Investigating Causal Associations of Diet-Derived Circulating Antioxidants with the Risk of Digestive System Cancers: A Mendelian Randomization Study

Abstract

Molecular mechanisms and observational studies have found that diet-derived antioxidants are associated with digestive system cancers, whereas there is a lack of causal evidence from randomized clinical trials. In this study, we aimed to assess the causality of these associations through a Mendelian randomization (MR) study. Single nucleotide polymorphisms of diet-derived circulating antioxidants (i.e., α- and γ-tocopherol, ascorbate, retinol, β-carotene, lycopene, and urate), accessed by absolute levels and relative metabolite concentrations, were used as genetic instruments. Summary statistics for digestive system cancers were obtained from the UK Biobank and FinnGen studies. Two-sample MR analyses were performed in each of the two outcome databases, followed by a meta-analysis. The inverse-variance weighted MR was adopted as the primary analysis. Five additional MR methods (likelihood-based MR, MR-Egger, weighted median, penalized weighted median, and MR-PRESSO) and replicate MR analyses for outcomes from different sources were used as sensitivity analyses. Genetically determined antioxidants were not significantly associated with five digestive system cancers, after correcting for multiple tests. However, we found suggestive evidence that absolute ascorbate levels were negatively associated with colon cancer in UK Biobank-the odds ratio (OR) per unit increase in ascorbate was 0.774 (95% confidence interval [CI] 0.608-0.985, p = 0.037), which was consistent with the results in FinnGen, and the combined OR was 0.764 (95% CI 0.623-0.936, p = 0.010). Likewise, higher absolute retinol levels suggestively reduced the pancreatic cancer risk in FinnGen-the OR per 10% unit increase in ln-transformed retinol was 0.705 (95% CI 0.529-0.940, p = 0.017), which was consistent with the results in UK Biobank and the combined OR was 0.747 (95% CI, 0.584-0.955, p = 0.020). Sensitivity analyses verified the above suggestive evidence. Our findings suggest that higher levels of antioxidants are unlikely to be a causal protective factor for most digestive system cancers, except for the suggestive protective effects of ascorbate on colon cancer and of retinol on pancreatic cancer.

Keywords: antioxidant; ascorbate; digestive system cancers; mendelian randomization; retinol.

Figure 1

Schematic overview of the MR study design. (A) Principles of this MR study. There are three principal assumptions in MR design, as follows: (Ⅰ) the relevance assumption—the selected instrument is predictive of the exposure; (Ⅱ) the independence assumption—the instrument is not associated with any confounders of the exposure and outcome; and (Ⅲ) the exclusion-restriction assumption—the instrument is only associated with the outcome through the exposure. (B) Study design and framework of this research [29,30,31,32,33,34,35,36].

Figure 2

The main MR analyses results of the causal effects of absolute circulating antioxidant levels on five digestive system cancers. If there was significant heterogeneity (p-value of Cochran′s Q Statistic < 0.05) we used the random-effect IVW model, otherwise we used the fixed-effect IVW model. The odds ratios were scaled per 10% unit increase in log-transformed α-tocopherol values, per unit increase in ascorbate, per 10% unit increase in ln-transformed retinol, per 10% unit increase in ln-transformed β-carotene, per unit increase in lycopene, and per unit increase in urate. Statistical significance was defined as Bonferroni-corrected threshold of p-value < 0.01 (0.05/5), and p-value between 0.01 and 0.05 was considered suggestive evidence (*) of associations.

Figure 3

The main MR analyses results of the causal effects of circulating antioxidant metabolites on five digestive system cancers. If there was significant heterogeneity (p-value of Cochran′s Q Statistic < 0.05), we used the random-effect IVW model, otherwise we used the fixed-effect IVW model. The odds ratios were scaled per 10% unit increase in log-transformed α-tocopherol, γ-tocopherol, and urate values, and per unit increase in log-transformed ascorbate and retinol values. Statistical significance was defined as Bonferroni-corrected threshold of p-value < 0.01 (0.05/5), and p-value between 0.01 and 0.05 was considered suggestive evidence (*) of associations.

Figure 4

The complementary MR analyses results of the causal effects of diet-derived circulating antioxidants on five digestive system cancers. The MR-Egger, weighted median, and penalized weighted median required the number of SNPs in the instrumental variable > 2, and MR-Egger method could not accurately estimate due to collinearity in MR analyses for α-tocopherols. The MR-PRESSO requires the number of SNPs in the instrument variable > 3. If the MR-PRESSO global test did not identify significant outliers in the genetic instrument, the MR-PRESSO did not require correction. The error bars indicate 95% CIs. Statistical significance was defined as Bonferroni-corrected threshold of p-value < 0.01 (0.05/5), and p-value between 0.01 and 0.05 was considered suggestive evidence (*) of associations. Shaded areas represent suggestive evidence from meta-analysis, combining UK Biobank and FinnGen.