This 16-week, randomized, double-blind, placebo-controlled trial was conducted between October 2022 and November 2023 at community and two hospital sites. The study protocol was approved by the Ethics Committee of the National Institute for Nutrition and Health, Chinese Centre for Disease Control and Prevention (No. 2021-015) and has been made publicly accessible on the clinical trial registry website (ChiCTR2100047666). All pre-specified primary (change in fasting plasma glucose) and secondary outcomes (changes in insulin, HOMA-IR, HbA1c, glycated albumin, hs-CRP, and IL-6) were explicitly listed in the trial registry record. The non-targeted metabolomic profiling reported in this manuscript was a pre-planned exploratory analysis, as indicated in the registered protocol. The trial was conducted in accordance with the principles of the Declaration of Helsinki. All participants provided written informed consent.

We enrolled 71 community-dwelling adults with pre-diabetes, defined as fasting plasma glucose (FPG) ≥5.6 mmol/L and/or glycated haemoglobin (HbA1c) 5.7%−6.5% (14), and concurrent hypomagnesemia. Hypomagnesemia was defined as plasma Mg concentration ≤ 0.80 mmol/L, a cutoff consistent with clinical guidelines and previous studies in similar populations (15). A complete CONSORT flow diagram of participant screening, enrolment, allocation, follow-up, and analysis is provided in Figure 1. Detailed inclusion and exclusion criteria are summarized in Supplementary Table S1. Key exclusion criteria included: (1) diagnosis of diabetes mellitus; (2) current use of medications known to affect glucose or mineral metabolism (e.g., diuretics, insulin, oral hypoglycemic agents).

Eligible participants were randomly assigned in a 1:1 ratio to either the magnesium supplement group or the placebo group. The randomization sequence was computer-generated by an independent statistician (W.WX) using block randomization (block size of four), stratified by age (< 65 vs. ≥65 years), sex, baseline plasma Mg concentration ( ≤ 0.70 vs. >0.70 mmol/L), and baseline FPG (< 6.1 vs. ≥6.1 mmol/L). Allocation concealment was ensured using sequentially numbered, opaque, sealed envelopes (SNOSE). The envelopes were prepared by the independent statistician and kept securely by the study pharmacist (P.M), who was not involved in participant recruitment or outcome assessment.

Participants, research staff (Y.JX) involved in participant management and outcome assessment, laboratory personnel, and data analysts were all blinded to group assignment throughout the trial. The magnesium oxide tablets (providing 360 mg of elemental magnesium per day) and the visually identical placebo tablets (composed primarily of starch) were manufactured, coded, and supplied by Amway Corporation. The blinding was maintained until all statistical analyses were completed.

Participants were instructed to take two tablets (either magnesium or placebo) once daily with a meal for 16 weeks. Compliance was monitored at each monthly visit via pill counts of returned medication bottles and review of participant medication diaries. Adherence was calculated as (number of tablets taken/number of tablets prescribed) × 100%. Good compliance was defined as taking ≥80% of the prescribed tablets. Any adverse events or side effects, particularly gastrointestinal symptoms (e.g., diarrhoea, abdominal discomfort), were recorded at each follow-up visit using a structured questionnaire.

Data were collected at baseline (Week 0), an interim visit (Month 2), and at the end of the intervention (Month 4). After a 10–12 hour overnight fast, venous blood samples were collected into appropriate vacutainers for plasma, serum, and whole blood separation. A spot urine sample was also collected.

Anthropometric measurements included body weight, height, waist circumference, and blood pressure (measured in triplicate after a 5-min rest). Dietary intake was assessed using a combination of tools to improve accuracy. At baseline, three consecutive days of 24-hour weighed dietary records were conducted, supplemented by a validated semi-quantitative food frequency questionnaire (FFQ) to assess habitual magnesium intake.

Plasma magnesium concentration was determined by inductively coupled plasma mass spectrometry (ICP-MS). Glucose metabolism markers, including FPG, insulin, C-peptide, HbA1c (by HPLC), and glycated albumin, were measured. Homeostasis model assessment of insulin resistance (HOMA-IR) was calculated. HOMA-IR was used to measure the insulin resistance (IR) of hepatic and was calculated as insulin (μU/mL) × fasting plasma glucose (FPG, mmol/L)/22.5 (14). Inflammatory markers [high-sensitivity C-reactive protein (hs-CRP) and interleukin-6 (IL-6)] were assayed by standard clinical biochemical methods and electrochemical luminescence, respectively. Liver and kidney function tests were performed for safety monitoring.

Plasma samples underwent non-targeted metabolomic analysis using ultra-high-performance liquid chromatography coupled with high-resolution mass spectrometry (UHPLC-MS). Chromatographic separation was performed on a Vanquish UHPLC system (Thermo Scientific, Massachusetts, USA), and mass spectrometric detection was conducted on a Q Exactive HF-X hybrid quadrupole-Orbitrap mass spectrometer (Thermo Scientific, Massachusetts, USA) in both positive and negative electrospray ionization modes. Metabolites were extracted with a methanol/acetonitrile solvent system. Data processing included peak alignment, identification, and normalization. Multivariate statistical analyses, including principal component analysis (PCA) and orthogonal partial least squares-discriminant analysis (OPLS-DA), were performed to assess group separation. Differential metabolites were initially screened using variable importance in projection (VIP) >1.0 from OPLS-DA and p < 0.05 from Student's t test. To control for false discoveries, we further applied a false discovery rate (FDR) correction (Benjamini–Hochberg method), with FDR < 0.05 as the significance threshold for differential abundance. Metabolite identification remained putative pending MS/MS validation. Pathway enrichment analysis was conducted using the Kyoto Encyclopaedia of Genes and Genomes (KEGG) database based on these putatively identified metabolites, and results are presented as exploratory and hypothesis-generating.

The primary efficacy endpoint was the change from baseline to Month 4 in FPG. Secondary endpoints included changes in insulin, HOMA-IR, HbA1c, glycated albumin, hs-CRP, and IL-6. Analyses were performed following the intention-to-treat (ITT) principle, including all randomized participants (n = 71). For the primary and continuous secondary outcomes, missing data at Month 4 were imputed using the last observation carried forward (LOCF) method as a primary approach. A sensitivity analysis using complete cases only was also performed.

Baseline characteristics were compared between groups using independent samples t-tests or Mann–Whitney U-tests for continuous variables and chi-square or Fisher's exact tests for categorical variables. For the primary and key secondary outcomes, between-group differences in the change from baseline to Month 4 were analyzed using analysis of covariance (ANCOVA), adjusting for the respective baseline value as a covariate. Model assumptions (normality of residuals and homoscedasticity) were assessed using Shapiro–Wilk and Levene's tests, respectively; no substantial violations were observed. Results are presented as adjusted mean differences with 95% confidence intervals (CIs). Within-group changes from baseline were assessed using paired t-tests.

For the exploratory non-targeted metabolomics analysis, orthogonal partial least squares-discriminant analysis (OPLS-DA) was used to visualize group separation. Model validity was evaluated using a permutation test (n = 200) with the Q² intercept p-value < 0.05 indicating robustness against overfitting. Differential metabolites were identified using variable importance in projection (VIP) >1.0 from OPLS-DA and false discovery rate (FDR)-corrected p < 0.05 (Benjamini–Hochberg method) to control for multiple comparisons across metabolites. Pathway enrichment analysis was conducted using the Kyoto Encyclopaedia of Genes and Genomes (KEGG) database and is presented as hypothesis-generating and descriptive in nature.

All reported p values are two-sided. A p value < 0.05 was considered statistically significant for primary and pre-specified secondary outcomes. Given the exploratory nature of the metabolomics analysis and the multiplicity of secondary/exploratory outcomes, findings from these analyses should be interpreted as hypothesis-generating. Statistical analyses were performed using SPSS Statistics (version 26.0, IBM Corp.) and R (version 4.3.0).

The sample size was calculated based on the primary endpoint (change in FPG). Based on a previous artical of magnesium supplementation in individuals with pre-diabetes (11), we assumed a between-group difference in FPG change of 0.66 mmol/L (SD = 0.65 mmol/L). To achieve 90% power with a two-sided alpha of 0.05, 25 participants per group were required. Anticipating a 20% dropout rate over 16 weeks, we aimed to recruit at least 30 participants per group, resulting in a final target sample size of 71 randomized participants.

All adverse events (AEs) were recorded throughout the study. The incidence of gastrointestinal AEs (e.g., diarrhoea, nausea, abdominal pain), a known side effect of magnesium oxide, was specifically monitored and compared between groups.

Of the 780 individuals initially screened, 71 eligible participants with pre-diabetes and hypomagnesemia were enrolled and randomized. During the 16-week intervention, six participants withdrew for personal reasons, resulting in 65 completers (34 in the Mg group, 31 in the placebo group). Overall compliance was 91.6%. The flow of participants is detailed in Figure 1. The intervention was well-tolerated, with no intervention-related adverse events reported in either group. All participant discontinuations were for unrelated reasons (Figure 1).

Baseline characteristics are presented in Table 1. While demographic variables (age, BMI, waist circumference, sex, and smoking status) were comparable between groups, significant baseline imbalances were noted for two metabolic parameters, including insulin (12.0 ± 7.7 vs. 16.6 ± 10.7 μIU/ml, p = 0.040) and HOMA-IR (3.3 ± 2.2 vs. 5.4 ± 5.1, p = 0.034). All subsequent analyses were adjusted for these baseline differences as detailed in the statistical methods.

Dietary intake at baseline, assessed by both a three-day 24-hour dietary weighed and a food frequency questionnaire (FFQ), showed no significant differences between groups in energy, macronutrients, or dietary magnesium intake (Table 2). Dietary calcium intake assessed by FFQ was higher in the magnesium group at baseline (p = 0.049). Dietary intake remained stable throughout the study, with no significant between-group difference in magnesium intake at the end of the intervention (p = 0.899), as illustrated in Figure 2.

Changes in primary and secondary biochemical outcomes from baseline to 4 months are summarized in Table 3 and detailed in Supplementary Table S2. The primary analysis using ANCOVA, adjusted for baseline values, sex, insulin, and HOMA-IR, showed that magnesium supplementation significantly increased serum magnesium levels compared to placebo (effect size: 0.056 mmol/L, 95% CI: 0.028 to 0.085; p < 0.001).

Regarding glycemic parameters, magnesium supplementation led to a significant reduction in fasting plasma glucose (FPG) compared to placebo (effect size: −0.497 mmol/L, 95% CI: −0.818 to −0.176; p = 0.003). The change in HOMA-IR favored the magnesium group but was not statistically significant in the adjusted model (effect size: −0.495, 95% CI: −1.417 to 0.426; p = 0.296). After applying False Discovery Rate (FDR) correction for multiple comparisons (Supplementary Table S3), the between-group difference in FPG remained significant (q < 0.05). No statistically significant between-group differences were observed for HbA1c, insulin, C-peptide, glycated albumin (GA), interleukin-6 (IL-6), or high-sensitivity C-reactive protein (hs-CRP) in either the primary or FDR-corrected analyses (Table 3; Supplementary Table S3). The percentage changes in key glucose metabolism indices are visually compared in Figure 3.

Serum samples from the magnesium group at baseline (M0) and 4 months (M4) underwent exploratory non-targeted metabolomic analysis. Quality control included pooled QC samples and internal standards for normalization. In PCA, QC samples clustered tightly and separately from experimental samples, indicating analytical stability (Supplementary Figure S1).

Orthogonal partial least squares-discriminant analysis (OPLS-DA) was applied to the data. The score plot showed separation between M0 and M4 samples (Figure 4A). A permutation test (n = 200) yielded a Q² intercept p-value < 0.05, supporting the robustness of the model.

Using thresholds of |log₂fold change| >0.584, VIP >1, and an unadjusted p < 0.05, 387 metabolites met criteria for altered relative abundance (220 down, 167 up). To reduce false discovery, we applied FDR correction (FDR < 0.05), resulting in 52 significantly altered metabolites (29 up-regulated, 23 down-regulated; Figure 4B). The top 20 metabolites by p-value, with their putative identifications, are listed (Supplementary Table S4).

These putatively identified metabolites were submitted to KEGG pathway enrichment analysis. Enriched pathways included “Lipid and atherosclerosis”, “Cholesterol metabolism”, “Insulin resistance”, and others (Figure 4C). Metabolites contributing to OPLS-DA separation are shown in a loading plot (Supplementary Figure S2). Given the exploratory nature of the analysis, these pathway associations are descriptive and not indicative of mechanistic causality.