Rapid-acting insulin injections were sourced from Novo Nordisk Pharmaceuticals Co., Ltd. Hematoxylin-eosin (H&E) stain was purchased from Thermo Fisher Scientific (USA), and Oil Red O from Sigma-Aldrich (USA). Commercial kits for triglycerides (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C) were acquired from MyBioSource, Inc. (China). Rat Tumor Necrosis Factor Alpha (TNF-α) and Rat Interleukin 6 (IL-6) ELISA Kits were obtained from Elabscience, Inc. (China).

The eight herbal medicines constituting the various SQ formulations were procured from Sichuan Xinhehua Traditional Chinese Medicinal Herbs Co., Ltd. A detailed breakdown of the composition and constituents of SQ is provided in Table 1. The pharmaceutical preparation process was as follows: Panax ginseng C.A.Mey and other herbal components were separately soaked in purified water for 30 min. Subsequently, Panax ginseng C.A.Mey was boiled for 30 min. After this initial boiling phase, it was combined with the remaining herbal components. Upon reaching boiling, the mixture was vigorously boiled for an additional 40 min. The decoction was then filtered to obtain the medicinal liquid. The filtrate, obtained after two rounds of boiling, underwent distillation, concentration, and subsequent freeze-drying, resulting in a dried residue constituted the extraction material for SQ compound. This dried extract was stored at −20°C and thoroughly mixed with distilled water before use. A saxagliptin (SAX) suspension was prepared by dissolving saxagliptin tablets (Production License No.: J20160069; AstraZeneca Pharmaceuticals Ltd.) in distilled water.

Thirty-eight eleven-week-old male spontaneous type 2 diabetes Goto-Kakizaki (GK) rats (332–374 g) (bred from a Wistar strain background) and twelve age-matched male non-diabetic Wistar rats (365–379 g) were obtained from the Cavens Laboratory Animal Co., Ltd (Changzhou, Jiangsu; Permit No.: SCXK(Su)2016-0010). All experimental rats were housed under controlled conditions, at a constant temperature of 21°C ± 2°C, relative humidity of 60% ± 10%, and a 12/12-hour light-dark cycle, with ad libitum access to food and water.

The study employed four experimental groups: blank control (CON, n = 12), untreated diabetic model (MOD, n = 12), positive control (saxagliptin, 0.83 mg/kg body weight) (SAX, n = 13), and diabetes with medium-dose SQ (1.44 g/kg body weight) (SQ, n = 13). Following an adaptation period, the Wistar rats were fed a diet composed of 20% protein, 10% fat, 70% carbohydrates at 342 kcal/100g, over 10 weeks. To establish the T2DM model, GK rats were fed a high-fat diet, consisting of 88.2% standard feed, 10% refined lard, 1.5% cholesterol, and 0.3% pig bile salt. After 2 weeks of dietary manipulation, induction of T2DM in the GK rats was confirmed by random blood glucose levels that exceeded 11.1 mmol/L. To simulate glycemic variability in the T2DM rats, a regimen of staggered peak glucose and insulin intraperitoneal injections was employed. The rats were administered an intraperitoneal infusion of a 25% glucose solution at a volume of 0.375 mL/(kg∙d) at 10:00 AM. This was followed by a subcutaneous injection of NovoRapid insulin solution, at a dosage of 0.7 units per 100 g of body weight (U/100g), at 14:30 PM. The CON group received equivalent volumes of intraperitoneal injections of 0.9% saline and subcutaneous sham injections of the same intensity. The rats in the SAX and SQ groups were administered with the assigned compound via gastric gavage, once daily, throughout the eight-week intervention phase. Rats in the CON and MOD groups received equivalent volumes of purified water. At the end of the experiment, all rats underwent a 12-hour fast with ad libitum access to water before being sacrificed. Blood, liver, and fresh feces samples were collected and immediately stored at −80°C for subsequent analysis.

Weekly recordings of blood glucose levels were conducted at five time points (8:30 AM, 10:30 AM, 14:30 PM, 16:30 PM, and 18:30 PM) to determine the glycemic fluctuations. Several parameters, including the largest amplitude of glycemic excursions (LABG), daily mean blood glucose (MBG), and the standard deviation of daily average blood glucose (SDBG), were employed to evaluate day-to-day glycemic variability among the different groups. The success of the diabetes glycemic variability model induction was confirmed in line with expert consensus (Danne et al., 2017), which recommends SDBG ≥2.0 mmol/L and LABG ≥4.4 mmol/L.

Animal body mass and FBG were determined. The OGTT was conducted at the end of the experiment, and involved a 14 h fast, followed by an oral 2.0 g/kg dose of glucose with pre- and post- (0, 30, 60, and 120 min post glucose) blood glucose quantification via a retro-orbital bleed and glucometer. The fasting insulin in serum (FINS) levels were determined using ELISA kits. The homeostasis model assessment-insulin resistance (HOMA-IR) was calculated via the following formula: The HOMA-IR = [fasting blood glucose (mmol/L)] × [fasting insulin (μIU/mL)]/22.5.

Biochemical serum lipid profiles, including TC, TG, HDL-C, and LDL-C, were quantified using a fully automated biochemical analyzer (Hitachi). In addition, specific ELISA assay kits were used to quantify TNF-α and IL-6 levels in plasma.

Anesthesia was performed by an intraperitoneal injection of 20% urethane solution at 0.5 mL/100g body weight. Immediately, fresh hepatic tissues were collected and weighed to calculate the liver organ index using the formula: Liver Mass/Body Weight × 100%. A portion of these liver tissues was fixed in 4% paraformaldehyde for 48 h and subjected to H&E staining for histological analysis. Histological images of the hepatic tissues were captured using an optical microscope (Leica Microsystems, Japan). Five frozen liver samples from each group were flash-frozen at −80°C for 48 h to prepare them for oil red O staining. Three tissue sections were selected from each sample to observe the aggregation of lipid droplets (red signal) and the changes in the positive signal in the liver under an optical microscope. Image Pro Plus 6.0 software was used to calculate the area of the positive oil red O signal.

16S rRNA sequencing employed a fecal genomic DNA kit (Tiangen et al., China). DNA extraction was performed by adopting the CTAB/SDS method from the colon contents samples. The amplification of the V3 and V4 hypervariable regions of the 16S rRNA gene was performed using specific primers, 515F and 806R. Subsequently, the sequencing library was constructed by employing the TruSeq DNA PCR-Free Sample Preparation Kit (Illumina, USA), and high-throughput sequencing was performed on the NovaSeq6000 sequencing platform. The bioinformatics analysis of the resulting microbiome data was carried out using QIIME 2 software.

Hepatic tissue samples (100 g) were homogenized and combined with a pre-chilled methanol/acetonitrile mixture (volume ratio 1:1) in Eppendorf (EP) tubes. Following thorough agitation and ultrasonication in an ice-water bath for 10 min, the samples underwent high-speed cold centrifugation at 12,000 g for 15 min at 4°C. Subsequently, 500 μL of the supernatant was extracted and placed in EP tubes, then dried in a vacuum concentrator. The dried extracts were reconstituted with 160 μL of a water/acetonitrile mixture (volume ratio = 1:1) and centrifuged at 12,000 g for 15 min at 4°C. The resulting supernatant was then subjected to liquid chromatography-mass spectrometry (LC-MS).

Metabolomic analysis was conducted using an ultra-high-performance liquid chromatography (UPLC) system (Waters Acquity I-Class, China) coupled with a quadrupole time-of-flight (QTOF) mass spectrometer (Waters Xevo G2-XS, China). Analytical separation was achieved using a Waters UPLC HSS T3 column (2.1 × 100 mm, 1.8 μm). The UPLC system was operated at a constant flow rate of 0.4 mL/min, with an injection volume of 1 μL for gradient elution. Eluent A (0.1% formic acid) and eluent B (0.1% formic acid in acetonitrile) were employed in positive ion mode. The cone voltage was set as 30V, and the ion source temperature was maintained at 100°C. Desolvation gas flow was set at 800 L/h. An auxiliary gas heating temperature of 500°C and a counter-flow gas rate of 50 L/h were applied.

Raw data acquired from the UPLC-MS were analyzed using the Masslynx 4.2 code. Before further analysis, the raw data underwent preprocessing by employing the Progenesis QI code. Metabolite identification was conducted by referencing the METLIN database and a custom library, followed by data normalization. Principal component analysis (PCA) and orthogonal projections to latent structures-discriminant analysis (OPLS-DA) were performed within the SIMCA-P+ software to further refine and select differential metabolites. Volcano plots were constructed by utilizing the ggplot two package in the R programming language. The correlation analysis between metabolites and differential microbial communities was conducted using the Pearson correlation coefficient.

Quantitative data were analyzed and presented as mean ± standard deviation. Statistical data analysis was conducted using SPSS 25.0, using one-way analysis of variance (ANOVA) and Student’s t-tests. Significance thresholds were set at p < 0.05 and p < 0.01, denoting different levels of confidence.

To assess glycemic variability during the modeling phase, the average blood glucose fluctuation trend was monitored at five specific time points within 24 h at the third, sixth, and tenth weeks. As illustrated in Figure 1 A-C, the GK rats exhibited significant peaks and troughs in blood glucose levels over 24 h, in sharp contrast to the stable blood glucose levels of the CON group. Additionally, elevated blood glucose levels in the MBG, SDGB and LAGE groups among diabetic rats (Figures 1D–F) confirmed the successful establishment of the diabetic glycemic variability model.

Following the establishment of the T2DM model, diabetic rats displayed symptoms of polydipsia, polyphagia, polyuria, lean body shape, weakened reaction, dull hair, and local skin rash. These metabolic phenotypes were further characterized. Figures 2A, B show that, despite similar food intake, diabetic GK rats experienced a gradual increase in body weight and higher blood glucose levels compared to the CON group. Intervention with SAX and SQ did not significantly affect body weight but did lower blood glucose levels in T2DM rats in later stages of the experiment, with reductions in FBG levels by 16.02% and 22.65%, respectively.

Additionally, compared to the CON group, diabetic rats demonstrated reduced glucose tolerance via OGTT (Figures 2C, D), increased insulin levels (Figure 2E), and higher HOMA-IR (Figure 2F), indicating marked insulin resistance. SAX and SQ treatments improved glucose tolerance in OGTT. Meanwhile, the serum FINS levels in the SAX and SQ groups decreased by 23.97% and 21.61%, respectively, along with a significant reduction in HOMA-IR values, indicating that these treatments regulated blood glucose homeostasis in diabetic rats.

Given the close link between dyslipidemia, chronic low-grade inflammation, and insulin resistance (Jaiswal et al., 2014), a comparative analysis of serum biochemical parameters (HDL-C, LDL-C, TC, TG) and plasma inflammatory factors (IL-6, TNF-α) was conducted. In the MOD group, diabetic rats exhibited significantly elevated levels of TC, TG, HDL-C, and LDL-C in serum (Figures 3A–D), along with increased levels of IL-6 and TNF-α in plasma (Figures 3E, F), signifying the presence of dyslipidemia and a severe inflammatory response induced by the high-fat diet. Compared to the MOD group, SAX and SQ interventions significantly reduced the serum levels of HDL-C, LDL-C, TC, TG, and plasma levels of TNF-α in the T2DM rats. Additionally, SAX treatment notably decreased plasma IL-6 levels. These outcomes suggest that SQ has a beneficial effect in correcting blood lipid abnormalities and mitigating inflammatory responses in diabetic rats.

The liver, critical for lipid metabolism, can be adversely affected by excessive fat intake. As illustrated in Figure 4A, in the MOD group, histological analysis revealed structural damage and lipid degeneration, as evidenced by small vacuoles within the cytoplasm. This damage is likely attributable to the direct impact of the high-fat diet on hepatic cells. Additionally, oil red O staining revealed numerous lipid droplets, accompanied by increased liver weight and liver coefficient in the MOD group (Figures 4B–D), providing further evidence of disordered hepatic lipid metabolism. Following intragastric administration of SAX and SQ, liver wet weight and coefficient were significantly reduced, and oil red O staining indicated fewer and smaller lipid droplets in the liver, demonstrating the efficacy of SQ in reducing lipid accumulation.

Alpha diversity analysis explored changes in bacterial diversity and abundance, as illustrated in Figures 5A–C. The ACE, Chao1, and Shannon indices collectively demonstrated a notable decrease in microbial diversity and abundance in the MOD group compared to the CON group. Following interventions with SAX and SQ, an increase in bacterial diversity and abundance was observed. Beta diversity analysis, aimed at assessing the overall structural changes in the gut microbiota, utilized principal Coordinate Analysis (PCoA). The analysis showed that the first two axes (PCo1 and PCo2) accounted for 27.97% and 12.87% of the total variance, respectively, indicating significant separation and distinct clustering among the samples (Figure 5D).

At the phylum level, intervention with SAX and SQ significantly altered the relative abundance of Firmicutes and Bacteroidetes, resulting in a decreased Firmicutes/Bacteroidetes (F/B) ratio, which is considered indicative of a healthier composition of the gut microbiome (Figure 5E). At the genus level, compared to the CON group, the MOD group exhibited higher relative abundances of Ruminococcus_1 and Phascolarctobacterium, along with lower levels of Akkermansia and Lachnospiraceae_NK4A136_group. Following treatment with SAX and SQ, these imbalances were corrected, shifting towards the bacterial profiles observed in the CON group, suggesting a normalization of gut microbiota (Figure 5F).

Linear Discriminant Analysis Effect Size (LEfSe) analysis, employing the Linear Discriminant Analysis (LDA) effect size, identified significant biomarkers with an LDA score >4, indicating substantial differences in observed variations. This analysis identified species such as Roseburia_hominis and Lactobacillus_acidophilus as biomarkers that are significantly associated with the SQ treatment groups. These outcomes underscore the potential of SQ to beneficially modulate the gut microbiota composition in diabetic rats by reducing the relative abundance of pathogenic bacteria such as Ruminococcus and enhancing beneficial bacteria, notably Roseburia and Lactobacillus (Figure 5G).

An untargeted metabolomic approach was utilized to explore the impact of T2DM with glycemic variability on liver metabolism. The distribution of hepatic metabolic profiles between the diabetic and control groups showed conspicuous separation, indicating significant alterations in liver metabolites associated with metabolic dysregulation within the T2DM rat model (Figure 6A). Our OPLS-DA analysis further validated this, with the score scatter plots exhibiting clear separation for the MOD and SQ groups, emphasizing the metabolic impact of SQ treatment (Figure 6B). Permutation tests for the PLS-DA model, conducted 200 times, demonstrated the reliability of the model by showing that all blue Q² points were positioned below the green R² point from left to right, confirming the model’s validity and the absence of overfitting (Figure 6C). Metabolites related to the progression of T2DM were identified using a combination of Variable Importance in Projection (VIP) scores and p-values or fold change (FC) values from a univariate analysis. Volcano plots highlighted differentially regulated metabolites as potential biomarkers between the groups. Specifically, biomarkers differentiating the CON and MOD groups included 3beta-6beta-Dihydroxynortropane and Pro Ile Ile Val, while Eriojaposide A and N2-Succinoylarginine were identified between the SAX and MOD groups (Figures 6D,E). 91 metabolites distinguished the SQ group from the MOD group, with significant alterations in metabolites such as trans-Cinnamic acid and Testosterone enanthate, highlighting their potential as biomarkers of the therapeutic effect of SQ (Figure 6F, Supplementary Table S1).

A bubble map, based on the KEGG pathway annotation of regulated metabolites, was used to explore the potential metabolic pathways impacted by SQ treatment (Figure 6G). The results indicated enrichment in 38 metabolic pathways following SQ intervention. This enrichment was particularly notable in Glycerophospholipid metabolism, followed by other pathways such as Choline metabolism in cancer, Bile secretion, Steroid biosynthesis, and Vitamin digestion and absorption. These metabolic pathways provide further evidence for the potential of SQ to improve glycemic variability in diabetes.

A correlation heatmap was utilized to visually depict the strength and direction of associations, aiming to uncover potential functional connections between intestinal bacteria and liver metabolism. As shown in Figure 6H, the analysis indicated that the relative abundances of Flavonifractor and Alistipes were positively correlated with Testosterone enanthate. In summary, SQ exhibited the capacity to influence the composition of the gut microbiota, thereby contributing to improvements in hepatic metabolite concentrations.