Genetic association between TNF-α promoter polymorphism and susceptibility to squamous cell carcinoma, basal cell carcinoma, and melanoma: A meta-analysis

Tumor necrosis factor-alpha (TNF-α) is a multifunctional pro-inflammatory cytokine that plays an important role in cancer development. We performed a metaanalysis to assess the relationship between single nucleotide polymorphisms in the TNF-α promoter region (rs1800629 and rs361525) and susceptibility to squamous cell carcinoma (SCC), basal cell carcinoma (BCC) and melanoma. After database retrieval, article selection, data extraction, and quality assessment, 20 articles comprising 4865 cases and 6329 controls were included in this study. rs1800629 was associated with an increased overall risk of SCC, lung SCC, and oral SCC in the AA vs G and AA vs GG+GA genetic models (all OR>1, Passociation<0.05). No increased risk of skin SCC, skin BCC or melanoma was observed (all Passociation>0.05). Rs361525 was not associated with overall SCC risk in the allele, heterozygote, dominant, recessive, or carrier model (all Passociation>0.05). Begg’s and Egger’s tests (PBegg>0.05; PEgger>0.05) demonstrated there was no significant publication bias. These data indicate that the AA genotype of TNF-α rs1800629, but not rs361525, is associated with an increased risk of SCC, suggesting it could potentially serve as a prognostic marker for predicting SCC risk.

The role of TNF-α gene mutations in the risk of squamous cell carcinoma (SCC) remains inconclusive. For instance, the rs1800629 polymorphism of TNF-α gene has been linked to the risk of esophageal SCC in northern Indian patients [5], but not in Kazakh patients [6]. TNF-α rs1800629 polymorphism has been associated with the risks of oral SCC in Taiwan [7], but not in northern Indian population, which has been linked with rs361525 polymorphism [8]. There was also no association between the rs1800629 polymorphism and lung SCC risk in the German population [9].
Skin cancer comprises cutaneous melanoma, skin SCC (SSCC), and skin basal cell carcinoma (SBCC) [10]. Allelic variants of TNF-α gene have been reported to contribute to the risk of skin cancer in certain populations. For example, the study by Rizzato et al. has indicated that TNF-α rs1800629 might affect the SBCC risk in Caucasian population [11]. The A allele or GA genotype of TNF-α gene rs1800629 polymorphism was also reported to influence the course of BCC in Polish population [12]. However, the role of TNF-α polymorphisms in skin cancer is still inconclusive. For example, Skov et al. reported that TNF-α release, but not rs1800629 polymorphism, was linked to the SBCC risk in Caucasian population [13]. To our knowledge, no metaanalysis has been previously performed to assess the link between TNF-α polymorphisms and the risk of skin cancer. Meta-analysis www.impactjournals.com/oncotarget Therefore, in this study, we carried out a comprehensive systematic review and meta-analysis to determine the association of TNF-α polymorphisms and the risk of skin cancer and different SCC diseases.

Characteristics of studies included in metaanalysis
Six databases, including PUBMED, Web of Science (WOS), EMBASE, WANFANG, CNKI, and SCOPUS, were electronically searched on January 17 th , 2017 to identify the eligible studies. The search details are shown in Supplementary Table 1. Flowchart of the search strategy and article selection for meta-analysis is shown in Figure 1. Briefly, 985 related articles were obtained from the above databases. After 241 duplicated articles were removed, 699 articles were excluded by screening the title and abstract. The eligibility of 45 full-text articles was then assessed, and 25 articles were excluded. The results are shown in Supplementary Table 2. Finally, 20 eligible articles with 4865 cases and 6329 controls were included for quantitative synthesis [1,[5][6][7][8][9][11][12][13][14][15][16][17][18][19][20][21][22][23][24]. All selected articles met the inclusion and exclusion criteria. We used the Newcastle-Ottawa Scale (NOS) to assess the quality of the studies. As shown in Supplementary Table 3, the NOS scores of all studies were equal to or greater than 7, indicating a high quality. After covariate adjustment in logistic regression, the characteristics and genotype distributions of included studies are shown in Tables 1 and 2. Association between TNF-α rs1800629 polymorphism and the risk of SCC Meta-analysis of 16 studies [1, 5-9, 14, 16-24] comprising 2836 cases and 5235 controls was performed to analyze the association between TNF-α rs1800629 polymorphism and the risk of SCC under allele model (A vs G), homozygote model (AA vs GG), heterozygote model (GA vs GG), dominant model (GA+AA vs GG), recessive model (AA vs GG+GA), and carrier model (carrier A vs G). Pooled analysis data are shown in Table 3 There are several types of SCC, including skin SCC (SSCC), esophageal SCC (ESCC), oral SCC (OSCC), and lung SCC (LSCC). We performed subgroup analyses of the above SCC types and different ethnicities under all models. As shown in Table 3 and Figure 2A, an increased overall SCC risk was observed in the Asian population under AA vs GG model (OR=3.67, 95% CI=1.89~7.16, P association <0.001). The increased risk of LSCC (OR=2.72, 95% CI=1.32~5.61, P association =0.007) and OSCC (OR=3.91, 95% CI=1.38~11.05, P association =0.010) was also observed (Table 3 and Figure 2B). Similar results were observed for the AA vs GG+GA model (Table 3 and Figure 3). No significant difference was found under other genetic models (all P association >0.05). These data indicate that the AA genotype of TNF-α rs1800629 polymorphism correlates with the higher susceptibility towards SCC.

Association between TNF-α rs361525 polymorphism and the risk of SCC
Meta-analysis of the relationship between TNF-α rs361525 polymorphism and susceptibility to SCC was also performed. Six studies comprising 962 cases      and 1116 controls were analyzed [7,8,14,20,21,24]. Data of pooled analysis indicated that there was no significant difference for overall SCC risk under all genetic models (Table 4 and Supplementary Figure 1; all P association >0.05). Subgroup analysis (

Association between TNF-α rs1800629 polymorphism and the risk of skin cancer
We then performed meta-analysis of the relationship between TNF-α rs1800629 and the risk of skin cancer, including SSCC, SBCC, and melanoma. Seven studies comprising 2710 cases and 2786 controls were included [11][12][13][15][16][17]. Data of pooled analysis indicated no significant difference under all genetic models (all P association >0.05, Table 5 and Supplementary Figure 2). Subgroup analysis (based PB and SBCC) also showed no significant difference (Table 5). However, only one case-control study was included in the subgroup analysis of melanoma [15] and SSCC [16] (Table 5). These data suggest that TNF-α rs1800629 polymorphism does not have a significant correlation with the risk of skin cancer.

Heterogeneity, publication bias and sensitivity analysis
Regarding the rs1800629 polymorphism and SCC risk, A vs G (I 2 value of 77.5 % and P heterogeneity <0.001), GA vs GG (I 2 =66.3 % and P heterogeneity <0.001), GA+AA vs GG (I 2 =71.4 % and P heterogeneity <0.001) and carrier A vs G (I 2 =64.0 % and P heterogeneity <0.001) data indicated a high degree of heterogeneity among the studies (Table 6). Thus, random-effect model was applied. In addition, fixed model was used in AA vs GG (I 2 =30.0 % and P heterogeneity =0.137) and AA vs GG+GA contrast (I 2 =23.2 % and P heterogeneity =0.203, Table 6).
For the rs361525 polymorphism and SCC risk, random-effect model was used for the overall SCC, due to the presence of overall significant heterogeneity  We also performed Begg's and Egger's tests to evaluate the potential publication bias among the included articles. The results indicate that publication bias can be ruled out for all comparisons (Table 6 and

DISCUSSION
In the present study, 16 case-control studies of TNF-α rs1800629 polymorphism [1, 5-9, 14, 16-24] and 6 case-control studies of rs361525 polymorphism [7,8,14,20,21,24] were included in the meta-analysis of TNF-α polymorphism and the risk of SCC disease. We found that an increased overall SCC risk was associated with the rs1800629 polymorphisms in the Asian population under AA vs GG, and AA vs GG+GA models, but not A vs G, GA vs GG, GA+AA vs GG, or carrier A vs G models. A significant difference between LSCC/OSCC risks and the rs1800629 polymorphism was found under the AA vs GG, and AA vs GG+GA models; this corresponds with previous data on the link of rs1800629 and the risk of upper aerodigestive tract or head/neck SCC [25,26]. However, in 2013, Chen et al performed a meta-analysis to analyze the association between rs1800629 and oral cancer, and observed a negative association between rs1800629 and OSCC [27]. Different selection criteria may contribute to this discrepancy. In our meta-analysis, two studies were excluded due to the requirement of Hardy-Weinberg equilibrium (HWE) or genotype data [28,29]. Regarding the ESCC risk and rs1800629 polymorphism, the negative result was found under all genetic models, which was in line with the data of Luo et al [30]. The rs361525 allele was reported to be significantly increased in healthy controls compared with cancer patients, indicating a protective function [31]. Here, no significant difference was detected for rs361525 and overall SCC risks under all genetic models, which was partly in accordance with the data of Gao et al regarding head and neck SCC [26] and Zhou et al for overall cancer [32]. In addition, seven case-control studies in Caucasian population were included for the analysis of skin cancer [11][12][13][15][16][17]. We failed to observe a significant association between TNF-α rs1800629 and skin cancer. In 2011, Nan et al also did not find any association between TNF-α gene variants and skin BCC or SCC in the Genome-Wide Association Studies (GWAS) from 2045 cases and 6013 controls of European population [33]. Although our results were validated by Begg's and Egger's tests, and by sensitivity analysis, the limitations in our meta-analysis should also be addressed. (1) Due to the limited number of studies published to date, only the common genetic polymorphisms of TNF-α, including rs1800629 and rs361525, were chosen. In addition, small sample size and/or limited genotype data in eligible articles affected our analysis. For example, there are two case-control studies of the association between TNF-α rs1800629 and melanoma risk [15,34]. However, one study was excluded due to the departure of HWE [34]. The frequency data of GA+AA combined genotype and GG genotype were extracted in one OSCC study [24]. (2) A considerable heterogeneity was observed in the meta-analysis of rs1800629/rs361525 and the SSC risks. SCC has many different etiologies, and stratified analyses by every SCC disease type were not performed. The variations of clinical characteristics, ethnicity, geographical location, habits, gender, age and population feature were not fully considered. In spite of the use of random-effect model, a limited number of studies was included in the subgroup analysis. For example, only one case-control study was included for the rs1800629 and the susceptibility to a specific SCC disease, including SSCC [16] and OPSCC [21]. The subgroup analysis of LSCC and rs361525 was based on only 2 case-control studies [14,20], and showed a positive correlation under A vs G model, GA vs GG model; GA+AA vs GG, and carrier A vs G model. It is possible that the GA genotype of rs361525 is associated with the decreased risk of LSCC. However, well-powered studies and stratified analyses by more factors are required to confirm our findings.
TNF-α is an important multifunctional proinflammatory cytokine, which is closely linked to the occurrence, progression, metastasis, prevention and therapy of many types of human cancer [35][36][37]. Alterations in TNF-α gene expression or TNF-α cytokine release lead to a variety of cancers [2,38]. Genetic variation has been considered as a disease susceptibility or resistance factor [2]. The rs1800629 G/A polymorphism, located in the promoter region (-308 site) of human TNF-α gene, can lead to the substitution from G common allele to A rare allele [2]. In vitro experiments showed that the "A" rare allele of rs1800629 could increase TNF-α transcription [39,40]. The frequency of "A" allele also positively correlates with high TNF-α levels in patients with oral cancer [28]. TNF-α rs1800629 was found to be positively associated with distant metastases of triple negative breast cancer patients [36]. However, no association was found between rs1800629 and TNF-α gene expression in gastric cancer patients [41,42]. Here, we observed a positive correlation between the AA genotype of rs1800629 and the risks of LSCC/OSCC. However, we did not find any significant association between the A allele and the SCC risks. It is possible that the "A" rare allele functions in an allele dosage-dependent manner. TNF-α was found to increase the efficiency of chemotherapy and radiotherapy against breast cancer cells [43]. The carriage of the A rare allele of rs1800629 may be involved in this process, through inducing TNF-α transcription and protein expression. It may be meaningful to analyze the effect of combined mutations of TNF-α and other genes, including TNF-beta and interleukin-6, on the carcinogenesis and SCC cancer therapy, since this may lead to the discovery of potential novel biomarkers for SCC.
In conclusion, our meta-analysis indicates that the AA genotype of TNF-α rs1800629 polymorphism may serve as a prognostic biomarker for SCC, especially for LSCC and OSCC in the Asian population. The rs361525 polymorphism does not seem to be a genetic risk factor for SCC. In conjunction with other studies, these results provide a scientific support for the prognostic value of TNF-α rs1800629 polymorphisms in predicting the SCC risk.

Database retrieval
The related articles published before January 17 th , 2017 were searched in the electronic databases, including PUBMED, WOS, EMBASE, WANFANG, CNKI and SCOPUS, without any language restrictions. The present meta-analysis followed "Preferred Reporting Items for Systematic Reviews and Meta-Analyses" (PRISMA) [44], as shown in Supplementary Table 4.

Article selection
Duplicated articles were removed by EndNote X7 software (Thomson Reuters). The following articles were excluded: 1) Reviews, theses, cases and trials; 2) Meeting abstracts and posters; 3) Cell or animal studies; 4) Other genes or diseases; 5) Meta-analyses; 6) Nonmutation data; 7) Data without detailed genotype; 8) Lack of control data; 9) Studies without specific oral cancer type information; 10) P value for HWE (P HWE ) was less than 0.05; 11) Studies with unselected mutation sites. The selected articles provide sufficient information regarding the genotypes for TNF-α polymorphisms in case and control groups. P HWE values were obtained by the chisquared test.

Data extraction and quality assessment
The authors extracted independently the following information: First author, publication year, country, ethnicity, number of cases/controls, source of controls, age (mean value), genotyping assay, gender (male %), SNP, genotype frequencies, disease type, χ 2 and P HWE values in control group. Newcastle-Ottawa Scale (NOS) system (http://www.ohri.ca/programs/clinical_epidemiology/ oxford.asp) was used to assess the quality of the included studies; NOS score ≥7 indicates a high quality study.

Statistical analyses
Mantel-Haenszel statistics was used to estimate the values of pooled odd radios (ORs) and 95 % confidence intervals (CIs); P association value less than 0.05 was considered statistically significant.Six genetic models, including allele, homozygote, heterozygote, dominant recessive, or carrier models were used. Cochran Q statistic and I 2 test were carried out to assess the potential heterogeneities between studies. When P heterogeneity value of Cochran Q statistic > 0.05 or I 2 value <50 %, the fixedeffect model was used. Otherwise, random-effect model was applied. To investigate the potential sources of heterogeneity, sensitivity analyses and subgroup analyses based on SCC disease type, ethnicity or source of controls were performed. Begg's test with pseudo 95 % confidence limits and Egger's test were also conducted to evaluate the potential publication bias. Stata/SE 12.0 (College Station, TX, USA) software was used for all statistical analyses.