Table 1 lists the demographic data and clinical characteristics of the sepsis cohort of patients. Intra-abdominal (50%) and pulmonary (26.3%) infections were the predominant origins of sepsis in our patients. The occurrence of complications, mortality within 90 days, SOFA score, APACHE II score, and ICU stay time were recorded as indicators of the severity of sepsis. Twenty-one (55.2%) samples were collected within 48 h of medical treatment. The majority (35/38, 92.1%) of patients had received short-term antibiotic treatment in a lower-grade clinic before being admitted to our ICU. Among the antibiotics, carbapenem was the most frequently used (20/38, 52.6%), followed by 3rd/4th generation cephalosporins (10/38, 26.3%).

Table 1 Clinical and demographic characteristics of study subjectsSepsis-related bacteria distinguished by microbiota comparison between sepsis patients and HCs

First, we compared the alpha diversity and beta diversity between the sepsis and HC groups. Sepsis patients had obviously lower alpha diversity in the gut than HCs (Shannon index 2.146 vs. 2.911, p < 0.01; Fig. 2A) and a disparate microbial structure compared with the HC group (ANOSIM test, p = 0.001; Fig. 2B). The distribution of prevalent bacteria in sepsis and HC group is shown in Fig. 2C. Sepsis patients showed expansion of Proteobacteria phylum and Enterococcaceae family, but a remarkable drop in the proportions of Firmicutes phylum and Lachnospiraceae family. LefSe revealed that the sepsis patients also exhibited significant reductions in the abundance of health-promoting flora, such as Blautia, Anaerostipes, Bifidobacterium, and Eubacterium_hallii_group [10,11,12,13]. Conversely, Enterococcus and Klebsiella, the common nosocomial infectious agents [14], were more enriched in sepsis patients than in HCs (Fig. 2C, D). Furthermore, compared to HCs, sepsis patients had a less intricate network of inter-bacterial relationships, in which Enterococcus showed exceptional inhibitory effects on other taxa (Fig. 2E). Taking into account the age discrepancy between the two groups, we repeated these comparisons in an age-matched manner. After controlling for the effect of age, the Shannon diversity index was found to be similar between the sepsis group and the control group (Additional file 3: Figure S1A). However, the sepsis group still displayed a distinct microbial composition compared to the control group (Additional file 3: Figure S1B, ANOSIM test, p = 0.001), as evidenced by changes in the relative abundances of Blautia, Eubacterium_hallii_group, and Enterococcus, which remained statistically significant (Additional file 3: Figure S1C).

Fig. 2

Comparison of microbial alterations in sepsis patients and HCs. A Student’s t-test showing differences in the Shannon diversity index of fecal samples from sepsis patients (n = 38) and HCs (n = 19). Data represent the median and quartiles  of each group; **p < 0.01. B PCoA of the sepsis and HC groups, with plots based on the Bray–Curtis distance. Each point represents a sample and the colors represent different groups. The results of the ANOSIM test to compare dissimilarity indexes among samples are shown above the plots. C Average relative proportions of the main phyla, families, and genera in the two groups. D LEfSe used to identify essential differences in bacterial abundance (family to species level) between the sepsis and HC groups. Only taxa with a significant LDA threshold value > 4 are shown. E Spearman’s correlation analysis to evaluate the abundance between the most common bacteria at the genus level in the two groups. Only correlations with a p value < 0.05 are shown. Color shading of nodes represents species abundance. Color gradation of lines represents the R value, with negative correlations shown in blue and positive correlations shown in red

Concerning the differences in backgrounds and medications among the sepsis cohort, we also analyzed the potential influence of these variables on microbial structure (Additional file 9: Table S2). According to the results, age, gender, the time interval from medical treatment to specimen collection and the origin of infection had no significant effect on the microbiota of the sepsis group. Antibiotic type, number of antibiotics, and use of proton-pump inhibitors did not cause apparent changes in the microbiota in our sepsis cohort, which might be explained by the short exposure time to these medications. The body mass index (BMI), which can be related to distinct dietary habits, might be expected to have an effect on the microbial composition.

Correlation between metabolic profile and the altered microbiota in patients with sepsis

Next, we performed non-targeted metabolomic profiling of feces samples from sepsis patients and HCs. Although a total of 449 annotated differential metabolites were identified in the derivation group, only 10 (2.23%) were more abundant in the sepsis group than the HC group, suggesting a possible weakening of metabolic activity in these patients (Figs. 1B and 3A). PCA analysis revealed a distinct metabolic profile in both positive and negative ion patterns between the two groups (Fig. 3B, Additional file 4: Figure S2A). We assessed the impact of various variables on the metabolome and found that the number of antibiotics used may affect the metabolic composition (Additional file 9: Table S2, Additional file 4: Figure S2B). Furthermore, we mapped the sepsis-altered metabolites to their biochemical pathways using metabolic enrichment and pathway analysis based on KEGG annotations (Fig. 3C). Among the 81 potential functionally differential metabolites, 38 of them were detected in the validation group, which we defined as sepsis-related metabolites (Fig. 1B). We also evaluated the relationship between sepsis-related metabolites and other parameters among sepsis patients and considered only the correlations confirmed in the validation group to be reliable (Fig. 1C).

Fig. 3

Differential gut metabolites of the sepsis patient cohort and their correlations with the microbiota. A Volcano plot showing the differential metabolites between sepsis patients and HCs. Green and red dots represent depleted and enriched metabolites, respectively, with the red line denoting a cut-off p value of < 0.05. B PCA of the stool samples from sepsis patients and HCs, plotted in positive-ion mode. C Bubble diagram illustrating the KEGG enrichment analysis. The size of each bubble represents the number of metabolites enriched in the pathway, and the color gradient indicates the significance of enrichment. Comparisons between the sepsis patients and HCs were observed after age matching (n = 15 per group). D Spearman’s correlation heatmap showing the test results for differential species in OTU level (LDA > 3.0) and sepsis-related metabolites among fecal samples in the derivation group (n = 25). Different colors indicate correlation level; *p < 0.05; **p < 0.01, ***p < 0.001. The 10 metabolites highlighted on the right were the most reliably correlated with microbial taxa after confirmation in the validation group

The differential metabolites appeared to be heavily involved in amino acid metabolism, including tryptophan, tyrosine, and arginine and proline metabolism. Involvement in imbalanced purine metabolism, biosynthesis of plant secondary metabolites, and ABC transporters was also observed. Considering the intimate contact between gut microbes and metabolites, we appraised the reliability of correlations between altered taxa and sepsis-related metabolites (Figs. 3D and Additional file 4: S2C). A variety of bacteria were found to be strongly associated with these key metabolites. Among the top five, four belong to Lachnospiraceae, indicating the potent metabolic regulatory function of this family of bacteria. Notably, OTU808 (Enterococcus faecium) showed an almost negative correlation with metabolites, which was similar to the above-mentioned inhibitory relationship between Enterococcus and other bacteria in the intestinal microenvironment of sepsis. Furthermore, we identified the 10 metabolites most closely related to bacteria: 3-methyldioxyindole, 5-hydroxyindoleacetic acid, dopaquinone, N,N-dihydroxy-l-phenylalanine, isonicotinic acid, deoxycholic acid, 2-hydroxycinnamic acid, 4-hydroxybenzaldehyde, thymine, and hexadecanedioic acid. Six of these are aromatic compounds, which might be derived from metabolites of aromatic amino acid (AAA), including tryptophan, tyrosine, and phenylalanine. Among them, 3-methyldioxindole, a metabolite involved in tryptophan metabolism, showed the most relevance to gut microorganisms found in patients with sepsis.

Agreement between gut microbial and metabolic structure and clinical severity of sepsis

The potential effects of the gut microbiome and metabolites on the severity of sepsis were examined next. As shown in Fig. 4A, patients in this group could be generally classified into three subgroups based on their gut microbial composition, also known as enterotypes. Lachnospiraceae and Ruminococcaceae prevail in enterotype 1 (E1); enterotype 2 (E2) is dominated by Enterococcus; and enterotype 3 (E3) is abundant in Bacteroides and Escherichia-Shigella (Figs. 4B and Additional file 5: Figure S3A). The comparison of enterotypes revealed that patients in E3 were the most critical, with a significantly larger proportion of APACHE II scores ≥ 18 compared to non-E3 patients (46% vs. 12%, p = 0.040; Fig. 4C). Additionally, E3 patients had a potentially larger proportion of SOFA scores ≥ 10 and ICU stays ≥ 30 days, as well as the highest mortality rate, but these differences did not reach statistical significance (Additional file 5: Figure S3B, Additional file 10: Table S3). Although E2 patients appeared to have lower bacterial diversity and richness relative to the other enterotypes (Additional file 5: Figure S3C), their disease was not more severe than the others.

Fig. 4

Parallels between microbial and metabolic profiles and clinical severity of sepsis. A PCoA of all sepsis patient samples (n = 38), with plots based on the Bray–Curtis distance. Each point represents a sample. According to the dispersion, the samples were classified into three enterotypes, shown in different colors. B LEfSe used to display the dominant bacteria in the three enterotype subgroups (family to species level): E1, n = 8; E2, n = 17; E3, n = 13. Only taxa with an LDA value > 4 are presented. Circles indicate phylogenetic levels; diameter and color of each circle represent its abundance and enterotype, respectively. C Column diagram showing the proportion of sepsis patients with an APACHE II score ≥ 18. D Linear correlation showing the relationship between the relative abundance of OTU773 and the length of ICU stay among surviving sepsis patients (n = 33). E, F Wilcoxon rank sum tests comparing the relative abundance levels of OTU808 and OTU773 between the sepsis patients with (n = 12) and without (n = 26) complications during their ICU stay (E), and between surviving (n = 33) and nonsurviving (n = 5) sepsis patients (F). G Spearman’s correlation heatmaps representing relationships between clinical parameters and the relative abundance of sepsis-related metabolites among surviving sepsis patients in the derivation group (n = 21). Distinct colors represent correlation levels; “#” indicates a reliable correlation

We also analyzed the correlations between the predominant bacterial taxa and clinical indices of surviving sepsis patients. Our analysis revealed that two OTUs, OTU773 from Bacteroides vulgatus and OTU822 from Rikenellaceae (s__unclassified_g__Alistipes), were significantly positively correlated with the length of ICU stay (Figs. 4D and Additional file 5: Figure S3D). The correlation between B. vulgatus and the length of ICU stay remained significant even after removing the potential effect of confounding variables in a multiple linear regression model (Additional file 11: Table S4). On the other hand, OTU808 (E. faecium) showed only a negligible correlation with this clinical parameter. Additionally, patients who experienced complications during the ICU period had significantly higher OTU773 (B. vulgatus) loads, but did not exhibit differences in the OTU808 (E. faecium) loads (Figs. 4E and Additional file 5: Figure S3E). Although not significant, there was a similar trend showing that OTU773 (B. vulgatus) was more abundant among the patients who died from sepsis (Fig. 4F). These results suggested a possible adverse effect of B. vulgatus on sepsis patients. Furthermore, we found that other bacteria, such as Akkermansia and OTU40 from Micrococcaceae (s__unclassified_g__Rothia), might have exerted beneficial effects on our sepsis patients. OTU40 was inversely correlated with APACHE II scores, while the abundance of Akkermansia, a genus known to be protective in many human diseases, was lower in the gut microbiota of non-surviving sepsis patients (Additional file 5: Figure S3D, F) [15].

Assessment of the reliability of correlations between gut metabolites and the disease severity of sepsis uncovered three compounds that were negatively correlated with length of ICU stay (Table 2 and Fig. 4G). Furthermore, we observed that two sepsis-related metabolites were reduced in the non-surviving patients (Table 2 and Additional file 4: Figure S2D). Four of the five metabolites were aromatic compounds and involved in amino acid metabolism, particularly metabolism of tryptophan and tyrosine. Interestingly, 5-hydroxyindoleacetylglycine was also positively correlated with the APACHE II score, revealing an intricate network of interactions between gut-derived metabolites and sepsis (Fig. 4G). We also observed that non-surviving patients had an enriched abundance of kynurenine (Additional file 4: Figure S2E), which is also a tryptophan metabolite but was not detected in the validation group. Interestingly, we observed that the abundance of these tryptophan metabolites remained comparable in patients receiving different numbers of antibiotics, suggesting that the perturbation of tryptophan metabolism may not be associated with the number of antibiotics administered (Additional file 4: Figure S2F). Our results overall suggested that tryptophan metabolites played a pivotal role in the progression of sepsis.

Table 2 Important metabolites associated with the gut microbiota or clinical severity of sepsis patientsChanges in the gut microbiota and intestinal gene expression in septic rats

Even if our assessment of stool specimens closely replicated the types of specimens that are routinely encountered in clinical practice, it is theoretically difficult to eliminate the interference of the multifarious clinical medications on the human intestinal microbiota. Therefore, we sought to corroborate the sepsis-related microbial features of our sepsis cohort of patients in an animal model. Having observed that our sepsis patients with intra-abdominal infection had a similar microbiome structure to those with lung infection (Additional file 3: Figure S1D), we established the classic CLP model of sepsis, which is caused by abdominal infection, in rats. Similar to the patient and HC groups, the septic rat group exhibited divergence from the sham operation group in the intestinal microbial composition (ANOSIM test, p = 0.008; Fig. 5A), although there was slight variation in the Shannon diversity index (Additional file 6: Figure S4A). Likewise, the proportions of essential probiotic bacteria, including Muribaculaceae and Butyricicoccus, dropped in septic rats, while opportunistic pathogens like Enterobacteriaceae and Enterococcus prospered (Fig. 5B) [16,17,18]. Notably, the abundance of B. vulgatus, which was associated with poor outcomes in our sepsis patients, was increased in the intestine of septic rats, despite the comparable loads of B. vulgatus in the stool specimens of sepsis patients and HCs.

Fig. 5

Bacterial portrait and intestinal transcriptomic adjustment in the septic rat. A PCoA of data from CLP sepsis model rats and control rats (n = 6 per group), with plots based on the Bray–Curtis distance. Each point represents one sample and the colors indicate different groups. The results of the ANOSIM test to compare dissimilarity indexes among samples are shown above the plots. B LEfSe performed to identify important differential bacterial taxa (family to species level) between the sepsis patient and HC groups (n = 6 per group). Only taxa with a significant LDA threshold value > 3 are shown. C KEGG pathway enrichment analysis of DEGs upregulated (left) and downregulated (right) in the septic colon (n = 6) relative to their expression in the colon of HCs. The top 15 pathways are shown. D Gene expression network generated by STRING and displayed via Cytoscape. Red (colon, left) and orange (ileum, right) rectangles represent upregulated genes and blue circles represent downregulated genes in the sepsis group. The size of each gene node is proportional to its BC value, and the higher the BC value, the closer the node to the center of the circle

To further illuminate the functions of key sepsis-related bacteria, we further evaluated the relationship between the bacterial abundance profile and the intestinal gene expression profile. Initially, we focused on transcriptomic variance in the septic intestine of rats. KEGG analysis was conducted to determine the functional enrichment of DEGs (Fig. 5C, Additional file 6: Figure S4B). For both the ileum and colon, masses of genes that were upregulated in sepsis were found to be involved in cytokine–cytokine receptor interaction and the chemokine signaling pathway, as reflected by the immune activation. Multiple cell adhesion molecules were downregulated in the septic colon, as illustrated by the weak functioning of the extracellular matrix (ECM)–receptor interaction with the transforming growth factor-β signaling pathway, which contributes equally to the recruitment of immune cells and the enhancement of intestinal permeability [19,20,21]. Another notable alteration was found in the lipid metabolism pathway. Numerous genes that were downregulated are pertinent to steroid hormone biosynthesis, the peroxisome proliferator-activated receptor (PPAR) signaling pathway and cholesterol metabolism. Cholesterols are precursors for several hormones that exert an anti-inflammatory function, and the PPAR signaling pathway can be activated by fatty acids and their derivatives, which effectively ameliorate inflammatory damage in sepsis [22,23,24]. Additionally, the pathway related to the metabolism of xenobiotics was also diminished in sepsis. This pathway responds to the activation of aryl hydrocarbon receptor (AhR) and pregnane X receptor (PXR) by gut-derived metabolites and subsequently interacts with nuclear factor-κB signal transduction to exert an anti-inflammatory function [25].

Next, a protein interaction expression network of the DEGs was conducted through STRING. The central DEGs are shown in Fig. 5D. Regarding core DEGs in the colon, Plg assists with ECM disruption and neutrophil migration, Il6 is involved in inflammation, and Apob and Apoe are components of lipoprotein carriers, which might modulate the immune process by binding endotoxin or other toxins [26, 27]. Regarding DEGs in the ileum, Cd44, Vcam1 and Cxcr4 serve the immune cell aggregation process, while Mmp9 degrades adhesive substances to promote cell movement [28,29,30,31]. Taken together, the intestinal transcriptome of sepsis revealed an inflammatory storm associated with various types of host response.

Distinct interactions of specific bacteria in the intestinal transcriptome of septic rats

Earlier, we proposed that Bacteroides and Enterococcus are important sepsis-related bacteria that likely have different effects on disease progression. Therefore, we carried out a correlation analysis of microbial abundance and intestinal gene expression to identify the potential roles of gut microbes in the pathophysiology of sepsis. The results for the colon are shown in Fig. 6. We were able to distinguish three bacterial clusters (clusters 1–3) and three gene clusters (clusters A–C). Cluster 1 chiefly contained Enterococcus and Enterobacteriaceae and showed a positive correlation with genes of Cluster A, which function in inflammatory response (Cybb, Tlr2) and ECM disruption (Mmp13, Spp1). Furthermore, Cluster 1 organisms from Enterobacteriaceae are more closely related to genes relevant to lipid metabolism (Apoc3, Apob) than those from Enterococcus. Cluster 2, which comprised Muribaculaceae, showed an utterly different association trend compared with Cluster 1. Bacteroides species gathered in Cluster 3 and interacted with a few genes from Cluster B outstandingly. Among these gene interactions, Mmp9, a well-studied metalloprotease involved in barrier disruption, was positively correlated with Bacteroides species, while Cth, which encodes cystathionine-γ-lyase, a major endogenous hydrogen sulfide-producing enzyme shown to suppress inflammation through IL-8 signaling [22, 23], was negatively correlated. Correlation analysis of the ileum also revealed three main bacterial clusters and three gene clusters (Additional file 6: Figure S4C). The compositions of Clusters 1 and 2 were similar to their colonic counterparts and were tightly correlated with DEGs from Clusters A and B, but in the opposite direction. Although Clusters A and B were both composed of immune and ECM genes, they failed to show any distinct effects in sepsis. The majority of Bacteroides species were still found in Cluster 3. Only a few significant correlations were found for Bacteroides, of which the most remarkable was a link with Cyba expression. Ultimately, the correlation analysis revealed that Bacterioides and Enterococcus might influence different host intestinal physiological processes in sepsis.

Fig. 6

Correlation analysis of differential species and central DEGs in the colon of CLP model rats. Spearman’s correlation analysis conducted to evaluate associations between the top 50 central DEGs (identified by BC value) and differential OTUs (LDA > 2) in the colon of septic rats (n = 6). The color gradient corresponds to the R value, where red represents the highest positive correlation and blue represents the lowest