For the preparation of SFH, the ingredients were used as followings: Panax ginseng C.A.Mey (30 g), Rheum palmatum L (30 g), and Aconitum carmichaelii Debeaux (60 g). The drugs were steeped in pure water for 30 min. The Aconitum carmichaelii Debeaux was decocted first and the remaining ingredients were added into the decocted Aconitum carmichaelii Debeaux. These three drugs were decocted together for 30 min and then collected the filtrate. Repeated the process and collect the filtrate. Finally, mixed the filtrate and concentrated with 1 g of the crude drug/mL (used as the mother liquid). Pure water was used to dilute the mother liquid to 0.5 g/mL (high dosage for the treatment) and 0.1 g/mL (low dosage for the treatment).

The quality control of SFH was achieved through ultra-performance liquid chromatography, tandem mass spectrometry MS/MS, and biosystems QTRAP. Briefly, 4 μL test solution was injected into the Agilent SB-C18 column (2.1 mm) at 40°C × 100 mm, with a flow rate at 0.35 mL/min. Mobile phase A and B were prepared as recommended. Supplied with an electrospray ionization (ESI Turbo) source, the QTRAP appliance was used for positive and negative ionization scanning. The ion spray positive mode voltage was set at 5,500 V and its negative mode was 4,500 V. The ion source gas I (GSI), gas I (GSII), and curtain gas (CUR) are set to 50, 60, and 25.0 psi, and the collision-induced ionization parameter was restricted to high.

Healthy male SPF C57BL/6J mice (18–22 g, 6 weeks old) were purchased from the Beijing Hua FuKang Biotechnology Co, animal Certificate No. SCXK (Beijing) 2019–0008. The animals were taken for tests under the guideline of animal management with a stable light source (12 h of alternating light) and kept in a stable room temperature and humidity. The animal experiments were approved by the Animal Care and Use Committee of Beijing Institute of Chinese Medicine (approval No. 2021030201). According to the references, the sepsis model was replicated using CLP (Rittirsch et al., 2009). Sham-operated animals were not subjected to cecum ligation and perforation, and other steps were consistent with CLP animals. The mice were injected with 2 mL sterile normal saline after surgery.

After 7 days of adaptive feeding, 100 mice were randomly separated into four groups. (1) Sham-operated group (Sham, n = 10), mice underwent sham surgery without supplementary medications. (2) CLP group (CLP, n = 30), mice that only underwent CLP surgery. (3) SFH low-dose group (SFH-L, n = 30), CLP-mice have given an extra 0.5 mL of SFH low-dose decoction, and finally, the (4) SFH high-dose group (SFH-H, n = 30), CLP-mice with 0.5 mL high-dose SFH decoction. Mice were orally administered a day after the operation, while in the rest groups, mice were supplied with an equal volume of saline. The mice were continuously treated with SFH or saline for 3 days. An additional 30 mice per group were used in the survival test to assess the survival rate over 7 days.

At 72 h after CLP operation, sodium pentobarbital (50 mg/kg) was used for anesthetization. For serum collection, a syringe was used to obtain blood from the heart by thoracotomy, and was then processed through a 15-min centrifugation at 3,000 r/min. The mice feces were placed in sterile lyophilization tubes and stored in liquid nitrogen immediately before transferring to the −80°C refrigerator.

Circulating amounts of TNF-α, IL-6, and IL-1β in each group were measured following the instructions of the corresponding ELISA test kit (MEIMIAN, Jiangsu, China).

Genomic DNA was extracted from fecal samples, and 1% agarose gel was used to monitor the DNA concentration and purity. After diluting the DNA to 1 ng/μL in sterile water, the V3-V4 region of the 16S rRNA gene was amplified for generating sequencing libraries by DNA PCR-Free Sample Preparation Kit (Illumina, United States). The sequence quality control was proceeded by means of Qubit@ 2.0 Fluorometer (Thermo) and Agilent Bioanalyzer system. In the final step, the sequence library was obtained through the Illumina platform.

Raw tags were acquired by stitching and filtering the reads of each sample to obtain clean tags (Caporaso et al., 2010; Magoc and Salzberg, 2011), which were further compared with the reference database (Edgar et al., 2011) for the removal of chimera sequences (Haas et al., 2011). Meaningful tags were obtained and further analysis was performed (Uparse software). Those sequences with more than 97% similarity were assigned to the same OTUs. The Silva Database and the Mothur algorithm were applied to annotate species. We managed to compute alpha and beta diversity. LDA score value at 4 was discarded in LEfSe analysis. To construct additional plots, we also applied the R software, which includes the following: rarefaction curves, PCoA plots, and differential plots highlighting group disparities.

Serum samples were kept at 4°C and vortexed. An appropriate amount of sample was accurately transferred to a tube with 400 μL methanol. After centrifuging for 10 min at 12,000 rpm at 4°C, the supernatant was then collected for subsequent concentration and dryness. We added a 150 μL of 2-chloro-l-phenylalanine with 80% methanol–water for redissolution. After removal, the supernatant was filtered through a 0.22 μm membrane and transferred into the detection bottle for LC–MS detection.

Maintained at 40°C, 0.25 mL/min flow rate, and 2 μL injection volume, the chromatography was applied by ACQUITY UPLC ^® HSS T3 (Waters, Milford, United States). Separation was conducted according to LC-ESI (+)-MS guidelines. As for another analysis, the analytes were carried out with acetonitrile and ammonium formate (5 mM). Instructions conducted separation.

Orbitrap Exploris 120 (Thermo, United States) was used for mass spectrometric detection of metabolites, with simultaneous MS1 and MS/MS acquisition. The parameters were standardized according to the manufacturer’s instructions.

The raw data were first transformed into a uniform format using the ProteoWizard package (Smith et al., 2006). R XCMS package (Navarro-Reig et al., 2015) was then exploited for deep procession and integration. Identification of metabolites was made by accuracy mass (<30 ppm). All the MS/MS data were matched with HMDB, mass bank, LipidMaps, mzcloud, and KEGG. The LOESS signal correction method based on QC samples corrects and eliminates systematic errors. The substances with RSD > 30% in QC samples were filtered out from the data quality control. Moreover, we evaluated the robustness of the OPLS-DA model. The permutation test considered p value and VIP (projection of variable importance of OPLS – DA model) were chosen as the screening criteria. p < 0.05 and VIP > 1 were considered statistically significant. MetaboAnalyst then analyzed the metabolites for pathway analysis (Xia and Wishart, 2011). The identified metabolites in metabolomics were then mapped to the KEGG pathway. The KEGG Mapper tool assisted us in visualizing metabolites and pathways.

All statistical analysis proceeded with SPSS Statistics (IBM Software, United States) and figures were drawn using GraphPad Prism 7 (San Diego, California, United States). The discrepancy among groups was done using one-way ANOVA and Student–Newman–Keuls tests. Kaplan–Meier survival curve analysis was used to illustrate mortality between groups. p < 0.05 was considered statistically significant. All compared groups in this study at least included three independent samples.

The main compounds were identified in SFH decoction by UPLC-MS. As illustrated, the test results of positive and negative ion flow diagrams were shown (Figure 1). The left diagrams were shown as Supplementary Figure 1. 12 major compounds in SFH, including Rheic Acid, Catechin, Emodin, Gallic acid, Benzoylmesaconine, Ginsenoside Rg1, Chrysophanol, Aconitine, Aloe emodin, Mesaconitine, Ginsenoside Rb1, Ginsenoside Re were identified and characterized in Supplementary Figure 2.

First, we observed the dose effect of SFH in CLP-induced septic mice. As shown in Figure 2A, the survival rate, around 35% within 3 days and 0% within a week after CLP, was drastically reduced compared to the Sham mice (p < 0.05). Kaplan–Meier survival analysis demonstrated that a high dose of SFH could significantly prolong the 7-day survival rate to 35%, which was only 15% in the SFH-L group (p < 0.05, Figure 2A). Therefore, high-dose of SFH was chosen for the subsequent experiments. To further explore the mechanism of SFH in reducing mortality in sepsis, we also measured the levels of pro-inflammatory mediators in mice serum. In CLP-induced septic mice, circulating TNF-α, IL-6, and IL-1β were significantly increased compared with the Sham group mice. At the same time, SFH treatment drastically decreased the levels of these inflammatory cytokines (p < 0.05, Figures 2B–D).

16S rDNA sequencing was used to depict the alterations in intestinal microbiome composition in SFH-treated septic mice. The rarefaction curve (Figure 3A) indicated that as the number of sequences rose, the curve flattened, indicating that the results of this sequencing were relatively reasonable. We also generated a rank abundance curve (Figure 3B), which corroborated the sequencing results. Changes in both diversity and richness of the gut flora were displayed by the Shannon index. The graphs indicated that the CLP could significantly lower the Shannon index in comparison to the Sham group (p < 0.001), whereas the low Shannon index could be restored by SFH induced by surgery (p < 0.05, Figure 3C). Moreover, PCoA results showed an apparent discrepancy in clusters among groups. When compared to the CLP group, the cluster of SFH was closer to the Sham group (Figure 3D). To conclude, these results indicated that CLP induced significant change in gut microbiota, and treatment with SFH significantly restored the disturbed gut microbiota in septic mice.

We then focused the study on investigating the effect of SFH on the bacterial abundance in septic mice. In each group, gut microbiota composition at the phylum level was displayed (Figures 4A,B). Firmicutes and Bacteroides represent the majority of intestinal flora in mice. The phyla of Proteobacteria and Campylobacterota were increased whereas Firmicutes was decreased in the CLP group. Interestingly, these alterations were reversed after the treatment of SFH. To identify the key phylotypes and biomarkers of gut microbiota among various groups, LEfSe analysis (LDA score > 4, p < 0.05) was performed. In the CLP group, the families of Helicobacteraceae and Enterobacteriaceae, including Escherichia_coli and Escherichia_Shigella, were abundant compared to the others (Figure 4C). However, treatment with SFH markedly increased the relative abundance of Blautia and Lachnospiraceae bacterium_28_4 (Figure 4D).

We further investigated the changes in serum metabolites after the septic mice were treated with SFH by metabolomics. OPLS-DA models were established based on metabolomics data in two ion modes from LC/MS. We found a considerable divergence between the CLP and SFH groups in both ion modes. In the positive ion mode, the Sham and CLP groups were well separated, indicating that sepsis significantly altered the mice metabolome (Figure 5A). Figure 5C also demonstrats that CLP and SFH groups were well separated, meaning that SFH significantly changed the metabolome of septic mice. The OPLS-DA substitution test plot fulfills any of the following points, showing that the findings are accurate and valid: the first point where all blue Q2 points are lower than the rightmost original blue Q2 point (the rightmost blue Q2 point in the plot is likely to overlap with the green R2 point in the top rightmost corner), the second point where the regression line crosses the horizontal coordinate or is less than 0. Therefore, the test results were reliable and valid (Figures 5B,D). A negative ion mode diagram is presented in Supplementary Figure 3. Moreover, metabolite expression was also different among the three groups. 196 critical metabolites were detected in the Sham and CLP groups, among which 146 were upregulated and the other 50 were downregulated. When comparing CLP and SFH groups, we found 111 different metabolites with statistical significance. More specifically, there were 54 upregulated and 57 downregulated metabolites (Table 1). By differential Venn diagram, 35 metabolites were commonly different (Figure 5E). The changes in differential metabolites are shown in Table 2. These results suggested that SFH restored the serum metabolites of sepsis mice similar to healthy mice.

Next, the critical metabolic pathways regulated by the Sham vs. CLP group and SFH vs. CLP were analyzed using MetPA software. Although sepsis-induced gut microbial imbalance is related with eight metabolic pathways, including central carbon metabolism in cancer, glucagon signaling, alanine, aspartate and glutamate metabolism, mineral absorption, TCA cycle, PPAR signaling pathway, galactose metabolism, and pyrimidine metabolism, SFH therapy can restore part of the metabolic balance. CLP surgery exerted an effect on the metabolic pathways including glucagon signaling, alanine, aspartate, glutamate metabolism, mineral absorption, and TCA cycle (Figure 6A), while SFH treatment probably influenced the central carbon metabolism, glucagon signaling, PPAR signaling pathway, galactose metabolism, and pyrimidine metabolism (Figure 6B).

Spearman correlation was performed to investigate the relationship between differential serum metabolites and microbiome at the genus level in the treatment groups. As illustrated, Bacteroides, Lachnospiraceae_NK4A136_group, Escherichia_Shigella, Blautia, Ruminococcus, and Prevotella were related to most of the metabolites (Figure 7). The metabolites, including troxilin B3, protocatechuic acid, imidazoleacetic acid, oxalacetic acid, antibiotic JI-20A, and gamma−Glutamylcysteine, were positively linked with Blautia (p < 0.05) and inversely correlated with Escherichia_Shigella (p < 0.05). However, 9,10-Epoxyoctadecenoic acid, arachidic acid, 11-Dehydrocorticosterone, and fructose 1,6–bisphosphate demonstrated a negative correlation with Blautia (p < 0.05), but were positively correlated with Escherichia_Shigella (p < 0.05). Furthermore, Prevotella was positively correlated with oxoglutaric acid, malonate, allose, imidazoleacetic acid, fructose 6-phosphate, oxalacetic acid, antibiotic JI-20A, and gamma-Glutamylcysteine (p < 0.05), but negatively correlated with hepoxilin B3, 9,10-Epoxyoctadecenoic acid, pyridoxine, arachidic acid, 11-Dehydrocorticosterone, and fructose 1,6-bisphosphate (p < 0.05). Lachnospiraceae_NK4A136_group was positively correlated with protocatechuic acid (p < 0.05). Ruminococcus was positively associated with protocatechuic acid, oxoglutaric acid, dibutyl phthalate, allose, imidazoleacetic acid (p < 0.05), and negatively correlated with 2-(Methylamino) benzoic acid, 5′-Dehydroadenosine, and pyridoxine (p < 0.05). The results indicated that SFH could alter the composition of gut microorganisms and regulate certain metabolites in vivo.