Circulating interleukin-6 and cancer: A meta-analysis using Mendelian randomization

Interleukin-6 (IL-6) plays a contributory role in the progression and severity of many forms of cancer; it however remains unclear whether the relevance between circulating IL-6 and cancer is causal. We therefore meta-analyzed published articles in this regard using IL-6 gene -174G/C variant as an instrument. Seventy-eight and six articles were eligible for the association of -174G/C variant with cancer and circulating IL-6, respectively. Overall analyses failed to identify any significance between -174G/C and cancer risk. In Asians, carriers of the -174CC genotype had an 1.95-fold increased cancer risk compared with the -174GG genotype carriers (P = 0.009). By cancer type, significance was only attained for liver cancer with the -174C allele conferring a reduced risk under allelic (odds ratio or OR = 0.74; P = 0.001), homozygous genotypic (OR = 0.59; P = 0.029) and dominant (OR = 0.67; P = 0.004) models. Carriers of the -174CC genotype (weighted mean difference or WMD = −4.23 pg/mL; P < 0.001) and -174C allele (WMD = −3.43 pg/mL; P < 0.001) had circulating IL-6 reduced significantly compared with the non-carriers. In further Mendelian randomization analysis, a reduction of 1 pg/mL in circulating IL-6 was significantly associated with an 12% reduced risk of liver cancer. Long-term genetically-reduced circulating IL-6 might be causally associated with a lower risk of liver cancer.

Forty-six of 87 study groups were conducted in Caucasians, 31 in mixed populations, 9 in Asians and 1 in African-Americans. Sixteen study groups focused on colorectal cancer, 14 groups on breast cancer, 8 groups on gastric cancer, 7 groups on prostate cancer, 6 groups on myeloma, 5 groups respectively on lung and liver cancers, 4 groups on lymphoma, 3 groups on oral cancer, 2 groups respectively on esophageal, basal cell, bladder, cervical, leukatmia, gallbladder and thyroid cancers, 1 group respectively on melanoma, ovarian, renal cell, leiomyoma and pancreatic cancers. Thirty-three of 87 study groups had total sample size of at least 500. Age was reported to be matched in 60 study groups and unmatched in 4 study groups between cancer patients and controls. Fifty-seven study groups had controls enrolled from general populations and 30 study groups from hospitals. Fifty-nine and 28 study groups followed a retrospective and prospective design, respectively. Sixty-five of 87 study groups had genotype distributions of -174G/C variant in Hardy-Weinberg equilibrium at a significance level of 5%, 19 study groups in Hardy-Weinberg disequilibrium, and 3 study groups without mutation. The frequency of IL-6 gene -174C allele ranged from 0.0% to 52.94% in cancer patients and from 0.0% to 51.15% in healthy controls. By contrast, this frequency was significantly lower in Asians than in Caucasians (cancer patients: 0.0% to 28.0% versus 17.90% to 52.94%; healthy controls: 0.0% to 26.75% versus 13.33% to 51.15%). Table 2 summarizes mean circulating IL-6 across -174G/C genotypes. All study groups involved Caucasians except for one with mixed descents. Seven of 9 study groups provided circulating IL-6 in cancer patients, and 1 study group respectively in healthy controls and in combined patients and controls.
Potential sources of heterogeneity were exploited by subgroup analyses according to ethnicity, cancer type, matching condition, source of controls, study design and sample size, respectively (Table 3). To avoid chance results, only subgroups with three or more study groups were considered in this meta-analysis. By ethnicity, no significance was attained in Caucasians under three genetic models, and contrastingly Asian carriers of -174CC genotype were observed to have 1.95-fold increased cancer risk compared with those with the -174GG genotype (95% CI: 1.95; 1.19 to 3.20; P = 0.009), even after the Bonferroni correction to control for the multiple testing (P < 0.05/3, here 3 refers to the total number of subgroups by ethnicity). There was no heterogeneity (I 2 = 17.9%) for this significant association.
By cancer type, effect estimates were significant only for liver cancer with the -174C allele confer-   Table 3. Overall and stratified risk estimates of IL-6 gene -174G/C variant for cancer risk under three genetic models. Abbreviations: OR, odds ratio; 95% CI, 95% confidence interval; NA, not available.
21.5%, respectively). Even after the Bonferroni correction, significance was still preserved for the allelic and dominant models (P < 0.05/9, here 9 refers to the total number of subgroups by cancer type). Effect estimates of -174G/C variant for cancer by source of controls, study design and sample size did not deviate significantly from the unity under three genetic models (all P > 0.05), and there was no material improvement in heterogeneity within these subgroups.
Predicted causality of circulating IL-6 for cancer. Under the principles of Mendelian randomization, a reduction of 1 pg/mL in circulating IL-6 was significantly associated with an 12% reduced risk of liver cancer (95% CI: 0.64 to 0.99). However in Asians, this association was totally reversed with 1 pg/mL reduced circulating IL-6 corresponding to an 17% increased cancer risk (95% CI: 1.03 to 1.68).
Considering that the unity was not included by above 95% CIs, it is safe to the reject the null hypothesis of none causal relevance between circulating IL-6 and certain cancer subtypes. Sensitivity analysis. Sensitivity analysis confirmed the overall differences in risk estimates for the prediction of IL-6 gene -174G/C variant for cancer risk and circulating IL-6 changes between -174G/C genotypes in both direction and magnitude by sequentially omitting each study once at a time and computing differential estimates for the remaining studies.

Discussion
In this meta-analysis of 80 qualified articles, we employed Mendelian randomization to test the completing hypothesis that the association between circulating IL-6 and cancer is causal. The most noteworthy finding of this study was that long-term genetically-reduced circulating IL-6 might be causally associated with a lower risk of liver cancer. To the authors' knowledge, this is the first comprehensive meta-analysis assessing the impact of long-term differences in circulating IL-6 on cancer risk. Currently, targeted anti-IL-6 antibody therapy has been successfully applied in several clinical trials and found to be well tolerated in cancer patients 85 . Evidence from epidemiological studies is accruing in favor of a contributory role of elevated circulating IL-6 in patients with advanced tumor stages of various cancers, such as non-small cell lung cancer, colorectal cancer and renal cell carcinoma [86][87][88] . Currently, whether the progression and severity of caner is due to elevated circulating IL-6 still remains an open question. Genetic association studies are deemed as more similar to randomized clinical trials than other types of observational epidemiological studies due to Mendelian randomization (Mendel's second law) 89 . We therefore utilized Mendelian randomization to assess whether the relevance between circulating IL-6 and cancer is causal by selecting the most frequently evaluated variant -174G/C in IL-6 gene as a genetic instrument to minimize residual confounding and reverse causation.
In this meta-analysis, risk estimates of IL-6 gene -174G/C variant with cancer were heterozygous between Caucasians and Asians. Considering the multifactorial nature of cancer, divergent genetic backgrounds or linkage disequilibrium patterns might be the most likely explanation for such divergence 90 . This is well exemplified in the present study with regard to the frequency of -174C allele, which was exceedingly lower in Asians than in Caucasians (Table 1). Even in some Asian populations, the mutation rate of this allele was zero 40,67,81 . Generally, it is not uncommon for the same variant playing a different role in cancer susceptibility across different populations. This is the principal limitation of this Mendelian randomization meta-analysis, that is, if the other flanking variants within or near IL-6 gene related to cancer risk are in linkage disequilibrium with -174G/C variant we have examined, this will confound our findings. What's more, such confounding is difficult to exclude completely; however it is unlikely that it would explain our finding that IL-6 gene -174CC genotype was associated with lowered circulating IL-6 without predicting a low cancer risk. There is also evidence that a variant may be in close linkage with another nearby causal locus in one ethnic population but not in another 91 . In view of this limitation, it is necessary to establish an ethnicity-specific database of candidate genes and variants in susceptibility to cancer 92 . Another limitation of this study is that excluding the pleiotropy of IL-6 gene -174G/C variant seems impractical for us since data on other inflammatory factors across -174G/C genotypes are rarely provided from most eligible articles, necessitating further confirmation using additional genetic variants and/or exposure outcomes.
It is also worth noting that IL-6 gene -174G/C variant exhibited heterozygous association with different forms of cancer in this meta-analysis. For example, the -174C allele was observed to confer a significantly protective effect against liver cancer, yet a risk effect for oral cancer with no attainable significance. The identification of -174G/C variant affecting the significant risk of liver cancer allowed us to employ Mendelian randomization to account for potential biases due to residual confounding and reverse causation. Consistent with the findings of other studies 8,9 , carriers of the -174C allele or -174CC genotype had lower circulating IL-6 than the non-carriers, supporting the plausibility of causal relevance between circulating IL-6 and liver cancer. Given the insufficient statistical power of this meta-analysis in some subgroups, far larger sample sizes than studied here will be required to produce enough power to evaluate the causality between circulating IL-6 and various forms of cancer.
Several limitations should be acknowledged in this meta-analysis. Firstly, only published articles were retrieved and the 'grey' literature (articles written in languages other than English) was not covered, leading to the possible existence of publication bias. However, the influence of publication bias on the gene-disease association is expected to result in an overestimation, rather than an underestimation. Secondly, as the majority of involved studies in this meta-analysis recruited cancer patients aged over 50 years for whom environmental factors are likely to contribute more prominently than a genetic component to the development of cancer, more large studies in a younger cancer population will of great interest. Thirdly, given the possible impact of drug regimens on circulating IL-6, the relationship between circulating IL-6 changes and IL-6 gene -174G/C genotypes might be biased, calling for further validation of this relationship in healthy controls. Fourthly, nearly all involved studies had circulating IL-6 measured only once, which cannot reflect its long-term level in the development of cancer. Fifthly, this meta-analysis was based on summarized data, rather than individual participant data, precluding further gene-to-environment interaction. Sixthly, only one variant in IL-6 gene was selected, and investigation on other candidate genes or polymorphisms involved in IL-6 regulation was highly encouraged, leaving a challengeable task to test whether this variant integrated with other risk determinants will enhance cancer risk prediction. The jury therefore must refrain from drawing a firm conclusion until large, well-designed studies to confirm our findings.
To sum up, there is possible evidence for causal association between long-term genetically-reduced circulating IL-6 and reduced risk of liver cancer. For practical reasons, we hope that this study will advance our understanding of the role of circulating IL-6 leading to the progression and severity of liver cancer. Further studies to elucidate the specific role of IL-6 in cancer pathogenesis are required.

Methods
The implementation of this meta-analysis complied with the guidelines outlined in the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) statement (Please see the supplementary PRISMA Checklist) 93 .
The titles and abstracts of all retrieved relevant articles were independently reviewed by two investigators (Chunhua Yang and Xuri Li). In case of uncertainty for rejection, the full text and supplementary materials if available were downloaded to check whether information on the topic of interest was provided. If more than one article from a study group was published, data from the most recent or complete article were abstracted. The eligibility of each retrieved article was assessed in duplicate and independently by two authors (Chunhua Yang and Xuri Li). Any uncertainty over the eligibility was adjudicated by further joint inspection of the articles.
Inclusion/exclusion criteria. Inclusion criteria were to test the hypothesis that IL-6 gene -174G/C variant was associated with cancer or circulating IL-6 and to provide detailed genotype or allele counts of this variant between cancer patients and healthy controls or the mean levels of circulating IL-6 across -174G/C genotypes. Articles were excluded if they assessed the progression, severity, phenotypic modification and response to treatment or survival of cancer, or if they lacked patients or controls, or if they were conference abstracts or proceedings, case reports or series, editorials, narrative reviews, and non-English articles.
Data extraction. Two authors (Chunhua Yang and Xuri Li) independently abstracted the following data from each qualified article according to a fixed protocol, including the first author's last name, publication year, ethnicity, cancer type, matching condition, study design, sample size, the genotype or allele counts of the -174G/C variant between patients and controls, mean level of circulating IL-6 for each genotype carriers expressed as mean ± standard deviation, as well as some baseline characteristics of study populations where available, including age, gender, body mass index, the percentages of smokers and drinkers. The unit of circulating IL-6 was uniformly transferred into pg/mL in this meta-analysis. Quality assessment. Criteria for quality assessment of the association between IL-6 gene -174G/C variant and cancer risk were in agreement with the standards formulated by Thakkinstian et al. 94 There were 7 criteria in total and summarized as a quality score, ranging from 0 (the worst) to 12 (the best). Quality assessment was independently conducted by two authors (Chunhua Yang and Xuri Li), and any disagreement was resolved by consensus.
Statistics. Risk estimates for the association of IL-6 gene -174G/C variant with cancer were expressed as odds ratio (OR) and its corresponding 95% confidence interval (95% CI), and for the changes of circulating IL-6 between genotypes of this variant as weighted mean difference (WMD) and its 95% CI.
Hardy-Weinberg equilibrium was tested by Chi-squared test. Differences at P < 0.05 were accepted as statistically significant. A random-effects model was employed to bring individual effect-size estimates together by using the DerSimonian and Laird method.
Heterogeneity between studies was quantified by the inconsistency index (I 2 ) statistic, which ranges from 0% to 100%. This statistic is defined as the percentage of the observed between-study variability that is due to heterogeneity rather than chance. A threshold of over 50% for I 2 statistic was treated as statistically significant heterogeneity.
Predefined subgroup analyses were explored to identify the potential sources of between-study heterogeneity according to ethnicity, cancer type, matching condition, source of controls (population-based controls and hospital-based controls), study design (prospective study and retrospective study) and sample size (<500 subjects and ≥500 subjects). To assess the contribution of each individual studies to pooled effect estimates, sensitivity analyses were undertaken by sequentially omitting each study one at a time and computing differential estimates for remaining studies.
Publication bias was assessed by the Begg's funnel plot and the Egger regression asymmetry test. The Egger test can identify the asymmetry of funnel plots by determining whether the intercept deviates significantly from zero in regressing the standardized effect estimates against their precision. A value of P < 0.10 was used to indicate statistical significance for Egger's test.
Risk estimates in Mendelian randomization analysis were calculated as the ratio of the coefficient of the association between IL-6 gene -174G/C variant and cancer to that of the association between -174G/C variant genotypes and circulating IL-6 as a reflection of the potential causal impact of circulating IL-6 on cancer.
All statistical analyses described above were completed with the StataCorp STATA version 12.0.