Methyl Donor Nutrient Intake and Incidence of Type 2 Diabetes: Results From Three Large U.S. Cohorts

The prevalence of type 2 diabetes is growing worldwide and continues to remain one of the major global public health challenges (1). In the U.S., the estimated economic cost of diabetes was $327 billion as of 2017 (2). Therefore, it is important to identify risk factors in the development of type 2 diabetes to develop targeted prevention strategies.

There is increasing evidence that DNA methylation may play a role in the development of metabolic disorders, including type 2 diabetes and its associated complications (3–5). DNA methylation can be influenced by environmental factors, such as diet (6). These dietary factors include intake of methyl donor nutrients involved in one-carbon metabolism, such as B vitamins (B₂, B₆, B₁₂, and folate), methionine, choline, and betaine. Intake of these nutrients may influence risk of type 2 diabetes via epigenetic changes (7). Only a few studies have investigated the association between methyl donor nutrients and type 2 diabetes. In a randomized controlled trial among 4,252 women at high risk of cardiovascular disease (CVD), Song et al. (8) reported no significant effect of combined supplementation with vitamin B₆, vitamin B₁₂, and folate on the risk of type 2 diabetes. However, in the same study a trend toward a protective effect in participants with family history of type 2 diabetes was reported. These findings suggest that vitamin B₆, vitamin B₁₂, and folate may be beneficial for a specific subgroup that is predisposed to development of type 2 diabetes. Additionally, several small cross-sectional and prospective studies have implicated vitamin B₁₂ deficiency in the development of gestational diabetes mellitus (9). There is a paucity of evidence for the association between intakes of these nutrients and type 2 diabetes risk in large prospective settings. Furthermore, whether the association is similar for food and supplemental sources remains unknown.

In the current study, we examined the association between intake of methyl donor nutrients, specifically vitamins B₂, B₆, and B₁₂ and folate, and risk of type 2 diabetes in three large prospective cohorts in the U.S. We also examined whether associations varied when these nutrients were consumed from foods or supplements. Other methyl donor nutrients, including methionine, choline, and betaine, were also examined in supplementary analyses.

This study included three prospective U.S. cohort studies: the Nurses’ Health Study (NHS), Nurses’ Health Study 2 (NHS2), and the Health Professionals Follow-Up Study (HPFS). NHS was initiated in 1976 and recruited 121,700 female registered nurses between the ages of 30 and 55 years from 11 U.S. states. NHS2 was initiated in 1989 and included 116,429 female registered nurses between the ages of 25 and 42 years from 14 states. HPFS was initiated in 1986 with 51,529 male health professionals aged 40–75 years. In all three studies, information on medical history, lifestyle factors, and occurrence of chronic diseases was collected through questionnaires at baseline and every 2 years during follow-up. The response rate per cycle was ∼90% for all cohorts. In this study, the 1984, 1991, and 1986 cycles were baselines for NHS, NHS2, and HPFS, respectively, because most covariates of interest were comprehensively measured from this cycle onwards. We excluded participants with baseline history of type 1 diabetes, type 2 diabetes, gestational diabetes mellitus, CVD, or cancer. We also excluded those with missing dietary data or implausible energy intakes at baseline (500–3,500 kcal/day for women and 800–4,200 kcal/day for men). The population for analyses consisted of 75,430 women from the NHS, 87,953 women from the NHS2, and 40,261 men from the HPFS. Study protocols for all cohorts were approved by the institutional review boards of Brigham and Women’s Hospital and Harvard T.H. Chan School of Public Health, and all participants provided informed consent.

Dietary data were collected every 2–4 years through a validated semiquantitative food-frequency questionnaire (FFQ) as previously described (10,11). Participants were asked how often they consumed a prespecified serving of ∼130 food items over the year preceding the questionnaire completion. Frequency categories included the following: “never or less than once/month,” “1–3 times/month,” “1 time/week,” “2–4 times/week,” “5–6 times/week,” “1 time/day,” “2–3 times/day,” “4–5 times/day,” or “≥6 times/day.” Participants also reported whether they regularly took supplements, what type (including multivitamins, vitamin B₆, vitamin B₁₂, niacin, B-complex vitamins, or other), and how often (from “2 or less per week,” “3–5 per week,” “6–9 per week,” or “10 or more per week”). We calculated dietary intakes by multiplying the consumption frequency of each food by the nutrient content of the portion specified. Nutrient values of foods and supplements were obtained from the Harvard University Food Composition Database, derived mainly from the U.S. Department of Agriculture. Intakes of vitamin B₂, vitamin B₆, vitamin B₁₂, and folate from foods and supplements were calculated separately from food and supplement sources and added together for estimation of total intakes of each nutrient. Methionine, choline, and betaine intakes were only estimated from food sources, as supplemental intake of these nutrients was very low in the study population.

Participants in all three cohorts were asked whether they were diagnosed with type 2 diabetes by a physician every 2 years. Participants who self-reported physician-diagnosed type 2 diabetes were sent a supplementary questionnaire for confirmation of diagnosis (12,13). We categorized participants as type 2 diabetes case subjects if they met at least one of the following criteria (according to the National Diabetes Data Group) (14): one or more classic symptoms (excessive thirst, polyuria, or frequent urination) plus fasting plasma glucose of ≥7.8 mmol/L or random plasma glucose of ≥11.1 mmol/L, at least two elevated plasma glucose levels on two different occasions (i.e., fasting plasma glucose of ≥7.8 mmol/L or random plasma glucose of ≥11.1 mmol/L and/or 2-h blood glucose concentrations of at least 11.1 mmol/L during oral glucose tolerance testing) in the absence of symptoms, or treatment with hypoglycemic drugs (insulin or oral hypoglycemic agent). For cases identified after 1998, we applied the American Diabetes Association criteria, in which the threshold for fasting plasma glucose was changed from ≥7.8 mmol/L to ≥7.0 mmol/L (15). Starting in 2010, HbA_1c ≥6.5% was added to the diagnosis criteria (16).

Height was measured in the cohort enrollment questionnaire (1976 for NHS, 1989 for NHS2, and 1986 for HPFS). Biennial follow-up questionnaires were used to ascertain and update information on weight, smoking, marital status, hypertension, hypercholesterolemia, level of physical activity, postmenopausal status and postmenopausal hormone use (NHS and NHS2 only), and oral contraceptive use (NHS2 only). Race was ascertained in 1992 and 2004 for NHS, 1989 and 2005 for NHS2, and 1986 for HPFS. Family history of type 2 diabetes in first-degree relatives was assessed in 1988 and 1992 in NHS; 1989, 1997, and 2001 in NHS2; and 1987 and 1992 in HPFS. Marital status was updated every 4 years starting in 1992 in NHS and 1991 in NHS2 and every 2 years from 1986 in HPFS. BMI was calculated as weight in kilograms divided by the square of height in meters. Intakes of total energy, alcohol, cereal fiber, animal protein, red meat, polyunsaturated fat, saturated fat, and Alternative Healthy Eating Index (AHEI) scores were estimated from the FFQs every 2–4 years.

We calculated a cumulative updated average for total, supplemental, and food intakes of each of the methyl donor nutrients for each cohort to better estimate long-term intakes and to reduce within-person variation. Missing data on dietary intake during follow-up were replaced by the cumulative average of previous assessments. We stopped updating diet upon report of cancer or CVD as changes in diet following disease diagnosis may confound the association between diet and type 2 diabetes (17). To limit the influence of outliers and to detect potential nonlinear associations, we computed quintiles of each nutrient. We calculated person-time for each participant from the baseline questionnaire return date (1984 in NHS, 1991 in NHS2, and 1986 in HPFS) until type 2 diabetes diagnosis, death, or end of follow-up (2016 for NHS and HPFS and 2017 in NHS2).

Hazard ratios (HRs) were estimated with Cox proportional hazard regression analyses with time-varying covariates for assessment of the association between quintiles of each methyl donor nutrient and risk of type 2 diabetes. Assessment for all nutrients included adjustment for total energy intake with use of the residual method (18). Age was included as the timescale (model 1). Model 2 additionally included adjustment for smoking (never, past, current: 1–14, 15–24, or ≥25 cigarettes/day), marital status (currently married, widowed, divorced or separated, or never married), family history of type 2 diabetes (yes or no), hypertension (yes or no), hypercholesterolemia (yes or no), level of physical activity (<3, 3–9, 9–18, 18–27, 27–42, or ≥42 MET h/week), total energy intake (quintiles), alcohol consumption (0, 0.1–5.0, 5.0–10.0, 10.0–15.0, or ≥15.0 g/day), race (White, Black, or other), BMI (<21, 21–23, 23–25, 25–27, 27–30, 30–33, 33–35, 35–40, or ≥40 kg/m²), postmenopausal status and hormone use in NHS and NHS2 (premenopausal or postmenopausal: never, current, or past hormone use), and oral contraception use in NHS2 only (current, past, never use). Model 3 additionally included nutritional factors that may be potential confounders in the association with diabetes: cereal fiber, animal protein, and polyunsaturated fat–to–saturated fat (PUFA:SFA) ratio. We also additionally adjusted models for B₁₂ from food for red meat intake. For analyses stratified by source, models 2 and 3 for methyl donor nutrients from foods were additionally adjusted for multivitamin use (yes or no) (models 4 and 5), whereas models 2 and 3 for methyl donor nutrients from supplements included additional adjustment for AHEI score (models 6 and 7). The proportional hazards assumption of the Cox model was checked with performance of a test for heterogeneity of exposure over time. We tested for effect modification by age, family history of type 2 diabetes, and alcohol consumption by including cross product terms in the most adjusted models and checking the P value for significance. To determine whether the folic acid fortification implementation in 1998 affected the association between folate intake and risk of type 2 diabetes, we performed several sensitivity analyses. First, we continuously updated folate intake, instead of cumulatively updating. Second, we separated analyses for time periods before and after 1998. Analyses were performed per cohort and then combined in meta-analysis with use of inverse variance–weighted fixed-effects models: the Cochrane Q statistic, the I² statistic, and between-study coefficient of variation were used for assessment of heterogeneity. All analyses were performed with SAS, version 9.4, for UNIX (SAS Institute). A two-sided P value of <0.05 was considered statistically significant for main analyses and a Bonferroni-corrected P < 0.003 for interaction testing (three potential modifiers and five exposures: 0.05 / 15 = 0.003).

Baseline characteristics of participants are presented in Table 1. Generally, participants with higher intake of methyl donor nutrients tended to smoke less, drink less alcohol, were more physically active, were more likely to take multivitamins, and had better diet quality. Median intakes of vitamin B₂, vitamin B₆, vitamin B₁₂, and methionine remained similar between baseline and during follow-up, whereas intake of folate notably increased due to the folate fortification that was implemented in 1998 (19) (Supplementary Table 1).

We documented 19,475 cases of type 2 diabetes during 4,900,181 years of follow-up. The associations of quintiles of energy-adjusted intakes of B vitamins with risk of type 2 diabetes are presented in Table 2. In the multivariable-adjusted meta-analysis, participants in the highest quintile (Q5) of total vitamin B₂ intake had lower risk of type 2 diabetes compared with those in the lowest quintile (Q1) (model 3, HR 0.93 [95% CI 0.89, 0.98]; P_trend = 0.02). With stratification by source, higher intake of vitamin B₂ from food was associated with a 16% lower type 2 diabetes risk, after adjustment for multivitamin use (model 5, Q5 vs. Q1, HR 0.84 [0.80–0.89]; P_trend < 0.0001), but supplemental vitamin B₂ intake was not associated with type 2 diabetes risk after we accounted for differences in overall diet quality (models 6 and 7).

A higher intake of total vitamin B₆ was also associated with lower risk of type 2 diabetes (model 3, Q5 vs. Q1, HR 0.93 [95% CI 0.89, 0.97]; P_trend = 0.07). However, with stratification by source, neither food (model 5, Q5 vs. Q1, HR 0.98 [0.93, 1.04]; P_trend = 0.54) nor supplemental vitamin B₆ (model 7, Q5 vs. Q1, HR 0.97 [0.92, 1.01]; P_trend = 0.44) was associated with type 2 diabetes.

We found no evidence of an association between total vitamin B₁₂ intake and type 2 diabetes risk (Table 2), but when analyses were stratified by source, participants in the highest quintile of vitamin B₁₂ from food sources had a higher risk of type 2 diabetes compared with those in the lowest quintile (model 5, HR 1.05 [95% CI 1.00, 1.11]; P_trend = 0.002). When we additionally adjusted for red meat intake this association was slightly attenuated (1.04 [0.99, 1.10]; P_trend = 0.004 for Q5 vs. Q1) (Supplementary Table 2). In contrast, supplemental vitamin B₁₂ intake was not associated with type 2 diabetes risk (model 7, Q5 vs. Q1, HR 0.96 [0.92, 1.00]; P_trend = 0.30).

For folate, participants in the highest quintile of total folate intake had lower type 2 diabetes risk than those in the lowest quintile in model 2 (Q5 vs. Q1, HR 0.92 [95% CI 0.87, 0.96]; P_trend = 0.88). This association did not remain after adjustment for other nutrients in model 3 (cereal fiber, animal protein, and PUFA:SFA ratio) found in foods high in folate (model 3, Q5 vs. Q1, HR 0.97 [0.93, 1.02]; P_trend = 0.87).

Associations of quintiles of methionine, choline, and betaine with risk of type 2 diabetes are presented in Supplementary Table 3. Similar to folate, higher methionine and betaine intake tended to be associated with higher risk of type 2 diabetes in model 2, but this association was mitigated after adjustment for the other nutrients in model 3. Choline intake was not associated with type 2 diabetes in adjusted models.

Cohort-specific associations were all similar (Supplementary Tables 3–7). We found no evidence of effect modification of the association between vitamin intake and type 2 diabetes risk by sex, alcohol intake, and family history of type 2 diabetes (P_interaction, all > 0.003); therefore, results are presented for the total population. In sensitivity analyses for the effect of folate fortification the results did not change (data not shown).

We hypothesized that food and supplemental intake of nutrients involved in one-carbon metabolism may influence risk for type 2 diabetes. Of these nutrients, higher intakes of total vitamin B₂ and total vitamin B₆ showed an inverse association with risk of type 2 diabetes. When we looked at food versus supplemental sources, only vitamin B₂ from food had a significant inverse association with incident type 2 diabetes, with a 16% lower risk. This observation reinforces recommendations to get nutrients from whole foods over supplements (20). This could be because nutrients from foods may be more bioavailable or work in synergy with other nutrients in the food matrix. It is not entirely clear why we no longer observed significant associations with B₆ when looking at food sources of the vitamin, but it is possibly due to reduced power in these subanalyses. Total intake of folate, vitamin B₁₂, methionine, betaine, and choline were not associated with risk of type 2 diabetes after we accounted for additional dietary factors.

There are several potential mechanisms that may explain a protective effect of vitamin B₂ and vitamin B₆ for type 2 diabetes risk. Like the other methyl donor nutrients, both play a role in one-carbon metabolism and therefore may influence DNA methylation. Such epigenetic changes can alter gene expression pathways related to metabolic disease, including glucose control (21). Additionally, both B vitamins have also been proposed to improve glucose metabolism and inflammation markers (22). Findings from clinical and animal studies suggest that vitamins B₂ and B₆ may decrease risk of type 2 diabetes through reduced inflammation, oxidative stress, and hyperglycemia (23–26). Further, results from animal studies suggest that supplementation of vitamin B₂ may be beneficial for glucose and insulin metabolism and also reduce diabetes complications (24,27). In an in vitro study, Bialy et al. (28) showed that treatment of adipocyte-macrophage cocultures with vitamin B₂ leads to a reduction of proinflammatory factors and an increase of anti-inflammatory factors.

Previous studies have provided conflicting evidence for an association between folate intake and type 2 diabetes risk. In a case-control study, Nilsson et al. (29) identified 251 CpG sites that were differentially methylated, of which the majority were hypomethylated, in type 2 diabetes case compared with control subjects. In the same study, circulating folate levels were positively correlated with the identified CpG sites, and folate levels were lower in participants with type 2 diabetes. Although causality cannot be inferred with this study design, results suggest that the association between folate and type 2 diabetes may be mediated through changes in DNA methylation. However, the lack of evidence for an association of folate and vitamin B₁₂ and weak association for vitamin B₆ that we observed is in line with findings from a randomized controlled trial among women at high risk of CVD, where no significant effects of supplementation of vitamin B₆, vitamin B₁₂, and folate on the risk of type 2 diabetes were observed (8). The authors did observe a reduction in type 2 diabetes risk among women with a family history of type 2 diabetes. We tested in our study population whether associations differed for participants with family history of type 2 diabetes, but no significant interaction was observed. In a more recent longitudinal analysis among Japanese women and men, dietary B₆ and folate were associated with lower type 2 diabetes risk over 5 years among women but not men, and B₁₂ was not associated with type 2 diabetes risk in either sex (30). Another recent longitudinal analysis in a Chinese population showed no association between baseline plasma B₁₂ concentration and incident type 2 diabetes over a median follow-up of 4.5 years (31). Although previous evidence is mixed for B₆ and folate, evidence from our study is strengthened by the longer follow-up time and large study population. In studies of B₁₂ investigators have more consistently found that there is no evidence of an association with type 2 diabetes.

Several limitations of this study should be considered. Due to the observational nature of the study design, residual confounding cannot be excluded. As dietary intake is self-reported through an FFQ, assessment is prone to measurement errors. There was low variation in vitamin intake, limiting the ability to detect associations for extreme high or extreme low intakes, and complete isolation of the effect of individual nutrients from their food sources is not possible. Therefore, observations may be related to the effect of foods rich in B vitamins, rather than the vitamins themselves, although we did try to adjust for other nutrients that have been associated with type 2 diabetes. Our study population was socioeconomically homogeneous, which may enhance internal validity but also reduces external validity. Therefore, results should be replicated in additional types of populations. However, strengths of this study include the large sample size with a high follow-up rate and its prospective design, which allowed us to assess diet before disease onset. Furthermore, the repeated dietary data enabled us to calculate cumulative average intake to represent a long-term diet. This reduces measurement error compared with estimating nutrient intake from a single assessment (32). We also had detailed information on nutrients from different sources and a wide range of information on lifestyle factors and other covariates, minimizing confounding.

To conclude, findings from this study suggest that higher intakes of vitamin B₂, especially from foods, are associated with a lower risk of type 2 diabetes. Overall, none of the methyl donor nutrients from supplements were associated with type 2 diabetes. This supports the dietary guidelines recommendation that nutrients from foods may be more beneficial over supplements whenever possible. Higher vitamin B₁₂ intake from food seems to be associated with higher risk of type 2 diabetes, which is likely due to consumption of animal products. Future studies are needed to replicate these findings in other populations.

Acknowledgments. The authors thank all of the participants in the NHS, NHS2, and HPFS for their tremendous contributions.

Funding. K.V.E.B. was supported by the Albert Renold Travel Fellowship from the European Foundation for the Study of Diabetes. C.M.S. and D.E.H. are supported by National Institutes of Health (NIH) Ruth L. Kirschstein National Research Service Award T32 CA 009001. J.-P.D.C. is a research scholar of Fonds de recherche du Québec – Santé (Quebec Health Research Funds). Cohort studies are supported by the following NIH grants: UM1 CA186107 (NHS), U01 CA176726 and U01 HL45386 (NHS2), and U01 CA167552 (HPFS).

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH.

Duality of Interest. J.-P.D.C. received speaker fees, consulting honoraria, and research funding from Dairy Farmers of Canada, outside the submitted work. S.N.B. is a scientific consultant for LayerIV for work outside the submitted manuscript. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. K.V.E.B., J.-P.D.C., T.V., O.H.F., Q.S., F.B.H., and S.N.B. contributed to the concept and design of the study. C.M.S., D.E.H., and K.V.E.B. conducted data analyses and wrote, reviewed, and edited the manuscript. All authors contributed to discussion and reviewed and edited the manuscript. All authors approved the final version of the manuscript. S.N.B. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.