Study design
We used two-sample MR analysis to identify the causal relationship between circulating adiponectin levels and prostate cancer. SNPs of adiponectin were selected as IVs from published GWAS analyses. For the MR study, three key assumptions must be met. First, the SNPs should be related to circulating adiponectin levels robustly. Second, the SNPs should be independent of confounders, such as body mass index (BMI), sex, age and so on. Third, the SNPs should affect prostate cancer only by adiponectin and the SNPs can’t have a direct correlation with prostate cancer [10]. (Figure 1)
We chose a GWAS meta-analysis of adiponectin including 67,739 individuals of European, Hispanic, African American and East Asian ancestry from a published study [11]. Some SNPs related with adiponectin robustly were screened from the GWAS for MR analysis. And two GWAS of prostate cancer from Prostate Cancer Association Group to Investigate Cancer Associated Alterations in the Genome (PRACTICAL) [12] and Japanese Encyclopedia of Genetic Associations by Riken (JENGER) [13] were used to estimate the causal relationship. Then we employed the inverse variance weighted (IVW) model with random-effects for our main effect estimation, alongside weighted median, MR-Egger and weighted mode models. In addition, leave-one-out and heterogeneity sensitivity analysis were performed to meet the key assumptions. Finally, two other GWAS of adiponectin whose individuals were European and East Asian respectively, were added into the study to avoid the bias of human species.
Exposure and outcome dataset
For the exposure dataset, we used the largest-scale GWAS meta-analysis of circulating adiponectin levels, which included 60,465 European, 1,435 Hispanic, 3,271 African American and 2,568 East Asian. The GWAS meta-analysis incorporated 28 data-sets from 25 previous studies (mean age: 20.0-73.9 years; mean BMI: 24.2-43.3 kg/m2; mean circulating adiponectin level: 2.8-29.7 ug/ml). The analysis of circulating adiponectin levels had been adjusted for age, sex, BMI and other study-specific covariates.
In addition, we chose two GWAS meta-analysis of adiponectin including 29,347 European and 12,125 East Asian ancestry respectively, for avoiding the bias of human species. The GWAS of European was compounded of 26 studies (mean age: 9.8-75.4 years; mean BMI: 24.2-43.3 kg/m2; mean circulating adiponectin level: 4.9-25.5 ug/ml) [14] and the GWAS of East Asian consisted with 10 data-sets including 5,403 Chinese, 3,973 Korean, 1,717 Filipino and 1,030 Japanese (mean age: 41.6-66.2 years; mean BMI: 17.2-29.0 kg/m2; mean circulating adiponectin level: 2.5-14.0 ug/ml) [15].
Considering the bias of human species, we employed two different-ethnical larger-scale GWAS meta-analysis of prostate cancer. The GWAS accessed from PRACTICAL had 79,148 cases and 61,106 controls of European ancestry and JENGER dataset had 5,408 cases and 103,939 controls of East Asian ancestry. Unfortunately, we did not obtain the demographic data from two GWAS. (Table 1)
Statistical analysis
For the adiponectin, we extracted significant genome-wide SNPs (IVs) from the GWAS. Then each SNP calculates the R2 indicating the proportion of phenotypic variance. And F-statistic of each SNP calculates based on R2 to evaluate the strength of IVs in adiponectin [16]. The SNPs are remained if the result of F-statistic is larger than 10. Moreover, the SNPs on the same chromosome were assessed in linkage disequilibrium with each other. It can be calculated in a website (https://ldlink.nci.nih.gov/) [17-20]. The SNPs would be excluded if the result of linkage disequilibrium r2 is larger than 0.01.
Two-sample MR analysis was performed to estimate the causal effect between serum adiponectin concentrations and prostate cancer. IVW method with multiplicative random effects is the main analysis to obtain the causal estimates [21]. Wald estimate is used to evaluate each genetic variant by the ratio of the SNP-outcome estimate over the SNP-exposure estimate, with standard error (SE) using Delta method. Furthermore, MR-Egger, weighted media and weighted mode methods were chosen as complementary analysis. The result of IVW is based on the hypothesis that there is no horizontal pleiotropy meaning no intercept in the axis of coordinate. But it is vulnerable if the hypothesis does not hold, which is against the assumption three of MR analysis. Therefore, MR-Egger model is utilized to estimate the precise intercept representing the average horizontal pleiotropy. And the slope of MR-Egger shows the pleiotropy-adjusted estimate [22]. Weighted media model is a greater choice if over 50% of the SNPs meet the hypothesis of no horizontal pleiotropy that each SNP is weighted equally to the inverse of its SD in the analysis.
As for the heterogeneity, the Cochrane’s Q statistics are used to evaluate the variance between SNPs in IVW model [23]. In addition, leave-one-out analysis calculates the causal estimates after excluding one SNP. It is visualized to assess whether some SNPs play a particularly significant role in the results among all the SNPs [24].
For avoiding the bias of confounders, the SNPs selected from the GWAS of adiponectin were checked whether they were related with any diseases or traits factors other than adiponectin by the PhenoScanner online dataset (http://www.phenoscanner.medschl.cam.ac.uk/phenoscanner, p<5×10-8, r2>0.8) [25, 26]. Then the SNPs associated with adiponectin only were chosen to estimate the causal effects by IVW.
In our study, We used RStudio software (version 4.0.2; http://www.rproject.org) and TwoSampleMR package of R (version 0.5.4; https://github.com/MRCIEU/TwoSampleMR) to accomplish the analysis. The threshold value of statistical significance is set as P<0.05 (two-tailed).