Ginsenoside Rb1 (>98% purity) was purchased from Desite Co., Ltd. Chengdu. The structure of Rb1 was shown in Figure 1A. The nuclear magnetic resonance (NMR) spectra obtained for Rb1 were shown in Supplementary Figures S1, 2. The HFD was purchased from Trophic Animal Feed High-Tech (Catalog number: TP 23300). Methanol and acetonitrile were provided by Sigma-Aldrich (St. Louis, MO, United States). Acetone and other analytical chemical solvents were obtained from Tianjin Chemical Factory (Tianjin, China).

Six-week-old C57BL/6J specific pathogen-free (SPF) male mice were raised in 5–6 per cage in rooms with constant temperature (25 ± 2°C), humidity (50 ± 20%), 12 h dark-light cycle, and free access to normal chow diet and water. After adaption for a week, mice (n = 18) were randomly separated into three groups, NC group and HFD group were fed with normal chow diet and high fat diet food, respectively, while Rb1 group received oral doses of ginsenoside Rb1 (200 mg/kg/day) and high fat diet food simultaneously. The food ingredients are listed in Supplementary Table S1. Mice were assessed weekly for body weight and food intake index. After 12 weeks of feeding, mice were euthanized, blood and faeces were collected at the end of experiments. All samples were collected and stored at −80°C until analysis. All procedures were performed in accordance with the guidelines of the Animal Care and Use Committee of Nankai University, Tianjin, China (2006–0009), and conformed to the Guide for the Care and Use of Laboratory Animals publish Ted by the National Institutes of Health (NIH Publications No. 8023, revised 1978). The Committee approval number is NO. SYXK (JIN) 2019-0001.

At the end of animal experiment, all mice were euthanatized with diethyl ether to collect blood. The blood samples were placed at room temperature for 15 min, and then were centrifuged at 3,000 rpm for 15 min to obtain serum. All serum samples were stored at −80°C for further experiments.

Total triglyceride (TG), total cholesterol (TC), low-density lipid cholesterol (LDL-C), high-density lipid cholesterol (HDL-C), and free fatty acids (FFA) in serum together with TG in liver were measured by Elisa kit (Nanjing Jiancheng Bioengineering Institute). All protocols were conducted according to manufacturer recommendations.

At week 10, mice were fasted for 12 h and glucose tolerance test (GTT) was performed after an intraperitoneal injection of glucose (1 g/kg body weight). Blood was collected at tail vein and the concentration was measured with an Accu-Check glucometer (Roche). Blood glucose levels were tested before (0 min) and after (15, 30, 60, 90 and 120 min) for glycaemia determination. At the end of week 11, animals were fasted for 4 h, and an insulin tolerance test (ITT) was performed after an intraperitoneal injection of insulin (0.7 UI/kg). Blood glucose concentrations were measured before (0 min) and after (15, 30, 60, 90 and 120 min) insulin injection.

Total RNA was extracted from liver. RNA was used for cDNA synthesis using a reverse transcription PCR kit (TransGen Biotech). Real-time qPCR was performed using Tip Green qPCR SuperMix (TransGen Biotech) with 1:10 diluted cDNA product from the transcription product. GAPDH (Glyceraldehyde-3-phosphate dehydrogenase) was used as housekeeping gene. Data was calculated according to 2^ΔΔCt method. Primer sequence for targeted mouse gene was available in Table 1.

Faeces DNA was extracted using PowerSoil DNA Isolation Kit (MoBio Laboratories, Carlsbad, CA) following the manual. Purity and quality of the faeces genomic DNA were checked on 0.8% agarose gels. The V3-4 hypervariable regions of bacterial 16S rRNA gene were amplified with the primers 338F (ACT​CCT​ACG​GGA​GGC​AGC​AG) and 806R (GGACTACHVGGGTWTCTAAT). The PCR products were purified using a QIAquick Gel Extraction Kit (QIAGEN, Germany), quantified using Real Time PCR, and sequenced at Allwegene Company, Beijing. Deep sequencing was performed on Miseq platform at Allwegene Company (Beijing). After the run, image analysis, base calling and error estimation were performed using Illumina Analysis Pipeline Version 2.6. The raw data were first screened, and sequences were removed from consideration if they were shorter than 200 bp, had a low-quality score (≤20), contained ambiguous bases or did not exactly match to primer sequences and barcode tags. Qualified reads were separated using the sample-specific barcode sequences and trimmed with Illumina Analysis Pipeline Version 2.6. And then the dataset was analyzed using QIIME. The sequences were clustered into operational taxonomic units (OTUs) at a similarity level of 97% (Edgar, 2013), to generate rarefaction curves and to calculate the richness and diversity indices. The Ribosomal database Project (RDP) Classifier tool was used to classify all sequences into different taxonomic groups (Cole et al., 2008). To examine the similarity between different samples, clustering analyses and principal component analysis (PCA) were used based on the OTU information from each sample using R (Wang et al., 2012). The evolution distances between microbial communities from each sample were calculated using the tayc coefficient and represented as an Unweighted Pair Group Method with Arithmetic Mean (UPGMA) clustering tree describing the dissimilarity (1-similarity) between multiple samples (Jiang et al., 2013). A Newick-formatted tree file was generated through this analysis. To compare the membership and structure of communities in different samples, heat maps were generated with the top 20 OTUs using Mothur (Jami, et al., 2013).

PICRUSt was used to performed the function profiling of microbial communities’ prediction (Langille et al., 2013). OTU table was normalized by 16S copy number predictions so that OTU abundance more accurately reflected the true abundances of the underlying organisms. Predicted functional genes were categorized into Clusters of Orthologous Groups (COG) and into Kyoto Encyclopedia of Genes and Genome (KEGG).

Fecal samples (100 mg) were supplemented with 10 μL of internal standards and 500 μL of methanol, and then extracted by vortexing for 1 min. After centrifuged at 12,000 rpm and 4°C for 10 min, the supernatant was detected using a GC–MS system (Agilent Technologies, Santa Clara, CA), equipped with a polar HP-FFAP capillary column. Injection volume was 2 μL and helium was used as a carrier gas at a constant flow rate of 2 ml/min through the column. The initial oven temperature was kept at 80°C for 1°min, raised to 170°C at a rate of 10°C/min, to 220°C at a rate of 20°C/min, and finally kept 220°C for 3 min. SCFAs standards were mixtures of acetic acid, propionic acid, butyrate, isobutyric acid, valeric acid, and isovaleric acid.

10 μL serum was mixed with 20 μL methanol, vortexed for 1 min, and then centrifuged at 4°C for 15 min. Supernatants were collected, and 2.5 μL supernatant or amino acids mixture standards solution was combined with 2.5 μL orthophthalaldehyde (OPA). High performance liquid chromatography (HPLC) Agilent 1,100 series system equipped with a diode array detector was used to analyze 5 μL mixed solution, fluorescence is monitored at excitation and emission wavelengths of 340 and 460 nm. The separation was performed on a Kromasil C18 reverse-phase column (250 mm × 4 mm, 5 μm), protected with an Ultrasep C18 guard column, which was eluted by a binary mobile phase consisting of water containing 0.1% formic acid (A) and acetonitrile (B) with a gradient. The levels of Val, Leu, Iso, histidine (His), serine (Ser), arginine (Arg), glycine (Gly), Tyr, alanine (Ala), phenylalanine (Phe), lysine (Lys), and tryptophan (Trp) were tested, and the chromatogram of amino acids were shown in Supplementary Figures S3, 4.

IPA and IS standard were mixed in 20 μL with 60 μL methanol, and 10 μL serum was mixed with 30 μL methanol. The liquid was vortexed for 1 min, and supernatants were then dried in a SpeedVac. The sediment was resuspended in 50% methanol-water and then underwent HPLC separation on a Kromasil C18 reverse-phase column (250 mm × 4 mm, 5 μm). Water with 0.1% formic acid and acetonitrile were used as mobile phases A and B. The fluorescence was monitored at excitation and emission wavelengths of 280 and 370 nm. The chromatogram of IPA and IAA were shown in Supplementary Figures S5, 6.

The data were expressed as the mean ± SEM. One-way analysis of variance (ANOVA) followed by Tukey’s post-hoc test was applied to assess intergroup variations in all other cases except gut microbiota results. One-way analysis of variance (ANOVA) followed by kruskal-wallis post-hoc test was applied to assess gut microbiota results. Spearman’s rank test was applied to assess the correlation results. The statistical analysis was performed using GraphPad (GraphPad Prism, United States). A p-value of <0.05 indicates a significant difference.

To investigate the protective effect of Rb1 on glucose control, we established a model of HFD-induced mice by raising C57BL/6J with high fat diet feeding for 12 weeks. Rb1 supplementation to HFD mice did not change body weight and energy intake when compared to HFD group (Figures 1B–D). Compared to NC group, liver weight was significantly higher while cecum weight was lower in mice fed with HFD. Significant lower liver weight and higher cecum weight were observed in mice fed with Rb1 as against HFD group (Figures 1E,F).

Contrasted with NC group, impaired glucose control and insulin resistance as well as higher liver TG, serum LDL-c, HDL-c, fasting glucose and free fatty acids were also observed in mice fed with HFD (Figure 2). In agreement with the observed reduction of liver weight, Rb1 supplementation reduced hepatic triglycerides accumulation (Figure 2A). However, Rb1 did not reverse HFD induced elevated levels of cholesterol and triglycerides in serum (Supplementary Figure S7). Moreover, Rb1-treated mice reduced serum fasting glucose, LDL-c and free fatty acid but increased HDL-c compared to HFD group (Figures 2B–E). In addition, significantly improved of glucose homeostasis and insulin sensitivity were observed by our intraperitoneal insulin tolerance test (IPITT) and intraperitoneal glucose tolerance test (IPGTT) data (Figures 2F–I).

Considering the role of liver in the development of obesity, expression levels of hepatic genes involved in glucose metabolism were evaluated. Rb1 treatment significantly down-regulated the uncoupling protein 2 (UCP2) expression in the liver to improve insulin sensitivity (Diano and Horvath, 2012; Kim, et al., 2019). Nuclear receptor subfamily one group H member 4 (Nr1h4) regulates amino acids catabolism and detoxification of ammonium by ureagenesis and glutamine synthesis (Massafra et al., 2017). In addition, the farnesoid X receptor (FXR, encoded by Nr1h4) would regulate the homeostasis of BAs, lipids and glucose, which has been reported to have important role in obesity and hypercholesterolemia (Watanabe et al., 2006). Our analysis revealed that the mRNA expression of Nr1h4 is significantly upregulated by Rb1 (Figure 3). Furthermore, the expression of Fiaf, a circulating inhibitor of lipoprotein lipase, was significantly increased after Rb1 treatment, which was accorded with the decreased liver weight (Bäckhed et al., 2004). Activators of both PPARα and PPARγ have been shown to improve insulin sensitivity and normalize impaired glucose tolerance in both humans and rodent models of insulin resistance (Francque et al., 2021). Rb1 increased the PPARγ expression compared to HFD though did not show significance. In liver, PPARα regulates genes involved in lipid and lipoprotein metabolism (Pawlak, et al., 2015). However, the administration of Rb1 reduced the expression of PPARα, which need to be explored further.

Previous study demonstrated the synergistic effect of polysaccharides and Rb1 on diabetic model, owing to poor bioavailability of Rb1, the glucose control mechanism related to gut microbes was considered (Li et al., 2018). High-quality gut microbiota sequences were obtained from all the fecal samples of mice by 16S rRNA gene sequencing analysis (Supplementary Figure S8). Microbial α-diversity analysis showed that Rb1 did not recover chao1 index and shannon index of HFD-induced mice (Figures 4A,B). Beta diversity analysis using PCA and nonmetric multidimensional scaling (NMDS) exhibited distinct clusters of NC, HFD, and Rb1 groups (Figures 4C,D), which implied that Rb1 treatment affected partial microbiota.

To examine the changes in greater detail, we next estimated differences using the perspective of cluster analysis. The relative abundance at the phylum, class, order and genus levels were shown in Figure 4E and Supplementary Figures 9A–C. Significant Rb1 induced changes were found for two phyla (Actinobacteria and Verrucomicrobia), four classes (Actinobacteria, Melainabacteria, Verrumcomicrobiae, and Bacilli), and four genera (Allobaculum spp. Reyranella spp. Eubacertium coprostanoligenes, and Akkermansia spp.). The relative abundance of phylum Actinobacteria was significantly lower in HFD group compared with NC group. Rb1 significantly enriched the abundance of phyla Verrucomicrobia and Actinobacteria. (Figures 5A,B). The abundance of three classes, Actinobacteria, Melainabacteria, and Bacilli was lowered in HFD group compared with NC group, while Rb1 supplementation significantly reversed this trend (Figures 5C–E). In addition, we observed that the Verrumcomicrobiae was markedly enriched in Rb1 group in comparison with both NC and HFD group (Figure 5F). The abundance of three genera, Allobaculum spp. Reyranella spp. and Eubacertium coprostanoligenes, was higher in HFD group compared to NC group, which was significantly reduced after Rb1 treatment (Figures 5G–I). Interestingly, Akkermansia spp. known as a probiotic to ameliorate obesity and immune response, was markedly increased by Rb1 (Figure 5J). To reveal the involvement of SCFAs, the main metabolites of gut microbiota, in the Rb1 treatment on HFD-induced mice, six SCFAs in serum were measured. All of the SCFAs were significantly decreased in the HFD-fed mice compared with NC group, which were not recovered by Rb1 supplementation (Supplementary Figure S10).

For the purpose of better understanding Rb1-mediated the gut microbiota changes, we performed the functional profiling of microbial communities’ prediction by PICRUSt. As shown in Figure 4F, the results revealed that the predicted KEGG pathways mainly enriched in amino acid metabolism, glycan biosynthesis and metabolism, and energy metabolism.

Amino acid metabolism, enriched in the predicted KEGG pathways by PICRUSt analysis, plays a vital role in glucose homeostasis, where most of the metabolites then enter into the citric acid cycle to synthesize glucose (Holecek, 2020). Increasing evidences suggested that amino acids are emerging as a new class of effective molecules in the etiology of obesity and diabetes mellitus (Hasanpour, et al., 2020). Among all amino acid analyzed, in contrast to NC group, levels of Iso, His, Ser, Tyr, Ala, and Trp in serum were significantly higher in HFD-fed rats. Meanwhile, Rb1 supplementation decreased levels of Iso, His, Leu, Tyr, Ala, and Trp (Figures 6A–E).

The metabolism of the aromatic amino acid Trp results in numerous effects on immune homeostasis, epithelia function, and gut motility. In addition, as the metabolites of Trp-like indoleacetic acid (IAA), IPA and IS are known to affect intestinal permeability and host immunity (Venkatesh et al., 2014; Huang et al., 2020). The results indicated that tryptophan microbial metabolites IPA was significantly lower in HFD group compared to NC group, and Rb1 supplementation markedly increased the level of IPA and decreased the level of IS (Figures 6F,G).

BCAAs in particular, are strongly associated with insulin resistance (Zeng et al., 2020). The Spearman’s correlation test was applied to investigate the correlation between amino acids and gut microbiota. As depicted in Figure 7A, 19 genera showed strong correlation with Iso and Leu levels in serum, four genera of which were positively correlated and 15 genera were negatively correlated. Among the 19 genera, Eubacterium coprostanoligenes was significantly correlated with Leu and Iso simultaneously.

We also performed Spearman’s correlation analysis of gut microbiota with Trp, IPA and IS. As seen in Figure 7B, 38 genera were correlated with Trp level and its metabolites, and we found that most genera that are negatively correlated with Trp are also positively correlated with IPA. Among 38 genera, Allobaculum spp. and Reyranella spp. were significantly altered by HFD and Rb1 treatment.

As shown in Figure 7C, 50 genera were correlated with the levels of Tyr, Ser, His, and Ala in serum. Allobaculum spp. was positively correlated with Tyr, Ser, and His levels in serum, and Reyranella spp. was negatively correlated with Tyr and Ala. Interestingly, we found that Ala was the only amino acid correlated with Akkermansia spp., and the relative abundance of Akkermansia spp., was significantly increased in Rb1-treated mice. Thus, Rb1 might increase insulin sensitivity by modulating Eubacterium coprostanoligenes, Allobaculum spp., Reyranella spp. and Akkermansia spp. to improve the metabolism of amino acids in mice. Moreover, the glucose-alanine cycle necessitates an examination of its quantitative contribution to overall glucose balance, which suggested that Rb1 could regulate homeostasis by normalizing the metabolism of alanine in obese mice (Felig, 1973).