TNF-α is a member of the TNF/TNFR cytokine superfamily, and is an intercellular communicating molecule involved in building transient or long-lasting multicellular structures
. It is also known to play critical and non-redundant roles in the innate and adaptive immune responses
, including the response to tumours
. TNF-α has been directly and indirectly linked to neoplasia and is involved not only in maintenance and homeostasis of the immune system, inflammation, and host defence, but also in pathological processes such as chronic inflammation, autoimmunity, and malignant disease
[35, 37]. TNF-α expression is mostly regulated at the transcriptional level
. Promoter polymorphisms in the TNF-α gene related to the pro- and anti-inflammatory response could directly influence production of TNF-α, thus causing inter-individual differences in immune responsiveness, which may influence the susceptibility of prostate cancer
Two variations in the promoter region of the TNF-α gene, namely, TNF-α-308G/A and TNF-α-238G/A, have been commonly studied. The G to A substitution at position −308 in the TNF-a promoter increases TNF-α transcription activity and the serum TNF-α level
. Indeed, the TNF-308A allele has been associated with malignant tumours such as gastric cancer, breast carcinoma, and hepatocellular cancer
[40–42]. The functional significance of the rare TNF-238A allele is not yet clear, but Kaluza et al. reported that this allele caused a significant decreased in the transcription of the TNF-α gene in human T and B cells
. The allele has also been associated with certain autoimmune and infectious diseases
[44, 45]. Several studies have observed the association between prostate cancer risk and TNF-α promoter polymorphisms, and TNF-α-308G/A and/or TNF-α-238G/A polymorphisms, but the results are controversial. These inconsistent results are possibly because of the small effect of the polymorphism on prostate cancer risk or the relatively low statistical power of the published studies. Meta-analysis could overcome these disadvantages because (1) it can investigate data for a large number of individuals; (2) it can estimate the effect of a genetic factor on disease risk; and (3) if a significant association is found, it can estimate whether the association is common among different backgrounds (such as population or age groups)
[46–49]. Therefore, we performed a meta-analysis to comprehensively evaluate the association between TNF-α-308G/A and/or TNF-α-238G/A polymorphisms and prostate cancer risk.
Our meta-analysis summarized for the first time all the available data on the association between TNF-α-238G/A polymorphism and prostate cancer risk, including a total of five studies, involving 1,967 prostate cancer cases and 2,004 controls. Our results demonstrated that the TNF-α-238G/A polymorphism was not significantly associated with prostate cancer risk not only in the overall population but also in the subgroup analyses stratified by ethnicity and source of controls.
With respect to TNF-α-308G/A polymorphism, 14 studies including 5,757 prostate cancer cases and 6,137 controls were found in our meta-analysis. The data suggested no significant association between TNF-α-308G/A polymorphism and prostate cancer risk in all genetic models in the overall populations, which is consistent with the previous findings made by Wang et al.
 and Wang et al.
. However, in the subgroup analysis according to source of controls, significantly increased prostate cancer risk was found in the healthy volunteer studies but not in hospital-based and population-based studies. However, the result may be underpowered because the sample size the healthy volunteer studies in this analysis is relatively small, and controls in these studies may not always be truly representative of the general population. Therefore, a methodologically preferable design such as a representative population-based study is needed to avoid selection bias and to increase the statistical power.
For the TNF-α-308G/A polymorphism, substantial heterogeneities between studies were observed in the additive model AG vs. GG (p < 0.001) and dominant model AA + AG vs. GG (p < 0.001) in the overall populations. To explore the source of heterogeneity, we employed meta-regression and subgroup analyses. Meta-regression analysis revealed no definite source of heterogeneity. Subgroup analyses by ethnicity showed that heterogeneity existed in Caucasian subjects and hospital-based studies in the additive and dominant models. To further investigate the heterogeneity, Galbraith plot analysis was used to identify the outliers that might contribute to the heterogeneity, and two outliers were found. When they were excluded from the analysis, the I2 values decreased obviously and PQ values were greater than 0.10 in the overall populations and in Caucasians. In addition, excluding the two studies did not significantly affect the results in the different comparison models in the overall population and subgroup analyses. The results indicated that the two studies might be the major source of the heterogeneity for the 308G/A polymorphism.
Substantial heterogeneities between studies were observed in the additive and dominant models for the TNF-α-238G/A polymorphism in the Caucasian population and hospital-based studies. As only five studies on the association between TNF-α-238G/A polymorphism and prostate cancer risk were included in our meta-analysis, meta-regression analysis and Galbraith plots analysis were not performed; therefore, these results require further investigation.
This meta-analysis has some limitations that must be considered. First, the overall outcomes were based on individual unadjusted ORs, while a more precise estimation should be adjusted by confounding factors such as smoking status, age, and environmental factors. Second, the sample sizes in this analysis were not adequate, especially the African-American populations; therefore, more subjects of different ethnicities would be required to accurately clarify whether ethnicity has a biological influence on cancer susceptibility. Third, the controls were not consistently screened across the studies analysed. Therefore, the control groups may have different risks of developing prostate cancer. Fourth, as only certain published studies were included in our meta-analysis, publication bias is very likely to occur although it was not shown in the statistical test.