Due to limited number of studies, we could not perform pooled analysis with these potential confounding factors. Hence, ndings in this meta-analysis should be interpreted with caution to these factors.
Nevertheless, there are strengths in our present meta-analysis study. More than half of the included studies were carried out in the Asia region, in which the gastric cancer was more prevalent. Moreover, the majority of gastric cancer patients in the primary studies identi ed were males. The current analysis re ected geographical and gender representativeness. The vast majority of included studies had evidence of HWE. Numerous studies had highlighted the issue of deviations from HWE in genetic association studies such as genotyping error, population admixture/substructure, among others [60][61][62]. Furthermore, for robustness of the ndings, we have attempted several subgroup analyses. There are strengths in this meta-analysis compared with published reviews in this eld [11,53,54,[60][61][62][63]. To be comprehensive, we have attempted the TSA technique, which is useful to adjust random-error risk. Moreover, we introduced TSA for con rmation of the estimates to assess a required information size. TSA plots indicated that there was su cient information to provide conclusive results. An add-on TSA approach to this eld will highlight to researchers the optimal sample size to make judgement of the effect estimates. This will help the researchers and policy makers to determine the need for future similar studies, which can save limited resources.

Conclusions
The current ndings suggest that TNF-α-308 gene polymorphism plays an important role as host genetic factor predisposing to gastric carcinogenesis, and it could be useful as a screening marker. As the relationship of gastric cancer risk is ethnic speci c, the consideration as a biomarker should be tailored to the speci c population group. To substantiate this, studies only from the Asian regions, using more reliable genotyping technique are recommended.   Trial sequential monitoring plot of TNF-α-308 in gastric cancer risk under dominant model