YGYZ was provided by Wangjing Hospital of China Academy of Chinese Medical Sciences, with raw materials sourced from Kangmei Pharmaceutical (China). The composition and dosage of each herb in YGYZ are detailed in Table 1. Herbal components were macerated in 10 volumes of distilled water (1 h), then decocted twice. The combined aqueous extracts were lyophilized to yield a 4.0 g/mL crude extract (in distilled water), stored at 4 °C.

For sample preparation, 0.2 g of the herbal extract was dissolved in 50% methanol, centrifuged for 5 min, and filtered through a 0.22 μm membrane to obtain the test solution. Chromatographic separation was performed on a Waters ACQUITY UPLC HSS T3 column (2.1 × 100 mm, 1.8 μm) maintained at 35 °C using a Waters Synapt G2-Si Q-TOF mass spectrometer. The mobile phase consisted of 0.1% formic acid in water (phase A) and acetonitrile (phase B) with a flow rate of 0.25 mL/min. The gradient elution program was as follows: 0–15 min, 100–80% A; 15–50 min, 80–0% A; 50–60 min, 0% A; 60–70 min, 0–100% A. Mass spectrometric analysis was conducted using a Q-Exactive system with an injection volume of 10 μL.

All animal experiments were approved by the Ethics Committee of the Medical Experimentation Center of the China Academy of Chinese Medical Sciences (Approval No. ERCCACMS21-2302-07). The sample size was determined following the ARRIVE 2.0 guidelines, based on effect size estimation from preliminary experiments and calculation of the model’s coefficient of variation, while adhering to the 3R principles of animal ethics. For information on animal numbers, refer to the figure legends. All experiments used BALB/c mice (male, 6–8 weeks, 18–20 g) that were purchased from Beijing Vital River Laboratory Animal Technology (License No. SCXK (Jing) 2021-0006). The animals were housed in specific pathogen-free facilities with controlled temperature (22 ± 2 °C) and humidity (50 ± 10%). After 7-day acclimation with ad libitum feed/water, mice were maintained under 12:12 light: dark cycles.

Animals were randomly divided into six experimental groups using a random number table. LM models were established using a spleen-preserving technique (Xu et al., 2020) wherein 100 μL (2 × 10⁶ cells/mL) of ct26-luc cell suspension (American Type Culture Collection, United States) was slowly injected into the spleen, while sham-operated controls received equal-volume PBS injections. Following a 3-day recovery period with prophylactic penicillin treatment (40,000 IU, i.p.), six groups received different intervention plans respectively: Group Sham (0.9% saline, 0.1 mL/10 g/d, p.o.), Group Model (0.9% saline, 0.1 mL/10 g/d, p.o.), Group Capecitabine (Cap, H20133365, Jiangsu Hengrui Medicine, China), (150 mg/kg/d, p.o.), YGYZ-high dose (YGYZ-H, 39.78 g /kg/d), YGYZ-medium dose (YGYZ-M, 19.89 g/kg/d), and YGYZ-low dose (YGYZ-L, 9.95 g/kg/d). The clinical dosage of YGYZ was 2.19 g/kg/d (total herb weight per dose was 153 g, based on a human average body weight of 70 kg). The mouse equivalent dose was calculated as 19.89 g/kg/d through body surface area conversion (Nair and Jacob, 2016). Experimental performers were blinded to group allocation until data analysis was completed. All treatments were administered orally in 0.1 mL/10 g volumes. Terminal endpoints included comprehensive evaluations followed by tissue collection.

Fecal microbiota transplantation (FMT) was conducted as previously described (Wu et al., 2021; Zong et al., 2023) with modifications. Before this, we conducted a preliminary experiment and found that a YGYZ-high dose of 39.78 g/kg/d demonstrated the most significant therapeutic effects and safety in mice. Therefore, fresh YGYZ-H donor feces were homogenized in anaerobic PBS (1:10 w/v), centrifuged (800 × g, 3 min), and filtered (70 μm) to obtain a microbiota-rich supernatant (25 mg/mL). Group FMT received 0.25 g/kg/day via daily gavage.

For in vivo bioluminescence imaging, mice were intraperitoneally injected with D-luciferin potassium salt (150 mg/kg, ADP104300004809391, Promega, United States) and anesthetized using isoflurane (R510-22-16, Shenzhen Ruiwode Lift Technology, China). Animals were positioned in prone orientation within the imaging chamber of a Bruker In-Vivo Xtreme imaging system (Bruker, United States). Signal intensity was quantified with the following spectral representation: red indicating high signal intensity and blue representing background-level signals.

Liver and kidney function were assessed in serum samples using standardized colorimetric kits (Jiangsu Zecheng Biotechnology, China) with the following references: alanine aminotransferase (ALT, 20192400200), aspartate aminotransferase (AST, 20192400201), blood urea nitrogen (BUN, 20192400212), and creatinine (CRE, 20192400214).

Liver tissues were immediately fixed and paraffin-embedded, then sliced with a 3 μm thickness and air-dried. Stained with hematoxylin for nuclei and with eosin for cytoplasm after returning to the blue. Finally, dehydrated and sealed for examination by a microscope and image collection.

Liver tissues were immediately fixed and paraffin-embedded. Serial sections (3 μm) were cut, deparaffinized in xylene, and rehydrated through a graded ethanol series. Antigen retrieval was performed using sodium citrate buffer (pH 6.0) at 95 °C for 15 min. Sections were then incubated overnight at 4 °C with primary anti-Ki-67 antibody (1:200, BS-23103R, Beijing Boaosen Biotechnology, China) in a humidified chamber. Immunoreactivity was visualized using DAB chromogen, followed by hematoxylin counterstaining and sealed. Ki-67-positive cells were quantified under an Olympus BX53 microscope (Olympus, Japan).

Paraffin-embedded liver sections (3 μm) were deparaffinized and rehydrated through graded ethanol. Antigen retrieval was performed using citrate buffer (pH = 6.0) at 95 °C for 15 min. Endogenous peroxidase activity was blocked with 3% hydrogen peroxide for 10 min. Sections were incubated with optimally diluted primary antibodies in a humidified chamber at 4 °C overnight. HRP-conjugated secondary antibodies were applied for 50 min at room temperature (RT). Following microwave-mediated epitope recovery, the third antigen was detected using fluorophore-conjugated secondary antibodies (50 min, RT, light-protected). Nuclear counterstaining was performed with DAPI, followed by autofluorescence quenching (Solution B, 5 min) and mounting with anti-fade medium. Multispectral images were acquired using a fluorescence microscope (Nikon, Japan). Antibodies purchased from BioLegend Corporation (United States) were used for target detection: APC anti-mouse CD3 Antibody (1:200, 100235), PE anti-mouse CD4 Antibody (1:200, 100407), FITC anti-mouse CD8 Antibody (1:200, 140403), APC/Cyanine7 anti-mouse/human CD11b Antibody (1:200, 101225), APC anti-mouse Ly-6G Antibody (1:200, 127613), FITC anti-mouse Ly-6C Antibody (1:200, 128005).

Liver tissues were enzymatically digested with 0.25% trypsin solution. The filtered cell suspension was centrifuged and resuspended to obtain single-cell preparations. Cells were stained with the following fluorescently conjugated antibodies from BioLegend Corporation (United States): CD11b, Ly6G, Ly6C, CD3, CD4, and CD8. An unstained control was included for compensation. After 30 min incubation at 4 °C protected from light, cells were washed and fixed in 500 μL PBS containing 1% paraformaldehyde. Flow cytometry (Changzhou Bidak Biotechnology, China) was performed, and FlowJo7.6.1 Software (United States) was used to analyze the results.

Liver tissue homogenates were centrifuged, and the resulting supernatants were collected and diluted uniformly. Cytokine levels were quantified using commercial ELISA kits (Beijing Bioassay Systems Biotechnology, China): Arg-1 (arginase-1, MD35262), TGF-β (transforming growth factor-β, MD132594-1), IL-10 (interleukin-10, MD106493), IL-6 (interleukin-6, MD106412). Absorbance was measured at 450 nm using a microplate reader. Cytokine concentrations (Arg-1, TGF-β, IL-10, and IL-6) were quantified in ng/mL or ng/L by interpolating absorbance values against a standard curve generated with recombinant proteins.

Fresh fecal samples were collected to extract total microbial DNA, with DNA concentration and purity assessed by NanoDrop (United States). Sequencing libraries were prepared with the TruSeq Nano DNA LT Library Prep Kit (Illumina, United States). Libraries were sequenced on Illumina NovaSeq following standard denaturation (NaOH) and dilution protocols. Microbiome analysis was performed using Quantitative Insights Into Microbial Ecology. Metabolic pathway information was determined through the MetaCyc database.¹

Fresh fecal and liver tissue samples were collected for analysis. BAs profiling was performed using an ACQUITY UPLC^® BEH C18 column (2.1 × 100 mm, 1.7 μm) (Waters, United States) coupled to an AB Sciex 6500+ QTRAP mass spectrometer. Mass spectrometric detection employed multiple reaction monitoring (MRM) with optimized transitions for the quantification of individual BAs (Supplementary Table S1).

Homogenized tissues were lysed in lysis solution, mixed, and then cells were broken by sonication at maximum power (3 × 10s) and centrifuged (12,000 rpm, 15 min) to collect the supernatant. Protein concentrations were determined using the BCA assay, with samples normalized to 5 mg/mL. Proteins were separated by 10% SDS-PAGE and transferred to PVDF membranes. After blocking with 5% non-fat milk for 1 h at RT, membranes were incubated overnight at 4 °C with primary antibodies (Affinity Biosciences, United States): anti-β-actin (1:5000, AF7018), anti-kruppel-like factor 15 (KLF15) (1:1000, DF12203), anti-STAT3 (1:2000, AF6294), and anti-pSTAT3 (Tyr705, 1:1000, AF3293). Following washes, membranes were probed with HRP-conjugated secondary antibodies (1:5000) for 1 h at RT. Protein bands were visualized using enhanced electrochemiluminescence (ECL).

Statistical analysis was performed using SPSS 26.0 (IBM, United States), and graphs were generated with GraphPad Prism 9.0 Software (United States). All continuous variables were first assessed for normality. For multiple-group comparisons of normally distributed data, one-way ANOVA was employed, followed by appropriate post-hoc tests. Non-normally distributed data were compared among multiple groups using the Kruskal-Wallis test with post-hoc analysis. Microbiome and metabolome data were corrected for multiple comparisons (Benjamini-Hochberg FDR). Correlation analysis was conducted using Pearson’s correlation coefficient. A two-tailed p < 0.05 was considered statistically significant for all analyses.

Phytochemical profiling of YGYZ by UPLC-HR-MS/MS identified 95 compounds (Supplementary Table S2; Figure 1), including 20 flavonoids, 15 phenylpropanoids, 14 organic acids/derivatives, 12 terpenoids, 5 alkaloids, 8 phenolic compounds, 6 oligosaccharide/polysaccharide derivatives, and 15 other classifications. Key representative compounds included Glycyrrhizic acid, Kokusaginine, Icariin, Icaritin, Bergapten, Sagittatoside B, Licurazide, Epimedin C, and Hesperetin.

To evaluate YGYZ’s preventive effects on LM, we established the model via intrasplenic injection of ct26-luc cells (Day 0), followed by 24-day YGYZ administration (Figure 2A). Both YGYZ-H and YGYZ-M reduced hepatic metastatic nodules compared to the Model group, with efficacy comparable to Cap (Figure 2B). From day 0 to 12 gavage, the YGYZ-H group exhibited higher body weight gain compared to the Model (Figure 2C). From day 12 to 24 of gavage, Model group gained the faster body weight than Sham, Cap, YGYZ-M, and YGYZ-H groups (Figure 2C). The liver weight of mice in the YGYZ-H, YGYZ-M, and Cap groups was reduced compared with that of the Model (Figure 2D). The proportion of Ki-67-positive cells in YGYZ-H, YGYZ-M, and Cap groups was reduced compared with Model, indicating suppressed tumor proliferation (Figure 2E). All of these results indicated that YGYZ inhibited the progression of LM. Compared with the Model group, YGYZ-H, YGYZ-M, and YGYZ-L increased the AST level, while YGYZ-H and YGYZ-M increased the BUN and CRE levels (Figure 2F). The results suggested that YGYZ preserved liver and kidney functions. Integrated analysis of both anti-metastatic efficacy and safety parameters established YGYZ-H as the most favorable group for follow-up experiments.3.2 YGYZ and FMT inhibited tumor progression and delayed LM.

The progression of LM was dynamically observed by the in vivo imaging system, and on the 12th day of intervention, the ct26-luc cells were clustered in the right splenic region, and on the 24th day of gavage, fluorescent signals were observed in the left hepatic region. The strongest fluorescent signals were observed in the Model on both day 12 and 24, and the intensity in the Cap, YGYZ-H, and FMT groups was weaker (Figure 3A), suggesting that all three groups inhibited the proliferation of ct26-luc cells and delayed tumor progression and metastasis.

HE staining revealed distinct metastatic progression phases. At day 12, the liver exhibited no observable metastases across all groups, demonstrating preserved hepatic architecture with orderly hepatocyte cords and intact portal triads. By day 24, the Model developed extensive metastatic lesions characterized by hypercellular tumor colonies displaying malignant features including nuclear pleomorphism, loss of cellular polarity, and prominent tumor vasculature. Therapeutic interventions significantly reduced the metastatic burden versus the Model (Figure 3B). Combining the results of small animal in vivo imaging and HE staining, the liver pre-metastatic stage on day 12 and the liver metastatic stage on day 24 could be set to further analyze the immune microenvironment (IME) at each stage.

Flow cytometry revealed that during the pre-metastasis period on day 12, the proportions of hepatic PMN-MDSCs and G-MDSCs were reduced in the Cap, YGYZ-H, and FMT groups, compared with the Model. On day 24 (LM period), the proportions of hepatic PMN-MDSCs were reduced in the Cap, YGYZ-H, and FMT groups, with no significant difference in the proportion of G-MDSCs (Figure 4A). Immunofluorescence staining showed that on day 12, the proportions of CD11b+Ly6G+ and CD11b+Ly6C+ cells were increased in the Model, which were decreased in the Cap, YGYZ-H, and FMT groups. On day 24, the proportions of CD11b+Ly6G+ and CD11b+Ly6C+ cells were elevated in the Model. CD11b+Ly6G+ cells were reduced in the Cap, YGYZ-H, and the FMT group, whereas no statistical difference was observed in CD11b+Ly6C+ cells (Figure 4B). The results implied that YGYZ depressed the MDSC subpopulation during the pre-metastasis and LM stages.

MDSCs-derived cytokine profiling indicated significant immunosuppressive activity within the hepatic microenvironment during the pre-metastatic phase (Day 12). Compared to Sham group, the Model group exhibited elevated levels of prototypic MDSC-secreted factors, including Arg-1, TGF-β, and IL-10. Cap, YGYZ-H, and FMT groups significantly attenuated these immunosuppressive mediators versus the Model (Figure 4D).

The CD4+/CD8+ T cell ratio is a critical indicator of immune function, with low ratios indicating immunosuppression. Flow cytometry revealed that on days 12 and 24, the CD4+/CD8+ T cell ratio decreased in the Model compared with the Sham group. On day 12, YGYZ-H and FMT groups raised the ratio compared with the Model. On day 24, the Cap, YGYZ-H, and FMT groups elevated the ratio compared with the Model, but no statistical difference was seen in the latter two groups (Figure 4C). The findings showed that the primary therapeutic advantage of YGYZ lies in its regulation of IME during the pre-metastatic stage.

To investigate differences in GM composition during the pre-metastatic stage, Venn diagram analysis was performed to identify shared and unique operational taxonomic units (OTUs) among five groups. A total of 179 OTUs were shared across all groups, while the Sham, Model, Cap, YGYZ, and FMT groups contained 2,304, 3,355, 1,103, 2,846, and 3,515 unique OTUs, respectively (Figure 5A). Principal coordinate analysis (PCoA) revealed clear separation among the five groups, with PC1 and PC2 explaining 7.1 and 6.2% of the total variance, respectively (Figure 5B). Analysis of the pre-metastatic phase GM demonstrated that the Cap group decreased the index of Chao1, Simpson, Observed species, Shannon, Faith’s PD, and Pielou’s evenness, suggesting that the abundance, diversity, evolutionary diversity, and evenness of the GM declined after the intervention of the chemotherapy (capecitabine). However, there was no difference in α-diversity (Chao1, Observed species, Shannon, Faith’s PD, and Goods coverage) between the YGYZ-H and FMT, with Sham and Model groups, indicating preservation of GM richness, evenness, and diversity (Figure 5C).

Compared to the Model group, both the YGYZ-H and FMT groups showed an increased abundance of Firmicutes and decreased Bacteroidetes, resulting in a higher Firmicutes/Bacteroidetes ratio at the phylum level (Figure 6A). At the class level, the abundance of Bacteroidia decreased, while Clostridia, Verrucomicrobiae, and Bacilli increased (Figure 6B). At the family level, the relative abundance of S24-7 and Coriobacteriaceae was reduced, whereas Bacteroidaceae, Lachnospiraceae, Verrucomicrobiaceae, and Lactobacillaceae were enriched in both treatment groups (Figure 6C). At the genus level, YGYZ-H and FMT exhibited decreased abundance of Adlercreutzia, Prevotella, and increased abundance of Bacteroides, Akkermansia, Lactobacillus, and Oscillospira. In contrast, Cap generated a distinct taxonomic profile divergent from other groups (Figure 6D).

To compare intergroup differences in microbial composition, a clustered heatmap was generated using the top 30 most abundant genera, revealing distinct taxa distribution patterns. Analysis of the heatmap showed that the Model group was enriched in Coprococcus, Clostridium, Alistipes, Sutterella, and Prevotella. Both YGYZ and FMT demonstrated the capacity to reduce these enriched GM genera while increasing the abundance of Akkermansia, Eubacterium, Dehalobacterium, Bifidobacterium, and Mucispirillum (Figure 7A). As Clostridium has been identified as a risk microbiota promoting the progression of CRC (Lin et al., 2025), we quantitatively analyzed their abundance at both family and genus levels. Comparative analysis revealed the Model group maintained higher Clostridium abundance compared to others, which notably reduced in YGYZ-H and FMT groups (Figure 7B). According to the annotation results from the MetaCyc metabolic database, the GM genomes were primarily enriched in biosynthetic pathways, with the highest abundances observed in the biosynthesis of amino acids, nucleotides, prosthetic groups, as well as fatty acids and lipids (Figure 7C). The results suggested an intrinsic connection between GM and BAs.

Orthogonal Projections to Latent Structures Discriminant Analysis (OPLS-DA) was used to analyze the differences between groups. At the hepatic BAs level, OPLS-DA revealed a separation trend between the Sham, Model, YGYZ-H, and FMT groups, indicating intergroup differences (Figure 8A). Hepatic BAs profiling identified 20 types with >50% detection rates during the pre-metastatic phase. Compared to Sham, the Model exhibited marked BA dysregulation while YGYZ-H and FMT groups restored near-normal profiles (Figure 8B). Compared with the Model, liver total and primary BAs levels were raised in the YGYZ-H and FMT groups, but no difference was seen in secondary BAs among the groups (Figure 8C). Besides, both YGYZ-H and FMT groups enhanced the content of ursodeoxycholic acid (UDCA), Nor cholic acid (NorCA), Taurocholic acid (TCA), Taurochenodeoxycholic Acid (TCDCA), and Tauro beta-Muricholic Acid (T-β-MCA) (Figure 8D).

At the fecal BAs level, OPLS-DA showed no clear separation among the four groups (Figure 8E). Fecal BAs profiling identified 24 metabolites with >50% detection rates during the pre-metastatic phase. The BAs levels of the Model were significantly different from Sham and YGYZ-H groups, which were most similar to the Sham group (Figure 8F). Compared with the Model, fecal total BAs levels were decreased in the YGYZ-H, and no difference was seen in primary and secondary BAs among the groups (Figure 8G). Compared to the Model, YGYZ-H reduced the content of NorCA, alpha-Muricholic acid (α-MCA), Tauro alpha-Muricholic acid (T-α-MCA), and UDCA (Figure 8H). The results suggested that YGYZ restored the metabolic balance of BAs in the enterohepatic circulation.

Significant correlations were observed between GM at the genus level and BAs profiles. At the hepatic BAs level, UDCA showed a positive correlation with Coprococcus, and TCDCA was positively correlated with Coprococcus but negatively associated with Clostridium (Figure 9A). In fecal samples, UDCA was positively associated with Akkermansia and Robinsoniella. α-MCA showed positive correlations with Anaerotruncus, Butyricicoccus, Akkermansia, and Robinsoniella. T-α-MCA was positively correlated with Oscillospira and Mucispirillum but negatively associated with Sutterella (Figure 9B).

Hepatic KLF15 is a central regulatory node for BAs metabolism (Han et al., 2015), and WB results indicated elevated levels of KLF15 protein expression in the Model against the Sham group. In comparison to the Model, YGYZ-H downregulated KLF15 protein expression, while no difference was seen in the FMT (Figure 8A). The results implied that the KLF15-inhibited function of YGYZ was a synergistic effect of small molecules absorbed into the liver, but not just mediated by GM. Abnormal BAs metabolism directly led to the activation of the IL-6/STAT3 signaling pathway, promoting hepatic inflammation and MDSCs proliferation and activation (Tengesdal et al., 2021). Our study demonstrated that the expression and phosphorylation level of STAT3 protein were most significant in the Model, and both the YGYZ-H and FMT groups depressed hepatic STAT3 and pSTAT3, with pSTAT3 being more pronounced (Figure 10A). The IL-6 protein was expressed significantly in the Model, which was downregulated in both the YGYZ-H and FMT groups (Figure 10B). These results hinted that YGYZ inhibited the IL-6/STAT3 signaling pathway and STAT3 phosphorylation via regulating GM metabolism.

Integrative analysis demonstrated that YGYZ preserved the enterohepatic BAs homeostasis through KLF15 downregulation and GM modulation. This led to suppression of MDSCs proliferation through IL-6/STAT3 pathway, exerting the effect of preventing and controlling LM by strengthening the IME of PMN (Figure 10C).