Fresh kiwifruits were collected from Shaanxi Province, China. Chemical standards of acrylamide (purity ≥99.0%) was procured from Aladdin Chemical Co. (Shanghai, China). Astragalus polysaccharides were procured from Xi'an Tianxingjian Natural Bio-products Co., Ltd (Xi'an, Shaanxi Province, China). Measurement kits for serum alanine transaminase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (ALP), and triglyceride (TG) were supplied by Nanjing Jiancheng Bioengineering Institute (Nanjing, Jiangsu Province, China). Dibasic sodium phosphate, sodium dihydrogen phosphate, formaldehyde, absolute ethanol, and dimethylbenzene were procured from Sichuan Xilong Scientific Co., Ltd., (Chengdu, Sichuan Province, China). All other chemicals of analytical grade were purchased from Sinopharm Chemical Reagent Co., Ltd., (Shanghai, China).

KFP were isolated from fresh kiwifruits as described previously with slight amendments (23). Briefly, fresh kiwifruits were crushed and the homogenate was filtered through four layers of gauze. The filter residue was collected, dried, and pulverized. KFP from pulverized residue were extracted in water (1:40, m/v). The homogenate was filtrated as before; the filtrate was collected for later use, and the filter residue was subjected to secondary water extraction as before. The two filtrates were combined and then centrifuged. The obtained supernatant was precipitated with 95% ethanol at 4°C overnight. The collected precipitate was deproteinized, purified by an anion-exchange chromatography of DEAE-52, dialyzed, and lyophilized to obtain KFP. The monosaccharide composition and molecular weight (Mw) of KFP were analyzed as reported previously (24). In this study, the average Mw of KFP was 7.60 × 10⁵ Da, and the molar ratio of monosaccharide composition was rhamnose: arabinose: galactose: glucose: galacturonic acid (1:4:11:40:10). Infrared spectroscopy results showed that KFP is a polysaccharide in the form of pyranose and it is connected by a β-glucoside bond.

Male, 6-week-old Balb/c mice (20 ± 2 g) were purchased from Xi'an Earnsville Biotechnology (Xi'an, Shaanxi, China) (Production license number: SCXK (jing) 2019-0010). Animals were reared in specific pathogen-free (SPF), standard animal facilities at 22 ± 2°C and 50 ± 10% humidity with caged free rearing, and 12 h light/dark cycle. The mice had ad libitum access to a standard diet (AIN-93M) and water. After adapting to the above conditions for 7 days, 60 mice were randomly divided into six groups (n = 10 per group): control group (CK), acrylamide model group (AM), low-dose KFP group (KFP-L), medium-dose KFP group (KFP-M), high-dose KFP group (KFP-H), and astragalus polysaccharide positive group (APS). The animal experiment design is depicted in Figure 1. Firstly, mice were intragastrically administered with distilled water (CK group) or 30 mg/kg·bw AA solution (in all other groups). After 1 h, the KFP groups were administrated with 100, 300, and 500 mg/kg·bw KFP solution in KFP-L, KFP-M, and KFP-H groups, respectively; the APS group was administrated with 500 mg/kg·bw APS solution. Meanwhile, CK and AM groups were given distilled water. AA, KFP and APS were administered to mice once a day. We determined the AA dose based on a previous study (25) and preliminary experiments. The entire experiment lasted 6 weeks. During this period, the water intake, food intake, and change in body weight of mice were recorded every week.

After 6 weeks of KFP intervention, all animals were fasted for 12 h and then anesthetized with isoflurane. Mice blood samples were collected and centrifuged at 3,000 rpm for 10 min at 4°C. The obtained serum was stored at −80°C for further analysis. Liver tissue was rapidly removed and weighed. A part of the liver was fixed in 4% paraformaldehyde for tissue sections. The other part was immediately stored at −80°C for further analysis. Also, the cecal contents were collected and stored at −80°C for gut microbiota analysis.

All animal protocols were approved by the Medical Ethics Review Committee of Xi'an Medical College, Xi'an, China. All operations were performed under anesthesia to minimize animal suffering.

The serum levels of ALT, AST, ALP, and TG were measured by commercial assay kits and an automatic biochemical apparatus (BS-460; Shenzhen Mindray Biomedical Electronics) following the manufacturer's instructions. The fresh liver tissues were fixed in 4% paraformaldehyde for more than 24 h, embedded with paraffin, sliced about 4 μm, and finally, stained with hematoxylin and eosin (H&E). The stained liver sections were examined under an optical microscope and the section images were collected by Pannoramic 250 digital slice scanner (3DHISTECH, Hungary). The liver sections were first observed at low magnification for gross lesions and then the selected areas were examined and imaged at 400 x magnification for specific lesions.

The cecal contents were thawed before DNA extraction using the OMEGA Soil DNA Kit (M5635-02) (Omega Bio-Tek, Norcross, GA, USA). The extracted DNA samples were quantified and subjected to PCR amplification of the bacterial 16S rRNA gene V3-V4 region using the forward primer and reverse primers 338F (5'- ACTCCTACGGGAGGCAGCA-3') and 806R (5'- GGACTACHVGGGTWTCTAAT-3'), respectively. Meanwhile, PCR amplicons were purified with Vazyme VAHTSTM DNA Clean Beads (Vazyme, Nanjing, Jiangsu province, China) and then quantified by Quant-iT PicoGreen dsDNA Assay Kit (Invitrogen, Carlsbad, CA, USA). After the quantification, individual amplicons were pooled in equal amounts and subjected to pair-end 2 × 250 bp sequencing on the Illumina MiSeq platform with MiSeq Reagent Kit V3 at Bioyi Biotechnology Co., Ltd (Wuhan, Hubei province, China). Microbiome bioinformatics was performed with QIIME2 2019.4 with slight modifications with the help of official tutorials (https://docs.qiime2.org/2019.4/tutorials/) (26). Sample alpha diversity was analyzed by QIIME2 and R packages. Alpha diversity include the Chao 1, Shannon and Simpson indices, which represent the richness and diversity of gut microbiota. Beta diversity was analyzed by principal coordinate analysis (PCoA) and nonmetric multidimensional scaling (NMDS) based on the Bray-Curtis metrics. Differential microbiota between groups was identified by linear discriminant analysis (LDA) and LDA effect size (LEfSe) with an LDA threshold of 2. PCoA, NMDS, and LEfSe analysis were evaluated by the R, Ape, Vegan, and Python LEfSe packages.

The serum samples were subjected to metabolism analysis as reported previously with minor modifications (27). Briefly, 50 μL serum was extracted with a 200 μL admixture of methanol and acetonitrile to precipitate proteins. The respective admixture containing isotope-labeled internal standard was vigorously vortexed, ultrasonicated, incubated, and then centrifuged. The supernatants were filtered through a 0.22 μm microfiltration membrane and then analyzed by Ultra High-Performance Liquid Chromatography (UHPLC) coupled with Q Exactive Hybrid Quadrupole-Orbitrap mass spectrometer (UHPLC-QE-MS). The quality control sample included an equal amount of all samples. Instrument detection was performed using a UHPLC system (Vanquish, Thermo Fisher Scientific) with a Waters ACQUITY UPLC BEH Amide column (2.1 × 100 mm, 1.7 μm) coupled to Q Exactive HFX mass spectrometer (Orbitrap MS, Thermo). The mobile phase-A consisted of 25 mmol/L ammonium acetate and 25 ammonia hydroxide in water (pH = 9.75); the mobile phase-B was acetonitrile. The Q Exactive HFX mass spectrometer is capable of primary and secondary mass spectrometry data acquisition in information-dependent acquisition (IDA) mode under the control of system software (Xcalibur, Thermo). The ESI source was set to positive and negative modes under the following conditions: capillary temperature, 350 °C; full MS resolution at 60,000; MS/MS resolution as 7,500; collision energy, 10/30/60 in NCE mode; spray voltage, 3.6 kV (positive mode) or −3.2 kV (negative mode).

The raw data were processed by ProteoWizard software and the self-written R package (the kernel: XCMS) for peak identification, peak extraction, peak alignment, and integration. SIMCA (V16.0.2, Sartorius Stedim Data Analytics AB, Umea, Sweden) was used to perform principal component analysis (PCA) and orthogonal projections to latent structures-discriminant analysis (OPLS-DA). The OPLS-DA model was verified by 200 permutation tests; the model quality was assessed by Q² and R² values. Besides, potential metabolite markers and pathways enrichment analysis were identified from the biochemical databases including HMDB (http://www.hmdb.ca/), PubChem (https://pubchem.ncbi.nlm.nih.gov/), Kyoto Encyclopedia of Genes and Genomes (KEGG; http://www.kegg.com/) and MetaboAnalyst (http://www.metaboanalyst.ca/).

Statistical analysis was performed using GraphPad Prism 8 and SPSS 19.0 software (SPSS Inc., Chicago, IL). Differences among the groups were assessed by one-way analysis of variance followed by multiple comparison tests using the least square method or Student's t-test. Data are shown as mean ± SD. The correlations between gut microbiome and metabolites were analyzed based on the Spearman rank correlation coefficient. The correlation coefficients were always between −1 and +1. If the absolute value of the correlation coefficient was closer to 1, the linear relationship was better.

We used KFP intervention at different doses (100, 300, or 500 mg/kg·bw daily for 6 weeks) against AA toxicity in mice. Animals were induced with AA toxicity at 30 mg/kg·bw of AA. The experimental design is shown in Figure 1.

To evaluate the effect of KFP on weight loss of AA-induced mice, the weight of each mouse was measured weekly (Table 1). There was no significant difference in the initial body weight of animals. In the 4th week of the study, the body weight of AM mice decreased significantly compared with other mice. Loss in body weight is a characteristic of AA toxicity, indicating the successful establishment of the AA-mice model. Meanwhile, at 3–6 weeks, the body weight of KFP-M mice was significantly higher than other AA-induced mice (p < 0.05), indicating the intervention effect of KFP. The change in the food and water intake of mice during the study are shown in Table 1. Consistent with body weight loss, the food intake of AA-induced mice was lower than CK mice (p < 0.05). There was no significant difference in water intake among the animals of the CK, APS, and KFP-M groups. However, compared with KFP-M mice, the water intake of AM and KFP-L mice was significantly higher. Concisely, these results showed that exposure to 30 mg/kg AA negatively affected body weight and food and water intake of mice. However, some of these adverse effects were alleviated by KFP intervention.

Compared with CK mice, the serum levels of AST, ALT, ALP, and TG were significantly higher in AA-induced mice. However, compared with AM mice, these were significantly lower in KFP and APS-treated mice (p < 0.05). Notably, KFP-M mice had much-improved biochemical parameters than other groups mice (Figures 2A–D). These results suggested that AA promoted liver damage in AM mice, while different doses of KFP and APS exerted a noticeable protective effect against AA-induced liver damage.

The effect of KFP or APS intervention on the liver histology of AA-induced mice is shown in Figure 3. In the CK group, there was no thickening of the liver tissue capsule; the hepatic lobules were not clearly divided; the hepatic cords were arranged neatly; the hepatocytes were arranged radially around the central vein, and the morphology of the hepatocytes was normal. The structure of arteries, veins, and bile ducts was relatively complete (Figure 3a). In contrast, AA treatment resulted in hepatocyte necrosis, increased lymphocyte count, obvious vacuolar degeneration of hepatocytes in the hepatic lobule, lighter staining of degenerated hepatocytes, sparse cytoplasm, and enhanced light transmittance (Figure 3b). KFP-L and KFP-H mice showed slight amelioration of the histopathological effects of AA toxicity, however, degeneration of some hepatocytes was evident (Figures 3c, e). Interestingly, APS and KFP-M mice showed a gradual return to normal histopathological morphology with hepatocyte regeneration (Figures 3d, f), demonstrating the hepatoprotective of KFP at the medium level (300 mg/kg·bw daily for 6 weeks). In all, liver pathological analysis was consistent with serum biochemical analysis.

Next, we examined if KFP hepatoprotective effects were associated with gut microbiota health. The cecal contents of AA-induce mice were analyzed after 6 weeks of KFP-M intervention for microbiota diversity and composition by 16S rRNA gene sequencing. α-diversity analysis was performed based on operational taxonomic units, including Chao1, Shannon, and Simpson indices. As shown in Figure 4A, the Chao 1 index showed no difference among CK, AM, KFP-M, and APS mice, indicating no negative effect of AA toxicity on the richness of mice gut microbiota. Meanwhile, the Shannon and Simpson indices of AM mice were lower than CK, KFP-M, and APS mice, demonstrating that AA toxicity disturbed the diversity of mice gut microbiota. However, KFP or APS intervention significantly improved the diversity of mice gut microbiota. Furthermore, PCoA and NMDS plots were used to evaluate the influence of AA, KFP, and APS on the composition and structure of mice gut microbiota (Figures 4B, C). The results showed a significant deviation between CK and AM mice, indicating that AA altered the composition of mice gut microbiota (Figures 4B, C). The gut microbiota structures of KFP-M and APS mice were entirely different from AM mice, suggesting that KFP-M or APS intervention significantly improved the gut microbiota health of AA-induced mice (Figures 4B, C). To further explore the specific differences, the taxonomic composition of the mice gut microbiome was analyzed at the phylum and genus levels. At the phylum level (Figure 4D), the gut microbiota mainly consisted of Firmicutes, Bacteroidetes, Proteobacteria, and Actinobacteria, accounting for more than 95% of the total bacteria. The Firmicutes were the most abundant bacteria in all mice. Compared with CK mice, in AM mice, the relative abundance of Firmicutes increased from 45.8 to 57.7% (^***p < 0.001) and Actinobacteria from 2.31 to 4.27% (^*p < 0.05), while the relative abundance of Proteobacteria decreased from 10.11 to 4.37% (^**p < 0.01) (Figure 4E). Notably, KFP-M or APS intervention could not reverse these changes. At the genus level, AM and KFP-M mice showed changes in the relative abundance of gut microbiota to varying levels (Figure 4F). In detail, compared with CK mice, AM mice showed prominently increased abundance of Allobaculum, Oscillospira, but decreased abundance of Lactobacillus, Ruminococcus and Bacteroides (Figure 4G; ^****p < 0.0001, ^***p < 0.001, ^**p < 0.01, ^*p < 0.05). However, KFP-M intervention significantly reduced the abundance of Oscillospira and increased the abundance of Lactobacillus in AA-induced mice (Figure 4G; ^****p < 0.0001, ^***p < 0.001, ^**p < 0.01, ^*p < 0.05). Intriguingly, gut microbiota composition did not change at the phylum level but improved at the genus level in KFP-M mice. Concisely, the results suggested that a medium dose of KFP could alleviate gut microbiota dysbiosis in AA-induced mice. LEfSe analysis combined with nonparametric Kruskal-Wallis and Wilcoxon rank sum tests and LDA effect size can help identify the biomarkers. The results of LEfSe analysis are shown in Figure 4H; there were 16, 9, 5, and 11 potential gut microbiota biomarkers (LDA > 2) in CK, AM, KFP-M, and APS mice, respectively.

Considering the remarkable protective effect of medium-dose KFP on weight, serum biochemical parameters (ALT, AST, ALP, and TG), and liver pathology of AA-induced mice, we next examined the metabolites of CK, AM and KFP-M mice by serum untargeted metabolomics. UHPLC-QE-MS was used to find the serum metabolite information in positive and negative ion modes. The metabolite trends across different treatment groups were visualized by PCA. The variance explained by the two principal components (PC) is shown on the X and Y axes in a two-dimensional score plot. In Figure 5A, PC1 and PC2 accounted for 16.2 and 11.8% of the variance in the positive mode. Likewise, in Figure 5B, PC1 and PC2 accounted for 17.8 and 13.3% of the variance in the negative mode. The CK and AM mice groups were clearly separated in PCA score plots, indicating remarkable and massive inter-group variation in serum metabolites. Moreover, PCA analysis showed good separation between the KFP-M and AM groups, indicating a clear difference in their metabolites. The OPLS-DA maximized the differences between groups. In the OPLS-DA model, S-plot can identify the characteristics of inter-group differences. The OPLS-DA score plots manifested clear separations of metabolic profiles between AM and CK groups in both positive and negative modes (Figures 5C, E); the model quality evaluation index under positive ion mode was R²Y = 0.999, Q² = 0.920 (Figures 5C, D), and R²Y = 0.996 and Q² = 0.927 under negative ion mode (Figures 5E, F). The metabolites of KFP-M and AM groups also showed a good separation in both positive and negative ion modes (Figures 5G, I); the model quality evaluation index was R²Y = 0.999 and Q² = 0.780 in the positive ion mode (Figures 5G, H), and R²Y = 0.974 and Q² = 0.785 in the negative ion mode (Figures 5I, J). These values with R²Y > Q² and Q² > 0.5 indicated the stability and reliability of the model.

To more intuitively observe the overall distribution of metabolites between CK vs. AM groups and AM vs. KFP-M groups, the visually changed compounds were screened and displayed by volcano plot based on the variable importance in the projection (VIP) values > 1.0, fold change (FC) > 2.0 or < 0.05, and p-values < 0.05. The volcano plots showed that compared with CK mice, in AA-induced mice, there were 116 significantly altered metabolites (71 upregulated and 45 downregulated) under the positive mode and 95 significantly altered metabolites (47 upregulated and 48 downregulated) under the negative mode (Figures 5K, L). After the KFP-M intervention, compared with AM mice, in total, 86 (13 upregulated and 73 downregulated) and 77 (17 upregulated and 60 downregulated) differential metabolites were identified under positive and negative modes (Figures 5M, N). To further identify critical metabolites associated with KFP-M intervention, potential metabolic biomarkers were identified through a heatmap combined with VIP scores, which were used to rank the contribution of individual metabolites between different groups. The metabolites with top 40 VIP scores were selected as potential biomarkers, all of them had VIP > 1 and p < 0.05. As shown in Figure 6A, these 40 compounds exhibited significant differences between CK and AM groups (p < 0.05), 20 of which increased, while the other 20 decreased in AM mice compared with CK mice. Meanwhile, the 40 compounds also exhibited significant differences between AM and KFP-M groups (p < 0.05; Figure 6B); 9 metabolites increased and 31 decreased in KFP-M mice compared with AM mice. In addition, among the screened 80 differential metabolites, 21 were common in CK, AM, and KFP-M mice.

Moreover, to explore the most correlative pathways and possible biological mechanisms of KFP-M intervention on AA toxicity, the functions of identified metabolites were further analyzed through the KEGG, PubChem, and HMDB databases. As shown in Figure 7A, 9 metabolism pathways were highly enriched in AM mice compared with CK mice, including nicotinate and nicotinamide metabolism, taurine and hypotaurine metabolism, pyrimidine metabolism, glycine, serine, and threonine metabolism, thiamine metabolism, arginine and proline metabolism, primary bile acid biosynthesis, biosynthesis of unsaturated fatty acids, and valine, leucine and isoleucine biosynthesis. Likewise, 5 significantly enriched pathways were observed in KFP-M mice compared with AM mice, including nicotinate and nicotinamide metabolism, alanine, aspartate and glutamate metabolism, histidine metabolism, D-arginine and D-ornithine metabolism, and D-glutamine and D-glutamate metabolism (Figure 7B). Additional details are shown in Supplementary Table S1 indicating 44 different metabolites and 39 metabolism pathways among CK, AM and KFP-M groups. In the nicotinate and nicotinamide metabolism pathway, nicotinic acid mononucleotide, niacinamide, 1-methyl-nicotinamide, and nicotinic acid were significantly increased in AM mice with CK mice. Whereas, the same were significantly decreased after KFP intervention (^***p < 0.001, ^**p < 0.01 and ^*p < 0.05). Meanwhile, nicotinic acid mononucleotide and nicotinic acid were also identified as important metabolites (Figure 6). Compared with the CK group, chenodeoxycholic, allocholic, and cholic acids were increased in AM mice, and while KFP treatment decreased chenodeoxycholic, cholic, and taurocholic acids of the primary bile acid biosynthesis pathway (^***p < 0.001, ^**p < 0.01 and ^*p < 0.05). Meantime, chenodeoxycholic and cholic acids were identified as important metabolites (Figure 6). Compared with CK mice, taurocholic acid of the taurine and hypotaurine metabolism pathway was increased in AM mice and decreased after KFP intervention. In sum, 38 metabolism pathways and 43 differential metabolites were alerted in AM mice compared with CK mice, of which, 21 metabolism pathways and 24 differential metabolites were alleviated by KFP intervention. Therefore, based on the relationships between metabolites and metabolism pathways, a schematic overview of the prominent differential metabolites correlated with enriched metabolism pathways was constructed in Figure 8. These results manifested that KFP-M intervention ameliorated the metabolism disturbances of AA-induced toxicity.

A correlation matrix was established to calculate the Spearman correlation coefficient for the comprehensive analysis of relationships between metabolites and gut microbiota (Figure 9). Twenty four gut microbiota correlated with 21 differential metabolites. For instance, g_Lactobacillus was extremely significant correlated with 17 differential metabolites (^**p < 0.01), including cholic acid, taurocholic acid, chenodeoxycholic acid, carnosol, nicotinic acid, and nicotinic acid mononucleotide; g_Parabacteroides showed extremely significant correlation with 14 differential metabolites (^**p < 0.01) including cholic acid, taurocholic acid, chenodeoxycholic acid, nicotinic acid, and nicotinic acid mononucleotide; g_Alistipes had extremely significant correlation with 13 differential metabolites (^**p < 0.01) including carnosic acid, cholic acid, and chenodeoxycholic acid; g_Odoribacter had extremely significant correlation with 11 differential metabolites (^**p < 0.01) including carnosic acid, cortisone, and chenodeoxycholic acid. Among the 21 differential metabolites, chenodeoxycholic acid was extremely significantly correlated with 10 gut microbiota (^**p < 0.01) including p_Firmicutes, p_Bacteroidetes, p_Proteobacteria, g_Lactobacillus, g_Bacteroides, g_Alistipes, g_Odoribacter, g_[Prevotella], g_Prevotella and g_Parabacteroides; carnosic acid was extremely significantly correlated with 10 gut microbiota (^**p < 0.01) including p_Firmicutes, p_Bacteroidetes, p_Proteobacteria, g_Lactobacillus, g_Bacteroides, g_Alistipes, g_Odoribacter, g_[Prevotella], g_Prevotella and g_Parabacteroides; cholic acid showed extremely significant correlation with 7 gut microbiota (^**p < 0.01) including g_Lactobacillus, g_Bacteroides, g_Alistipes, g_[Prevotella], g_Prevotella and g_Parabacteroides and g_Roseburia. Overall, the correlation analysis suggested that metabolites and gut microbiota could affect each other.