The glucose-lowering effects of α-glucosidase inhibitor require a bile acid signal in mice

Male C57BKS-Lepr^−/− (db/db) mice were purchased from the Model Animal Research Centre of Nanjing University (Nanjing Jiangsu, China). Nr1h4^tm1Gonz/J (also known as Fxr-KO) mice were purchased from the Jackson Laboratory (RRID: IMSR_JAX:004144; Bar Harbor, ME, USA) and maintained on a C57BL/6J background. Only male animals were used in this study. Mice were maintained in a facility in Shanghai Jiao Tong University School of Medicine (Shanghai, China).

After purchase, 6-week-old db/db mice were maintained in the facility and subjected to chow diet (ReadyDietech, Shenzhen, China) for 2 weeks. Only mice that grew to have a bodyweight ≥25 g and a random blood glucose ≥15 mmol/l were included for further study. Eight-week-old db/db mice with similar bodyweights and random glucose levels were assigned to the AGI treatment group or the control group. For the AGI group, AGI (Acarbose; Huadong Medicine, Hangzhou, China) was mixed with the chow diet at a set concentration of 1000 mg/kg to make the daily dosage around 100 mg/kg bodyweight [22, 23]; mice were fed the AGI-mixed diet for 24 days. For the control group, vehicle treatment was provided for 24 days, which was the same chow diet as the AGI group but did not contain the AGI.

At 8 weeks of age, Fxr-KO mice and their wild-type littermates were fed an HFHS diet (ReadyDietech; 32% of kJ from fat, 25% of kJ from sucrose) for 10 weeks. See electronic supplementary material (ESM) Table 1 for details of diet formulas. After this, both genotypes were assigned to the AGI treatment or control group with HFHS diet, for another 4 weeks.

All mouse experiments were repeated three times in different litters, each time with 5–8 mice per group. Randomisation and blinding were not carried out for the animal experiments.

After treatment, all mice were euthanised using 10% (wt/vol.) chloral hydrate (10 μl/g body weight) after overnight (16 h) fasting. Blood samples were collected and the liver, ileum and pancreas of mice were harvested and either fixed in 4% (wt/vol.) paraformaldehyde or snap frozen and stored at −80°C.

Mice were fed ab libitum and housed in specific pathogen-free isolators at 23 ± 1°C with a 12 h light/dark switch. All animal protocols were approved by the Animal Care Committee of Shanghai Jiao Tong University School of Medicine.

After 6 h fasting (from 08:00 hours to 14:00 hours), 22-week-old Fxr-KO mice and their littermates were subjected to IPGTTs, in vivo glucose-induced insulin secretion (GIIS) tests (measured by ELISA) and ITTs [24]. See ESM Methods for more details.

Bacterial DNA was extracted from colon content and faeces of mice as previously reported [25] and subjected to PCR amplification and Illumina MiSeq sequencing (San Diego, CA, USA). See ESM Methods for more details.

BAs were extracted from plasma, cecum content, liver and ileum tissues of mice and analysed by ultra-performance liquid chromatography-tandem mass spectrometry (UPLC-MS/MS) using the Waters Acquity UPLC system (Waters, Milford, MA, USA) coupled to a Triple Quad 5500 tandem mass spectrometer (AB Sciex, Framingham, MA, USA), as described previously [8].

Body composition (fat and lean mass) of mice was assessed by an animal whole body composition analyser (EchoMRI 100H, Houston, TX, USA), according to the manufacturer’s instructions.

Triacylglycerol and total cholesterol in both liver and plasma of mice were measured by the Quantification Colorimetric/Fluorometric Kit (Biovision, Milpitas, CA, USA), as previously reported [25]. See ESM Methods for further details.

Frozen mouse liver tissue sections with 5–10 μm thickness were subjected to Oil Red O and haematoxylin staining, using a standard protocol [26].

Selected continuous pancreas sections from mice were immunolabelled with insulin and the area of staining in each section was analysed. See ESM Methods for further details.

Pancreatic insulin content was extracted from murine samples by acidic ethanol and assayed by a mouse insulin ELISA kit (Alpco, Salem, NH, USA). See ESM Methods for further details.

Mouse islets were isolated and handpicked to perform GIIS assays ex vivo [24]. After incubation with low glucose (3.3 mmol/l) or high glucose (16.7 mmol/l) for 1 h, the supernatant was collected or islets were lysed in acidic ethanol. The insulin level in supernatant and islet extracts were assayed by a mouse insulin ELISA kit (Alpco). See ESM Methods for further details.

Immunolabelling of insulin, glucagon, MafA and urocortin-3 (UCN3) in mouse pancreas sections, and image processing were performed as previously described [24]. See ESM Methods for further details; antibody details are listed in ESM Table 2.

Real-time quantitative RT-PCR was used to determine the relative levels of gene transcription in murine samples. Expression levels were normalised to the housekeeping gene 36b4 (also known as Rplp0). See ESM Methods for further details. All primers used in this study were listed in ESM Table 3.

Tissues of mice were lysed, quantified, and blotted, as described previously [25]. See ESM Methods for further details; antibody details are listed in ESM Table 2.

The extracted RNA from isolated mouse primary islets was subjected to library construction and sequencing based on the BGISEQ-500 platform [27]. See ESM Methods for further details.

The two-step hyperglycaemic clamp was employed to evaluate changes in beta cell function [28] in individuals with type 2 diabetes before (n = 10) and after (n = 9) AGI treatment. These individuals were enrolled in a 3 month intervention study to examine the glucose-lowering and gut microbiota effects of AGI (ClinicalTrials.​gov. registration no. NCT01758471) [8]. Of the 106 participants in the original study, 20 individuals (n = 10 in the AGI group, n = 10 in the control group) additionally volunteered to take part in the clamp study; n = 19 completed the clamp study. All the study participants gave additional informed consent specifically for the clamp study. The trial conformed to the provisions of the Declaration of Helsinki and was approved by the ethics committees at each participating centre. See ESM Methods and ESM Table 6 for further details of the experimental procedures and participant characteristics.

A subset of db/db mice were subjected to an experiment comparing the effects of AGI treatment after antibiotic pre-treatment. Antibiotic pre-treatment of db/db mice was employed as a proxy for inducing germ-free status before AGI treatment. See ESM Methods for further details.

Statistical calculations were conducted with SPSS v11.0 software (SPSS, Chicago, IL, USA). Two-group comparisons were analysed by two-tailed Student’s t test. Multiple groups and repeated measurements were analysed by single-tailed two-way ANOVA followed by Fisher’s least significant difference (LSD) post hoc test. Two-tailed Mann–Whitney U tests were used for analysing all BA metabolome data (including data from plasma, liver, ileum and cecum content samples of mice) and the gut microbiota data. All results are expressed as the mean±SEM. A p value of <0.05 was considered statistically significant.

First, we treated db/db mice with AGI for 24 days (Fig. 1a). AGI consistently reduced body weight and blood glucose in db/db mice vs untreated controls (Fig. 1b, c) during the intervention, similar to its effects in humans [8, 29]. Body composition and plasma and liver lipids were all improved by the end of treatment, except for total cholesterol in the liver (ESM Fig. 1). With the same bacterial load (Fig. 1d), 16S ribosomal RNA sequencing revealed significant differences in gut microbiota composition between AGI treated and untreated mice (Fig. 1e), such as enriched phylum Actinobacterium and Proteobacterium and depleted phylum Bacteroides (Fig. 1f). Though not exactly the same, AGI-induced changes in relative abundance of Bacteroides and Bifidobacteriaceae in the db/db gut microbiome were similar to previous reports in human participants (Fig. 1g) [8]. The BA pool sizes were enlarged in the liver and plasma of db/db mice treated with AGI vs untreated controls, but these were not significantly different in cecum content (Fig. 1h–j). The enhanced primary/secondary BA ratio (PBA/SBA) in BA pools in the blood of db/db mice after AGI treatment vs untreated controls (Fig. 1k–m, ESM Fig. 2a and ESM Table 4) was consistent with what has been observed in human participants and what has been linked with improved metabolic parameters [8, 30]. In addition, compared with the control, the known FXR antagonist [31], taurine-conjugated β murine cholic acid (TβMCA), increased in cecum content but decreased in the blood and liver with AGI treatment (Fig. 1n–p), in line with changes in taurine-conjugated cholic acid (TCA)/TβMCA ratio (indicating FXR activation; Fig. 1q–s), suggesting that the FXR was activated in the liver and inhibited in the distal ileum. Accordingly, FXR-targeting genes, such as those encoding cytochrome P450, family 8, subfamily b, polypeptide 1 (CYP8B1) in the liver, which promotes cholic acid synthesis, and fibroblast growth factor 15 (FGF15) in the distal ileum, which inhibits BA synthesis, were both downregulated by AGI vs untreated controls (ESM Fig. 2b). Furthermore, hepatic expression of Cyp7a1, which encodes the key enzyme regulating BA synthesis (cytochrome P450, family 7, subfamily a, polypeptide 1 [CYP7A1]) was not different between untreated controls and AGI-treated mice (ESM Fig. 2d), probably due to the converged effect of downregulated gut FGF15 and activated hepatic FXR [15, 31]. In contrast, genes regulating ileum BA reabsorption, such as those encoding the sodium-dependent cholesterol transporter (Asbt; also known as Slc10a2) and organic solvent tolerance protein β (Ostβ; also known as Slc51b), were upregulated by AGI (ESM Fig. 2c), indicating increased gut BA reabsorption. Taken together, AGI induced similar changes in gut microbiome and BA compositions in db/db mice as those observed in humans. We further found that enterohepatic BA circulation, rather than liver BA synthesis, could possibly be more greatly affected by AGI treatment.

To further test if BA signalling was responsible for the metabolic benefits of AGI, we examined the effects of AGI in Fxr global KO (Fxr- KO) mice on a C57BL/6J background or their wild-type littermates that were fed on an HFHS diet (details of the study design are illustrated in Fig. 2a). Obesity and hyperglycaemia were induced in mice before AGI treatment (ESM Fig. 3a, b). The metabolic improvements induced by AGI treatment of wild-type vs untreated wild-type mice, such as improved fasting blood glucose, glucose tolerance and blood triacylglycerol levels, were attenuated in AGI-treated Fxr-KO mice vs untreated KO mice (Fig. 2d–h). Of note, though the effect was lessened, glucose tolerance was still significantly improved in Fxr-KO mice treated with AGI compared with untreated mice (Fig. 2e, f). Plasma total cholesterol was different depending on genetic background of the mice (p < 0.05 for AGI treated or untreated Fxr-KO vs respectively treated/untreated wild-type mice), rather than AGI treatment (p > 0.05 for AGI treated wild-type vs untreated wild-type and AGI treated KO vs untreated KO mice) (Fig. 2h), which is in line with previous studies [32, 33]. Further, improved insulin sensitivity (Fig. 2i) along with lower in vivo GIIS (Fig. 2j) with AGI treatment (the ‘so-called’ insulin-sparing effect [8, 29] of AGI) was also notably attenuated in AGI-treated KO mice.

The role of FXR in regulating insulin sensitivity and liver metabolism is well recognised [16, 33]. AGI reduced liver weight and hepatic triacylglycerol accumulation (Fig. 2k–m) with reduced expression of the gene encoding cell death-inducing DNA fragmentation factor alpha-like effector protein a (Cidea; Fig. 2o) and reduced gluconeogenesis via inhibition of the genes encoding glucose-6-phosphatase, catalytic subunit (G6pc; p < 0.05) and phosphoenolpyruvate carboxykinase 1, cytosolic (Pepck; also known as Pck1; p = 0.053) (Fig. 2p); these changes were abrogated in Fxr-KO. Thus, metabolic phenotyping suggests that the glucose- and lipid-lowering effects of AGI are partially dependent on FXR activity, which could be due to FXR-dependent improvements in insulin hypersecretion and hepatic metabolism.

We then sought to characterise the signature of BA pools and alterations in signalling in the HFHS-fed mice treated by AGI. The BA profiles in plasma, liver, ileum and cecum content of mice were assayed (Fig. 3a, ESM Table 5) and significantly increased plasma and liver BA pool sizes and downregulated cecum pool sizes were observed with AGI treatment vs untreated controls (Fig. 3b–e). These changes were similar but to a greater extent than those observed in db/db mice (Fig. 1h–j). The PBA/SBA ratio in blood, liver and cecum content were all increased by AGI treatment, whilst the 12α-hydroxylated/non-12α-hydroxylated BA ratio was only increased in the liver but not in other compartments vs controls. In addition, the unconjugated/conjugated BA ratio in the blood and ileum were increased upon AGI treatment vs controls (Fig. 3f–i); this trend was also observed in the liver of HFHS-fed mice but did not reach significance (Fig. 3g). Also, like in db/db mice, the changes in the proportion of TβMCA (Fig. 3j–m) and the ratio of TCA/TβMCA (Fig. 3n–q) in BA pools of different compartments were consistent with those in db/db mice (Fig. 1n–s), i.e. the TCA/TβMCA was enhanced in the blood and liver but was lower (non-detectable) in cecum content with AGI treatment vs controls (Fig. 3n–q). Compared with controls, Fxr-KO mice showed similar changes in BA pool as wild-type mice following treatment with AGI, further supporting the notion that the alterations in microbial BA metabolism following treatment with AGI might underlie the AGI-induced BA pool repartition.

We then examined FXR/BA signalling in the enterohepatic tissues. Similar to the findings in db/db mice, AGI treatment exerted diverse effects on hepatic and ileum FXR activity and impacted on host BA transportation. We found elevated transcription of the gene encoding bile salt export protein (Bsep; also known as Abcb11) and protein levels of small heterodimer partner (SHP), and decreased transcription of Cyp8b1 and the gene encoding cytochrome P450, family 7, subfamily b, polypeptide 1 (Cyp7b1) in liver (Fig. 4a–c), suggesting activated hepatic FXR. Of note, the expression of Cyp7a1, the gene encoding key BA synthesis enzyme, remained unaltered in the liver following AGI treatment (Fig. 4c). Meanwhile, expression of downstream signals of FXR, Shp (also known as Nr0b2) and Fgf15, were diminished in the distal ileum (Fig. 4d). The BA reabsorption genes Asbt and Ostβ were increased in the ileum, whilst genes encoding the BA importers sodium taurocholate cotransporting polypeptide (Ntcp; also known as Slc10a1) and organic anion transporting polypeptide (Oatp; Slco1a1) were decreased (Fig. 4e) and the expression of a BA exporter (Bsep) was enhanced (Fig. 4a) in the liver after AGI treatment. Deletion of Fxr nearly ablated all the molecular changes observed with AGI treatment, except for those of Cyp8b1, Cyp7b1 and Oatp.

From these findings, we proposed that the BA pool repartitioning after AGI treatment, i.e. decreased biotransformation of PBA to SBA, altered TCA/TβMCA in enterohepatic tissues, enhanced gut BA reabsorption and enlarged BA pool size in the liver and circulation, may alter BA signalling in the liver and ileum (via inhibition of FXR in the gut and activation of FXR in the liver), exerting opposite effects on hepatic BA synthesis, but similar effects with regard to improving metabolic homeostasis [16, 34, 35].

We went on to investigate whether FXR also mediated the insulin-sparing effect of AGI. In AGI-treated wild-type mice, we found significantly reduced beta cell mass vs untreated wild-type controls (Fig. 5a) and decreased beta cell proliferation, manifesting as reduced percentage of Ki67-labelled beta cells (Fig. 5b–c). No changes in pancreatic islet structure (ESM Fig. 4a) and the ratio of beta/alpha cells were found with AGI treatment (Fig. 5d) vs untreated controls. Pancreatic insulin content showed a minor but significant elevation after AGI treatment (Fig. 5e). On the other hand, though in vivo treatment of AGI decreased levels of ex vivo insulin secretion under high glucose conditions in primary islets (basal insulin secretion showed a trend of decrement with no significance) (Fig. 5f), it did not affect ex vivo GIIS responses (Fig. 5g) as compared with islets from untreated wild-type mice. However, the AGI-related islet changes were nearly all absent in AGI-treated Fxr-KO mice, suggesting a role for islet FXR signalling in regulating the adaptation of the endocrine pancreas in response to nutrient overload. Notably, AGI rescued islet structure and insulin levels in db/db mice (ESM Fig. 4b,c), with similar effects of decreasing beta cell proliferation (ESM Fig. 4d,e) along with recovery of beta cell MafA and UCN3 levels (key proteins that preserve the mature function of beta cells; ESM Fig. 4f–i).

Though not being able to acquire data on how beta cells proliferate in human participants, we measured beta cell function in ten AGI-treated individuals with type 2 diabetes during a two-step hyperglycaemic clamp (all participants had enrolled onto a clinical trial [ClinicalTrials.​gov registration no. NCT01758471]; details are provided in the ESM Methods and ESM Table 6). Despite mitigated insulin secretion response during the oral glucose tolerance test (ESM Table 6), the clamp data showed a better acute insulin secretion response to a meal test post vs pre-treatment (ESM Table 7, ESM Fig. 5) [8], although this was not significant (p = 0.077). Overall, the downregulated beta cell mass, proliferation and absolute insulin secretion with AGI treatment may be beneficial to beta cell preserve in diabetes and may also require FXR signalling.

To further understand the mechanism underlying the FXR-dependent response of pancreatic islets to AGI treatment, we performed RNA sequencing on isolated primary islets from different mouse groups. In total, there were 465 differentially expressed genes (DEGs) that were induced by AGI treatment in wild-type mice vs untreated wild-type mice and 497 in Fxr-KO mice vs untreated Fxr-KO mice (fold change ≥2.0, q ≤ 0.001) and only 157 genes overlapped in these two comparisons, as shown in the Venn diagram in Fig. 6a. We hypothesised that the genes overlapping between DEG gene sets in untreated vs AGI-treated wild-type mice and untreated wild-type vs untreated Fxr-KO mice, excluding those that were different between untreated vs AGI-treated Fxr-KO mice, were those targeted by AGI and dependent on FXR activity. In total 119 DEGs of this kind were further analysed using gene ontology (GO) enrichment (Fig. 6b). Regulation of cell proliferation was among the 15 most significantly annotated GO terms (Fig. 6c). The 17 DEGs in this GO were all downregulated after AGI treatment in wild-type mice and included the gene for baculoviral inhibition of apoptosis protein (IAP) repeat-containing 5 (Birc5), a gene encoding survivin, which is known for its role in promoting pregnancy- and surgery-induced beta cell proliferation [36, 37]. Though not included in the annotation of cell proliferation, expression of MKi67 (gene encoding Ki67) was also significantly decreased after AGI treatment in wild-type mice and did not change in Fxr-KO mice (Fig. 6c), which paralleled with the Ki67 immunostaining results in islets (Fig. 5b). However, though genes regulating insulin secretion biosynthesis (i.e. Ins1 and Ins2), exocytosis (i.e. Slc30a8 and Slc2a2) and beta cell identity (i.e. Mafa, Foxa2 and Pdx1) were not assigned as DEGs (fold change <2.0), they were slightly but significantly altered by AGI treatment in wild-type mice vs untreated mice (q ≤ 0.001; ESM Fig. 6a) [38‐40]. How these changes may affect beta cells in the long term awaits further study. Islet immunolabelling of MafA showed no obvious changes between the mouse groups (ESM Fig. 6b). In addition, FXR protein levels were comparable in wild-type mice treated with and without AGI and absent in treated and untreated Fxr-KO mice (ESM Fig. 6c). Thus, the transcriptome analysis suggested that the AGI-induced inhibition of the cell proliferation signalling pathway in pancreatic islets might lead to decelerated beta cell hyperplasia and cell mass expansion in an FXR-dependent manner.