Background
Numerous studies have addressed the effects of micronutrients, particularly vitamins and minerals, on various health outcomes, including cancer. Observational studies have frequently reported the benefits of micronutrient supplementation on cancer risks [
1‐
7], whereas randomized controlled trials (RCTs) have often reported a null effect [
8‐
10]. However, observational studies are inherently prone to confounding and reverse causation, and RCTs are expensive and often insufficiently powered for cancer outcomes, requiring long-term follow-up and large cohort sizes. A systematic review conducted in 2013 reported a paucity of fair- or good-quality studies assessing the associations between micronutrient supplementation and cancer and concluded that there was a lack of evidence to support micronutrient supplementation for cancer prevention [
9]. The Mendelian randomization (MR) approach attempts to overcome these limitations by using genetic variants as instrumental variables (IVs) to assess the potential causal association between risk factors and disease [
11].
In the past decade, many health benefits of micronutrients reported in observational studies have been shown to have null causal associations in MR studies. For example, although vitamin D is a promising micronutrient with statistically significant beneficial effects on various malignant, cardiovascular, metabolic, and other diseases [
12], over 60 MR studies published in the previous decade found no effect of genetically predicted vitamin D concentrations on most health outcomes [
13]. Numerous MR studies reporting associations between various micronutrients and the risk of various cancers [
14‐
17] revealed that only a few exposure-outcome pairs were potentially genuine associations. Furthermore, MR methods are not consistent across studies, making it difficult to compare the robustness of associations. According to a recent systematic review, most MR studies assessing cancer outcomes did not adequately perform sensitivity analyses assessing the pleiotropy of MR associations, such as MR-PRESSO, resulting in potentially biased estimates [
18], and many previous MR studies reporting micronutrient-cancer associations [
15‐
17,
19‐
22] chose IVs under linkage disequilibrium (LD) thresholds less strict than the conventionally used threshold of
r2 < 0.001, resulting in potentially biased estimates. To overcome these limitations and clarify the presence and robustness of causal associations, we performed exposure-wide and outcome-wide MR analyses of 14 micronutrients and 48 cancer outcomes.
Discussion
In this large-scale MR analysis of 672 micronutrient-cancer associations, we provided the most comprehensive and updated atlas of associations between micronutrients and cancer. We discovered that two associations, the association of magnesium with the risk of breast cancer and the association of vitamin B12 with the risk of colorectal cancer, were robust in terms of statistical significance, sensitivity, and replicability in different cohorts. Cancer subset analysis revealed that magnesium was associated with the luminal A-like breast cancer subtype. No specific micronutrients were beneficial in preventing cancer overall, which is consistent with the findings of previous RCTs [
9].
Genetically predicted 1 SD higher levels of magnesium were associated with 1.281 higher odds of breast cancer in the UKB and FinnGen meta-analyses, and 1.235 and 1.319 higher odds of overall breast cancer and luminal A-like breast cancer, respectively, in the BCAC analysis. These findings are consistent with the results of a previous MR study [
15] that used an earlier version of the BCAC summary statistics [
62], which reported that a 1 SD higher genetically predicted magnesium level was robustly associated with 1.17 higher odds of breast cancer and 1.2 higher odds of estrogen receptor-positive breast cancer, respectively. Although we could not find RCTs or observational studies assessing magnesium levels and breast cancer outcomes, the consistent results and similarity in the effect sizes in different datasets support the validity of our findings. Evidence from in vitro and animal studies has suggested plausible pathways wherein high magnesium concentrations can promote tumor growth and metastasis [
63‐
65].
Genetically predicted 1 SD higher levels of vitamin B12 were associated with 1.22 increased odds of colorectal cancer in the UKB and FinnGen meta-analysis and 1.115 increased odds of colorectal cancer in the analysis using cancer consortium data. This result is consistent with that of a previous MR study that used colorectal cancer GWAS meta-analysis of GECCO and other consortium data [
16], which reported 1.12 higher odds per 1 SD increase of genetically predicted vitamin B12 level of colorectal cancer (specifically 1.1 higher odds of colon cancer and 1.21 higher odds of rectal cancer). Previous RCTs reported conflicting results [
66,
67]. A long-term follow-up study of RCT of 2524 Caucasian participants compared 2–3 years daily supplementation of folic acid (0.4 mg)/vitamin B12 (0.5 mg) and placebo and reported a higher risk of colorectal cancer in the treatment group (hazard ratio = 1.77, 95% CI = 1.08 to 2.90) [
66]. The median age of the participants was 74 years, and the median follow-up period was 78 months, resulting in 68 colorectal cancer cases. Another four-arm RCT assessed 6837 Caucasian patients with ischemic heart disease, with a median age of 63 years. Patients were administered folic acid (0.8 mg), vitamin B12 (0.4 mg), or vitamin B6 (40 mg) [
67] for a median of 39 months and were followed for an additional 38 months, resulting in 95 colorectal cancer cases. In this study, the incidence of colorectal cancer did not differ according to the supplements received. One limitation of these trials was their low power. The latter trial [
67] reported 629 cases of any cancer and the power to detect any cancer incidence difference between the two groups was 61% [
67]. It can be expected that the power to detect differences in colorectal cancer incidence was much lower, given that there were only 95 colorectal cancer cases. Additionally, both trials assigned both vitamin B12 and folic acid to treatment arms, and the participants either had elevated homocysteine level [
66] or ischemic heart disease [
67], making it hard to generalize the results to the effect of vitamin B12 on the general population. Another key limitation of these RCTs was that the follow-up duration was insufficient to detect an effect on the incidence of colorectal cancer (77–78 months), given that the progression of colorectal adenoma to colorectal cancer may take 10–15 years [
68]. A long-term study over 10 years may be beneficial for identifying the true association between vitamin B12 supplementation and colorectal cancer.
For breast, colorectal, lung, ovarian, and prostate cancers, the associations between micronutrient levels were assessed in both the combined UKB and FinnGen cohorts, as well as the cancer consortia cohort. However, there were considerable inconsistencies between the two results. For example, in the UKB plus FinnGen cohort, vitamin C was found to be associated with a decreased risk of colorectal cancer (OR = 0.822, 95% CI = 0.698 to 0.968,
P = 0.0187) (Fig.
3), which aligns with the results of a previous MR study [
69]. However, this association was not replicated in the cancer consortia analysis in our study. Similarly, vitamin B12 was associated with an increased risk of non-invasive ovarian cancer in the cancer consortia dataset analyses (OR = 1.348, 95% CI = 1.106 to 1.643,
P = 0.0031) and was statistically significant after Bonferroni’s correction for the number of exposures (
P < 0.0036) (Fig.
4). This was consistent with a previous MR study [
22], but the result was not replicated in the UKB plus FinnGen meta-analysis dataset in our study. Only one association, that of iron and colorectal cancer, was statistically significant in both analyses. Nevertheless, this association showed a borderline p-value (0.01 <
P < 0.05) in both results and a sensitivity analysis was not feasible because it was supported by only two SNPs. This discrepancy indicates that the seemingly robust associations reported in previous MRs need to be carefully assessed to confirm their true causality.
Associations detected on MR may raise awareness regarding the potential harm of micronutrient supplementation, which should be considered when conducting RCTs. For example, the present study and another MR [
15] identified high serum magnesium levels as a risk factor for breast cancer; clinicians conducting RCTs with magnesium supplementation should be cautious about breast cancer as a potentially negative outcome and carefully consider whether to include individuals with a high risk of developing breast cancer. Additionally, results from robustly conducted MR may serve as secondary evidence in clinical decision-making before sufficiently powered RCTs are fully conducted. Results from our MR and previous MR studies [
15,
16] may imply that excessive intake of magnesium or vitamin B12 (via diet or supplements) can potentially be harmful, especially for individuals without nutritional deficiency and who have a high risk of developing breast cancer or colorectal cancer.
The strengths of our study include its extensive scope, examining over 600 potential causal associations in a consistent manner and analyzing cancer outcome data from the UKB, FinnGen, and various cancer consortia. Additionally, the MR design is inherently less likely to be biased compared to those in classical observational studies, and our MR analysis also reflected the effect of lifelong exposure to micronutrients, thereby assessing long-term risks that may not be moderated by relatively short-term interventions [
70].
Nevertheless, this study has several limitations. First, the number of SNPs was small, ranging from 1 to 9, and some exposures had less than four genome-wide significant SNPs; thus, tests for potential pleiotropy could not be performed. Second, although we used the largest available micronutrient and cancer GWASs available to our knowledge to report the most powered associations, not all 308 associations using the UKB plus FinnGen data were sufficiently powered, partly because of our stringent IV selection method. Null associations with low power should be interpreted cautiously to avoid false negative results. Third, because full summary statistics data were not available for most exposures, bidirectional MR analysis was not possible, and some exposure GWASs of micronutrients were adjusted for cancer risk factors (i.e., mediators) and cancer status, potentially leading to collider bias [
71]. For example, the GWASs of vitamins A1 and E were adjusted for BMI, cholesterol level, and cancer status (Additional file
2: Table S1). Fourth, we could not assess micronutrients, such as vitamin K, with no appropriate GWAS for MR analyses that may affect cancer outcomes [
72]. Fifth, participants in the UKB and FinnGen are likely to represent well-nourished populations without nutritional deficiency, and the observed associations may differ in populations with nutritional deficiency [
8], which were not assessed here. Associations may also vary according to sex [
9], and sex-stratified MR analyses were not available because of the absence of GWAS results for exposure. Sixth, we meta-analyzed the UKB and FinnGen cohorts; however, the ancestry of the UKB and FinnGen cohorts may have been slightly different, and participants in the FinnGen cohort had fewer close genetic relatives than those in the UKB cohort [
73], potentially leading to heterogeneity in the association effect between the cohorts. Additionally, we restricted the study sample to individuals of European ancestry to minimize the population stratification bias. This, in turn, prevents our findings from generalizing to other ancestries. Finally, two-sample MR assumes that the relationship between exposure and outcome is linear; thus, we might not have detected true nonlinear relationships between micronutrients and cancer.
Acknowledgements
The authors acknowledge the efforts of cancer consortia for providing high-quality GWAS summary statistics for researchers. The breast cancer genome-wide association analyses for BCAC and CIMBA were supported by Cancer Research UK (PPRPGM-Nov20\100002, C1287/A10118, C1287/A16563, C1287/A10710, C12292/A20861, C12292/A11174, C1281/A12014, C5047/A8384, C5047/A15007, C5047/A10692, C8197/A16565) and the Gray Foundation, The National Institutes of Health (CA128978, X01HG007492- the DRIVE consortium), the PERSPECTIVE project supported by the Government of Canada through Genome Canada and the Canadian Institutes of Health Research (grant GPH-129344) and the Ministère de l’Économie, Science et Innovation du Québec through Genome Québec and the PSRSIIRI-701 grant, the Quebec Breast Cancer Foundation, the European Community’s Seventh Framework Programme under grant agreement n° 223175 (HEALTH-F2-2009-223175) (COGS), the European Union's Horizon 2020 Research and Innovation Programme (634935 and 633784), the Post-Cancer GWAS initiative (U19 CA148537, CA148065 and CA148112—the GAME-ON initiative), the Department of Defence (W81XWH-10-1-0341), the Canadian Institutes of Health Research (CIHR) for the CIHR Team in Familial Risks of Breast Cancer (CRN-87521), the Komen Foundation for the Cure, the Breast Cancer Research Foundation and the Ovarian Cancer Research Fund. All studies and funders are listed in Zhang et al. [24] (Nat Genet, 2020). We want to acknowledge the participants and investigators of the FinnGen study.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.