Region-Specific Effects of Metformin on Gut Microbiome and Metabolome in High-Fat Diet-Induced Type 2 Diabetes Mouse Model

Compared to that in the control group, weight gain was significantly slowed down in the high-fat diet-induced T2DM mouse model treated with metformin via gavage (). In addition, fasting blood glucose, fasting insulin levels, and the insulin resistance index (HOMA-IR) were also improved in metformin-treated T2DM mice, as shown in–E. These results suggest that metformin had a significant therapeutic effect on the high-fat diet-induced T2DM mouse model. Supplementary Figure S1A,B Supplementary Figure S1C

The microbial composition of the metformin-treated and control groups was analyzed by comprehensive sequencing of the full-length 16S rRNA gene from 75 microbial samples from various gastrointestinal regions in metformin-treated and control mice, including the proximal and distal small intestines, cecum, colon, and feces. Due to insufficient sequencing data for two samples, this study ultimately excluded these two samples and focused on detailed microbiome analyses of the remaining 73 samples. The scheme of the experimental procedure is shown in Figure 1A,B.

We identified a total of 786,902 high-quality reads across all 73 samples, which were clustered into 436 distinct operational taxonomic units (OTUs). Notably, in the proximal small intestine, 59 (26.2%) OTUs were common to both groups, while 87 (38.7%) and 79 (35.1%) OTUs were exclusive to the control and treatment groups, respectively (); in the distal small intestine, there were 77 (33.6%) shared OTUs, with 90 (39.3%) and 62 (27.1%) unique OTUs for the control and treatment groups, respectively (); in the cecum, 198 (55.9%) OTUs were shared, with 134 (37.9%) and 22 (6.2%) unique OTUs for the control and treatment groups, respectively (); in the colon, there were 190 (40.7%) shared OTUs, with 143 (54.1%) and 18 (5.1%) unique OTUs for the control and treatment groups, respectively (); and in the feces, 195 (54.3%) OTUs were shared, with 152 (42.3%) and 12 (3.3%) unique OTUs for the control and treatment groups, respectively (). Supplementary Figure S2A Supplementary Figure S2B Supplementary Figure S2C Supplementary Figure S2D Supplementary Figure S2E

This study further delved into the microbial composition at the phylum and genus levels, as demonstrated in Figure 1C–G. At the phylum level, Firmicutes and Verrucomicrobia were predominant in the small intestine of metformin-treated mice, contrasting with the control group’s dominance of Firmicutes. The cecum, colon, and feces of metformin-treated mice were characterized by the predominance of Bacteroidetes and Verrucomicrobia, whereas the control group showed a predominance of Bacteroidetes and Firmicutes. At the genus level, Faecalibaculum dominated in the proximal and distal small intestine of both treated and control mice, whereas Romboutsia was more abundant in the control group, and Akkermansia was more abundant in the metformin-treated group. Additionally, the control group exhibited a higher abundance of Alistipes and Mucispirillum in the cecum, Alistipes and Lachnoclostridium in the colon, and Faecalicatena and Romboutsia in feces. Post-metformin treatment, Akkermansia became the dominant genus in the cecum, colon, and feces.

Within-sample (α) phylogenetic diversity, as measured by the Shannon index, showed no significant differences between the proximal and distal small intestine, nor between the cecum, colon, and feces. However, there was a marked disparity in alpha diversity between the small intestine and other gastrointestinal sites, with lower diversity in the small intestine and significantly higher alpha diversity in the cecum, colon, and feces (Figure 2A). A notable difference in alpha diversity between the metformin-treated and control groups was also observed (Figure 2B). Further examination of the impact of metformin on alpha diversity across different intestinal sites revealed significant differences in the distal small intestine, cecum, colon, and feces following treatment, while the proximal small intestine remained unaffected (Figure 2C).

Beta diversity was ascertained through ordination analysis. Principal component analysis (PCA) revealed distinct clustering of samples from the cecum, colon, and feces, which differed markedly from samples obtained from both regions of the small intestine (Figure 3A). Moreover, a significant divergence was observed between the metformin-treated and control groups, with Akkermansia, Romboutsia, Clostridium, Muribaculum, and Enterorhabdus emerging as the predominant identifiers in the metformin-treated group (Figure 3B). In the proximal small intestine, specifically after metformin treatment, Akkermansia, Turicimonas, and Lactobacillus were identified as the most prominent identifiers (Figure 3C). In the distal small intestine, the principal identifiers distinguishing the metformin-treated samples were Akkermansia, Romboutsia, Clostridium, Lactobacillus, and Enterorhabdus (Figure 3D). The cecum’s metformin-responsive identifiers included Akkermansia, Romboutsia, Peptococcus, Tyzzerella, and Gabonia (Figure 3E). In the colon, Akkermansia, Mailhella, Desulfovibrio, Longibaculum, and Butyricimonas were identified as key metformin-associated identifiers (Figure 3F). Similarly, in the feces, Akkermansia, Peptococcus, Oscillibacter, and Parabacteroides were the most significant identifiers post-metformin treatment (Figure 3G).

To delineate the disparate microbial compositions of the metformin-treated and control groups across four intestinal regions and the feces, we employed a detailed LEfSe analysis (LDA > 2 and p-value < 0.05), as shown in Supplementary Figure S3. Metformin had a stronger effect on the abundance of bacteria in the cecum, colon, and feces. In the proximal small intestine of the metformin-treated group, 25 bacterial species exhibited significant changes compared to the control group, with the majority belonging to Firmicutes (15/25, 60%) and Proteobacteria (5/25, 20%) at the phylum level, and the most affected families were Lachnospiraceae and Desulfovibrionaceae (Supplementary Figure S3A). In the distal small intestine, 35 bacterial species showed significant variations post-metformin treatment, predominantly within Firmicutes (19/35, 54.3%), Proteobacteria (6/35, 17.1%), and Bacteroidetes (5/35, 14.3%), with Lachnospiraceae, Ruminococcaceae, Lactobacillaceae, Desulfovibrionaceae, and Eggerthellaceae being the most affected families (Supplementary Figure S3B). In the cecum, 75 bacterial species were significantly altered, mostly within Firmicutes (47/75, 62.7%) and Bacteroidetes (13/75, 17.3%), with Lachnospiraceae, Ruminococcaceae, Rikenellaceae, and Clostridiaceae being the predominantly affected families (Supplementary Figure S3C). In the colon, 77 bacterial species were significantly changed following metformin treatment, with the largest changes observed in Firmicutes (49/77, 63.6%) and Bacteroidetes (15/77, 19.5%) and the most impacted families being Lachnospiraceae, Ruminococcaceae, Erysipelotrichaceae, Desulfovibrionaceae, and Rikenellaceae (Supplementary Figure S3D). In the feces, 84 species exhibited significant alterations, primarily in Firmicutes (61/84, 72.6%) and Bacteroidetes (10/84, 11.9%), with Lachnospiraceae, Ruminococcaceae, and Clostridiaceae being the most altered families (Supplementary Figure S3E).

Further analysis of the expression changes in the gut microbiota at the species level revealed that, following metformin treatment, there were a total of 85 bacterial species exhibiting significant differences across various intestinal segments and feces (LDA > 2 and p-value < 0.05), as shown in Figure 4. The variation in bacterial species was less pronounced in both parts of the small intestine compared to the cecum, colon, and feces. Following metformin treatment, there was a predominant increase in the expression of bacterial strains from the Akkermansiaceae, Lactobacillaceae, Tannerellaceae, and Erysipelotrichaceae families. Conversely, strains exhibiting a decline in expression were largely associated with the Lachnospiraceae and Ruminococcaceae families. Of these, Akkermansia muciniphila, Lactobacillus johnsonii, Lactobacillus reuteri, Parabacteroides distasonis, Turicimonas muris, and Roseburia hominis exhibited significantly higher expression in all intestinal regions and feces post-metformin treatment. Additionally, Staphylococcus sciuri expression increased in all intestinal regions but decreased in the feces after metformin administration. Post-metformin treatment, there was a notable decline in the expression of most bacterial strains in the cecum, colon, and feces, with strains Clostridium lactatifermentans, Acetatifactor muris, Bacteroides uniformis, Alistipes putredinis, Oscillibacter valericigenes, and Butyricimonas virosa experiencing the most substantial changes, whereas in the small intestine, expression levels predominantly stayed constant.

We quantified 415 metabolites that covered a wide range of significant metabolites in the intestine using targeted metabolomic techniques, encompassing 22 distinct classes of metabolites (Supplementary Table S1). The most abundant categories were amino acids and peptides (76 metabolites), followed closely by bile acids and fatty acids (65 and 63 metabolites respectively), as shown in Figure 5A.

Utilizing orthogonal partial least squares discriminant analysis (OPLS-DA), a method renowned for enhancing group differentiation [31], we discerned notable disparities in metabolite profiles between metformin-treated and control groups. This was evident in various intestinal regions and the feces, as shown in Supplementary Figure S4A, with no overfitting observed (Supplementary Figure S4B), thus validating the model’s precision.

The comparative analysis uncovered pronounced metabolomic alterations across various intestinal regions and the feces following metformin treatment (Supplementary Figure S5). Screening criteria for differential metabolites were VIP ≥ 1, fold change ≥ 1.2 or ≤0.83, and q-value < 0.05. In both regions of small intestine, there were hardly any changes in metabolites. However, in the cecum, colon, and feces, a substantial number of metabolites exhibited significant alterations, with the highest quantity of metabolic changes occurring in the cecum and colon (Figure 5B–F). In the proximal small intestine, notable decreases were observed in diosmetin, cimetidine, and threonic acid levels post-metformin treatment, with no significant increases reported (Figure 5B). Similarly, in the distal small intestine, lithocholic acid, nordeoxycholic acid, and isoallolithocholic acid levels significantly decreased, with no notable increases (Figure 5C). However, in the cecum, the number of differential metabolites was significantly higher, with 120 metabolites altered significantly post-treatment, including 18 metabolites, like mandelic acid, 3-hydroxybutyric acid, and arabinonic acid, exhibiting pronounced increases and 102 metabolites showing significant decreases (Figure 5D). The colon revealed alterations in 71 metabolites, with 38, like 3,4-dihydro-2H-1-benzopyran-2-one and 2-deoxyglucose, significantly increasing and 33 significantly decreasing (Figure 5E). Fecal analysis showed 55 metabolites with significant changes, 18 of which, such as O-phosphoryl-ethanolamine, increased significantly, while 37, including lithocholic acid and hydroretrocortine, decreased significantly (Figure 5F).

By categorizing these differential metabolites, we observed a predominant decrease in bile-acid-class metabolites, particularly in the cecum, colon, and feces, post-metformin treatment. In contrast, benzenoid-class metabolites were significantly increased in the cecum, colon, and feces. Additionally, it is noteworthy that the highest number of significantly reduced metabolites was seen in the cecum, while the colon had the highest number of significantly increased metabolites. Specifically, in the cecum, metabolites from classes like amino acids and peptides, carbohydrates, benzenoids, fatty acids, organic acids, and organoheterocyclics primarily decreased, whereas, in the colon, metabolites from amino acids and peptides, fatty acids, carbohydrates, and organic acids predominantly increased. Furthermore, distinct from other areas, in the colon post-metformin treatment, there was a noticeable increase in most fatty-acid-class metabolites (Figure 5G).

After metformin treatment, the main categories of differential metabolites and the expression differences in all differential metabolites within these categories were detailed in the cecum, colon, and feces (Figure 6). Notably, bile acid metabolites, particularly lithocholic acid, exhibited significant alterations in the cecum, colon, and feces post-treatment. After metformin treatment, in the cecum, colon, and feces, most metabolites belonging to benzenoids, carbohydrates, and fatty acids significantly increased, with the most notable changes observed in the colon. For example, in the colon, the most pronounced increases were seen in benzenoid-class metabolites such as L-3-phenyllactic acid, phenyllactic acid, and phenylglyoxylic acid; in carbohydrates, arabinonic acid and 2-deoxyglucose showed the most significant upward trends; and among fatty-acid-class metabolites, 2-hydroxy-4-(methylthio)butanoate and 3-methyladipic acid exhibited the most marked increases. These data suggest that although metformin had the broadest impact on the metabolites in the cecum, primarily exerting an inhibitory effect, its stimulatory effects on metabolites were mainly manifested in the colon.

Using correlation heatmaps, this study delves into the relationships between the top 20 most significantly altered microbiota and metabolites following metformin treatment, as shown in Figure 7.

We observed notable variations in the correlations between altered microbiota and metabolites after metformin treatment across different regions of the intestine, including the proximal small intestine, distal small intestine, cecum, colon, and feces. In our previous study, through microbiome analysis, we found that the abundance of Akkermansia was significantly higher in all intestinal regions and fecal samples after treatment with metformin, and Akkermansia was identified as the predominant identifier in all intestinal regions and fecal samples (Figure 1 and Figure 3). Furthermore, several studies have highlighted that Akkermansia muciniphila plays an important role in treatment with metformin [32,33,34,35]. In the proximal small intestine, no metabolites exhibited positive interactions with Akkermansia muciniphila (Figure 7A). In the distal small intestine, the metabolite positively correlated with Akkermansia muciniphila was hydrocinnamic acid (Figure 7B). Compared to those in the small intestine, the interactions between the microbiota and their metabolites in the cecum, colon, and feces were more complex following metformin treatment.

In the cecum, the types of metabolites associated with Akkermansia muciniphila changed, showing positive correlations with metabolites such as 2-deoxyglucose, 3,4-dihydro-2H-1-benzopyran-2-one, 3-hydroxybutyric acid, 3-hydroxyphenylacetic acid, 4-hydroxy-3-methylbenzoic acid, and arabinonic acid (Figure 7C). However, compared to that in the cecum, in the colon following metformin treatment, the number of metabolites associated with Akkermansia muciniphila significantly increased, showing positive correlations with metabolites including 2-deoxyglucose, 2-hydroxy-4-(methylthio)butanoate, 3,4-dihydro-2H-1-benzopyran-2-one, 3-hydroxybutyric acid, 3-hydroxyphenylacetic acid, 4-hydroxy-3-methylbenzoic acid, arabinonic acid, and phenylglyoxylic acid (Figure 7D). In the feces after metformin treatment, the number of metabolites associated with Akkermansia muciniphila decreased, and the correlated metabolites shifted to 2-deoxyglucose, 2-hydroxy-4-(methylthio)butanoate, 3-hydroxyphenylacetic acid, 3-methyloxindole, 4-hydroxy-3-methylbenzoic acid, arabinonic acid, and phenylglyoxylic acid (Figure 7E). These data indicate significant differences in the types of metabolites associated with Akkermansia muciniphila across different intestinal regions after metformin treatment, with the colon having the highest diversity of positively correlated metabolites.

C57BL/6J mice exhibit a high susceptibility to high-fat diets, leading to the development of obesity and type 2 diabetes mellitus (T2DM) in experimental models. Therefore, C57BL/6J mice were selected for this study. C57BL/6J mice obtained from Beijing Vital River Laboratory Animal Technology Co. were kept in a controlled environment with SPF conditions, including a temperature of 21 ± 2 °C and humidity between 40% and 70%. A 12-h light cycle (8 am to 8 pm) was implemented. The mice were accommodated in individually ventilated cages (Fengshi, China), housing 3–4 mice each. They were given an unrestricted amount of a 60% high-fat diet (D12492, Research Diets), as well as free access to water.

T2DM was induced by maintaining the mice on a 60% high-fat diet for 8 weeks. Targeted metabolomics analyses require a minimum of 6 replicates. Therefore, subsequently, we randomly divided the mice into two groups: one group of mice was gavaged with metformin (D150959, Sigma, Castleford, UK) at a dose of 300 mg/kg/day (n = 8), and the other group was gavaged with an equal amount of sterile water daily as a control (n = 7). This program was continued for 4 weeks along with the high-fat diet. The experimental design is illustrated in Figure 1A. The body weights of the mice were recorded every two days during metformin treatment. At the end of metformin treatment, all experimental mice were fasted for 12–16 h, followed by the measurement of fasting blood glucose levels in venous blood (Roche) and the detection of fasting insulin levels in the serum using an ELISA kit (90080, Crystal Chem, Grove Village, IL, USA). The insulin resistance index (HOMA-IR) was calculated using the following formula: HOMA-IR = fasting blood glucose (mmol/L) × fasting insulin level (mIU/L)/22.5.

For all experimental mice, including high-fat-diet-induced T2DM mice treated with metformin (n = 8) and gavage vehicle-treated control mice (n = 7), the contents of their proximal small intestine, distal small intestine, colon, cecum, and feces were collected. Therefore, eight and seven biological replicates of the experimental and control groups per intestinal segment were used for full-length sequencing analysis of 16S rRNA genes and targeted metabolomics analysis, respectively. The detailed sample collection protocol is described below: feces were gathered from all mice before euthanasia and stored in sterile tubes at −80 °C. The mice were then euthanized under CO₂ anesthesia, and their intestinal regions (proximal small intestine, distal small intestine, cecum, and colon) were isolated using sterile scissors. The contents of each segment were extracted with sterile forceps and directly transferred into sterile tubes. These samples were immediately frozen at −80 °C for subsequent analysis. A schematic of the sample collection sites is depicted in Figure 1B.

DNA extraction from the collected samples was performed using the QIAamp Fast DNA Stool Mini Kit (51604, Qiagen, Venlo, The Netherlands), adhering to the manufacturer’s protocol. DNA concentrations were measured with an enzyme laboratory apparatus using the DNA BR kit (Q33230↗, Invitrogen, Carlsbad, CA, USA). For each sample, 10 ng of genomic DNA was subjected to PCR amplification of the full-length 16S rRNA gene using the universal primers 27F (5′-AGRGTTYGATYMTGGGCTCAG-3′) and 1492R (5′-RGYTACCTTGTTACGACTT-3′). The PCR products underwent 1% agarose gel electrophoresis, followed by purification using magnetic beads. Subsequently, the purified samples were sequenced utilizing the Pacbio platform.

To obtain clean data, the raw sequencing data underwent processing. Initially, cutadapt software (v2.6) was used to intercept primer and junction contamination in the reads that aligned with the primers, resulting in the extraction of fragments within the target region. Next, filtering was performed using readfq (v1.0) [58]. A method was employed to lower the quality of reads by analyzing windows of 30 bp each. If the average quality value within a window fell below 20, the end sequences of the reads were trimmed from the start of the window. Reads with a final length less of than 75% of the original length were then discarded. Furthermore, reads containing N and low-complexity reads were removed, and the resulting clean data underwent quality control using iTools Fqtools fqcheck (v0.25). The software FLASH (v1.2.11) was utilized to merge read pairs from double-end sequencing into a single sequence by identifying overlaps and to extract tags from the high-variance region. The splicing parameters included a minimum matching length of 15 bp and an allowable mismatch rate of 0.1 for the overlap region [59]. Following this, the combined tags were grouped into operational taxonomic units (OTUs) through the utilization of the program USEARCH (v7.0.1090) [60]. This process entails clustering with UPARSE at a 97% similarity threshold to acquire the characteristic sequences of OTUs and subsequently eliminating chimeric sequences produced during PCR amplification with the assistance of UCHIME (v4.2.40). OTU representative sequences were compared to all tags using the usearch_global method to calculate the abundance statistics of OTUs in each sample. The representative OTU sequences were then annotated against the NCBI database (20170709) using BLCA software (v2.1) [61], with a confidence threshold of 0.6 to refine the annotation results. The refined OTUs were subsequently utilized for further analyses.

Venn diagrams of OTUs were plotted using the VennDiagram package in R (v4.2.2) software. Shannon indices for alpha diversity were calculated using Mothur (v1.31.2)’s summary.single command [62], and then the results were imported into R (v4.2.2) software and the ggplot2 package was used to read and visualize these alpha diversity indices. Principal component analysis (PCA) and visualization were performed using the FactoMineR package in R (v4.2.2) software. In a Linux setting, LEfSe was utilized to identify the significant bacterial taxa that distinguished the metformin-treated group from the control group based on criteria of LDA > 2 and a p-value less than 0.05. Strains with significant differences and grouping names were imported into R (v4.2.2) software and visualized using the ggplot2 package.

A comprehensive quantitative analysis was conducted on 415 metabolites that covered a wide range of significant metabolites in the intestine (Supplementary Table S1). The initial phase involved the meticulous preparation of intestinal and fecal samples through a process of homogenization, dissociation, and centrifugation. The subsequent UPLC-MS analyses employed a Waters ACQUITY UPLC I-Class Plus system (Waters, Milford, MA, USA) in conjunction with a high-precision QTRAP6500 Plus mass spectrometer (SCIEX, Framingham, MA, USA). The chosen column for this procedure was the BEH C18 (2.1 mm × 10 cm, 1.7 μm, Waters), utilizing a mobile phase comprising 0.1% formic acid in water (Liquid A) and a 30% isopropanol–acetonitrile mixture (Liquid B). The QTRAP 6500 Plus instrument was operated in both positive and negative ion-switching modes, effectively identifying metabolites as they were eluted from the column. Detailed analyses of chromatographic peak areas and retention times were conducted using Skyline (v.21.1.0.146) software [63]. The derived results underwent further refinement using metaX software (v3.2), facilitating the extraction of pertinent compounds and quantitative data suitable for rigorous analysis. The metabolite identification results were informatively annotated using the HMDB database. During the mass spectrometry uptake of the samples, a certain number of quality control samples was added for data quality assessment. OPLS-DA analysis was conducted on metabolite data with the ropls package in R (v4.2.2) software to obtain VIP values. The fold change and q-value of metabolites in the treatment group were determined relative to those the control group, primarily utilizing the Benjamini–Hochberg (BH) approach. Differential metabolite screening was conducted using the VIP value from the OPLS-DA model and the fold change and q-value in univariate analysis. The criteria for screening were VIP ≥ 1 in the OPLS-DA model, fold change ≥ 1.2 or ≤0.83, and q-value < 0.05. The results of the differential metabolites were visualized using the ggplot2 package in R (v4.2.2) software.

Spearman’s rank correlation analysis was utilized to investigate the interactions between the top 20 differential metabolites and the top 20 differential microbiota across various intestinal regions (proximal small intestine, distal small intestine, cecum, colon, and feces) in both metformin-treated and control groups. After adjusting for multiple testing using the false discovery rate (FDR) correction, the analysis selectively showed only correlations with q-values less than 0.05. The correlation results were visualized using the corrplot package in R (v4.2.2) software.

The data that support the findings of this study are deposited in the Bioproject (National Center for Biotechnology Information (NCBI)) at https://submit.ncbi.nlm.nih.gov/subs/sra/SUB14377791/overview↗ (accessed on 1 June 2024), reference number [PRJNA1100315].