A trial sequential meta-analysis of TNF-α –308G>A (rs800629) gene polymorphism and susceptibility to colorectal cancer

Purpose: Tumor necrosis factor-α (TNF-α), secreted by the activated macrophages, may participate in the onset and progression of colorectal cancer (CRC). The association of TNF-α –308 G>A (rs1800629) single-nucleotide polymorphism (SNP) with CRC risk has been investigated by many studies but the results are inconclusive. A trial sequential meta-analysis was performed for precise estimation of the relationship between TNF-α –308 G>A gene polymorphism with CRC risk. Methods: Medline (PubMed), EMBASE (Excerpta-Medica) and Google Scholar were mined for relevant articles. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated to estimate the significance of association. Results: The pooled analysis indicated no risk associated with TNF-α –308 G>A SNP and overall CRC risk in five genetic comparison models, i.e. allelic (A vs. G: P = 0.524; OR = 1.074, 95% CI = 0.863–1.335), homozygous (AA vs. GG: P = 0.489; OR = 1.227, 95% CI = 0.688–2.188), heterozygous (AG vs. GG: P = 0.811; OR = 1.024, 95% CI = 0.843–1.244), dominant (AA+AG vs. GG: P = 0.630; OR = 1.055, 95% CI = 0.849–1.311) and recessive (AA vs. AG+GG: P = 0.549; OR = 1.181, 95% CI = 0.686–2.033). Subgroup analysis revealed that TNF-α –308 G>A SNP is associated with reduced risk of CRC in Asian ethnicity. The study showed no publication bias. Conclusions: No association of TNF-α –308 G>A SNP with overall CRC risk was found. This SNP is likely to be protective against CRC in Asian population when compared with Caucasian population. Larger prospective-epidemiological studies are warranted to elucidate the roles of TNF-α –308 G>A SNP in the etiology of CRC and to endorse the present findings.


Introduction
Approximately 608,000 people lose their life to colorectal cancer (CRC) worldwide [1]. According to the World Health Organization (WHO), millions of people will suffer from symptomatic as well as approximately the equal number from asymptomatic cases of CRC disease in the next decade.
CRC is a very heterogeneous and polygenic disease at a molecular level. It may be the result of interaction among different factors like environmental and genetic [2]. Early genome-wide association studies have shown contribution of many new single-nucleotide polymorphisms (SNPs) to increased

Search for relevant literature
An online search was done on different databases like PubMed (Medline), Google Scholar and EMBASE covering all research studies published. The search strings used to retrieve the hits were: Tumor necrosis factor OR tumor necrosis factor-alpha OR TNFA OR TNFα OR TNF-α OR TNF gene (polymorphism OR variant OR mutation) AND colorectal cancer susceptibility OR risk (last updated on February 2018). The relevant studies about genetic association were extracted after perusing their titles and abstracts. The publications suiting the above discussed preset eligibility criteria were considered for further examination. The references of the retrieved reports were also searched for additional relevant reports.

Criteria for inclusion and exclusion of studies
To keep the heterogeneity in check and right interpretation of the study, following criteria were followed to include the published reports in current meta-analysis: (a) only case-control studies assessing association between TNF-α -308 G>A gene polymorphism and CRC risk, (b) the study must have recruited clearly defined and confirmed CRC patients and CRC free controls, (d) genotype frequency in cases and the controls should be reported, (e) language of these studies should be English and (f) should have used statistically relevant data collection and analysis methods. Additionally, if the case-control studies derived the cases from the same population, the study having larger number

Data extraction
The quality of the data extracted was assessed by two investigators (R.K.M. and M.A.K.) individually following a standard protocol. Preset inclusion/exclusion as well as the sequential exclusion criteria of the unsuitable studies outlined in the data-collection form was strictly adhered to ensure the accuracy of the collected data. Any disagreement between the investigators about the quality of collected data was first subjected to a consensus and then finally settled with an open discussion with the arbitrator (S.H.). The data extracted from the retrieved publications consisted first author name, the country of origin, year of publication, number and source of cases and controls, type of study type, genotype frequencies and association with CRC.

Quality assessment using Newcastle-Ottawa Scale
Quality assessment of the selected studies was done independently by two investigators, namely A.H. and N.A. This evaluation was done by following the Newcastle-Ottawa Scale (NOS) of quality assessment [31]. The major aspects used for NOS quality assessment criteria were: (a) selection of subjects: 0-4 points, (b) subject comparability: 0-2 points and (c) clinical outcome: 0-3 points. The extracted case-control studies securing 5 or more stars were considered having moderate to good quality [31,32]. In case, if any difference occurred on any item between the above two investigators, the issue was fully discussed and solved by a detailed discussion in the presence of third investigator (S.A.D.) participated as adjudicator.

Statistical analysis
Pooled ORs and their corresponding 95% CIs were used to appraise the risk association between the TNF-α -308 G>A gene polymorphism and susceptibility to CRC. Heterogeneity was assessed using the chi-square-based Q-test and was considered significant if the P-value was less than 0.05 [33]. The collected data from single comparison was calculated using a fixed effects model [34], in case of no heterogeneity. However, the random-effects model [35] was employed for pooling of the data. Further, I 2 statistics used to estimate the interstudy variability in which larger values showed a higher degree of heterogeneity [36]. Hardy-Weinberg equilibrium (HWE) in the controls was calculated using chi-square test. Whereas Egger's regression test showing the funnel plot asymmetry was used to measure significance of publication bias, if any. Further to this, ethnicity was adopted to perform the subgroup stratified analysis, when data were available. The Comprehensive Meta-Analysis (CMA) software program Version 2.0 from Biostat (NJ), U.S.A. was selected to conduct all the statistical calculations involved in the present meta-analysis.

Trial sequential analysis
According to the Cochrane handbook, the meta-analyses are acceptable if it includes all the eligible trials. However, it may lack sufficient evidences. The meta-analysis may contain systematic errors (bias) or random errors (play of chance), which can be reduced using novel statistical analysis program named 'Trial Sequential Analysis' (TSA) tool, made by Copenhagen Trial Unit, Center for Clinical Intervention Research, Denmark). TSA calculates required information size as well as adjusts the threshold for the statistical significance and finally calculates the robustness of present conclusion [37][38][39]. Briefly, a TSA monitoring boundary crossed with Z curve confirms the presence of robust evidence. In such case further trials are not needed. However, Z curve not crossing the monitoring boundaries suggest that the trial should continue. Trial Sequential Analysis (version 0.9, http://www.ctu.dk/tsa/) was used in current study.

Literature search
Two investigators (viz. R.K.M. and M.A.K.) individually examined every title and abstract of the retrieved studies using the designated online web-databases search in a sequential order. The full-text of each study apposite for the inclusion was also recovered. To evaluate the aptness of the study for the inclusion in this pooled analysis, one researcher (R.K.M.) systematically examined all the full-text retrieved publications. Afterwards, the second researcher (M.A.K.) performed the same procedure of text evaluation independently by selecting randomly 10% of the full-text articles. During the study selection process, complete agreement was found between the above stated two researchers regarding the study exclusion and selection criteria. After the selection of the final set of the eligible studies, another researcher (S.A.D.) extrapolated the pertinent data from all the included studies. This step of data extrapolation was

Properties of the reports included in the present study
Fifteen articles were selected after systematic literature search done on PubMed, EMBASE and Google Scholar. The retrieved texts were scrutinized by perusing complete texts for their potential relevance for the current meta-analysis ( Figure 1). Reports showing TNF-α gene polymorphism to estimate survival in CRC patients or using CRC variants as indicators for prognosis were excluded at the onset. Likewise, the reports analyzing TNF mRNA levels or subsequent protein expression were also excluded. The studies with case-control or cohort that design only reporting frequency of all the three genotypes were included for the current meta-analysis. Additionally, all the references cited in the retrieved articles were also scanned to identify other potential case-control studies. Finally, the 15 original publications were found eligible after applying the stringent inclusion and exclusion criteria ( Table 1). The genotypes distribution, P-values of HWE and susceptibility towards colorectal risk have been given in Table 2. The selected studies (15 in number) were examined for the overall quality following the NOS. Maximum number of the studies included (>80%) scored 5 stars or more, showing a modest to decent quality ( Table 3).

Assessment of publication bias
Begg's funnel plot and Egger's test showed no publication bias among all the comparison models (Table 4) in every genetic model and the allelic contrast ( Figure 2).

Test of heterogeneity
The chi-squared-based Q-test and I 2 statistics showed the substantial amount of heterogeneity in all the genetic models leading to the use of random-effects model to process the data (Table 4).

Quantitative synthesis
All the 15 studies pooled together amounted to 3116 confirmed CRC cases and 4480 healthy controls for the evaluation of overall association between the TNF-α -308 G>A SNP and CRC risk. The overall ORs showed no statistically significant association with high or low risk between TNF-α -308 G>A gene polymorphism and CRC risk in neither genetic models (A vs. G: P = 0.524; OR = 1.074, 95% CI = 0.863-1.335), homozygous (AA vs. GG: P = 0.489; OR =        (Figure 3).

Sensitivity analysis
Sensitivity analysis used to evaluate the impact of each individual study on the pooled ORs revealed no significant influence on the pooled OR by any individual study (Figure 4).

Subgroup analysis: association of the TNF-α -308 G>A SNP and risk of CRC in Caucasian and Asian population
A stratified subgroup analysis based on the ethnicity of the enrolled subjects was performed to explore the effect of ethnicity (Caucasian and Asian) on the association of TNF-α -308 G>A SNP and the risk of CRC onset.

Subgroup analysis of Caucasian population
Nine case-control studies contain 2130 controls and 1846 cases. These controls and cases were included for subgroup analysis of Caucasian population. The analysis showed significant heterogeneity in three genetic models (Table 5) (Supplementary Figure S1). The conducted analyses using random and fixed models observed no significant association of CRC susceptibility in all the genetic models, i.e. allele model

Subgroup analysis of Asian population
Like Caucasian population, six studies having 2329 controls and 1284 cases were included in the subgroup analysis of Asian population. The analysis showed no publication bias but heterogeneity was observed in two genetic models (Table 6

Trial sequential analysis of TNF-α -308 G>A SNP and risk of CRC
TSA (taking the data of the dominant model) was used for the current analysis to check if further trials are required ( Figure 7A). Same result was observed after the subgroup analysis based on the Caucasian ( Figure 7B) and Asian ( Figure 7C) population.

Discussion
The SNP analysis is useful in genomic DNA screening. It is specifically important in case of CRC because these markers are not affected by disease activity and remain unchanged over time. Research probing genetic associations are powerful methods for identifying low penetrance susceptibility genes that can affect biological process and provide linkage analysis when investigating complex disease like CRC. Translating this information into routinely applied   diagnostics would assist in better understanding of the CRC etiology, possibly leading to novel and better clinical practice with many benefits for the patient. Inflammatory response harmonizes host response against infection participating in repair of tissue, in case of tissue damage. A chronic inflammation many a time alters the immune system and leads to carcinogenesis [40]. Over the time, the cytokines are receiving overdue attention due to their property of mediating and regulating the immune response including inflammation. The cytokines may regulate the pro-inflammatory and anti-inflammatory network to stimulate signaling pathways involved in malignancy development [41].
CRC onset and progression is linked with innate immune processes and inflammation in intestine. Pro-inflammatory genes have key role in maintenance and growth of CRC [42]. TNF-α is a pro-inflammatory cytokine made by activated immune cells leading to suppression of tumor proliferation [43]. TNF-α also activates antitumor the natural killer cells and CD8 T cells [44].
CRC onset and tumor progression are preceded by inflammation. This has developed a curiosity in scientific community to understand the molecular signaling pathways that connect TNF-α with the development and survival of CRC tumor cells. The expression of TNF-α like other cytokines is tightly regulated at the transcriptional level and also at post-transcriptional level. The -308 G>A SNP is located inside the regulatory regions of TNF-α gene. TNF-α expression and secretion both are influenced by this polymorphism.
Many scientists have published their works on CRC, but the molecular and biological mechanism of relationship between TNF-α gene polymorphism and risk of CRC is not completely understood. Till date, the reports of TNF-α (-308 G>A) SNP in relation with CRC risk lack consensus. Many clinical case-control studies have reported both positive as well as negative association of CRC and TNF-α (-308 G>A) SNP. Hence, larger sample size with pooled and subgroup analysis is demand of the time to evaluate the potential role of TNF-α -308 G>A polymorphism as a genetic risk factor for CRC infection. The combined ORs from many early reports that lead to large sample size with required statistical robustness also lower the random errors [45]. The meta-analyses address a wide variety of clinical problems using early published data. This meta-analysis included 15 eligible case-control studies comprising 3116 cases and 4480 healthy controls and analyzed the pooled ORs and P-value to appraise the precise relationship between the TNF-α -308 G>A SNP and CRC risk. NOS quality assessment showed nearly every study scoring five or more than five stars suggesting good to moderate quality of extracted data. The current meta-analysis shows no link between the TNF-α -308 G>A SNP and CRC susceptibility by any genetic model in overall population analysis. The speculations based upon the findings tell that numerous polymorphic sites present in the promoter and the coding regions of TNF-α gene might serve to keep this gene under tight control and influence the expression of TNF-α. Hence, the haplotype combinations might be conserved in certain population to protect against pathogens. Furthermore, the gene reporter assay also testified that A allele of -308 polymorphism does not influence TNF-α gene transcription [46]. Many early reports show that -308 G>A polymorphism leads to different transcription rate in TNF-α production [47,48].
A stratification analysis of ethnicity was performed considering the fact that polymorphism frequencies might differ among ethnic groups. The separate race-specific meta-analysis done in Caucasian and Asian populations show that TNF-α -308G>A SNP is protective against the CRC risk in Asian but not in Caucasians population. The increased plasma concentration of TNF-α, which is a result of single TNF-α -308 A allele, might not influence CRC risk. The most likely reason might be the population stratification within involved studies, especially when both allelic frequencies and incidence of disease vary across ethnic groups. The range of A allele frequency is from 2 to 9% in Asians, 8 to 10% in South Americans and 10 to 23% in Europeans.
An early meta-analysis by Min et al. showed increased risk of CRC [49]. But, they found increased risk with only homozygous model and less significant risk under heterozygote model in overall population. Furthermore, they have not provided the frequencies of each genotype of included studies and any risk involved in the subgroup analysis. After applying the stern inclusion criteria, the current meta-analysis was added new studies that led to increase in the number of included subjects in both CRC and controls. The analysis was also stratified along race namely Asian and Caucasian. The results of the current analysis tend to be more precise in estimating the relationship between the TNF-α -308 G>A SNP and risk of CRC in overall population as well as ethnicity than previous ones.
The previous findings suggested that susceptibility towards CRC is polygenic in nature that indicates the possibility that many genes are participate in determining the resistance or susceptibility to CRC. Consequently, the complex nature of CRC and multifaceted nature of the immune system, TNF-α -308 G>A SNP is not the sole reason for the predisposition to CRC, but this polymorphism may interact with other polymorphisms present in linkage disequilibrium of this gene to cause risk.
Despite being many advantages of the present aforementioned study, some limitations must also be mentioned, namely: first, interstudy heterogeneity was observed in the overall comparison from each genetic model. We minimized the likelihood of this problem by performing data analysis by using random-effects model. When studies were stratified by ethnicity, low heterogeneity was found in both Caucasian and Asian populations. Hence, we considered that the racial differences and ethnic origin of the study population, inadequate selection criteria of the subjects, and small sample size of each included study might be responsible for the foremost source of heterogeneity. Second, the present study included the reports published in the English language only. Further, PubMed-Medline, EMBASE and Google Scholar electronic databases were used for the study and subsequent data retrieval. It is possible that some relevant studies published in language other than English or indexed on different databases, are not included. Third, the present study did not have any information on gene-environment interactions due to inadequate data available on this matter. Fourth, the calculations used unadjusted assessment of ORs, which may influence the results.
Despite above limitations, the present study does have many strengths: First, it has included large number of subjects that gave powerful evidence to reach on precise and robust conclusion. It would greatly improve the understanding on the role TNF-α -308 G>A SNP in CRC pathogenesis. Second, absence of publication bias and sensitivity analysis suggests the reliability of the results and whole study. Furthermore, all the included studies were of good to modest quality to fulfill the preset needful criteria required by NOS quality assessment.

Conclusions
The current meta-analysis indicates that TNF-α -308 G>A SNP has no role in CRC progression. It also suggests that individuals with TNF-α -308 G>A genetic variant have comparatively less CRC risk among Asians. It is a suggestive limited piece of evidence that -308 A allele might reduce CRC risk. As, TNF-α plays a significant role in immune response, further larger case-control studies are warranted to make the conclusions more comprehensive. Taken as a whole, the present study would greatly help the scientists in understanding the relationship of TNF-α -308 G>A SNP and CRC risk across the world.