Circulating adipokine concentrations and risk of five obesity‐related cancers: A Mendelian randomization study

Abstract Obesity is considered a chronic inflammatory state characterized by continued secretion of adipokines and cytokines. Experimental and epidemiological evidence indicates that circulating adipokines may be associated with the development of obesity‐related cancers, but it is unclear if these associations are causal or confounded. We examined potential causal associations of specific adipokines (adiponectin, leptin, soluble leptin receptor [sOB‐R] and plasminogen activator inhibitor‐1 [PAI‐1]) with five obesity‐related cancers (colorectal, pancreatic, renal cell carcinoma [RCC], ovarian and endometrial) using Mendelian randomization (MR) methods. We used summary‐level data from large genetic consortia for 114 530 cancer cases and 245 284 controls. We constructed genetic instruments using 18 genetic variants for adiponectin, 2 for leptin and 4 for both sOB‐R and PAI‐1 (P value for inclusion<5 × 10−8). Causal estimates were obtained using two‐sample MR methods. In the inverse‐variance weighted models, we found an inverse association between adiponectin and risk of colorectal cancer (odds ratio per 1 μg/mL increment in adiponectin concentration: 0.90 [95% confidence interval = 0.84‐0.97]; P = .01); but, evidence of horizontal pleiotropy was detected and the association was not present when this was taken into consideration. No association was found for adiponectin and risks of pancreatic cancer, RCC, ovarian cancer and endometrial cancer. Leptin, sOB‐R and PAI‐1 were also similarly unrelated to risk of obesity‐related cancers. Despite the large sample size, our MR analyses do not support causal effects of circulating adiponectin, leptin, sOB‐R and PAI‐1 concentrations on the development of five obesity‐related cancers.

using 18 genetic variants for adiponectin, 2 for leptin and 4 for both sOB-R and PAI-1 (P value for inclusion<5 × 10 −8 ). Causal estimates were obtained using twosample MR methods. In the inverse-variance weighted models, we found an inverse association between adiponectin and risk of colorectal cancer (odds ratio per 1 μg/ mL increment in adiponectin concentration: 0.90 [95% confidence interval = 0.84-0.97]; P = .01); but, evidence of horizontal pleiotropy was detected and the association was not present when this was taken into consideration. No association was found for adiponectin and risks of pancreatic cancer, RCC, ovarian cancer and endometrial cancer. Leptin, sOB-R and PAI-1 were also similarly unrelated to risk of obesity-related cancers. Despite the large sample size, our MR analyses do not support causal effects of circulating adiponectin, leptin, sOB-R and PAI-1 concentrations on the development of five obesity-related cancers.

| INTRODUCTION
A substantial body of evidence has shown that excess adiposity is associated with a greater risk of developing many common cancers. 1,2 The biological pathways linking adiposity with cancer development are incompletely understood, but likely involve alterations in insulin signaling, sex hormone pathways and adipose tissue-derived inflammation. 3,4 Obesity is considered as a chronic inflammatory state characterized by continued infiltration of adipose tissue by macrophages and other immune cells leading to increased or decreased adipose secretion of adipokines (such as adiponectin, leptin and plasminogen activator inhibitor-1 [PAI-1]) that may be linked to cancer development.
Adiponectin lowers secretion of inflammatory cytokines, improves insulin sensitivity and inhibits cell growth and angiogenesis, but is downregulated in obesity. 5,6 Multiple epidemiological studies

What's new?
Chronic inflammation attributed to obesity may influence cancer development. However, little is known about the relationship between oncogenesis and changes in adipokine secretion stemming from immune cell infiltration in adipose tissue. Here, large-scale Mendelian randomization analysis was used to assess possible causal associations of adipokine concentrations influenced by genetic variation and risk of five obesity-related cancers, including renal cell carcinoma and colorectal, pancreatic, ovarian and endometrial cancer.
In general, no association was detected between adipokines and the five malignancies, suggesting that adipokine levels have no causal influence on these cancers. have investigated the association between circulating adiponectin concentration and cancer risk with inverse relationships sometimes reported for endometrial, colorectal, renal cell carcinoma (RCC) and pancreatic cancer. 3,[7][8][9][10] Epidemiological studies that examined the association between circulating leptin concentration, which has proinflammatory effects, and obesity-related cancers have yielded inconsistent results. 9,11,12 It is unclear if these mixed results were due to most studies failing to measure soluble leptin receptor (sOB-R) concentrations, which may regulate the biological effects of circulating leptin concentration. Within the European Prospective Investigation into Cancer and Nutrition (EPIC) study, circulating concentration of sOB-R was inversely associated with colorectal cancer, even after statistical adjustment for leptin concentrations, suggesting that sOB-R may have an independent role in colorectal cancer development. 11 It is currently unknown if sOB-R is similarly associated with other obesity-related cancers as these studies have not been conducted. In addition, few studies have examined the association between circulating PAI-1 concentration (elevated in obesity) and cancer outcomes, although positive associations were found for colorectal cancer in the EPIC-Italy and Women's Health Initiative (WHI) studies. 9,13 These previous observational epidemiological studies are vulnerable to residual confounding and reverse causality which make causal inference challenging. An alternative approach is Mendelian randomization (MR) that uses genetic variants robustly associated with the exposure of interest as instrumental variables to allow causal inference for the effect of an exposure on an outcome. 14 MR analyses are largely free of conventional confounding and reverse causality as genetic variants are randomly assigned, and fixed, at conception.

| Statistical power
Power calculations were performed based on the method suggested by Brion et al. 31 We fixed the type-I error rate at 0.05. Based on the aforementioned cancer case and control numbers, and assuming an R 2 of 3.0% (variance explained by the selected variants for circulating adiponectin), our study had 80% power to detect an odds ratio (OR) of 0

| Statistical analysis
We employed a fixed-effects inverse-variance weighted (IVW) MR method. 32 For causal estimates from MR studies to be valid, three main assumptions must be satisfied: (a) the selected genetic variants used in the instrument are robustly associated with adipokine concentrations, (b) the genetic variants are not associated with any confounder of the adipokines and cancer relationship and (c) the genetic instrument should not affect the outcome independently of its effect on adipokine concentration. Assumption 1 was likely to be satisfied as only variants associated with adipokines at a genome-wide significance level were used. For assumption 2, we acquired information for the association of the selected variants used in the instruments with other traits from the Phenoscanner. 33 A series of statistical tests were performed to investigate the potential violation of MR assumption 3 and to assess the possible influence of horizontal pleiotropy on the causal estimates. We estimated the Cochran's Q statistic that quantifies the heterogeneity in effect sizes attributed to the selected genetic variants. When there was evidence for heterogeneity, we performed a random effects IVW approach in order to take into account this source of uncertainty. 34 MR-Egger regression was also used 35 and the estimator from the weighted median approach. 36 We conducted sensitivity analyses with variants associated with adiposity measures or insulin resistance excluded. We also restricted our analyses to cis-acting variants. For adiponectin, we used rs17366568 variant in the ADIPOQ gene; for leptin, we used rs10487505 variant in the LEP gene; for PAI-1, we used rs2227631 variant in the SERPINE1 gene; while for sOB-R, all four genetic variants (ie, rs17415296, rs4655537, rs17412403 and rs7535099) used are located in the LEPR gene.
For adiponectin and leptin, in sensitivity analyses, we also conducted analyses using selected variants unadjusted for BMI, to examine if collider bias may have influenced our results. In this scenario, we also accounted for pleiotropic effects acting via BMI using data from a recent GWAS of the GIANT consortium and the UK-Biobank 37 in a multivariable MR framework. 38 3 | RESULTS

| Adiponectin
In the IVW models, we found an inverse association between adiponectin and risks of colorectal cancer (OR per 1 μg/mL increment in adiponectin concentrations: 0.90 (95% confidence interval [CI] = 0.84-0.97); P = .01), with similar association found for men and women, colon cancer and rectal cancer ( Figure 1 and Table S2). Near identical results were found when we used summary estimates of adiponectin unadjusted for BMI (Table S3). However, evidence of pleiotropy was detected and using robust MR methods (ie, MR-Egger and Weighted median test) results were attenuated toward the null for all models T A B L E 1 Number of cancer cases and controls and statistical power in Mendelian randomization study of adipokines and risk of cancer  and Table S4). Similar results were found when analyses were restricted to the rs10487505 variant in the LEP gene (Table S4). No associations were also found for sOB-R with any cancer type ( Figure 1 and Table S5). For both leptin and sOB-R, similar results were generally found by cancer subsite, subtype and sex, and no evidence of heterogeneity or pleiotropy was detected.   Table S6). There was no association between genetically predicted circulating PAI-1 concentration and any of the cancers when rs2227631 variant in the SERPINE1 gene was used as the genetic instrument (Table S6) 12 Similarly, in a WHI case-cohort study, a positive association between serum leptin concentration and colorectal cancer risk was found, even after adjustment for insulin concentrations. 9 An analysis in the EPIC study found no association for circulating leptin, but did observe an inverse association for sOB-R with colorectal cancer. 11 The null results we found for leptin and sOB-R are largely consistent with previous epidemiological evidence for pancreatic cancer, 46 but inconsistent for RCC and endometrial cancer, for which the few prospective studies conducted have generally found positive associations for circulating leptin concentration. 43,47 We found no association between genetically determined PAI-I Our results are consistent with the findings of a recent MR study that showed no association between adiponectin, sOB-R and PAI-1 concentrations and breast cancer risk. 49 Though it is not possible to prove the validity of some of the MR assumptions with summarized data, we performed various sensitivity analyses and investigated potential associations with secondary phenotypes of interest, including BMI and insulin resistance. Importantly, our results were similar when we restricted the genetic instruments to include cis-acting variants, suggesting that pleiotropy did not markedly influence our findings.

| Plasminogen activator inhibitor-1
Power calculations indicated that our analyses were adequately powered to detect effect sizes comparable with prior observational studies that reported associations between adipokine concentrations and these cancers, 7,[11][12][13]43,47 with the exception of some of the pancreatic cancer and cancer subtype models for which GWAS casecontrol numbers were relatively low. 39