Metagenomic analysis reveals crosstalk between gut microbiota and glucose-lowering drugs targeting the gastrointestinal tract in Chinese patients with type 2 diabetes: a 6 month, two-arm randomised trial

Abstract

Aims/hypothesis: The use of oral glucose-lowering drugs, particularly those designed to target the gut ecosystem, is often observed in association with altered gut microbial composition or functional capacity in individuals with type 2 diabetes. The gut microbiota, in turn, plays crucial roles in the modulation of drug efficacy. We aimed to assess the impacts of acarbose and vildagliptin on human gut microbiota and the relationships between pre-treatment gut microbiota and therapeutic responses.

Methods: This was a randomised, open-labelled, two-arm trial in treatment-naive type 2 diabetes patients conducted in Beijing between December 2016 and December 2017. One hundred participants with overweight/obesity and newly diagnosed type 2 diabetes were recruited from the Pinggu Hospital and randomly assigned to the acarbose (n=50) or vildagliptin (n=50) group using sealed envelopes. The treatment period was 6 months. Blood, faecal samples and visceral fat data from computed tomography images were collected before and after treatments to measure therapeutic outcomes and gut microbiota. Metagenomic datasets from a previous type 2 diabetes cohort receiving acarbose or glipizide for 3 months were downloaded and processed. Statistical analyses were applied to identify the treatment-related changes in clinical variables, gut microbiota and associations.

Results: Ninety-two participants were analysed. After 6 months of acarbose (n=44) or vildagliptin (n=48) monotherapy, both groups achieved significant reductions in HbA1c (from 60 to 46 mmol/mol [from 7.65% to 6.40%] in the acarbose group and from 59 to 44 mmol/mol [from 7.55% to 6.20%] in the vildagliptin group) and visceral fat areas (all adjusted p values for pre-post comparisons <0.05). Both arms showed drug-specific and shared changes in relative abundances of multiple gut microbial species and pathways, especially the common reductions in Bacteroidetes species. Three months and 6 months of acarbose-induced changes in microbial composition were highly similar in type 2 diabetes patients from the two independent studies. Vildagliptin treatment significantly enhanced fasting active glucagon-like peptide-1 (GLP-1) levels. Baseline gut microbiota, rather than baseline GLP-1 levels, were strongly associated with GLP-1 response to vildagliptin, and to a lesser extent with GLP-1 response to acarbose.

Conclusions/interpretation: This study reveals common microbial responses in type 2 diabetes patients treated with two glucose-lowering drugs targeting the gut differently and acceptable performance of baseline gut microbiota in classifying individuals with different GLP-1 responses to vildagliptin. Our findings highlight bidirectional interactions between gut microbiota and glucose-lowering drugs.

Trial registration: ClinicalTrials.gov NCT02999841 FUNDING: National Key Research and Development Project: 2016YFC1304901.

Keywords: Glucose-lowering drugs; Gut microbiota; Type 2 diabetes.

© 2022. The Author(s).

Figures

Fig. 1

Major clinical outcomes in newly diagnosed type 2 diabetes patients after 6 month treatment with acarbose or vildagliptin. (a–r) Bar charts show changes in HbA1c (a); FPG (b); PPG (c); HOMA-IR (d); Pins (e); Fins (f); weight (g); BMI (h); L2-L3 VFA, L2-L3 SFA, L4-L5 VFA and L4-L5 SFA (i–l); CCK (m); GLP-1 (n); leptin (o); adiponectin (p); ghrelin (q); and PYY (r), in the acarbose (light green) or vildagliptin (light blue) treatment group. Wilcoxon signed-rank test, *BH-adjusted p<0.05. The y-axis indicates the delta (Δ) post-minus pre-treatment value of each variable. Data are presented as mean+SEM (detailed BH-adjusted p values are presented in Table 1). Individual data points are shown on the graph in grey

Fig. 2

Changes in the gut microbial structure induced by glucose-lowering treatment. (a) Comparisons of alpha diversity (Shannon index at the species level) between four groups. Wilcoxon signed-rank test for comparisons between pre- and post-treatment groups with the same agent; Wilcoxon rank-sum test for comparisons between groups with acarbose or vildagliptin, *p<0.05, ***p<0.001. (b) Non-metric multidimensional scaling (NMDS) plot illustrating the Bray–Curtis dissimilarities of the gut microbial species composition in pre- and post-treatment samples. (c) Bar plots of pseudo F-statistic values showing the magnitudes of microbial dissimilarities for within- and between-treatment groups. PERMANOVA (N=999 permutations), ***p<0.001. (d) Taxonomic cladogram showing significantly altered microbial taxa in patients treated with 6 months of acarbose or vildagliptin (Wilcoxon signed-rank test, BH-adjusted p<0.05; see details in ESM Tables 2, 3). (e) Gut microbial species consistently respond to acarbose and vildagliptin treatment. Wilcoxon signed-rank test, *BH-adjusted p<0.05. Colour bars indicate pre–post treatment effect sizes estimated from Wilcoxon signed-rank tests on the Clr-transformed RAs of species in the two treatment groups. Effect size >0: dark green and dark blue indicate the higher RAs in pre-treatment groups with acarbose (Acar base) and vildagliptin (Vild base), respectively; effect size ≤0: light green (Acar M6) and light blue (Vild M6) indicate the higher RAs in post-treatment groups. The effect size is calculated as the Z statistic divided by the square root of the sample size. The dashed line indicates an absolute value of effect size at 0.3. (f) Heatmap showing significantly altered species in four treatment arms, including 6 month treatment with acarbose (Acar base vs M6; n=42) or vildagliptin (Vild base vs M6; n=41) in the current study, and 3 month treatment with acarbose (Acar base vs M3; n=51) or glipizide (Glip base vs M3; n=43) in a previous study of Chinese type 2 diabetes patients [5]. The colour key indicates pre–post treatment effect sizes. Wilcoxon signed-rank test, * indicates BH-adjusted p<0.05. Acar, acarbose; Vild, vildagliptin

Fig. 3

Longitudinal associations between clinical variables and microbial abundances. (a, b) Heatmaps resulting from Wald statistics of the longitudinal associations of clinical variables with the 19 species (a) and 25 pathways (b) with consistent responses to 6 month treatment with acarbose or vildagliptin. All the metabolic pathways are ranked in the same order as presented in ESM Fig. 5. Wald statistics are calculated based on multivariate regression models using GEE, adjusting for sex and age. *BH-adjusted p<0.05, **BH-adjusted p<0.01, ***BH-adjusted p<0.001. Blue indicates species/pathways of higher abundances in pre-treatment groups. Red indicates species/pathways of higher abundances in post-treatment groups

Fig. 4

Links between baseline gut microbiota and post-treatment GLP-1 response. (a) Density curve of the PC% from baseline of fasting active GLP-1 in response to vildagliptin treatment. The dotted line represents a median value of 50.18%. Low response (LR): ≤50.18%; high response (HR): >50.18%. (b, c) Boxplots showing the comparisons of PC% from baseline (b) and baseline values (c) of GLP-1 and six type 2 diabetes-related variables between the two response subgroups. The p values were calculated using ANCOVA with adjustment for age and sex. *p<0.05, **p<0.01, ***p<0.001. (d) sPLS-DA to select baseline microbial features driving the separation of samples between the two subgroups. Individual samples from different subgroups are presented on a scatter plot using different colours (LR: blue; HR: orange) and 95% confidence ellipses. (e) Bar plot representing the contributions of the ten selected microbial features for the first sPLS-DA component. (f) Heatmap resulting from the coefficients of partial Spearman correlations (adjustment for age and sex) between PC%-GLP-1 and baseline abundances of ten selected microbial features in the two groups. *p<0.05. (g) Scatterplots showing the correlation between the PC%-GLP-1 and the sPLS-DA-based predicted probability in the vildagliptin group and the correlation between the PC%-GLP-1 and the predicted probability in the acarbose group. The p values and ρ values were calculated by Spearman’s rank correlation