An atlas of associations between 14 micronutrients and 22 cancer outcomes: Mendelian randomization analyses

Abstract

Background: Micronutrients, namely vitamins and minerals, are associated with cancer outcomes; however, their reported effects have been inconsistent across studies. We aimed to identify the causally estimated effects of micronutrients on cancer by applying the Mendelian randomization (MR) method, using single-nucleotide polymorphisms associated with micronutrient levels as instrumental variables.

Methods: We obtained instrumental variables of 14 genetically predicted micronutrient levels and applied two-sample MR to estimate their causal effects on 22 cancer outcomes from a meta-analysis of the UK Biobank (UKB) and FinnGen cohorts (overall cancer and 21 site-specific cancers, including breast, colorectal, lung, and prostate cancer), in addition to six major cancer outcomes and 20 cancer subset outcomes from cancer consortia. We used sensitivity MR methods, including weighted median, MR-Egger, and MR-PRESSO, to assess potential horizontal pleiotropy or heterogeneity. Genome-wide association summary statistical data of European descent were used for both exposure and outcome data, including up to 940,633 participants of European descent with 133,384 cancer cases.

Results: In total, 672 MR tests (14 micronutrients × 48 cancer outcomes) were performed. The following two associations met Bonferroni significance by the number of associations (P < 0.00016) in the UKB plus FinnGen cohorts: increased risk of breast cancer with magnesium levels (odds ratio [OR] = 1.281 per 1 standard deviation [SD] higher magnesium level, 95% confidence interval [CI] = 1.151 to 1.426, P < 0.0001) and increased risk of colorectal cancer with vitamin B12 level (OR = 1.22 per 1 SD higher vitamin B12 level, 95% CI = 1.107 to 1.345, P < 0.0001). These two associations remained significant in the analysis of the cancer consortia. No significant heterogeneity or horizontal pleiotropy was observed. Micronutrient levels were not associated with overall cancer risk.

Conclusions: Our results may aid clinicians in deciding whether to regulate the intake of certain micronutrients, particularly in high-risk groups without nutritional deficiencies, and may help in the design of future clinical trials.

Keywords: Cancer; Mendelian randomization; Micronutrient; Mineral; Vitamin.

Fig. 1

Minimum detectable odds ratios assuming statistical power of 80% for the 672 micronutrient-cancer associations. * The number of cancer cases and total participants are listed for cohorts reporting cancer outcomes, and the number of total participants is listed for cohorts reporting micronutrient levels

Fig. 2

Results of Mendelian randomization associations of micronutrients with cancer outcomes from combined UK Biobank and FinnGen cohort. * Bonferroni-corrected significance threshold by the number of exposures. ** Bonferroni-corrected significance threshold by the number of exposures-outcome pairs

Fig. 3

Statistically significant associations between micronutrient levels and cancer outcomes in UK Biobank and FinnGen meta-analysis. Associations are listed in ascending order of p-values in the inverse variance-weighted model. The forest plot represents the pooled odds ratio under the inverse variance-weighted model. For the association supported by 1 instrumental variable (vitamin B6-esophagus cancer association), the Wald ratio was reported instead of the pooled inverse variance weighted odds ratio. A weighted median analysis was not available for the associations supported by two or fewer instrumental variables. Details of all the statistical results are shown in Supplementary table 3. Abbreviations: CI, confidence interval; NSNP, number of single-nucleotide polymorphisms. ** The association showed Bonferroni-corrected significance by the number of exposure-outcome pairs (P < 0.00016)

Fig. 4

Statistically significant associations between micronutrient levels and cancer outcomes in cancer consortia studies. Associations are listed in ascending order of p-values in the inverse variance-weighted model. The forest plot represents the pooled odds ratio under the inverse variance-weighted model. For the association supported by 1 instrumental variable (vitamin B6-triple negative breast cancer association), the Wald ratio was used instead of the pooled inverse variance weighted odds ratio. A weighted median analysis was not available for the associations supported by two or fewer instrumental variables. Details of all the statistical results are shown in Supplementary table 3. Abbreviations: CI, confidence interval; HER2, human epidermal growth factor receptor 2; NSNP, number of single-nucleotide polymorphisms.