Microbiome-mediated colonization resistance to carbapenem-resistant Klebsiella pneumoniae in ICU patients

Demographic characteristics

Initially, 478 patients admitted to the ICU were screened, and 254 patients were excluded for not meeting the inclusion criteria or meeting the exclusion ones. A total of 224 patients were included and then followed up, and their stool samples were collected during ICU hospitalization. However, 56 patients did not defecate during ICU stay. Ultimately, 168 patients were enrolled, and 1248 stool samples were collected from them. During ICU stay, 107 patients were negative by CRKP screening, and 61 patients were positive; however, 43 patients were excluded due to incomplete sample collection, and five patients were lost during follow-up. Ultimately, 18 patients and their stool samples (at admission and after CRKP colonization) were used for subsequent analysis. Propensity score matching was conducted to match samples from the CRKP-non-convert group (CRKP-N), resulting in the use of 18 pairs of samples in the final study.

The demographic, clinical, and laboratory characteristics of the ICU patients are summarized in Table 1. The CRKP-positive patients showed a longer time in the ICU (p = 0.001) and hospital stay (p = 0.031). We also noted that the CRKP-positive patients had a lower survival ratio (60-day survival: 61.11 vs 77.78%, p = 0.278), but the difference was not significant compared to CRKP-negative patients, which is likely due to the small sample size. Additionally, 18 healthy individuals were also recruited, and their demographic information is presented in Table S1.

The feature of CRKP isolated from stool samples in the CRKP-positive group

All isolates from 18 CRKP-positive patients were initially screened using Simmons’ Citrate Agar Incositol (SCAI) medium agar containing 4 mg/L meropenem and 32 mg/L linezolid and subsequently identified as K. pneumoniae via matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI-TOF MS). To further elucidate their characteristics, whole-genome sequencing was performed. As detailed in Table S2, all isolates were confirmed as K. pneumoniae, with the majority belonging to sequence type ST11 (17/18, 94.4%), while only one isolate was identified as ST15 (1/18, 5.6%). Additionally, 16 isolates were assigned to capsular type KL64 (16/18, 88.8%), with the other remaining two isolates belonging to KL47 (1/18, 5.6%) and KL19 (1/18, 5.6%), respectively. Regarding carbapenemase-encoding genes, bla_KPC-2 was detected in all isolates, whereas one isolate also carries bla_NDM-1. Furthermore, all isolates were resistant to meropenem, with MICs ranging from 128 to >512 mg/L.

The microbiome features of different ICU patient groups

To investigate the characteristics of the gut microbiome in different ICU patient groups, we employed 16S rRNA sequencing to analyze the fecal microbiome. The results showed that the ICU admission (ICU-A) group had a lower alpha diversity, as measured by the observed species, Shannon, Simpson, and Chao1 indices, compared to the healthy control group (HCG) (Fig. 1A–D). However, no significant differences in alpha diversity were observed between CRKP-non-convert admission (CRKP-NA) and CRKP-positive-convert admission (CRKP-PA) at the time of admission to the ICU (Supplementary Fig. S1A–D), indicating that these two groups at admission were comparable.

Moreover, significant differences in alpha diversity were found only in the observed species and Chao1 indices between CRKP-N and CRKP-positive conversion (CRKP-P). Notably, CRKP-P exhibited higher indices in both metrics, contrary to expectations (Fig. 1E–H). Nonetheless, the CRKP-P group still had a lower alpha diversity compared with healthy individuals in terms of Shannon and Simpson index (Fig. 1I–L). In contrast, no significant differences were detected in the Shannon and Simpson indices between those two groups (Fig. 1F, G). Additionally, no significant differences in the alpha diversity were observed between CRKP-NA and CRKP-N (Fig. S1E–H), as well as between CRKP-PA and CRKP-P (Fig. S1I–L). These findings suggest significant fluctuations in the alpha diversity of the gut microbiome in healthy individuals, ICU patients, and ICU patients with CRKP colonization.

To further elucidate the differences in microbiome composition among the different groups, principal coordinates analysis (PCoA) was performed (Fig. 1M–P). Significant differences were observed between the ICU-A and HCG groups (Adonis R²: 0.0437, p = 0.024), indicating distinct microbiome compositions between these groups. Similarly, significant differences were found between the CRKP-PA and CRKP-P groups (Adonis R²: 0.0978, p = 0.003). However, the differences between the CRKP-NA and CRKP-PA groups showed borderline significance with an Adonis R² value of 0.053 and p value of 0.056. In contrast, no significant differences were observed between the CRKP-NA and CRKP-N groups (Adonis R²: 0.0264, p = 0.454), suggesting similar microbiome compositions between these two groups.

CRKP was associated with alterations in the abundance of specific taxonomy in ICU patients

Taxonomic analysis revealed significant differences at the genus level among the HCG, CRKP-NA, CRKP-N, CRKP-PA, and CRKP-P groups. The health control population was predominantly composed of Bacteroides, Faecalibacterium, Blautia, Bifidobacterium, Coprococcus, Roseburia, Prevotella, Ruminococcus, and Eubacterium. However, in patients of the ICU-A group, the top 10 prevalent genera were Enterococcus, Bacteroides, Klebsiella, Parabacteroides, Lactobacillus, Bifidobacterium, Ruminococcus, Alistipes, Eubacterium, and Clostridium. Notably, Klebsiella was listed among the top 10 ranked genera in this group (Fig. 2A). In ICU patients in the CRKP-P group who tested positive for CRKP, the most abundant genera were Enterococcus, Klebsiella, Bacteroides, Parabacteroides, Alistipes, Enterobacter, Ruminococcus, Eubacterium, Lactobacillus, and Clostridium. However, in the CRKP-PA group, the top ten genera were Enterococcus, Bacteroides, Parabacteroides, Bifidobacterium, Lactobacillus, Eubacterium, Alistipes, Ruminococcus, Klebsiella, and Clostridium (Fig. 2B).

A The gut microbiome of healthy individuals (HCG) and ICU patients (ICU-A) displays difference in the composition of gut microbiome at genus level. B The gut microbiome of CRKP-negative patients (CRKP-N) and their admission samples (CRKP-NA), CRKP-positive patients (CRKP-P), and their corresponding baseline samples upon admission (CRKP-PA) at the genus level. C–F Lefse analysis identified differential genera between HCG and ICU-A (C), CRKP-P and CRKP-N (D), CRKP-P and CRKP-PA (E), and CRKP-PA vs. CRKP-NA (F). G Sankey diagrams provided a more intuitive visualization, showing significant differences in Klebsiella between CRKP-PA and CRKP-P, while no significant changes were observed between CRKP-NA and CRKP-N. H The abundance of Bifidobacterium longum significantly decreased in CRKP-P compared with CRKP-PA. I Lactiplantibacillus plantarum abundance was significantly decreased in the CRKP-P compared with the CRKP-N. The Kruskal–Wallis test was used for significance testing.

To identify marker species between different groups, linear discriminant analysis effect size (Lefse) analysis was conducted. The analysis also revealed that 25 ASVs were significantly between ICU-A and HCG groups, including Enterococcus, Lactobacillus, Clostridium, Dorea, Ruminococcus, Roseburia, Coprococcus, Bifidobacterium, Blautia, and Faecalibacterium (Fig. 2C). However, when comparing CRKP-P to the CRKP-N group, more than 38 ASVs showed differences between groups including Klebsiella, Enterococcus, Alistipes, Escherichia_Shigella, Bacteroides, Akkermansia, Citrobacter, Barnesiella, Bifidobacterium and Lactobacillus, among others (Fig. 2D). When comparing the stool microbiome features in patients before (CRKP-PA) and after (CRKP-P) being detected as CRKP-positive, 14 ASVs were found to differ between the two groups, including Klebsiella, Akkermansia, Leptotrichia, Pseudomonas, Dermabacter, Turicibacter, and Bifidobacterium, among others (Fig. 2E). Not surprisingly, patients who developed CRKP-positive or negative group, more than 57 ASVs showed significant differences in their corresponding samples at the time of admission (CRKP-PA vs. CRKP-NA), which including Bifidobacterium, Alistipes, Collinsella, GEMMIGER, Leuconostoc, Coprococcus, Blautia, Odoribacter, Ruminococcus, Roseburia, Clostridium, Weissellam, Faecalibacterium,Anaerofustis, Eggethella, Dorea, Lactococcus, Citrobacter, and Coprobacillus, among others (Fig. 2F). Importantly, the Sankey chart clearly showed that Klebsiella had a significant increase in CRKP-P compared to its CRKP-PA at admission, while there was little change in Klebsiella abundance in patients who were consistently CRKP-negative (Fig. 2G). While comparing the CRKP-NA and CRKP-N groups, only four ASVs were found to be depleted in the CRKP-N group (Fig. S2).

Importantly, we found several featured genera/species when comparing CRKP-positive and negative samples. In the HCG and ICU-A groups compassion, Lactobacillus had an LDA score of 4.2 and Bifidobacterium had an LDA score of 4.5. Furthermore, in the CRKP-PA and CRKP-P group comparison, one strain of B. longum had a linear discriminant analysis (LDA) score of 4.23 and higher relative abundance in the CRKP-PA group (Fig. 2H). Notably, in the comparison of negative (CRKP-N) and positive (CRKP-P) samples, Klebsiella had the highest LDA score, reaching 5.07, and one strain of K. pneumoniae having an LDA score of 4.02 in CRKP-P. Simultaneously, one strain of L. plantarum in the CRKP-N group had an LDA score of 4.63 and higher relative abundance compared with CRKP-P (Fig. 2I).

CRKP colonization alters the function of the gut microbiota in ICU patients

To further investigate the changes in gut microbiome function after CRKP colonization, we performed a PICRUSt2 functional prediction analysis. Bray–Curtis principal coordinates analysis (PCoA) method was applied to analyze the abundance of KEGG pathways predicted by Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt2) at levels L1, L2, L3, and MetaCyc pathways. Our results showed that the predicted functions between the HCG and ICU-A groups exhibited significant differences. Notably, almost all pathways with significant changes displayed opposing abundance trends between the two groups, with pathways upregulated in HCG being downregulated in ICU-A, and conversely (Fig. S3). Meanwhile, the CRKP-P group showed significant differences from the other groups in all pathways (Fig. 3A–E and Fig. S4). Similarly, we found that the CRKP-P group had significantly different pathways in all predicted functions, which were opposite or significantly different from those of the other three groups. At the KEGG L2 level, pathways such as Cancer: overview, Cancer: specific types, cellular community-prokaryotes, neurodegenerative disease, and Xenobiotics biodegradation and metabolism were significantly higher in the CRKP-P group, while pathways such as Cell growth and death, Endocrine and metabolic disease, endocrine system, replication and repair, and translation were significantly lower (Fig. 3E). Similar results were observed at the KEGG L2 level and MetaCyc (Fig. S4). However, these drastic changes were not observed in CRKP-N and CRKP-NA (Fig. 3D, E). These findings indicate that ICU patients who were colonized with CRKP exhibited significant changes in their gut microbiome function.

A–D CRKP colonization led to significant changes in the gut microbiota function of ICU patients. Bray–Curtis distance matrix based PCoA and PERMANOVA analysis indicated significant differences between CRKP-negative patients (CRKP-N) and CRKP-positive patients (CRKP-P) (A), CRKP-positive patients (CRKP-P) and their corresponding baseline samples upon admission (CRKP-PA) (B), CRKP-NA and CRKP-PA (C), while no significant difference was observed between CRKP-NA and CRKP-N (D). E KEGG level 2 pathway enrichment analysis showed a significant and contrasting trend in functional pathways between CRKP-P and the other sample groups. The Kruskal–Wallis test was used for significance testing.

CRKP colonization induces metabolic profile shifts in ICU patients

We were interested in determining whether changes in the gut microbiome would lead to shifts in gut metabolic products. Therefore, we conducted a non-targeted metabolomics analysis. The metabolome profiles of the HCG, ICU-A, CRKP-NA, CRKP-N, CRKP-PA, and CRKP-P groups were distinct, as revealed by OPLS-DA (Fig. 4). KEGG pathway enrichment analysis revealed that there were significant differences between HCG group and ICU-A group in the main enrichment top 25 pathways, including metabolism for amino acids (e.g, tyrosine, β-alanine, and tryptophan), nucleotides (e.g., purine and pyrimidine), vitamin (e.g., vitamin B6 and thiamine), carbohydrates (e.g., glycolysis/gluconeogenesis and galactose), lipids (e.g., arachidonic acid and ether lipid), and xenobiotic, as well as five other related metabolic pathways (Fig. 4A–C).

A OPLS-DA analysis indicated significant differences in the untargeted metabolomic profiles between healthy individuals and ICU patients. B The top 25 enriched metabolites displayed in panel (B). C A random forest algorithm identified the top 20 most important differential metabolites. Similarly, significant differences were observed between CRKP-negative patients (CRKP-N) and CRKP-positive patients (CRKP-P) (D–F), CRKP-positive patients (CRKP-P) and their corresponding baseline samples upon admission (CRKP-PA) (G–I), as well as CRKP-negative patients (CRKP-N) and their admission samples (CRKP-NA) (J–L).

Further comparison of the differences between CRKP-P and CRKP-N revealed correspondingly enriched pathways. The metabolic pathways related to amino acid metabolism (e.g., 2-aminoacrylic acid, D-aspartic acid, D-serine, L-homoserine, and L-lysine), carbohydrate metabolism (1-deoxy-1-(N6-lysino)-D-fructose), lipid metabolism (e.g., But-2enoic acid, N,N-dimethylsphingosine, and D-tocotrienol), and vitamin metabolism (riboflavin and pyridoxal), nucleotide metabolism, xenobiotic metabolism (4-aminophenol, 1-hydroxy-6-methoxypyrene, and picrotoxinin), and nitrogen metabolism (1-aminocyclopropanecarboxylic acid) remained the most significantly changed pathways (Fig. 4D–F).

Additionally, in the comparison of the differences between CRKP-PA and CRKP-P revealed enriched pathways such as primary bile acid biosynthesis (tauroursodeoxycholic acid), nicotinate and nicotinamide metabolism (2-hydroxyadenine), pyruvate metabolism (palmitic acid and (R)-3-hydroxy-tetradecanoic acid), biotin metabolism (alanylglycine and palmitic acid), ether lipid metabolism (LysoPC(16:0) and 1-(5Z,8Z,11Z,14Z-eicosatetraenoyl)-sn-glycero-3-phosphate), and tyrosine metabolism (4-(2-furanylmethylene)-3,4-dihydro-2h-pyrrole) (Fig. 4G–I). However, fewer significantly different metabolic pathways were observed in the comparison between CRKP-NA and CRKP-N groups, including amino acid metabolism, primary bile acid biosynthesis, and nitrogen metabolism (Fig. 4J–L).

In vitro experiments verified the inhibitory effect of the altered microbes on CRKP

Our investigation of the fecal microbiome and non-targeted metabolomics revealed significant alterations in the microecological structure and function, as well as the composition of metabolites in CRKP-positive ICU patients. We selected L. plantarum 21790 and B. longum 6188, which exhibited notable differences in samples from the ICU patients, to verify their capacity to inhibit CRKP in vitro.

First, we demonstrated that two clinically isolated strains of K. pneumoniae 020003 and K. pneumoniae 020120 could grow in the Reinforced Clostridium Medium (RCM) but could only survive and not proliferate in the Man, Rogosa and Sharpe broth (MRS). Subsequently, we investigated the effects of B. longum 6188 and L. plantarum 21790 on K. pneumoniae proliferation by adding 10–30% of their overnight culture supernatants to co-culture with K. pneumoniae. We observed a concentration-dependent inhibition of K. pneumoniae growth, with the 30% L. plantarum 21790 culture supernatants being the most effective, reducing K. pneumoniae to undetectable levels within 6 h of co-culture. Meanwhile, the 30% B. longum 6188 culture supernatant exhibited inhibitory activity against K. pneumoniae, although its effect was not as pronounced as that of L. plantarum 21790. The inhibitory effect of other concentrations of B. longum 6188 culture supernatants decreased over time and was completely lost within 24 h (Fig. 5A–E).

A Experimental design for CRKP in different concentrations of supernatant of probiotics in vitro. L. plantarum 21790 cultured with DeMan, Rogosa and Sharpe (MRS) medium, and B. longum 6188 cultured with Reinforced Clostridium Medium (RCM). B, C The inhibitory effects of supernatant from L. plantarum 21790 culture mediums on two clinical CRKP isolates (K. pneumoniae 020120 (B) and 020003 (C) demonstrated a significant concentration-dependent response. D, E The inhibitory effects of supernatant from B. longum 6188 culture mediums on two clinical CRKP isolates (K. pneumoniae 020120 (D) and 020003 (E)) demonstrated a significant concentration-dependent response. F Experimental design for the co-culture of probiotics and CRKP in filtered supernatant of healthy stool. An ex vitro model using a healthy donor gut microbiome (BLK) was employed. K. pneumoniae 020003 was introduced into this system (CON), alongside the supplement of L. plantarum 21790 (LPCO), B. longum 6188 (BLCO), or both strains concurrently (COPR). G Both L. plantarum 21790 and B. longum 6188 exert significant inhibitory effects on CRKP. H Experimental design for effects of probiotics on decolonization of CRKP in vivo. K. pneumoniae 020003 into mice after 1 week of treatment with meropenem (MEM) in water, and mice were supplemented with PBS (Ctl), 1.0 × 10⁹ CFU/ml of L. plantarum 21790 (LP), or B. longum 6188 (BL). I–L The CRKP decolonization effects of L. plantarum 21790 (BL) and L. plantarum 21790 (LP) on days 12 (I), 14 (J), 18 (K), and 25 (L). A reusable two-factor analysis of variances was used for comparison between different concentration groups and the control group.

Ex vivo experiments verified the inhibitory effect of the probiotics on CRKP

To further confirm the inhibitory effects of B. longum 6188 and L. plantarum 21790 on K. pneumoniae proliferation, we simulated a CRKP-positive fecal environment by adding K. pneumoniae to fecal samples from healthy individuals and then co-culturing these samples with B. longum 6188 and L. plantarum 21790 (Fig. 5F). Our results validated that the supplement of B. longum 6188 and L. plantarum 21790 provided resistance to K. pneumoniae colonization, significantly reducing the abundance of K. pneumoniae in the culture system, consistent with our previous observations. Similar results were observed when both B. longum 6188 and L. plantarum 21790 were added to the culture system (Fig. 5G).

In vivo experiments verified the inhibitory effect of the altered microbes on CRKP

To validate our findings from clinical samples and in vitro studies, we established a CRKP-positive mouse model by administering mice with carbapenem antibiotics and subsequently gavaging mice with K. pneumoniae 020120 (Fig. 5H). We then investigated the effects of orally administered B. longum 6188 and L. plantarum 21790 on CRKP suppression in the mouse model. In the B. longum 6188 and L. plantarum 21790 treatment groups, no significant inhibitory effects on K. pneumoniae 020120 were observed on days 12 and 14 (Fig. 5I, J). However, on days 18 and 25, the fecal CRKP loads were significantly reduced, and B. longum 6188 exhibited a better inhibition effect than L. plantarum 21790.

Lactobacillus and Bifidobacterium played a crucial role in restoring the normal gut microbiome, thereby enhancing the host’s resistance to CRKP colonization and accelerating the clearance of CRKP from the gastrointestinal tract. To further investigate the impact of gut microbiota dysbiosis on CRKP colonization, we used MEM to induce dysbiosis in mice, followed by CRKP administration. Notably, CRKP was slowly cleared from the mice after colonization. However, upon resupply of MEM, CRKP abundance rapidly increased. In contrast, when CRKP was co-administered with MEM, the abundance of CRKP in the stool continued to increase until MEM was removed, at which point CRKP abundance began to decrease gradually. Interestingly, in mice that did not receive CRKP gavage, CRKP abundance in the feces increased in a time-dependent manner with MEM treatment (Fig. 6A, B). To further confirm the importance of a healthy gut microbiome in providing resistance to CRKP colonization, we performed fecal microbiota transplantation (FMT) in mice after CRKP gavage. As expected, FMT significantly accelerated the clearance of CRKP compared to mice that did not receive FMT from healthy mice (Fig. 6C, D), and this effect was even more significant after the MEM was stopped. These results suggest that a healthy gut microbiome provides important resistance to CRKP colonization, and that restoration of the gut microbiome through supplementation with specific bacteria or probiotics is one potential method.

A Experimental design for effects of antibiotics on CRKP colonization in vivo. The abundance of CRKP in the gut of mice was monitored by comparing different antibiotic treatments. The posttreatment group (PstTret group, showed in blue color): Mice were first treated with PBS for 10 days, followed by meropenem (MEM) treatment for 23 days; The intermittent treatment group (InterTret group, showed in green color): Mice were treated with MEM for 3 days before introducing K. pneumoniae 020003, then MEM was removed until day 11, at which point MEM treatment was resupplied. The pretreatment group (PreTret group, showed in red color): Mice were treated with MEM for 3 days before introducing K. pneumoniae 020003, with MEM treatment continuing until day 11, after which MEM was removed. The abundance of CRKP in mouse feces was measured daily. B CRKP abundance in feces in different groups during days 5–33. C Experimental design for effects of fecal transplantation (FMT) on CRKP decolonization in vivo. Control group (Ctl, showed in grayish blue): Mice was feeded to provide healthy fecal samples for FMT group during experiment. Phosphate Buffered Saline (PBS group, showed in purple): Mice were first treated with MEM for 14 days, 200 ul PBS was gavaged during days 8–28 as negative control. FMT group (showed in pink): Mice were first treated with MEM for 14 days, 200 ul FMT from the Ctl group was gavaged during days 8–28. About 200 μL of 1.0 × 10⁵ CFU/ml K. pneumoniae 020003 was gavaged on day 8. D CRKP abundance in feces in different groups during days 9–28.