Rationally-defined microbial consortia suppress multidrug-resistant proinflammatory Enterobacteriaceae via ecological control

Persistent colonization and outgrowth of pathogenic organisms in the intestine may occur due to long-term antibiotic usage or inflammatory conditions, which perpetuate dysregulated immunity and tissue damage1,2. Gram-negative Enterobacteriaceae gut pathobionts are particularly recalcitrant to conventional antibiotic treatment3,4, though an emerging body of evidence suggests that manipulation of the commensal microbiota may be a practical alternative therapeutic strategy5–7. In this study, we rationally isolated and down-selected commensal bacterial consortia from healthy human stool samples capable of strongly and specifically suppressing intestinal Enterobacteriaceae. One of the elaborated consortia, consisting of 18 commensal strains, effectively controlled ecological niches by regulating gluconate availability, thereby reestablishing colonization resistance and alleviating antibiotic-resistant Klebsiella-driven intestinal inflammation in mice. Harnessing these microbial activities in the form of live bacterial therapeutics may represent a promising solution to combat the growing threat of proinflammatory, antimicrobial-resistant bacterial infection.

Klebsiella species comprise a major etiology of nosocomial infections 3,4 .We previously isolated multidrug-resistant Klebsiella strains from IBD patients, including the K. pneumoniae Kp-2H7 strain, which can expand and persist in the intestine in the setting of antibiotic-induced dysbiosis and promote T helper 1 (T H 1) cell-mediated in ammation 17 .We set out to identify human gut commensals that promote decolonization of Kp-2H7 using the strategy outlined in Extended Data Fig. 1a.Germ-free (GF) mice were monocolonized with Kp-2H7 and orally inoculated with a stool sample from one of ve healthy Japanese donors (A, F, I, J, or K) 7 days later.E cacy of Kp-2H7 decolonization by FMT from each donor was examined by longitudinally quantifying faecal Kp-2H7 abundance.FMT from all donors resulted in a 3-4 log reduction in Kp-2H7 abundance (Fig. 1a).We selected stool samples from donors F, I, and K for follow-up analysis, and cultured them using six different types of media.We isolated 37 strains (31 unique strains when deduplicated) from donor F, 41 strains from donor I, and 46 strains from donor K based on 16S rRNA gene sequencing followed by whole-genome sequence analysis (Table S1).Isolate amplicon sequence variants (ASVs) accounted for 82%, 60%, and 48% of total microbiota sequences from donors F, I, and K, respectively (Extended Data Fig. 1b).A mixture of bacterial strains isolated from each donor was inoculated into Kp-2H7-monocolonized mice and Klebsiella decolonization capacity was queried.The mixture of 31 strains from donor F (designated as F31-mix) was most effective, and the magnitude and kinetics of Kp-2H7 reduction were comparable to those induced by donor F FMT (Fig. 1b, c).
To identify a minimal effector consortium from F31-mix, mice monocolonized with Kp-2H7 (day -7) were treated with F31-mix (day 0) and then given ampicillin via the drinking water (day 32 to 63) to perturb microbiota homeostasis (Fig. 1d and Extended Data Fig. 2a).Of note, Kp-2H7 carries b-lactamase genes and is resistant to ampicillin 17 .The faecal abundance of each of the 31 strains was longitudinally quanti ed by qPCR.Ampicillin treatment resulted in a transient surge in Kp-2H7 abundance, whereas the 31 strains showed variable trajectories (Fig. 1d and Extended Data Fig. 2a).The majority of Bacillota (formerly Firmicutes) strains exhibited an inverse abundance pattern compared to Kp-2H7, whereas Bacteroidota strains remained largely unchanged (Fig. 1d and Extended Data Fig. 2a).These results suggested that the Bacteroidota strains might not be contributing to Kp-2H7 suppression.Therefore, we divided the F31 strains into two groups: six Bacteroidota strains (F6-mix) and 25 other strains (F25-mix).F25-mix reduced Kp-2H7 abundance to a similar extent as did F31-mix, whereas F6-mix had no effect (Fig. 1b and Extended Data Fig. 3a).To further hone in on the effector strains among F25-mix, we excluded 5 strains (f34, f29, f13, f16, f11) that failed to colonize or were cleared following ampicillin treatment.In addition, we excluded a Coprococcus and a Ruminococcus strain (f14 and f10, respectively) that exhibited similar trajectories to Kp-2H7 (Fig. 1d and Extended Data Fig. 2a).A Spearman's rank correlation test indicated that most of the remaining 18 strains were signi cantly inversely associated with Kp-2H7 abundance (Extended Data Fig. 2b).We thus tested the activity of these 18 strains together (F18-mix) and observed robust Kp-2H7 decolonization, with similar magnitude and kinetics to mice treated with the parental F31-mix (Fig. 1b and 1e).In contrast, administration of the remaining 13 strains (F13-mix) that were excluded from F18-mix was far less effective (Fig. 1b and 1e).In an attempt to further narrow down the minimal effective consortia, we next generated seven derivatives of F18-mix by subtracting various combinations of bacterial species, which ranged in size from 12 to 17 strains.These derivative consortia exhibited varying capacities to decolonize Kp-2H7 in vivo, though none was as effective as the full F18-mix (Fig. 1b, Fig. 1f, and Extended Data Fig. 3b).Speci cally, F18-mix was subdivided into four phylogenetic groups and derivative subsets lacking either group A (4 Blautia strains), B (6 Lachnospiraceae strains), C (5 Bacillota strains), or D (3 strains from other phyla) were tested for their ability to decolonize Kp-2H7.Compared to the full F18-mix, all derivative subsets exhibited reduced decolonization capacity (Fig. 1b and Fig. 1f).As subtraction of group D strains showed the greatest effect, we also tested three different 17-strain mixtures, each excluding one of the three group D strains (f37_Escherichia coli, f35_Fusobacterium ulcerans, or f01_Bi dobacterium longum) from F18-mix.These three 17-mixes demonstrated a signi cantly reduced ability to decolonize Kp-2H7 as compared to the full F18-mix (Fig. 1b and Extended Data Fig. 3b).An alternate F18-mix (altF18-mix) generated by replacing f37_E.coli and f01_B.longum with F13-mix-derived f34_Veillonella and f09_Parabacteroides also exhibited signi cantly reduced activity (Fig. 1b), indicating that the observed suppressive effect is not merely a function of strain number but rather depends on consortium composition.Together, these results suggest that the F18 members act cooperatively and that all phylogenetic components are required to achieve maximal Kp-2H7 suppression.
We additionally investigated the effects of donor F-, K-, and I-derived consortia on gram-positive pathogens, including vancomycin-resistant Enterococcus faecium (VRE) and Clostridioides di cile, which are also listed as high-priority multidrug-resistant threats 4 .K46-mix was highly effective against both VRE (ATCC 700221) and C. di cile (ATCC BAA1382), whereas F18-, F13-, and I41-mixes were not (Fig. 2a).Interestingly, F31-mix (which comprises F18-and F13-mixes) was as e cacious as K46-mix at decolonizing VRE, though this effect was blunted against C. di cile (Fig. 2a).Collectively, these results indicate that the process of narrowing down F31-mix to F18-mix led to the selection of commensals preferentially able to decolonize Enterobacteriaceae.Kp-2H7 is a strong inducer of intestinal T H 1 cells and can act as a colitogenic pathobiont in the context of a genetically susceptible host, such as interleukin-10-de cient (Il10 -/-) mice 17 .Therefore, we examined whether F18-mix-mediated Kp-2H7 decolonization can prevent the induction of colitis.GF Il10 -/-mice were monocolonized with Kp-2H7 and either F18-mix or F13-mix was orally administered 7 days later.Similarly to wild-type mice, Il10 -/-mice treated with F18-mix but not F13-mix showed a 3-4 log reduction in Kp-2H7 abundance (Extended Data Fig. 4b).In contrast to F13-mix, F18-mix signi cantly reduced histological scores of colitis, levels of faecal lipocalin-2 and calprotectin (which serve as sensitive biomarkers of intestinal in ammation), and T H 1 cell frequency (Fig. 2b-e).Therefore, F18-mix is capable of reducing Klebsiella burden and alleviating IBD-like in ammation.

Previously uncharacterized mechanisms involved in Klebsiella decolonization
Next, we sought to elucidate the mechanisms underlying F18-mix-mediated Klebsiella suppression.Since Kp-2H7 is known to induce interferon-g (IFNg)-expressing T cells in the intestine 17 , we reasoned that F18mix might be controlling Klebsiella colonization via regulation of the IFNg response.However, the capacity of F18-mix to decolonize Kp-2H7 was not affected in IFNg receptor-de cient (Ifngr1 -/-) mice (Extended Data Fig. 5a).Similarly, F18-mix was able to e ciently reduce intestinal Kp-2H7 burden in mice de cient in Toll-like receptor signaling (Myd88 -/-Trif -/-) or lymphocytes (Rag2 -/-gc -/-) (Extended Data Fig. 5b), implicating a mechanism independent of the canonical innate and adaptive immune systems.We also assessed intestinal epithelial cell transcriptomic responses in GF mice colonized with effective (F31or F18-mix) versus less-effective (F13-mix) microbial consortia.However, we could not discern consistent shifts in genes or pathways that were speci cally modulated by the effective consortia, including those encoding previously-reported putative anti-Klebsiella gene products like antimicrobial peptides and PPARg-regulated molecules [23][24][25] (Extended Data Fig. 5c).Therefore, while host-intrinsic factors could still play a role in F18-mix-mediated Kp-2H7 decolonization, we decided to focus our line of inquiry on interbacterial interactions.
Caecal suspensions from mice colonized with F31-or F18-mix, but not F13-mix, strongly suppressed the growth of Kp-2H7 upon in vitro anaerobic coculture (Extended Data Fig. 6a).This suppressive effect was abrogated when coculture was performed under aerobic conditions or when the caecal suspension was ltered or heat-inactivated, suggesting that live F18-mix activity is required.We next searched for effector consortia-derived molecules that may have bacteriostatic or bactericidal effects on Kp-2H7.Liquid chromatography-mass spectrometry (LC-MS) revealed that mice colonized with effective consortia (F31or F18-mix) contained higher caecal levels of 4-hydroxybenzoic acid (4-HBA), cholic acid, acetate, and butyrate than did mice colonized with less-effective consortia (F13-mix, or F18-mix minus phylogenetic groups A, B, C, or D) or Kp-2H7 alone (Extended Data Fig. 6b).These molecules have previously been implicated in colonization resistance mechanisms against Enterobacteriaceae 26-28 .In our hands, however, 4-HBA, acetate, and butyrate exerted only weakly suppressive effects, requiring a high concentration of 100 mM (and, in the case of acetate and butyrate, low pH) to inhibit Kp-2H7 growth in vitro.Cholic acid was completely ineffective at all concentrations tested (Extended Data Fig. 6c).Furthermore, butyrate supplementation via tributyrin feeding did not signi cantly enhance the suppressive activity of F18-mix in vivo (Extended Data Fig. 6d).
We next pro led the Klebsiella transcriptome in mice colonized with Kp-2H7 versus Kp-2H7+F18-mix.Cocolonization with F18-mix signi cantly altered the Kp-2H7 transcriptional landscape, particularly pathways involved in carbon metabolism, amino acid metabolism, and the phosphotransferase system (Extended Data Fig. 8a, b), suggesting competition for nutrients and substrates, consistent with previous studies showing that nutrients affect intestinal colonization of Enterobacteriaceae 12,[32][33][34][35] .To identify Kp-2H7 genes that affect intestinal tness upon co-colonization with F18-mix, we employed a transposon mutagenesis approach.We generated a highly saturated transposon mutant library with approximately 8 x 10 5 distinct mutations in Kp-2H7, which is estimated to contain 100+ mutants per gene (Fig. 3a).GF mice were inoculated with a Kp-2H7 pool containing all transposon mutants (Kp-TPs) followed by oral gavage with F18-mix or F13-mix, and faecal samples were collected longitudinally and subjected to transposon insertion sequencing (Tn-seq) 36,37 (Fig. 3a).After treatment with F18-mix, 194 Kp-2H7 mutants displayed signi cantly reduced tness, many of which were de cient in carbohydrate and amino acid metabolism (Fig. 3b, Table S2).Speci cally, mutations in genes involved in the metabolism of gluconate, glucose, fructose, and galacturonate accelerated Kp-2H7 decolonization in F18-mix-treated mice relative to untreated and F13-mix-treated mice (Table S2), suggesting that F18-mix competes with Kp-2H7 for these carbon sources.
The DgntR mutant was just as resistant to F13-mix treatment as WT Kp-2H7, but exhibited enhanced sensitivity to F18-mix treatment over WT Kp-2H7 (Extended Data Fig. 9c).Notably, DgntK Kp-2H7 showed a 3-log reduction following treatment with the otherwise less effective F13-mix (Fig. 3e).These results support the notion that controlling gluconate metabolism is one of the mechanisms involved in the suppression of Klebsiella by F18-mix.
To further investigate the importance of gluconate, we next assessed the concentration of various carbon sources in the faeces of GF mice by quantitative LC-MS and found that gluconate was the most abundant (Fig. 3f).Compared to GF mice, gluconate concentration dropped by 1 log in Kp-2H7monocolonized mice and by 2 logs in F18-mix-colonized mice, but did not change substantially in F13mix-colonized mice (Fig. 3g).Dietary gluconate supplementation signi cantly diminished the anti-Klebsiella suppressive effect of F18-mix (Fig. 3h), whereas gluconate deprivation reduced Kp-2H7 load in F18-mix-treated mice (Fig. 3h) and in monocolonized mice (Extended Data Fig. 10a).When Kp-2H7monocolonized mice were sequentially inoculated with one of the F18 strains every ve days over a 95day period, we observed a cumulative decline in faecal gluconate that mirrored the reduction in Kp-2H7 burden (Extended Data Fig. 10b, c).Together, these results indicate that F18-mix suppresses Klebsiella through community action-mediated competitive reduction in gluconate availability, though we cannot exclude the contribution of additional factors.
Notably, gluconate preference is a relatively speci c feature of Enterobacteriaceae (Klebsiella and Escherichia), as the tested Pseudomonas, Campylobacter, Enterococcus faecium, Clostridioides di cile, and Clostridium perfringens strains failed to e ciently consume gluconate in vitro (Fig. 3i), potentially explaining why F18-mix treatment was selectively effective against Enterobacteriaceae (Fig. 2a).

Gluconate pathway genes in patients with IBD
Of the F18 strains, 8 effectively consumed gluconate in vitro (Fig. 4a).A mixture of these 8 gluconateutilizing strains (F8-mix) substantially reduced Kp-2H7 load in a gnotobiotic setting, though to a slightly lesser extent than did the full F18-mix (Fig. 4b), further supporting the gluconate competition hypothesis.We examined the genomes of F18 strains (Table S1) and found that 3 Blautia, 2 Enterocloster, and 1 E. coli strains carry gene clusters encoding enzymes and transporters putatively involved in gluconate metabolism (Fig. 4a, Extended Data Fig. 11).In contrast to the "classical" gluconate kinasedependent metabolic pathway found in Enterobacteriaceae, the gene clusters identi ed in the Blautia and Enterocloster strains encode an "alternative" pathway that utilizes gluconate dehydratase.In this pathway, gluconate is rst dehydrated to 2-keto-3-deoxygluconate (KDG), then phosphorylated into KD6PG by KDG kinase (KDGK), and eventually cleaved into pyruvate and GA3P by Eda 41,42 (Fig. 4c).We next queried the presence of alternative gluconate pathway genes in the genomes of our donor F-, K-, and I-derived culture collections (comprising 101 isolates) (Table S1).Classical gluconate operon genes were identi ed in several Enterobacteriaceae, Bi dobacterium, and Megasphaera species, whereas alternative gene clusters encoding both gluconate transporter and gluconate dehydratase homologues were identi ed in Blautia, Ruminococcus, Enterocloster, and Faecalibacterium species (Extended Data Fig. 11, Table S3).The carriage of gluconate pathway gene clusters, but not the transporter or dehydratase/kinase alone, was associated with effective gluconate consumption in vitro (Extended Data Fig. 11).
Finally, we examined faecal gluconate levels as well as the abundance and prevalence of species carrying gluconate operon genes in paediatric patients with UC in active (mild versus moderate to severe) and inactive states from the PROTECT cohort 43 (Extended Data Fig. 12a).Intensity of faecal gluconate (C 6 H 12 O 7 ) as predicted by LC-MS was used as a proxy for intestinal concentration, and was positively associated with levels of faecal calprotectin (Extended Data Fig. 12b).Supporting the idea that gluconate confers improved tness to Enterobacteriaceae, metagenomic species (MSPs) within Citrobacter freundii, Klebsiella oxytoca, K. pneumoniae, and E. coli that carry gluconate kinase and transporter genes were signi cantly more prevalent and abundant in UC patients in active versus inactive disease states (Fig. 4d-g).In addition, classical gluconate operon-encoding Veillonellaceae and Selenomonadaceae family members, such as Megasphaera massiliensis and Megamonas funiformis, were signi cantly enriched in UC patients with moderate/severe versus inactive disease (Fig. 4d-g).In contrast, alternate gluconate operon-encoding Blautia, Clostiridium, and Faecalibacterium were more abundant in patients in inactive disease states.We also examined the adult IBD cohort HMP2 (ref. 14) (Extended Data Fig. 12a).IBD was associated with expansion of gluconate kinase gene-carrying Enterobacteriaceae species (Extended Data Fig. 12c).In particular, E. coli, C. freundii, and K. pneumoniae consistently emerged as signi cantly more prevalent in disease in the HMP2 cohort.Overall, enrichment of gluconate operon genes in proin ammatory Enterobacteriaceae is associated with active disease and may contribute to IBD ares, whereas enrichment of alternative pathways in commensals is associated with disease remission and may enable nutrient competition that ultimately suppresses proin ammatory pathobionts and ameliorates gastrointestinal pathology.

Discussion
Enterobacteriaceae species, such as K. pneumoniae and E. coli, have been linked to both IBD-associated in ammatory pathology as well as multidrug-resistant nosocomial infections for which there are limited therapeutic options 3,4  sequencing were as follows: F27 primer: 5'-AGRGTTTGATYMTGGCTCAG-3'; R1492 primer: 5'-TACGGYTACCTTGTTACGACTT-3'.Individual isolates in the culture collection were grouped as "strains" if their 16S rRNA gene sequences shared >98.0%homology.
To prepare the bacterial mixture for inoculation, isolated strains were individually cultured in the appropriate broth at 37°C for 1-2 days (mGAM broth was used for culturing the F18 strains).Bacterial density was adjusted based on absorbance at 600 nm values, and equal volumes of the cultured strains were mixed and centrifuged at 3000 × g for 10 min at 4°C to concentrate vefold.Thereafter, GF mice were administered 200 mL of the bacterial mixture/mouse (approximately 1-2 × 10 Unless otherwise stated, mice were fed a high-calorie diet (CL-2; CLEA Japan, Inc.).To evaluate the effect of dietary gluconate supplementation, a chemically de ned diet (AIN93G; Oriental Yeast Co., Ltd) supplemented with 0%, 2.5% or 10% gluconate was used.To examine the effectiveness of each of the 18 strains, C57BL/6 GF mice were inoculated with Kp-2H7 (2 × 10 8 CFU/mouse) by oral gavage, followed by oral administration of each strain of the F18-mix one by one every ve days for 95 days.Faecal samples were collected every ve days to count the CFU of Kp-2H7 as well as to quantify the levels of gluconate.
Bacterial DNA extraction, quantitative real-time PCR, and 16S rRNA gene sequencing The frozen faecal samples were thawed and 50 µL of each sample was mixed with 350 µL TE10 (10 mM Tris-HCl, 10 mM EDTA) buffer containing RNase A ( nal concentration 100 µg/mL, Invitrogen) and lysozyme ( nal concentration 3.0 mg/mL, Sigma).The suspension was incubated for one hour at 37°C with gentle mixing.Puri ed achromopeptidase (Wako) was added to a nal concentration of 2,000 unit/mL, and the sample was further incubated for 30 min at 37°C.Then, sodium dodecyl sulphate ( nal concentration 1%) and proteinase K ( nal concentration 1 mg/mL, Nacalai) were added to the suspension and the mixture was incubated for one hour at 55°C.Thereafter, puri ed DNA was obtained from the samples using the Maxwell® RSC cultured cell DNA kit, according to the manufacturer's protocol.For quantifying the amount of bacterial DNA, real-time qPCR was performed using the Thunderbird SYBR qPCR Mix (TOYOBO) and LightCycler 480 (Roche).The primer pairs used in this study are listed in Table S4.
16S rRNA gene sequencing was performed using MiSeq according to the Illumina protocol.PCR was performed using primers 27Fmod (5'-AGRGTTTGATYMTGGCTCAG-3') and 338R (5'-TGCTGCCTCCCGTAGGAGT-3') to amplify the V1-V2 region of the 16S rRNA gene.Amplicons (approximately 330 bp in size) generated from each sample were puri ed using AMPure XP magnetic beads (Beckman Coulter).DNA was quanti ed using the Quant-iT Picogreen dsDNA assay kit (Invitrogen) and In nite M Plex plate reader (Tecan), according to the manufacturer's instructions, and then stored at 4°C.The pooled amplicon library was sequenced using the MiSeq Reagent Kit v2 (500 cycles) and MiSeq sequencer (Illumina; 2 × 250-bp paired-end reads).After demultiplexing the 16S sequence reads based on the sample-speci c index, primer sequences were trimmed by Cutadapt v. 3.3 46 .The trimmed reads were uploaded to the DADA2 R package v.4.0.3 (ref. 47) to construct amplicon sequence variants (ASVs) using the lterAndTrim function with the following parameters: maxN = 0, truncQ = 2, maxEE = 2, and truncLen = c (200,180).Possible chimeric reads were removed with the removeBimeraDenovo function of the DADA2.The taxonomic assignment of each ASV was determined by similarity searching using the GLSEARCH program.For determining taxonomy of ASV sequences that originated from human faecal samples, 16S RefSeq from NCBI, RDP 48 , CORE 49 , and GRD (https://metasystems.riken.jp/grd/)were used as the reference database.The sequences of isolates were compared to ASV detected in the faecal microbiome of donors F, I, and K, and those matching >99% were determined to be their corresponding ASVs.

Ex vivo caecal suspension culture
Caecal contents from GF or F31-, F18-, and F13-mix colonized mice were anaerobically resuspended in water at a concentration of 100 mg/mL.Caecal contents were either ltered through a 0.22 µm lter (Millex Millipore) after centrifuging at 10,000 × g for 5 min, heat-killed at 105 °C for 30 min, or left untreated.Thereafter, a diluted overnight culture of Kp-2H7 (10 3 CFU in 10 µL) was added to 200 µL of each caecal suspension.After incubating at 37°C for 48 hours under aerobic or anaerobic conditions, samples were serially diluted and plated on a selection agar plate (DHL with 30 mg/L ampicillin and 30 mg/L spectinomycin) for counting Kp-2H7 CFU.

Bacterial growth monitoring
The wild type, ΔgntK, or ΔgntR Kp-2H7 strain was cultured in M9 minimal medium for 24 hours at 37°C, which was diluted 100 times with sterile water.A 10 µL culture dilution was inoculated into 200 µL of M9 medium with 0.4% of glucose or gluconate as the sole carbon source or without carbon.To examine the effect of metabolites on Kp-2H7 growth, 10 mL of Kp-2H7 culture dilutions were inoculated into 200 µL of M9 medium containing varying concentrations of 4-HBA (4-hydroxybenzoic acid) (100, 10, 1, or 0.1 mM), cholic acid (500, 100, 20, or 4 µM), and acetate or butyrate (100, 25, 6.25, 1.56 or 0.39 mM).The pH of acetate and butyrate was adjusted to either 5.0 or 7.0.Bacterial growth was monitored by measuring absorbance at 600 nm every 30 minutes using a microplate reader [Sunrise Thermo (Tecan) for anaerobic conditions and In nite 200 PRO (Tecan) for aerobic conditions] at 37°C with a 100-second shaking before each time point.

Transcriptome analysis of epithelial cells
Total RNA was isolated from colonic epithelial cells using NucleoSpin RNA (MACHEREY-NAGEL), according to the manufacturer's instructions.Libraries for RNA sequencing were prepared using TruSeq Stranded mRNA Library Prep (Illumina Inc.), according to the manufacturer's instructions.The libraries were sequenced using NovaSeq 6000 (Illumina Inc.) with the mode of 150-bp paired-end.The sequenced paired-end reads were quality-controlled using Trimmomatic 57 version 0.39 with "2:30:10 LEADING:3 TRAILING:20 SLIDINGWINDOW:4:15 MINLEN:5" options and FASTX-Toolkit version 0.0.13 (http://hannonlab.cshl.edu/fastx_toolkit/index.html) with "-q 20 -p 80" options.Unpaired reads were excluded from further analyses.The remaining quality-controlled reads were mapped to the mouse reference genome (mm10) using STAR 58 version 2.7.2b.The mapped reads were counted for each gene using featureCounts 59 version 1.5.2 with "-t exon -p -B -Q 1" options.the transcripts per million (TPM) values of each gene in each sample were calculated.The differential expression analysis was performed using DESeq2 60 version 1.28.1, and the p-values were corrected by Benjamini-Hochberg (BH) method to maintain the false discovery rate (FDR) below 5%.

Transcriptome analysis of Kp-2H7
Total was isolated from faecal samples using NucleoSpin RNA (MACHEREY-NAGEL), according to the manufacturer's instructions.Libraries for RNA sequencing were prepared using TruSeq Stranded mRNA Library Prep (Illumina Inc.) and sequenced using HiSeq X (Illumina Inc.) with the mode of 150-bp paired-end.To analyse the transcriptome pro les of Kp-2H7 in the presence or absence of F18-mix, a reference genome was created by concatenating the genome sequence of Kp-2H7 with the genome sequences of the F18-mix.The sequenced paired-end reads were quality-controlled using Trimmomatic 57 version 0.39 with "2:30:10 LEADING:3 TRAILING:20 SLIDINGWINDOW:4:15 MINLEN:5" options and FASTX-Toolkit version 0.0.13 (http://hannonlab.cshl.edu/fastx_toolkit/index.html) with "-q 20 -p 80" options.Unpaired reads were excluded from further analyses.The remaining reads were mapped to the mouse (mm10) and PhiX reference genomes using minimap2 version 2.17-r941 with "-N 1 -a" options 61 .
Then, the reads unmapped to the mouse genome were extracted to obtain quality-controlled reads for subsequent analyses.The quality-controlled reads were mapped to the concatenated reference genome using bowtie2 62 version 2.3.4.1.Uniquely mapped reads were counted for each Kp-2H7 gene.The differential expression analysis was performed using DESeq2 60 version 1.28.1 with BH-correction method to maintain the FDR below 5%.The heatmap was obtained from the variance-stabilizing transformations values obtained from the DESeq2 output.
For real-time qPCR analysis, cDNA was synthesized using ReverTra Ace qPCR RT Master Mix (TOYOBO), and qPCR was performed using Thunderbird SYBR qPCR Mix (TOYOBO) on a LightCycler 480 (Roche).
Transposon sequencing GF mice were colonized with the pool of 8 x 10 5 Kp-2H7 transposon mutants.Faecal samples were collected on day 0, 4, 10, and 28 following colonization, suspended in PBS (50 mg/mL) containing 20% glycerol, and cultured overnight at 37°C on LB agar plates containing kanamycin (90 mg/L).Kp-2H7 mutant colonies were scraped together and DNA was extracted by the method described above.
Transposon sequencing was carried out according to the method described by Kazi et al. 63 .Brie y, genomic DNA was fragmented via sonication.Then, a poly-C tail was added to the 3' end of the DNA fragment by terminal deoxynucleotidyl transferase.The transposon junctions were ampli ed using a biotinylated primer, which was then enriched using streptavidin beads.By performing a second nested PCR, a single barcode was added to each sample.The libraries were sequenced using HiSeq 2500 (Illumina Inc.) with the mode of 50-bp single-end.The sequenced reads were quality-controlled using Trimmomatic 57 version 0.39 with "2:30:10 LEADING:3 TRAILING:20 SLIDINGWINDOW:4:15 MINLEN:5" options and FASTX-Toolkit version 0.0.13 (http://hannonlab.cshl.edu/fastx_toolkit/index.html) with "-q 20 -p 80" options.Unpaired reads were excluded from further analyses.The remaining reads were mapped to the mouse reference genome (mm10) using minimap2 version 2.17-r941 with "-N 1 -a -x sr" options 61 .Then, the reads unmapped to the mouse genome were extracted to obtain quality-controlled reads for subsequent analyses.The quality-controlled reads were mapped to the Kp-2H7 assembled genome using bowtie2 version 2.4.2.The mapped reads were counted for each gene using featureCounts3 version 1.5.2 with "-t CDS -p -B -Q 1" options, and the TPM of each gene was calculated as the relative abundance of a gene mutant in a sample by assuming that each transposon mutant has a single insertion.The differential abundance mutants were detected by Welch's t-test for log-scaled TPM with BH-correction method to maintain the FDR below 5%.

Generation of Kp-2H7 mutants
The Kp-2H7 deletion mutants of were generated using the Quick and Easy E. coli Gene Deletion Kit (Gene Bridges, Heidelberg) according to the manufacturer's protocol.Brie y, Kp-2H7 cells were transformed with the pRED/ET plasmid harbouring the tetracycline-resistant gene by electroporation.Bacteria with pRED/ET were selected on LB plates containing tetracycline (30 mg/L) at 30°C.Thereafter, these cells were incubated in LB broth with appropriate antibiotics at 30°C until absorbance at 600 nm reached 0.2, followed by an additional hour of incubation with 0.3% L-arabinose at 37°C to induce the expression of the recombinant proteins.These cells were used to prepare electrocompetent cells and were transformed with the linear DNA fragment (the FRT-PGK-gb2-neo-FRT cassette)-anked homology arms.The functional cassettes were generated by PCR, according to the manufacturer's protocol.The primers with homology arms are listed in Table S4.The electroporated cells were incubated in 1 mL of LB broth for three hours at 37°C.Gene deletion strains were selected on LB agar plates with kanamycin (90 mg/L) after overnight growth at 37°C.The double or triple knockout strains were generated by removing the kanamycin selection marker through electroporation of the FLP expression plasmid (707-FLPe) and repeating the above-mentioned protocol.The deletions were con rmed by DNA sequencing.

Isolation of lymphocytes and ow cytometry
Lymphocytes were collected from the large intestines and analysed according to previously described protocols 17,64 .Brie y, the intestines were dissected longitudinally and washed with PBS to remove all luminal contents.All samples were incubated in 15 mL of Hanks' balanced salt solution (HBSS) containing 5 mM EDTA for 20 min at 37°C in a shaking water bath to remove epithelial cells.Thereafter, after removal of any remaining epithelial cells, muscular layers and fat tissues using forceps, the samples were cut into small pieces and incubated in 10 mL of RPMI1640 containing 4% foetal bovine serum (FBS), 0.5 mg/mL collagenase D (Roche Diagnostics), 0.5 mg/mL dispase II (Roche Diagnostics), and 40 μg/mL DNase I (Roche Diagnostics) for 50 min at 37°C in a shaking water bath.Thereafter, the resultant digested tissues were washed with 10 mL of HBSS containing 5 mM EDTA, resuspended in 5 mL of 40% Percoll (GE Healthcare), and underlaid with 2.5 mL of 80% Percoll in a 15-mL Falcon tube.
Percoll gradient separation was performed by centrifugation at 850 × g for 25 min at 25°C.Lymphocytes were collected from the interface of the Percoll gradient and washed with RPMI1640 containing 10% FBS, and then stimulated with 50 ng/mL PMA and 750 ng/mL ionomycin (both from Sigma) in the presence of Golgistop (BD Biosciences) at 37°C for four hours.After labelling of the dead cells with Ghost Dye Red 780 Viability Dye (Cell Signaling Technology), the cells were permeabilized and stained with anti-CD3e (BUV395; Biolegend), CD4 (BUV737; Biolegend), CD8a (PE/Cy7; Biolegend), TCRβ (BV421; Biolegend), and IFN-γ (FITC; Biolegend) using the Foxp3/Transcription Factor Staining Buffer Kit (Tonbo Biosciences), according to the manufacturer's instructions.All data were collected on a BD LSRFortessa (BD Biosciences) and analysed using Flowjo software (TreeStar).CD4+ T cells were de ned as a CD4+ TCRβ+ CD3e+ subset within the live lymphocyte gate.

Measurement of lipocalin-2 and calprotectin
The faecal pellets Il10 Non-targeted metabolomics analysis C57BL/6 GF mice were monocolonized with Kp-2H7, followed by oral administration of bacterial mix.Caecal contents were collected on day 28 after administration of isolated bacterial mix and stored at -80℃ until use.Frozen caecum contents were homogenized by shaking with metal corn using a multi beads shocker as previously described 66 .Then, the samples were suspended in 400 µL of methanol per 100 mg caecum content, and a 40 mL aliquot was subjected to the single layer extraction and untargeted LC-QTOF/MS analysis 66 .SCFAs were simultaneously extracted and derivatized from 20 µL of the suspension by using penta uorobenzyl bromide alkylation reagent (Thermo Fischer Scienti c, Waltham, MA, USA), and analysed by GCMS as previously described 67 .Water-soluble metabolites were extracted by rst mixing 4 µL of the suspension, 196 µL of methanol, 200 µL of chloroform, 70 µL of water, and 10 µL of internal standards mix [100 µM of cycloleucine, 500 µM of citric acid-d4, and 1.0 mM of ornithine-d7 (Cambridge isotope laboratories, Andover, MA, USA)].After vortexing for 1 min and centrifugation at 15000 × g for 5 min at 4℃, 100 µL of supernatant was evaporated to dryness.The dried samples were derivatized via methoxyamination, trimethylsilylation, or tert-butyldimethylsilylation, and then analysed by GC-MS/MS using Smart Metabolite DatabaseTM (Shimadzu Corp., Kyoto, Japan) or GC-MS operated in selected ion monitoring mode, as described previously 68 .Bile acids were extracted from 4 µL of the suspension mixed with deuterium-labelled internal standard mix [1.0 µM of cholic acid-d4, 1.0 µM of lithocholic acid-d4, 1.0 µM of deoxycholic acid-d4, 1.0 µM of taurocholic acid-d4, and 1.0 µM of glycocholic acid-d4 (Cayman Chemical)] using the Monospin C18 column (GL science).The column was washed with 300 µL of water (x2) and 300 µL of hexane (x1).Bile acids were eluted with 100 µL of methanol, then subjected to LC-MS/MS analysis using an UPLC I class (Waters) with a linear ion-trap quadrupole mass spectrometer (QTRAP 6500; AB SCIEX) equipped with an Acquity UPLC BEH C18 column (50 mm, 2.1 mm, and 1.7 μm; Waters).Samples were analysed with a mobile phase consisting of water/methanol/acetonitrile [14:3:3 (vol/vol/vol)] and acetonitrile, both containing 5 mM ammonium acetate, for 4 min, which was changed to 40:60 after 12 min, to 5:95 after 2 min, and then held for 2 min; with ow rates of 300 μL/min.Bile acids were detected by multiple-reaction monitoring in negative mode.

Measurement of carbohydrate levels
To evaluate bacterial gluconate utilization in vitro, isolated strains were cultured in mGAM broth or RCM containing 300 µM gluconic acid for 48 hours at 37°C under anaerobic conditions.Supernatant of each  71 .A targeted screening of these bins with DIAMOND BLASTP version 0.9.14 72 was conducted to identify genes associated with gluconate transport and metabolism, retaining hits with an e-value <0.01 and sequence identity ≥60%.MSPs were categorized based on the combinations of gluconate-related genes detected.A differential abundance analysis was performed on TPM-normalized and Centred Log-Ratio-transformed MSP counts to control for sequencing depth, gene length, and compositional biases.Statistical signi cance was ascertained through a non-parametric Mann-Whitney U test accompanied by Benjamini-Hochberg correction.Effect sizes (r), calculated as the test statistic divided by the square root of the sample size, along with bootstrapped con dence intervals, were computed to account for unbalanced group sizes, offering insights into the robustness and directionality of the observed effects.
For the PROTECT cohort, comparative analyses were executed on randomly chosen cross-sectional samples from children manifesting mild UC (n = 32) or moderate to severe UC (n = 23), against inactive UC (n = 39).To validate the robustness of the ndings, these analyses were iteratively repeated with varying seed values for random sample selection from longitudinal data pools of mild (n = 64), moderate/severe (n = 57), and non-IBD samples (n = 119).Within the HMP2 cohort, inclusion was also limited to cross-sectional samples accompanied by available calprotectin data.In response to the attenuated metagenome disease signal observed in the study cohort 73 , a targeted in ammation-speci c selection approach was utilized instead of choosing the cross-sectional data via repeated random sampling.For the CD and UC sub-cohorts, the sample with maximal calprotectin value per patient was included (CD = 41, UC = 26).Conversely, for the non-IBD control group, the cross-sectional sample with the minimal calprotectin value per patient was chosen (n = 24).Statistical analyses were conducted using R software version 4.2.1 (Ubuntu 20.04.5 LTS).

Untargeted stool metabolomics and gluconate intensity estimation
Untargeted stool metabolomics of faecal samples from the PROTECT cohort was performed using LC-MS in negative mode, and calprotectin was measured by ELISA.Brie y, hydrophilic interaction liquid chromatography (HILIC) analyses of water-soluble metabolites in the negative ionization mode were conducted using Shimadzu Nexera X2 U-HPLC (Shimadzu Corp.) coupled to a Q Exactive Plus mass spectrometer (Thermo Fisher Scienti c).Metabolites were extracted from plasma or stool (30 µL) using 120 µL of 80% methanol containing inosine-15N4, thymine-d4, and glycocholate-d4 internal standards (Cambridge Isotope Laboratories).The samples were centrifuged (10 min, 9,000 × g, 4°C), and the supernatants were injected directly onto a 150 × 2.0 mm Luna NH2 column (Phenomenex; Torrance, CA).

Statistical analyses
Statistical were performed using GraphPad Prism software (GraphPad Software, Inc.).Kruskal-Wallis test and FDR method of Benjamini and Hochberg were used for multiple comparisons during CFU comparisons.Mann-Whitney U test with Welch's correction was used for comparisons between the two groups.Spearman's rank correlation was used to investigate the correlation between the relative abundance of Kp-2H7 and isolated strains.Kp-2H7 mutants (Kp-TPs) were obtained.Kp-TPs were pooled together and administered to GF mice, followed by oral administration of F18-mix or F13-mix.Faecal samples were collected and sequenced at All masses detected in HILIC negative mode were matched via adduct subtraction and molecular formula match to compounds downloaded from the Human Metabolome Database (HMDB) on 10/10/2022.The measured m/z values were adjusted for [M-H]-adducts, and molecular formulae matching to within 5 ppm were selected as candidate identi ers.In cases where multiple molecular formulae matched the adduct-adjusted mass (as a result of multiple potential adducts), the one with a minimal ppm difference was selected.Out of 4,461 detected features (m/z, retention time pairs) a single feature 195.0512 m/z @ 4.34 min resolved to the formula C 6 H 12 O 7 (delta ppm = 0.89), related to a group of ve compounds with canonical structure O=C(O)C(O)C(O)C(O)C(O)CO, which includes L-gluconic acid (HMDB0000625).

Figure 2 Control
Figure 2

Figure 4 Association
Figure 4 44terobacteriaceae, which may further aggravate dysbiosis and thus impair colonization resistance.In this study, we adapted a top-down gnotobiotic approach44to elaborate a de ned microbial consortium consisting of 18 effector bacterial strains from a healthy individual, which is capable of decolonizing Enterobacteriaceae strains.This F18-mix exerts potent anti-Enterobacteriaceae effects, likely through multiple mechanisms including restriction of nutrient availability and reshaping of ecological niches within the intestine.Each microbiota member possesses a unique nutritional program, which in turn determines local nutrient availability and thus niche de nition.Our results, together with previous studies 35,45, suggest that Enterobacteriaceae has a hierarchy of carbon preferences with gluconate being one of the most-preferred carbon sources in the intestine.When faced with F18-mix-mediated gluconate restriction, Klebsiella compensates by metabolically switching to utilize other unpreferred carbon sources.However, it is likely that F18-mix can effectively consume several of these alternative carbon sources as well, thereby further restricting nutrient availability to Klebsiella.Although more research will be needed to fully elucidate the rules governing effective competition (including other nutritional dependencies and the role of interspecies cooperation, among others), our ndings establish a solid foundation for the development of microbiota-directed therapeutics to suppress Enterobacteriaceae pathobionts via ecological control.Overall, the F18-mix strains represent a promising candidate for clinical development as a rationally-de ned microbial consortia to treat prevalent infectious and in ammatory diseases. Delarations Immunotherapeutics.J.M.N, and B.O. are employees of Vedanta Biosciences.S.C. was an employee of Vedanta Biosciences at the time of her contributions.All other authors declare no competing interests.Author Contributions K.H. and M.F.planned experiments, analysed data, and wrote the paper together with T.K., M.M.P., K.Y-M., D.R.P., S.S., Y.A., N.O., Y.K., M.A., A.N.S., D.P., and R.J.X.; K.Y-M.designed the in vivo Kp-2H7 decolonization experiment; M.F., T.K., and K.Y-M.conducted bacterial and animal experiments supported by, N.H., T.O., T.T., K.A., K.S., N.N., S.C., J.M.N., B.O., T.T., and K.A.; S.S., M.U., N.O, and M.A. conducted metabolomic analyses; M.M.P., D.R.P., Y.A., Y.K., W.S., M.H., M.S., V.B., and R.J.X.performed microbiome and bioinformatic analysis.manufacturer's protocol.DNA sequencing was performed by Euro ns.Sequences were aligned using the BLAST program of NCBI and the Ribosomal Database Project (RDP) databases.Primers used for DNA . Broad-spectrum antibiotics are often employed to treat multidrug-resistant 9CFU of total bacteria) by oral gavage.The bacterial mixture was administered into GF mice (200 mL/mouse, approximately 1-2 × 10 9 CFU of total bacteria) by oral gavage.
Colon tissue samples were dissected longitudinally and swiss-rolled, xed with 4% paraformaldehyde, embedded in para n, sliced to 5µm sections and stained with hematoxylin and eosin.The degrees of colitis were graded by The Mouse Colitis Histology Index 65 .The histological slides were evaluated blind by two investigators.
culture broth was collected, and the concentration of gluconate was measured by the ExionLC AD and SCIEX Triple Quad 6500+ LC-MS/MS system.To evaluate carbon level in faeces, each faecal sample was suspended in water (50 mg/mL), and the carbon levels in the culture supernatant were measured by LC-MS/MS.The measurement conditions for gluconate, glucuronate, and galacturonate were as follows: chromatographic separation was performed using the Intrada Organic Acid column, 150 × 2 mm (Imtakt); column temperature was 40℃; and the volume of each injection was 2 μL.The mobile phase comprising