This study enrolled 100 adults aged 30–70 years with newly diagnosed, treatment-naïve T2DM (NDT2D) from the endocrinology outpatient clinics of Pinggu Hospital (Beijing, China) between December 2016 and December 2017. The age range was selected to encompass the peak T2DM incidence window, while excluding individuals with monogenic diabetes (age <30 years) and polypharmacy confounders among older adults (age >70 years) (30, 31).

Inclusion criteria:1) Newly diagnosed T2DM per 1999 WHO criteria (32), confirmed by a 75-g oral glucose tolerance test (OGTT), with disease duration ≤12 months and no prior glucose-lowering medication use (or ≤1 month of prior use followed by a 3-month washout). 2) Body mass index (BMI) 24.0 -30.0 kg/m², classified as overweight (24.0-27.9 kg/m²) or obese(≥28.0 kg/m²) per Chinese obesity criteria (33). 3) HbA1c 7.0–9.0% (53–75 mmol/mol), aligned with efficacy-safety thresholds for oral monotherapy in the Chinese Diabetes Guidelines (2013) (34).

Excluded criteria comprised pregnancy, lactation, or intention to conceive; gastrointestinal diseases; infectious liver diseases (hepatitis B and C); endocrine diseases affecting glucose metabolism; psychiatric disorders; alcohol dependence; neoplasia; or any other severe disease that could affect the outcomes of the study. None of the participants had consumed or injected medicines or food containing antibiotics, hormones, or probiotics for at least 1 month before the initiation of the study and had a history of weight loss, dyslipidemia, hyperglycemia, or hypertension.

All participants underwent on-site screening, including informed consent (detailing gastrointestinal risks of acarbose), medical history review, anthropometric measurements and biochemical analysis. Participants were urban residents of Pinggu District with comparable dietary and lifestyle behaviors.

This study involved biomarker exploration and data mining derived from a previously conducted exploratory, randomized, controlled, open-label intervention trial (without a placebo arm) (28) ( Figure 1 ). The sample size for this exploratory trial (50 per group) was determined based on prior human gut microbial intervention studies (22–51 subjects per group) (35–37), accounting for potential attrition.

Eligible participants were randomly assigned (1:1) to the acarbose or vildagliptin group via simple randomization. A computer-generated random number sequence was created using SAS 9.0 by an independent statistician. Allocation concealment was ensured using opaque, sealed envelopes. The study treatments were as follows (1): Acarbose group (n = 50): acarbose (Bayer Medical and Health Co., Ltd., 50 mg/tablet) was initiated at 50 mg three times daily (TID) with the first meal for weeks 1-2, then escalated to 100 mg TID from week 3 onward to mitigate gastrointestinal adverse effects (e.g., bloating, flatulence). (2) Vildagliptin group (n = 50): vildagliptin (Swiss Novartis Limited, 50 mg/tablet) was administered as 50 mg twice daily (BID) before meals. All participants received telephone reminders 1–2 days prior to each visit, completing 5 telephone follow-ups and 3 on-site visits. During each visit, all participants received comprehensive lifestyle guidance according to the Chinese Diabetes Guidelines (2013) (34), which included daily caloric reduction (400–500 kcal), saturated fat restriction (<30% total fat), and ≥150 minutes/week of moderate-intensity exercise (e.g., brisk walking). Medication adherence was monitored through dosing diaries, pill counts, and returned packaging. Participants with 80–120% adherence were included in the Per Protocol Population (PPS).

Blood specimens were collected via venipuncture at baseline and at 3 months (3M) and 6 months (6M) post-intervention. Glycosylated hemoglobin (HbA_1c) was measured using cation-exchange high-pressure liquid chromatography (HPLC) method (Adams A1c HA-8160; Arkray, Kyoto, Japan). Plasma concentrations of fasting plasma glucose (FPG) and postprandial 2-h glucose (PPG) after a routine 75-g OGTT were measured using the glucose oxidase method. Fasting insulin (Fins) and postprandial 2-hour insulin (Pins) levels were determined using an electrochemiluminescence immunoassay (Elecsys 2010 system, Roche Diagnostic Ltd, Basel, Switzerland). C-peptide levels were measured using Direct Chemiluminescence (Atellica IM series; Siemens Healthcare Diagnostics).

HOMA-IR score was calculated using the following formula:

HOMA-β score was calculated using the following formula:

Levels of TMAO, choline, L-carnitine, betaine and γ-butyrobetaine were quantified using stable isotope dilution liquid chromatography–tandem mass spectrometry, as described previously (38). Briefly, 20 μL plasma were aliquoted into a 1.5-mL tube and mixed with 80 μL 10 μM internal standard composed of d9-TMAO, d9-choline, d9-carnitine, d11-betaine, and d9-γ-butyrobetaine in methanol. Protein in the samples was precipitated and the supernatant was recovered following centrifugation at 20,000 g at 4°C for 10 min, and the resulting supernatants were analyzed by injection onto a silica column (2.0*150 mm, Luna 5u Silica 100A; Cat. No. 00F-4274-B0, Phenomenex, Torrance, CA) at a flow rate of 0.4 mL/min using a LC-20AD Shimadazu pump system, SIL-20AXR autosample interfaced with an API 5500Q-TRAP mass spectrometer (AB SCIEX, Framingham, MA). A discontinuous gradient was employed to separate the analytes by mixing solvent A (0.1% propanoic acid in water) with solvent B (0.1% acetic acid in methanol) at different ratios, starting from 2% B linearly to 95% B, over 5.0 min, then held for 1.0 min, and then back to 2% B. Analytes were monitored using electrospray ionization in positive-ion mode with multiple reaction monitoring (MRM) of precursor and characteristic product-ion transitions of TMAO at m/z 76→58, d9-TMAO at m/z 85 →66, choline at m/z 104 →59.8, d9-choline at m/z 113.2→68.9, carnitine at m/z 162.1→103, d9-carnitine at m/z 171.1→102.8, betaine at m/z 118 →59, d11-betaine at m/z 129.1→65.9, γ-butyrobetaine at m/z 146.1→87, d9-γ-butyrobetaine at m/z 155.1→87, respectively.

The fasting serum lipid profile, including total cholesterol (TC), triglyceride (TG), high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C) levels, was ascertained using an automated biochemical analyzer. Serum concentrations of gut hormones, including active glucagon-like peptide-1 (GLP-1), ghrelin, peptide YY (PYY), and leptin, were quantified in the fasting state using standard enzyme-linked immunosorbent assay (ELISA) kits (EGLP-35K, EZGRT-89K, EZHPYYT66K, Merck Millipore, USA). The radioimmunoassay method (RK-069-04, Phoenix Peptide, USA) was used to measure serum cholecystokinin (CCK). Each participant was subjected to a series of anthropometric assessments, including body weight, height, waist and hip circumference, and systolic and diastolic blood pressure (SBP and DBP, respectively). BMI was calculated as the weight divided by height squared (kg/m²). The waist-to-hip ratio (WHR) was calculated as waist circumference divided by hip circumference. All anthropometric measurements were taken by registered nurses at the baseline, 3M, and 6M visits.

We analyzed the differences between the vildagliptin and acarbose treatment arms in clinical variables at baseline. Shapiro–Wilk test for determining whether continuous variables are normally distributed. Normally distributed quantitative variables are described as mean and standard deviation (SD); the independent samples t-test was used for intergroup comparisons. Non-normally distributed quantitative variables are presented as median (25^th, 75^th percentiles), and the Mann–Whitney U test was used to compare the two groups. Longitudinal changes in gut microbial metabolites (TMAO, L-carnitine, betaine, choline, and γ-butyrobetaine) across three timepoints (baseline, 3 months [3M], and 6 months [6M]) were evaluated using the Friedman test, a non-parametric repeated-measures analysis. Post hoc pairwise comparisons were conducted with Dunn’s test, adjusted for multiple testing via Bonferroni correction. To compare the degrees of posttreatment changes in clinical variables of different treatment group, we normalized the changes (Δ) of each variable from the baseline examination to 6M of the drug intervention. For each individual, change from baseline of a given variable was calculated using the following equation: . where Base_variable and M6_variable are the values of a given variable from the same individual at baseline and 6M in the acarbose and vildagliptin groups, respectively. To investigate the correlation between the levels of gut microbial metabolites and glucose metabolic variables during treatment with different agents, Spearman bivariate correlation analysis was conducted between the changing glucose metabolic characteristics and plasma levels of TMAO, L-carnitine, betaine, choline, and γ-butyrobetaine at 6M. All analyses were two-tailed, with p < 0.05 considered statistically significant, and performed in SPSS 25.0 and GraphPad Prism 9.5.0. A correlation heatmap was drawn using ChiPlot (https://www.chiplot.online/) (accessed 10 October 2024).

This study was approved by the Ethics Committee of the Peking University Health Science Center (2015PHB175-01; Date: 18 November 2015) and was conducted in compliance with the Declaration of Helsinki. The study was registered at ClinicalTrials.gov (NCT02999841). All participants provided written informed consent.

One hundred eligible patients with NDT2D (54 men and 46 women) were randomly assigned (1:1) to either the acarbose or vildagliptin treatment group. Seven participants were lost to follow-up at the 3-month visit and eight at the 6-month visit after starting the intervention. In total, 92 participants completed the 6-month trial: 48 in the acarbose group and 44 in the vildagliptin group. No serious drug-related adverse events were observed.

Baseline clinicodemographic characteristics were comparable between the study groups ( Table 1 ). No significant differences were observed in age (p= 0.434) or sex distribution (p= 0.108). Glycemic parameters (HbA_1c, FPG, PPG, Fins, Pins, HOMA-β, HOMA-IR), obesity measures (body weight, BMI, waist circumference, and hip circumference), and lipid profiles (TC, TG, LDL-C) were balanced between the two groups (p>0.05 for all; Table 1 ). However, HDL-C levels were significantly higher in the vildagliptin group (p=0.030). Additionally, there were no significant intergroup differences in blood pressure, intestinal hormone levels, or adipokine levels (p>0.05 for all; Table 1 ). More detailed information has been published (28).

Longitudinal analysis of plasma gut microbiota metabolites revealed no baseline intergroup differences in TMAO, L-carnitine, betaine, choline, or γ-butyrobetaine concentrations (p > 0.05 for all; Table 2 ).At 6 months, TMAO concentrations were significantly lower in the acarbose group versus vildagliptin (p < 0.05), while other metabolites showed no intergroup divergence (p>0.05 for all; Table 2 ).

Intragroup comparisons using the Friedman test demonstrated time-dependent changes. In the acarbose group, plasma TMAO concentrations decreased significantly at 6 months compared to baseline and 3 months (Bonferroni-adjusted p < 0.05; Figure 2a ). L-carnitine and γ-butyrobetaine concentrations increased at 3 and 6 months versus baseline (Bonferroni-adjusted p < 0.05; Figures 2b, e ), while betaine and choline remained stable (Bonferroni-adjusted p > 0.05; Figures 2c, d ).

In the vildagliptin group, plasma TMAO concentrations decreased significantly at 6 months compared to baseline (Bonferroni-adjusted p < 0.05; Figure 3a ). L-carnitine increased at 6 months versus baseline (Bonferroni-adjusted p < 0.05; Figure 3b ), while γ-butyrobetaine concentrations increased progressively at both 3 and 6 months (Bonferroni-adjusted p < 0.05; Figure 3e ). In contrast, betaine and choline concentrations remained stable (p > 0.05; Figures 3c, d ). The details of the statistical analysis are available in the Supplementary Material .

In our prior study, both acarbose and vildagliptin demonstrated significant improvements in glycemic control, lipid profiles, and blood pressure in newly diagnosed type 2 diabetes patients after 6 months of intervention (28). Building on these findings, we investigated correlations between gut microbial metabolites and glucose metabolic variables in the current work.

Spearman bivariate correlation analysis at 6 months demonstrated significant associations in the acarbose group. ΔTMAO correlated positively with changes in BMI (r = 0.323), waist circumference (r = 0.423), PPG (r = 0.338), Fins (r = 0.436), fasting C-peptide (r = 0.328), HOMA-IR (r = 0.484), and leptin (r = 0.396) (p < 0.05). ΔBetaine correlated with Δghrelin (r = 0.524, p < 0.001), and both ΔL-carnitine and Δγ-butyrobetaine correlated with Δfasting C-peptide (r = 0.319, p = 0.037; r = 0.360, p = 0.017; Figure 4 , Supplementary Table 3 ).

By contrast, no significant correlations were observed between changes in gut microbiota metabolites and metabolic parameters in the vildagliptin group (p > 0.05). Notably, a strong positive correlation between ΔL-carnitine and Δγ-butyrobetaine was consistent across both groups (r = 0.644, p < 0.001; r = 0.771 p < 0.001; Figure 4 , Supplementary Table 3 ), indicating coordinated regulation of these metabolites.

The gut microbiota plays a central role in TMAO biosynthesis. Bioinformatics suggested that gut microbiota can convert carnitine, betaine, and choline to TMA, through the utilization of three TMA-lyase complexes, namely cntA/B, yeaW/X, and cutC/D (39). Choline, abundant in egg yolk and red meat (40), is predominantly metabolized by Lachnoclostridium and Intestinibacter (Firmicutes) via cutC/D, genera enriched in cardiometabolic disorders (39, 41). Active L-carnitine is mainly found in meat and dairy products (42). For L-carnitine catabolism, Proteobacteria (e.g., Escherichia and Shigella) directly cleave its C-N bond via cntA/B-encoded oxygenases in the cecum (43, 44), while Emergencia timonensis (Actinobacteria) employs an alternative pathway involving γ-butyrobetaine as a transient intermediate through uncharacterized oxidoreductases (45–47). Owing to the presence of the “enterohepatic axis”, TMA can be rapidly oxidized to TMAO by hepatic FMO3 and subsequently released into the circulation (8). In the present study, the observed reduction in TMAO and accumulation of L-carnitine and γ-butyrobetaine suggest that both acarbose and vildagliptin impede the metabolic pathway of carnitine → γ-butyrobetaine → TMA, albeit through distinct mechanisms. To elucidate the impact of these two drugs on the TMAO metabolic pathway, it is important to consider the complex interplay between pharmacological interventions and the gut microbiome.

As a first-line α-glucosidase inhibitor (AGI), acarbose delays small intestinal carbohydrate absorption, thereby reducing postprandial hyperglycemia. Undigested carbohydrates reaching the colon undergo microbial fermentation, which reshapes the gut ecosystem (48, 49). This microbiota-dependent effect may underlie its therapeutic benefits in T2DM. In mice fed with either a high-starch or high-fiber diet, acarbose increased Bacteroidaceae and Bifidobacteriaceae while reducing Akkermansia muciniphila and Bacteroidales S24-7, accompanied by elevated SCFAs (25, 50). In humans, however, our prior work revealed host-specific divergence: 6-month acarbose treatment decreased gut microbial alpha diversity and Bacteroides species, but enriched Streptococcus (S. sanguinis, S. salivarius) and Bifidobacterium ( 28).

Vildagliptin, a DPP-4 inhibitor, enhances endogenous GLP-1 activity to improve glycemic control (51). While its glucose-lowering effects are partially microbiota-mediated, the specific taxonomic shifts differ across studies. In animal studies, sitagliptin increased Bacteroidetes abundance in mice fed a high-fat diet, correlating with improved glucose homeostasis (29). In contrast, our previous study yielded differing results: vildagliptin enriched Bifidobacterium adolescentis and Intestinibacter bartlettii, while reducing Paraprevotella spp., Parabacteroides distasonis and Bacteroides spp.(e.g. B. finegoldii, B. caccae) ( 28). Notably, a cross-sectional analysis observed a heightened abundance of these two Bacteroides species in treatment-naïve patients with T2DM as opposed to those with prediabetes or normal glucose tolerance (52), underscoring their role in diabetic dysbiosis.

Although these two drugs differ in their gastrointestinal mechanisms of glycemic control, our study indicates that they elicit comparable effects on the gut microbiota, particularly with respect to the proliferation of B. adolescentis and the reduction of several species belonging to the phylum Bacteroidetes (B. caccae and B. finegoldii). This overlap may explain their similar metabolic outcomes. Acarbose’s stronger TMAO-lowering effect could stem from its direct impact on microbial carbohydrate fermentation, which selectively depletes TMA-lyase-producing taxa. Conversely, vildagliptin’s milder effect might involve indirect GLP-1-mediated modulation of gut ecology. Crucially, baseline TMAO levels were comparable between groups, ruling out preexisting differences and emphasizing drug-induced microbial reprogramming.

After 6 months of treatment with acarbose, the blood glucose levels of participants were effectively controlled, including decreases in FPG, PPG, HbA_1c and HOME-IR, with significant decreases in body weight and BMI (28). IR is defined as a state of decreased insulin sensitivity and/or responsiveness. This study, we used HOMA-IR to quantify the extent of IR in an individual and calculated it by combining the FPG and fin levels. A reduction in the HOMA-IR is typically indicative of enhanced insulin sensitivity. In the present study, we found a positive correlation between the changes in TMAO and PPG, Fin, fasting C-peptide, and HOMA-IR after 6-month acarbose treatment, suggesting a mechanistic interplay between gut microbiota-derived TMAO and IR.

Experimental studies corroborate these clinical observations. Dietary TMAO supplementation exacerbates impaired glucose tolerance in high-fat diet-fed mice by disrupting hepatic insulin signaling and inducing adipose tissue inflammation, thereby elevating HOMA-IR indices (53). FMO3 is an enzyme expressed primarily in the liver and catalyzes the conversion of TMA to TMAO (8). In vitro studies have confirmed that FMO3 is repressed by insulin; however, its expression is increased in obese and insulin-resistant male mice, mirroring a similar increase observed in humans. Notably, the experimental suppression of FMO3 in insulin-resistant mice effectively inhibited FoxO1, a key metabolic regulatory node, and completely circumvented the progression of hyperglycemia, hyperlipidemia, and atherosclerosis (54). In parallel with these findings, another mouse study revealed a sophisticated mechanism through which TMAO exacerbates metabolic dysfunction. This study indicates that TMAO binds to and activates the endoplasmic reticulum stress kinase PERK (EIF2AK3), thereby triggering the transcription factor FoxO1 in a PERK-dependent cascade. This signaling pathway culminates in metabolic impairment. Interventions targeting TMAO reduction, whether via microbiota modulation or FMO3 inhibition, downregulate PERK-FoxO1 signaling, thereby ameliorating metabolic dysfunction (16). These findings collectively implicate TMAO as a mediator of IR through multiple pathways, yet key questions remain unresolved. Whether TMAO reduction directly enhances insulin sensitivity or merely reflects acarbose-induced microbiota remodeling requires causal validation through fecal microbiota transplantation studies.

Thus, based on the results of the present study, we clarify that both acarbose and vildagliptin treatments significantly reduced TMAO levels in newly diagnosed T2DM patients, with a more pronounced reduction observed in the acarbose group. Furthermore, the decline in TMAO levels demonstrated a significant association with improvements in insulin resistance parameters.

Our study also has several strengths. Primarily, this two-arm RCT employed stringent inclusion criteria and meticulous follow-up protocols, ensuring high-quality data collection. Additionally, we comprehensively investigated TMAO dynamics by analyzing not only TMAO itself but also its four key precursors (L-carnitine, betaine, choline, and γ-butyrobetaine), an approach rarely adopted in prior studies. Third, by leveraging longitudinal blood sampling, this work represents the first exploration of acarbose and vildagliptin effects on TMAO metabolism in a Chinese T2DM population, enhancing the translational relevance of our findings.

Despite these advances, there are some limitations. First, as an exploratory RCT designed to investigate gut microbiota-mediated mechanisms, the sample size (n = 50/group) was determined based on prior studies, rather than a formal power analysis. While this design aligns with the exploratory nature of our hypothesis-generating objectives, future trials with larger cohorts are needed to validate these findings and assess subgroup-specific effects. Second, although our 6-month data demonstrate progressive TMAO reduction, the study duration may not represent the optimal therapeutic window. The long-term efficacy of glucose-lowering drugs on TMAO dynamics—including whether suppression stabilizes or continues to decline—remains unknown. Future trials should extend follow-up to 9–12 months to clarify the durability of TMAO reduction and its relationship to glycemic control and cardiovascular outcomes. Third, while microbial shifts correlated with TMAO reduction, our analysis did not pinpoint specific taxa responsible for TMAO metabolism (e.g., cntA/B- or cutC/D-harboring bacteria) or establish causality. Mechanistic validation through animal studies, fecal microbiota transplantation (FMT), or targeted metagenomic sequencing of TMA-lyase genes is essential to confirm these associations.