Relationships between minerals’ intake and blood homocysteine levels based on three machine learning methods: a large cross-sectional study

Homocysteine (Hcy) is formed from methionine and elevated levels (≥15 µmol/L, hyperhomocysteinemia, hHcy) are linked to increased disease risk via hypomethylation, oxidative stress, and cellular damage. hHcy prevalence in China is high (37.2% in a meta-analysis). While B vitamins (B6, B12, folate) are well-known regulators of Hcy metabolism, evidence also suggests minerals may influence Hcy via roles in enzymes and as cofactors (e.g., Paraoxonase 1, betaine-Hcy methyltransferase). Prior studies have shown associations between single minerals (e.g., calcium, zinc, selenium) and Hcy/hHcy, but people consume minerals as mixtures with potential synergistic/antagonistic interactions, and the overall impact and relative contributions remain unclear. The study hypothesizes that mixed intake of multiple minerals is associated with reduced hHcy and aims to quantify both joint and individual mineral effects.

Limited prior studies assessed specific minerals and Hcy: dietary calcium intake was negatively associated with serum Hcy in postmenopausal women; plasma zinc concentrations showed a non-linear relationship with hHcy risk; higher blood selenium was linked to lower hHcy prevalence. Mechanistically, minerals act as enzyme components/cofactors in Hcy metabolism (e.g., calcium-dependent Paraoxonase 1; zinc-containing betaine-Hcy methyltransferase). However, evidence on the combined effect of multiple minerals is scarce, and few studies have applied mixture modeling approaches in nutrition, underscoring the need to assess joint mineral effects and individual contributions using machine learning methods.

Design and population: Cross-sectional analysis of SSACB baseline (2016) from two Shanghai districts (Songjiang, Jiading). Communities and neighborhood committees were sampled; trained staff conducted questionnaires, FFQ, exams, and biospecimen collection. Inclusion: adults 20–74 years (n=44,890). Exclusions: missing FFQ, implausible energy intake (<800 or >4000 kcal/d men; <500 or >3500 kcal/d women), missing key info. Final n=38,273. Ethics: Fudan University IRB (2016-04-0586); informed consent; Declaration of Helsinki. Exposure assessment: A validated FFQ (29 food groups) collected intake; reliability for foods 0.36–0.54 and nutrients 0.39–0.60; validity for foods 0.20–0.41 and nutrients 0.12–0.42. Intakes of calcium, phosphorus, sodium, potassium, magnesium, iron, copper, zinc, selenium, manganese and total energy were computed via the Chinese Food Composition Table (2nd ed.). Mineral intakes were log-transformed for normality. Outcome assessment: Morning fasting venous blood Hcy measured by fluorescence chromatography. hHcy defined as Hcy ≥15 µmol/L. Hcy was analyzed as log-transformed continuous and as dichotomous hHcy (1/0). Covariates: Selected via Directed Acyclic Graph approach. Variables included sex, age, education, marital status, smoking (never/cessation/current), alcohol drinking (≥3 drinks/week for >6 months vs no), tea drinking (same threshold), energy intake, physical activity (METs; quartiles), BMI (kg/m²; categories: <18.5, 18.5–23.9, 23.9–27.9, ≥28), hypertension, coronary heart disease, diabetes, and intakes of vitamins B6 and B12. Anthropometrics followed WS/T424-2013; BMI calculated from measured height and weight (average of three measures). Statistical analysis: (1) Descriptive comparisons between non-hHcy and hHcy using t-tests and chi-square tests. (2) Single-mineral associations with Hcy (multiple linear regression) and hHcy (multivariate logistic regression), adjusting for all covariates; Benjamini–Hochberg adjusted p-values. (3) Mixture models: WQS regression (training/validation split 4:6; component weight threshold >0.1 considered influential); Quantile g-computation (Qg-comp.boot to test linearity; Qg-comp.noboot linear model with quintiles; weights with positive/negative directions, threshold >0.05); BKMR (allows non-linearities and interactions; posterior inclusion probability, PIP, threshold 0.5 for importance). Given BKMR computational limits, a 10% stratified subsample by hHcy, sex, and age group was used (PROC SURVEYSELECT in SAS 9.4). All models adjusted for the same covariates; mineral intakes log-transformed. Software: SAS 9.4 and R 4.3.2; two-sided P<0.05.

- Study sample: 38,273 adults (13,726 males; 24,547 females); mean age 55.9±11.1 years. hHcy prevalence 21.6%. - Traditional models (adjusted): • Multiple linear regression for Hcy: Higher intakes of calcium, phosphorus, potassium, magnesium, iron, zinc, copper, and manganese were associated with lower Hcy (all adjusted p<0.05). Sodium and selenium were not significant after adjustment. • Logistic regression for hHcy: Higher intakes of calcium (OR 0.793; 95% CI 0.734–0.858; adj p=0.0002), phosphorus (0.538; 0.454–0.637; p=0.0002), potassium (0.741; 0.658–0.835; p=0.0002), magnesium (0.682; 0.585–0.795; p=0.0002), iron (0.812; 0.689–0.958; p=0.0173), zinc (0.579; 0.483–0.695; p=0.0002), selenium (0.827; 0.746–0.916; p=0.0004), and copper (0.851; 0.786–0.920; p=0.0002) were associated with lower hHcy risk; manganese (0.886; 0.774–1.014; p=0.0881) and sodium (0.992; 0.941–1.045; p=0.7570) were not significant. - WQS mixture model: Mixed minerals were negatively associated with Hcy (estimate −0.201; SE 0.062; p=0.001) and hHcy (estimate −0.074; SE 0.023; p=0.002). Calcium had the highest weight in the joint effect (e.g., mean weight ≈0.70 for Hcy, 0.63 for hHcy). - Qg-comp mixture model: Joint effect estimates were small and not statistically significant for Hcy (−0.024; SE 0.137; p=0.8874) and hHcy (−0.006; SE 0.030; p=0.8453). Iron had the highest positive weight (e.g., ~0.801 for Hcy and 0.628 for hHcy), while negative weights were led by certain minerals (reported as manganese or magnesium depending on outcome/context in figures), indicating heterogeneity in directionality across minerals. - BKMR (10% stratified subsample): The mixture was negatively associated with both Hcy and hHcy across higher percentiles of exposure (p<0.05). PIP rankings indicated importance of most minerals except sodium. Highest PIPs: for Hcy, phosphorus (0.7232) followed by zinc (0.6574), manganese (0.6156), magnesium (0.6082), calcium (0.5928), selenium (0.5832), potassium (0.5436), iron (0.5290); for hHcy, zinc (0.6904) followed by phosphorus (0.6802), iron (0.6106), potassium (0.5838), copper (0.5442), magnesium (0.5034). Univariate exposure–response functions generally showed negative associations for several minerals when others were held at median. Evidence of interactions among some minerals (notably fewer between zinc and phosphorus) was observed for hHcy risk.

Findings support the hypothesis that higher mixed mineral intake is associated with lower blood Hcy and reduced hHcy risk. Traditional single-mineral models identified inverse associations for multiple minerals but cannot adequately account for high intercorrelations, non-linearities, and interactions. Mixture models (WQS, Qg-comp, BKMR) collectively indicated a protective joint effect of minerals, with varying individual contributions: calcium featured prominently in WQS; iron had high positive weights in Qg-comp; and BKMR emphasized the importance of phosphorus and zinc (via PIPs) and suggested interactions among minerals. These results align with biological plausibility: minerals serve as cofactors or constituents of enzymes in Hcy metabolism (e.g., calcium-dependent Paraoxonase 1; zinc-containing betaine-Hcy methyltransferase; selenium effects on methionine synthase activity). The consistency across different modeling paradigms that address mixture effects strengthens the inference that balanced mineral intake may support healthy Hcy metabolism. Nevertheless, discrepancies in the magnitude and direction of individual mineral weights across methods highlight the complexity of nutrient mixtures and the need for triangulation and further mechanistic studies.

Higher intake of a mixture of minerals is associated with lower blood Hcy levels and reduced hHcy prevalence. Individual minerals contribute differently to the overall protective effect. Ensuring a balanced diet that provides adequate minerals may help maintain healthy Hcy levels. Future research should clarify mechanisms, incorporate both dietary intake and biomarker (blood) measures of minerals, and evaluate impacts on additional health outcomes using longitudinal designs.

- Cross-sectional design limits causal inference. - Dietary assessment via FFQ may be prone to recall error; reported reliability/validity coefficients were modest. - Internal exposure (blood mineral concentrations) was not measured; relationships between intake and circulating mineral levels are unknown in this study. - Sex imbalance (more females; higher hHcy prevalence in males) may influence results despite adjustment. - Chinese Food Composition Table includes a limited set of minerals; unassessed minerals may contribute to outcomes. - BKMR conducted on a 10% stratified subsample due to computational constraints, which may affect precision of mixture effect estimates.