Exclusive enteral nutrition initiates individual protective microbiome changes to induce remission in pediatric Crohn’s disease

SUMMARY

Exclusive enteral nutrition (EEN) is a first-line therapy for pediatric Crohn’s disease (CD), but protective mechanisms remain unknown. We established a prospective pediatric cohort to characterize the function of fecal microbiota and metabolite changes of treatment-naive CD patients in response to EEN (German Clinical Trials DRKS00013306). Integrated multi-omics analysis identified network clusters from individually variable microbiome profiles, with Lachnospiraceae and medium-chain fatty acids as protective features. Bioorthogonal non-canonical amino acid tagging selectively identified bacterial species in response to medium-chain fatty acids. Metagenomic analysis identified high strain-level dynamics in response to EEN. Functional changes in diet-exposed fecal microbiota were further validated using gut chemostat cultures and microbiota transfer into germ-free Il10-deficient mice. Dietary model conditions induced individual patient-specific strain signatures to prevent or cause inflammatory bowel disease (IBD)-like inflammation in gnotobiotic mice. Hence, we provide evidence that EEN therapy operates through explicit functional changes of temporally and individually variable microbiome profiles.

In brief

Häcker et al. showed that exclusive enteral nutrition (EEN) in pediatric Crohn’s disease (CD) creates temporally and individually variable microbiome profiles. They showed that medium-chain fatty acids link EEN to changes in the microbiota of CD patients. Finally, protective microbiota functions were validated in transfer model systems.

Graphical Abstract:

INTRODUCTION

Crohn’s disease (CD), one of the main entities of inflammatory bowel diseases (IBDs), is a chronic and relapsing inflammatory condition of the gastrointestinal tract.¹ Epidemiological studies have demonstrated a significant increase in the incidence of CD in Westernized countries, particularly among the pediatric age group.^2,3 The pathogenesis of IBD involves host genetics and environmental influences, among which diet and the intestinal microbiome are key etiologic factors.^4,5

Exclusive enteral nutrition (EEN) is the first-line treatment for pediatric CD^6,7 and also shows efficacy in adult patients.⁸ EEN refers to polymeric liquid formulas, which are mostly devoid of fibers.⁹ Reintroducing regular diet after 6–8 weeks of EEN often results in recurrence of inflammation and disease activity.¹⁰ EEN has been shown to drive changes in gut microbiota composition,¹¹ but the high relapse rate after cessation of EEN suggests the presence of a CD-conditioned microbiome, which is able to re-establish after dietary intervention.¹² CD is associated with reduced bacterial diversity, compositional and functional changes, and the disruption of metabolic circuits between microbiome and host.^13–15 Despite recent findings that EEN mediates protective functions in experimental models of colitis,¹⁶ human studies revealed incomplete recovery of the intestinal microbiome after antibiotic interventions.¹⁷ Thus, functional evidence for EEN-mediated changes in the human microbiome as an underlying cause of therapeutic efficacy is still lacking.

Here, we identify EEN-protective signatures using integrated microbiota and metabolite analysis of treatment-naive pediatric CD patients. Medium-chain fatty acids (MCFAs) directly target bacteria from EEN-protective signatures, providing a direct mechanistic link between diet and microbiota changes. Further, we established a combined gut chemostat-to-mouse transfer approach to functionally validate EEN-mediated protective changes in the microbiome of CD patients. We demonstrate that EEN directly shapes microbiomes to mediate protective functions in treatment-naive CD patients, operating through explicit patient-to-patient changes of temporally and individually variable strain profiles.

RESULTS

Microbiota and metabolite profiling identify individual responses to EEN therapy in pediatric CD patients

We established a pediatric IBD cohort and longitudinally followed 78 IBD patients, showing no disease-specific but individual bacterial and metabolite profiles in fecal samples (Figures 1A and S1A–S1C). To gain insight into mechanisms of EEN therapy, we focused explicitly on 20 newly diagnosed and treatment-naive CD patients receiving EEN as induction therapy (CD-EEN group, 9 females, age at study inclusion 12.5 ± 3.1 years) (Figures 1A and S1A; Table 1). EEN was highly effective and significantly improved disease activity and inflammatory markers at the end of EEN treatment, resulting in clinical response and induction of remission in 20/20 and 18/20 patients, respectively (Figure 1B). During follow-up, patients received methotrexate (6/20), anti-TNF (13/20), or combination therapies (9/13) as maintenance treatment. Relapse after EEN was observed in 8/20 patients over a 1-year follow-up (Figure 1B). Two patients switched from Modulen IBD to Neocate Junior during EEN due to suspected cow’s milk protein allergy.

Figure 1. Fecal microbial and metabolic changes in response to EEN.

Patients of the pediatric IBD cohort, with the focus on 20 newly diagnosed and treatment-naive CD patients, receiving EEN induction therapy. Longitudinal collected samples (n = samples, N = patients) that were used for 16S rRNA amplicon and metagenomic sequencing as well as metabolome analysis are shown.

Paired samples (n = 60) were used for data integration of 16S rRNA and metabolome data (samples highlighted in blue = taken under EEN, orange = taken PostEEN).

(B) Weighted pediatric Crohn’s disease activity index (wPCDAI) over time (EEN shaded blue), remission achieved below 12.5 points (dashed line). pE, PreEEN, EE, EEN, PE, PostEEN.

(C) Beta-diversity analysis of the fecal microbiota samples. The circular dendrogram shows differences in beta-diversities from the longitudinal microbial profiling based on generalized UniFrac distances between the 291 stool samples of the 20 patients. Indicated in color code are the patients and samples collected during EEN. Taxonomic composition of fecal samples at the phylum level is shown as color-coded stacked bar blots around the dendrogram. Branch color indicates results of unsupervised clustering. Patient color code found in (B).

(D) Visualization of untargeted metabolome samples from 18 newly diagnosed pediatric CD patients receiving EEN therapy: complete clustering with Euclidean distance of longitudinal samples from 18 CD-EEN patients (n = 145), patient color code found in (B).

(E) Beta-diversity MDS plot samples during EEN (n = 89, active: 55, inactive: 34) vs. PostEEN (n = 188, active: 41, inactive: 147).

(F) Analyses of untargeted metabolomics with representative PCA plot comparing samples during EEN (n = 35, active: 21, inactive: 14) and PostEEN (n = 106 active: 21, inactive: 85). Axes indicate principal components (PC), with PC1 representing the most variation (%) and PC2 representing the second most variation (%).

(E and F) Coefficients were tested with a t test, reported p values are Bonferroni-corrected; D, diet; A, activity; D:A, interaction term between diet and activity.

(G and H) Heatmaps depicting the abundances of the top selected feature from the correlation network Figure S2. Selected sPLS-DA features were subsetted by their mean importance (see STAR Methods: data integration of 16S rRNA amplicon data with metabolomics and metagenomics for details). All values shown are feature-wise Z scores. Information on association with EEN (blue), PostEEN (orange), sample collection under which diet (EEN blue), PostEEN (orange), community association (1–5) are indicated; (G) zOTUs,*BONCAT responders Figure 2, (H) metabolites.

Table 1.

Patient characterization

		CD (N = 45)	CD-EEN (N = 20)		UC/IBDU (N = 33, 29/4)
Sex	male	24 (53%)	11 (55%)	N/A	18 (55%)
	female	21 (47%)	9 (45%)	N/A	15 (45%)
Age at diagnosis (years)		11.9 (3.5)	12.4 (3.2)	N/A	9.6 (4.3)
Age at study inclusion (years)		13.3 (3.3)	12.5 (3.1)	N/A	10.8 (4.2)
First-degree family history of IBD		8 (18%)	3 (15%)	N/A	7 (21%)
Mean height (Z score)		−0.06 (1.05)	0.17 (1.06)	N/A	0.24 (1.12)
Mean weight (Z score)		−0.51 (1.29)	−0.57 (1.32)	N/A	−0.23 (1.08)
Mean body mass index (Z score)		−0.75 (1.49)	−0.95 (1.44)	N/A	−0.57 (1.03)
Disease characteristics
Location—lower GI1	none	1 (2%)	1 (5%)	proctitis	0
	ileum only	8 (18%)	3 (15%)	left sided	4 (12%)
	colon only	8 (18%)	5 (25%)	extensive	7 (21%)
	ileocolonic	28 (62%)	11 (55%)	pancolitis	22 (67%)
Location—upper GI1	proximal (n, %)	34 (76%)	18 (90%)	N/A	N/A
	distal (n, %)	11 (24%)	6 (30%)	N/A	N/A
Behavior1	nonstricturing-nonpenetrating	37 (82%)	17 (85%)	never severe	23 (70%)
	stricturing	5 (11%)	2 (10%)	ever severe	10 (30%)
	penetrating	1 (2%)	1 (5%)	N/A	N/A
	stricturing and penetrating	2 (4%)	0	N/A	N/A
Perianal involvement		9 (20%)	4 (20%)	N/A	N/A
Evidence of growth delay		11 (24%)	4 (20%)	N/A	N/A
Extraintestinal manifestations2		8 (18%)	5 (25%)	N/A	5 (15%)
Previous surgery		5 (11%)	0	N/A	0
Disease activity
Median wPCDAI		40 (30 to 52.5)	43.8 (38.1 to 51.3)	PUCAI	40 (25 to 57.5)
Inactive (wPCDAI <12.5)		3 (7%)	0 (%)	inactive (PUCAI <10)	3 (10%)
Mild (wPCDAI 12.5 to 40)		23 (51%)	8 (40%)	mild (PUCAI 10 to 34)	11 (33%)
Moderate (wPCDAI 42.5 to 57.5)		11 (24%)	8 (40%)	moderate (PUCAI 35 to 64)	12 (36%)
Severe (wPCDAI ≥57.5)		8 (18%)	4 (20%)	severe (PUCAI R 65)	7 (21%)
Baseline laboratory data
Mean CRP (mg/dL)		2.7 (3.5)	3.7 (4.3)	N/A	1.4 (2.6)
Mean ESR (mm/h)		31.4 (21.6)	36.8 (21.3)	N/A	27.1 (23.0)
Elevated inflammatory marker3		31 (69%)	18 (90%)	N/A	19 (58%)
Mean albumin (g/dL)		4.1 (0.7)	4.1 (0.5)	N/A	4.3 (0.6)
Hypoalbuminemia (<3.5 g/dL)		6 (13%)	2 (10%)	N/A	2 (6%)
Mean hemoglobin (g/dL)		11.9 (1.8)	11.7 (1.6)	N/A	11.4 (2.6)
Anemia (<10 g/dL)		6 (13%)	3 (15%)	N/A	8 (24%)
Median fecal calprotectin (mg/L) (IQR)		954 (166 to 3,041)	2,068 (825 to 7,362)	N/A	1,141 (491 to 2,972)
Treatment
Newly diagnosed		30 (67%)	20 (100%)	N/A	16 (48%)
EEN therapy duration in days		N/A	52 (19)	N/A	n.a.
Previous therapies	EEN	11 (%)	N/A	N/A	n.a.
	5-ASA	2 (4%)	N/A	N/A	4 (12%)
	Steroids	8 (18%)	N/A	N/A	9 (27%)
	IM	7 (16%)	N/A	N/A	1 (3%)
	Previous biologics	3 (7%)	N/A	N/A	1 (3%)
Ongoing therapies at time of study recruitment	5-ASA	4 (9%)	N/A	N/A	13 (39%)
	Steroids	1 (2%)	N/A	N/A	4 (12%)
	IM	4 (9%)	N/A	N/A	3 (9%)
	Anti-TNF	5 (11%)	N/A	N/A	3 (9%)

Data are N (%), mean (SD), or median (IQR 25^th to 75^th percentile). 5-ASA, 5-aminosalicylic acids; BMI, body mass index; CRP, C-reactive protein; EEN, exclusive enteral nutrition; ESR, erythrocyte sedimentation rate; IBD, inflammatory bowel disease; IM, immunomodulator (e.g., azathioprine andmethotrexate); PUCAI, pediatric ulcerative colitis activity index; TNF, tumor necrosis factor; wPCDAI, weighted Pediatric Crohn’s Disease Activity Index.

¹As per Paris classification.

²Five patients had joint manifestations, two had eye manifestations, eight had skin manifestations, one had autoimmune scle-rosing cholangitis, and one had autoimmune hepatitis.

³CRP R 0.5 mg/L or ESR R 25 mm/h.

Analysis of samples before (PreEEN), during (EEN), and after EEN therapy (PostEEN) (microbiota profiling: 291 samples, metabolomic analysis: 145 samples) revealed patient-specific responses (Figures 1C and 1D). Bacterial richness based on zero-radius operational taxonomic units (zOTUs) changed individually during formula feeding (Figure S1D) but was lower during active CD compared with remission (Figure S1E). Samples of EEN-treated patients are scattered throughout dendrograms of microbiota and metabolome profiles (Figures 1C and 1D). Despite considerable overlap in multi-dimensional scaling (MDS) plots based on microbial profiles, linear mixed models (LMMs) identified only diet as significantly contributing to the first MDS dimension (x axis) (Figure 1E, n = 277, p value < 0.0001; Figure S1F). Principal component analysis (PCA) plots of metabolic profiles also showed significant separation according to diet in PC2 but no separation according to disease activity (Figure 1F, n = 141, p value < 0.0001; Figure S1G).

Integration of microbiota and metabolome data identifies key features of EEN exposure

To identify key differences between fecal microbiota and metabolite profiles of patients collected under EEN (therapy, partially with still-active disease) and PostEEN conditions (receiving regular diets but completely in remission), a network analysis incorporating matched zOTU taxonomic profiles and untargeted metabolomics from 60 samples was performed. The identified interlinked features clustered in five distinct network communities (Figure S2). In summary, for the top selected features from the correlation network, higher mean importance was found for zOTUs associated with EEN compared with zOTUs associated with PostEEN (Figure 1G). The top eleven zOTUs associated with EEN were found within all communities (Figure 1G, green gradient), including several zOTUs from the genera Enterocloster and Clostridium (formerly Lachnoclostridium; Lachnospiraceae; Table S1), which still showed individual abundances in each patient (Figure S2D). Of the 22 top selected metabolite features, the majority were associated with PostEEN, whereas six metabolites were associated with EEN with higher mean importance (Figure 1H). Of the total of ten metabolites associated with EEN, the majority of those were MCFAs (palmitoleic acid, decanoic acid [DA], caprylic acid, and lauric acid [LA]), which were identified as metabolite features present in Modulen IBD (Figures S2C and S2E).

Medium-chain fatty acids selectively induce translational activation of fecal bacteria

Three MCFAs (LA, DA, and octanoic acid [OA]) were identified as signature metabolites of EEN, likely originating from IBD Modulen. To determine which gut bacteria can be activated by these metabolites, we employed a combination of bioorthogonal noncanonical amino acid tagging (BONCAT) and fluorescence-activated cell sorting (FACS) to identify translationally active bacteria in the presence of MCFAs under ex vivo anaerobic incubation (Figure 2A). We selected three donors (MP1–MP3, model patient [MP]) within the CD-EEN group based on clinical activity, limited drug exposure and chose four different samples along EEN therapy (Table S2, STAR Methods, Figures 4A, 4C, 4I, and S7F). Confocal microscopy visualization and quantification indicated that a significantly higher proportion of gut bacteria became translationally active after exposure to MCFAs compared with the solvent control (Figures 2B, 2C, S3, and S4). The MCFA-active fraction of the community (BONCAT labeling and FACS sorting) exhibited significant differences compared with the unsorted community and between patients, as shown in the MDS ordination (p = 0.001, Figure 2D). Ethanol-mediated variability is shown as additional control (Figure S4C). This separation highlights distinct microbial communities to be present after MCFA exposure. We identified different zOTUs enriched in the active fraction after exposure to MCFAs compared with the ethanol control (Figures 2E and S4). Different species from the Enterocloster genera were significantly enriched in response to MCFAs, especially in MP2 at EEN remission stage (Figure 2E). This includes Enterocloster aldensis and two subspecies of Enterocloster boltae, which aligns with the bacterial signature identified from the clinical cohort (Figure 1G). Additional enriched taxa included Eggerthella lenta in MP2 (remission); Escherichia coli, Clostridium scindens, Dialister invisus, Sporobacter spp., and Eubacterium spp. (MP2 relapse); and E. coli in MP1 and MP3 as well as Lachnospiraceae in MP1. By contrast, several additional zOTUs were depleted in the active fraction after MCFA exposure, including Clostridium scindens, Thomasclavelia ramosa, Hungatella hominis, Peptostreptococcus anaerobius, Ruthenibacterium lactatiformans, and Phocaeicola vulgatus in MP2 at EEN remission stage and Anaerostipes hadrus and Hominenteromicrobium mulieris in MP2 under relapse (Figures 2E and S4). In summary, of the 16 zOTUs associated with EEN from the network analyses, 8 zOTUs responded to MCFA exposure (labeled with * in Figures 1G and S2C).

Figure 2. MCFA-induced activation of translationally active gut bacteria.

(A) Experimental design. MP1, MP2, and MP3 stool samples, previously stored in glycerol, were incubated under anaerobic conditions and in the presence of medium-chain fatty acids (MCFAs) (lauric acid [LA], decanoic acid [DA], and octanoic acid [OA]) and of the cellular activity marker (L-azidohomoalanine [AHA]). Translationally active bacterial cells labeled with alkyne-modified dye were sorted with BONCAT-FACS. The taxonomic profiles of sorted and unsorted cells were identified by 16S rRNA genes with amplicon sequencing.

(B) Representative confocal microscopic images of one stool sample (MP2) stimulated with OA. Pink: active cells (BONCAT-Cy5); blue: all cells (DAPI). Scale bar is 10 μm.

(C) Translationally active bacterial cells quantified in MP1, MP2, and MP3 patients. p values were calculated by Kruskal-Wallis test and Dunn test for multiple comparisons. *p < 0.05, **p < 0.01. ***p < 0.001, ****p < 0.0001. Error bars represent standard deviation of the mean.

(D) Beta-diversity MDS plots show clustering of samples based on model patients and sorted-unsorted groups (perMANOVA p = 0.001 for both comparisons). MP1 = orange, MP2 = green, MP3 = turquoise, sorted cells = square, and unsorted cells = circle.

(E) Relative abundances of significantly enriched and depleted zOTUs in MP1, MP2, and MP3. *zOUTs identified in network analyses Figures 1G and S2. p values were calculated by ANOVA and Tukey’s test for multiple comparisons. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. Boxplot: boxplot medians (center lines), interquartile ranges (box ranges), whisker ranges.

Figure 4. Functional validation of gut-microbiota-dependent EEN efficacy using ex vivo fermentation and gnotobiotic humanized mice.

(A) Experimental scheme: FMT 1, patient sample is transferred into GF mice at the age of 8 weeks for 4 weeks. Same patient sample is also transferred (FMT 2) in an ex vivo gut chemostat model and treated with medium simulating EEN (EEN-like, devoid of fibers) or a regular diet (fiber-rich diet, FRD). Fermented sample is transferred (post ex vivo) into GF mice (age of 8 weeks) for 4 weeks (FMT 3—EEN-like medium conditioned, FMT 4—FR medium conditioned).

(B and C) Transfer of patient sample: patient’s disease state with wPCDAI and fecal calprotectin levels, and therapy response over time with maintenance therapy (MTX, anti-TNF) and sampling time point.

(B) Model patient (MP) 1, sample (red arrow) collected PreEEN in active disease.

(C) MP2, sample (blue arrow) collected in EEN-induced remission.

(D) Mouse histopathology scoring of colon Swiss rolls of FMT 1, 3, and 4 from MP1.

(E) Mouse histopathology scoring of colon Swiss rolls of FMT 1, 3, and 4 from MP2.

(F and G) Mouse transfer experiments from FMT 3 and 4: representative (mean of group) H&E-stained sections of colonic Swiss rolls and corresponding higher magnifications for Il10^−/− mice (scale bars, 500 μM; HS, histopathology score; respective controls in extended data, Figure 3) with Tnf gene expression levels (fold of control) in distal colon.

(F) MP1.

(G) MP2.

(H) Experimental scheme: patient sample from MP3 is transferred into GF mice (FMT 1) at the age of 8 weeks, for 4 weeks. Same patient sample is also transferred in an ex vivo gut chemostat model and treated with EEN-like medium. Fermented sample is transplanted (post ex vivo) into GF mice at the age of 8 weeks, for 4 weeks, getting either chow diet (FMT 3) or EEN-like purified diet (EEN-like) 1 week prior to FMT (FMT 4).

(I) MP3 sample for transfer (EEN): patients’ disease state with wPCDAI and fecal calprotectin, therapy response over time, maintenance therapy (MTX, anti-TNF), and sampling time point of transferred fecal sample (blue arrow); gray arrow indicates endoscopy performed in the patient PostEEN and the picture taken during endoscopy shows active ileitis. J Mouse histopathology scoring of colon Swiss rolls of FMT 1, 3, and 4 from MP3.

(K) Mouse transfer experiments from FMT 3 and 4 from MP3: representative (mean of group) H&E-stained sections of colonic Swiss rolls and corresponding higher magnifications for Il10^−/− mice (scale bars, 500 μM; HS, histopathology score; respective controls in extended data Figure 3) with Tnf gene expression levels (fold of control) in distal colon.

(A and H) Created with BioRender.com. Data are represented by (D)–(G), (J), and (K) mean ± SD of six biological replicates. p values were calculated by Mann-Whitney test (F, G, and K) or two-way ANOVA with Tukey multiple pairwise comparisons test (D, E, and J). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Microbial species, strains, and functions change dramatically at EEN cessation

Considering the various changes of community profiles associated with EEN or PostEEN, we next investigated dynamics within the patients along EEN therapy. Cessation of EEN resulted in an elevated turnover in zOTU composition (Figure 3A), resulting in higher dissimilarity—even after controlling for greater time periods—between pairs of samples spanning the transition off of EEN compared with during or post-EEN alone (p < 10⁻³ for both comparisons by permutation test). Using shotgun meta-genomes for 96 CD-EEN patient samples (n = 19 patients), we applied StrainFacts¹⁸ to resolve taxonomic differences below the level of species. Like zOTUs, and despite the much smaller number of comparisons, we detected the same elevated turnover in the composition of strains during the transition off of EEN relative to the PostEEN period (p < 10⁻³, Figure 3B). Overall, we observed high strain-level diversity within and between CD-EEN patients. For most of the 369 species, individual strains were largely specific to a single patient and appeared in multiple samples from their host, whereas some samples contained multiple distinct strains of a given species (e.g., Figures 3D–3F).

Figure 3. Species and strain turnover during the transition off of EEN.

(A and B) Turnover of taxonomic composition during the transition from EEN to a PostEEN diet of both (A) zOTU and (B) strain compositions. The three lines reflect cubic spline regression of Bray-Curtis dissimilarities against time, averaged across subjects, and offset by an independent intercept for each pair type.

(C) Volcano plots showing COGs with significant changes in normalized abundance at the transition off of EEN (Wilcoxon test at false discovery rate [FDR] < 0.1). Here, 54 COGs were at significantly greater relative depth during PostEEN (orange points) and 529 during EEN (blue).

(D–F) Composition over time of strains within (D) E. bolteae (red box), (E) E. clostridioformis (blue box), and (F) E. coli (black box), with all three abundant in the CD-EEN patients and all three species having notable strain dynamics during and after EEN. Strain genotypes and fractions were estimated from shotgun metagenomic data. Fractions of abundant strains (colored bars) are shown across samples from individual subjects (P01, P02…), before EEN (‘‘pE’’ labels on x axis), during (‘‘EE’’), and after (‘‘PE’’). *Samples used as donor samples in other experiments. Strain fractions for the species sum to 1; minor strains are summed together and shown in gray. An analogous plot for B. uniformis is in Figure S4D.

Besides taxonomic turnover, we also detected 583 protein coding gene families (COGs) with substantial changes in their normalized abundance (54 increased and 529 decreased) at the transition off of EEN (Figure 3C). Families increased during EEN included COG3845 (mean log₂ PostEEN/EEN of −0.58) and COG4603 (−0.58), components of the ABC-type guanosine uptake system. Increases in COG4813 (+2.1), annotated as trehalose utilization protein, and COG3867 (+1.0), arabinogalactan endo-1,4-beta-galactosidase, indicate that microbes encoding these proteins respond to host diet and may play a role in metabolite transformations. These findings are consistent with changes in carbohydrates and sugar acids found in the metabolite analysis (Figure S2C).

Next, we sought to identify species with high prevalence and abundance and with large changes in relative abundance or extensive strain turnover at the end of EEN. Using an indicator score based on these features (Figure S2A; Table S3), we flagged 46 species as highly dynamic. We identified pairs of zOTUs and metagenomic species (Table S4) and matched 37 (80%) of the top scoring species (Figure S5B; Table S3). Here, we highlight and contrast the strain dynamics of two closely related species, Enterocloster bolteae (formerly Lachnoclostridium bolteae) and E. clostridioformis (formerly C. clostridioforme). Both of these species were matched to zOTU5 (Figure S5C) and showed dynamic strain composition over patient time series (Figures 3D and 3E). In some patients, EEN and PostEEN samples contained similar strains of each species, but the dominant strain changed; in others, a new strain appeared PostEEN that was absent or undetectable during EEN. Interestingly, the relative abundance of E. bolteae, but not E. clostridioformis, was significantly affected by EEN cessation (p = 0.018 and 0.9, respectively, by Wilcoxin test; Table S3). However, although both showed a similar strain turnover overall, E. clostridioformis, but not E. bolteae, had significantly elevated strain turnover at the transition from EEN to PostEEN (p = 0.004 and 0.202, respectively, by permutation test: n = 999). This effect was also seen in nine other common species (Table S3), including E. coli (Figure 3F). By contrast, some species like Bacteroides uniformis revealed notably lower strain diversity within samples and across time (Figure S5D).

The distinction in response to EEN among strains and between closely related species reinforces the importance of strain-level characterization in individuals. Altogether, metagenomic analyses demonstrate that gut microbial communities are dynamic at and below the species level and highly influenced by EEN, motivating us to explore their relationship to the clinical efficacy of EEN.

EEN-like diet triggers protective changes in fecal microbiota

To further evaluate the functional impact of EEN on patient-derived fecal microbiota, we used continuous culture systems in a gut chemostat model and fecal microbiota transfer (FMT) into germ-free (GF) wild-type (WT) and Il10^−/− mice. Il10^−/− mice are a microbiota-dependent IBD mouse model responsive to human fecal microbiota, as previously shown for adult CD patients¹⁹ (Figure 4A). The previously used samples from MPs for the BONCAT experiments were also selected for the transfer experiment (Figures 4B, 4C, and 4I; Table S2, STAR Methods).

Microbiota transfer of the PreEEN sample of MP1 (MP1-FMT 1, Figure 4B, red arrow) induced colonic tissue inflammation in all gnotobiotic Il10^−/− mice (histopathology score of 5.3 ± 1.0) but not in WT control mice (Figure 4D p < 0.001, histopathology score of 1.3 ± 0.5; Figure S6A H&E), mirroring the active disease state of this patient. Following transfer of MP1 fecal material into the gut chemostat model (MP1-FMT 2, Figure 4A), dietary exposure was mimicked using combinations of fiber-free EEN-like (including 10% Modulen IBD of the 1 kcal/mL preparation) and fiber-repleted (FR) medium, respectively (Figures S7A and S7B). We then transferred microbiota (MP1-FMT 3) from the chemostat model exposed to EEN-like medium and thereby reversed the inflammatory phenotype of MP1 in Il10^−/− mice (Figure 4D, histopathology score of Il10^−/− mice = 2.5 ± 0.8; WT mice = 1.6 ± 0.5; FMT1 vs. FMT 3 Il10^−/− p < 0.001; Figures 4F H&E and S6A). Conversely, under FR medium conditions, MP1-FMT 4 induced IBD-like tissue pathology in four out of six Il10^−/− mice (Figure 4D, histopathology score = 3.7 ± 1.0; Figures 4F H&E and S6A) compared with disease-free WT mice (Figure 4D histopathology score = 1.4 ± 0.5, p = 0.0087; Figures 4F H&E and S6A). Tnf gene expression confirmed active tissue inflammation with significantly elevated expression levels in Il10^−/− compared with WT mice (Figure 4F p = 0.0022 dist. [distal] colon; Figure S6D p = 0.0152 prox. [proximal] colon). Thus, ex vivo EEN-like intervention recapitulated therapeutic functional changes to the microbiota.

Transfer of the fecal microbiota from MP2 failed to induce inflammation in Il10^−/− mice (MP2-FMT 1, Figures 4E H&E and S6B), resembling the inactive disease state of the patient. Because the stool sample of MP2 was retrieved under EEN therapy, we started the gut chemostat on EEN-like medium (MP2-FMT 2; Figure S7C). In contrast to MP2-FMT 3 (Figure 4E histopathology score in in Il10^−/− = 2.0 ± 0.6; WT = 1.0 ± 0.6; Figures 4G H&E and S6B), transfer of FR-conditioned microbiota (MP2-FMT 4) triggered IBD-like colitis, including histopathologic changes and elevated Tnf tissue expression (Figure 4E, histopathology score in Il10^−/− = 4.8 ± 1.9 vs. WT = 1.5 ± 0.5, p (Il10^−/− FMT1 vs. FMT4) = 0.0004, p (Il10^−/− FMT3 vs. FMT4) = 0.0008; Figure 4G H&E, Tnf gene expression p = 0.0152 dist. colon; Figure S6D p = 0.0022 prox. colon). Thus, both transfer experiments (MP2-FMT 1 and MP2-FMT 3) demonstrate the protective effects of EEN. Reintroduction of fiber (MP2-FMT 4) reverted this protective function, indicative for the risk of relapse in MP2 after EEN cessation (Figure 4C).

Next, we introduced an EEN-like mouse diet 1 week prior to FMT (MP3-FMT 4) to evaluate extended dietary effects in gnotobiotic WT and Il10^−/− mice (Figure 4H). MP3 had elevated fecal calprotectin levels of 2,828 mg/L (Figure 4I, blue arrow) and ongoing mucosal inflammation with endoscopic signs of disease activity (Figure 4I, gray arrow), despite achieving clinical remission under EEN therapy. Accordingly, FMT 1 from MP3 induced severe inflammation in all recipient Il10^−/− mice (Figure 4J histopathology score in Il10^−/− = 7.3 ± 0.8 vs. WT = 1.3 ± 0.5, p = 0.0022; Figure S6C H&E). In contrast to MP1 and MP2 (Figures 4D and 4E), EEN-like medium in the chemostat model did not change the inflammatory phenotype in IBD-susceptible mice (MP3-FMT 3, Figure 4J histopathology score in Il10^−/− = 6.0 ± 1.3 vs. WT = 1.6 ± 1.0, p = 0.0006; Figures 4K H&E and S6C; Tnf p = 0.0022 for dist. colon and prox. colon; Figure S6D). However, introducing EEN-like diet decreased the level of disease severity after MP3-FMT 4 in most Il10^−/− mice (Figure 4J histopathology score in EEN-like-fed Il10^−/− = 3.2 ± 3.1 vs. WT = 1.3 ± 0.5, p = n.s.; Figure 4K H&E, Tnf p = n.s dist. colon; Figures S6C and S6D, Tnf p = 0.0022 for prox. colon) and significantly reduced disease activity compared with the chow-fed group (Figure 3J FMT 3 vs. FMT 4, histopathology score in EEN-like-fed Il10^−/− = 3.2 ± 3.1, p = 0.0271, Il10^−/− FMT 1 vs. FMT 4 p = 0.0005, Figures 4K H&E and S6C). An additional sample of MP2 after clinical relapse was tested under conditions of the extended EEN-like feeding (Figure S7F, orange arrow), and, contrary to MP3, the MP2 relapse sample was resistant to EEN-induced protection (Figures S7E–S7L). These findings suggest that sustained EEN-like feeding increased the chance to prevent recurrence but efficacy of repeated EEN decreases with a second course.

Individual strain signatures in gnotobiotic Il10^−/− mice convey protective function of patient’s microbiota in response to diet

To investigate bacterial strain profiles responsible for the mouse disease phenotypes (Figure 4), we performed metagenomic analyses of all 54 gnotobiotic Il10^−/− mice colonized with fecal communities of MP1–MP3 (Figures 5A–5C, triangles: red—active disease, green—inactive disease; Figure S8A). Out of 100 strains detected in MP1, 64 were specific to mice with active disease, whereas only three were specific to inactive disease (Figure 5A, lowest Venn diagram [Vd]). Strains in MP2 mice were skewed toward being specific to inactive disease or shared (Figure 5B, lowest Vd), consistent with lower disease activity in MP2 and fewer mice with active disease. Strains in MP3 mice were skewed toward being specific to active disease, probably due to a higher number of MP3 mice having active disease (Figure 5C, lowest Vd).

Figure 5. Strains associated with active or inactive disease in Il10^−/− mice.

(A–C) Transfer overview of donor sample (gray) from FMT 1 (black), FMT 3 (EEN-conditioned, blue), and FMT 4 (FR-conditioned, brown) into Il10^−/− mice. Mouse inflammation endpoints are shown (triangles: red = active, green = inactive). Venn diagrams tally strains detected in inflamed (red) and non-inflamed (green) Il10^−/− mice or found in both conditions (yellow) for (A) MP1, (B) MP2, and (C) MP3 for each FMT and combined FMT 1, 3, and 4.

(D–F) LEfSe (linear discriminant analysis effect size) analysis of differentially abundant bacteria at species level based on metagenomic data in the recipient inflamed (active, red) or non-inflamed (inactive, green) Il10^−/− mice with abundance of strains in each condition. Different strains are labeled with different numbers. (D) MP1. (E) MP2. (F) MP3.

(G) Venn diagrams shows shared or individual species (left) or strains (right) detected in MP1, MP2, and MP3. *zOTUs found in network analyses Figures 1G and S2.

Motivated by these differences, we next sought to identify species, strains, and metabolites that distinguished active vs. inactive Il10^−/− mice. Dozens of species showed significant differential abundance after the transfer experiments of MP1, MP2, and MP3 (40, 16, and 11, respectively Figures 5D–5F), including taxa from the network analyses and BONCAT experiments. Considering all microbiota transfer experiments, we detected 181 species, including 24 species shared between MP1–MP3 transfer experiments into Il10^−/− mice (Figure 5G, left Vd). By contrast, none of the identified species shared any common strain (Figure 5G, right Vd), demonstrating that active vs. inactive disease states in Il10^−/− mice are mediated by individually variable strain communities. Linear discriminant analysis effect size (LEfSe) determined 47 species that were differentially abundant between mice with active vs. inactive disease, each with individual strain repertoires in samples from the three different MPs (Figure S9A). In the combined analyses, E. clostridioformis (zOUT5) and Bacteroides thetaiotaomicron (zOUT2 and zOTU66) were most dominant in mice with inactive disease, and Akkermansia muciniphila (zOUT65) and Escherichia coli (zOTU4 and zOUT8) were most dominant in mice with active disease (Figure S9A). In addition, 56 species were significantly different in abundance, comparing mice under chow vs. EEN-like diet (Figure S9B), with previously highlighted strains abundant in both groups, as mice colonized with ex vivo EEN-conditioned microbiota received either chow or EEN-like diet. Importance of strain-level diversity is obvious, as several species were associated with both active and inactive mice in the different MP transfers, but the different strains were responding differently in the model systems (Figures 5D–5G and S9A).

SPLS-DA of untargeted metabolomics from mouse colon content separated active and inactive mice in the respective transfer experiments independent of dietary exposure (Figures S8B–S8D and S9C–S9F). Several bile acids were highlighted as differentially abundant in the untargeted measurement. To confirm the exact structures, a targeted assay was used to quantify the primary taurine conjugated and secondary unconjugated BA (Figures S8B and S8C). Targeted analyses of Isobutyrate showed highest concentration in mouse feces samples colonized with EEN-conditioned microbiota on an EEN-like diet (Figure S8E). SPLS-DA of the combined, untargeted metabolomics dataset confirmed distinct clusters for active and inactive disease independent of MP1–3 donor samples (Figures S9F and S10), suggesting that individual strain profiles convey protective or aggressive responses in the IBD-susceptible host (Figure S11). These results clearly demonstrate that despite the MPs having personalized strain composition (Figure 5G, right Vd), disease activity in gnotobiotic mice receiving FMTs was consistently associated with a number of metabolic and species-level taxonomic differences. Furthermore, the differentially abundant species in all humanized mice exhibited unique patterns linked to EEN and its cessation in the CD patient cohort (Figure S12), emphasizing both the clinical relevance of our mouse FMT experiment findings and the individual microbial fingerprint of each patient.

In summary, we provide evidence that EEN impacts the individual gut microbiome–EEN therapy in the CD-EEN patients and selected fecal samples from three individuals exposed to MCFAs and EEN-like exposure in a non-host and host environment–leading to individual microbiota changes that are functionally relevant and harboring the capacity to induce or not to induce inflammation in mice. Despite the limited selection of donor samples, of the 16 EEN-associated zOUTs from the network analyses, 7 zOTUs were found in significantly different abundant taxa in the mouse experiments (Figures 5D–5F) and 8 zOTUs were identified in the BONCAT experiments (Figures 2E and S4D), suggesting a direct link of MCFAs in EEN to changes in the fecal microbiota of CD patients. The microbiota transfer studies in the gut chemostat model provide proof of concept that microbiota changes are functionally relevant and that diet shapes individual microbiota signatures harboring the capacity to induce or not to induce inflammation in mice. In conclusion, we provide evidence that EEN therapy creates explicit functional changes of temporally and individually variable microbiome profiles in CD patients. These data challenge the concept that protective microbiome signatures are universal to patients and therapeutic interventions; instead, we postulate individually diverse microbiome signatures with shared functional properties under these therapeutic conditions.

DISCUSSION

We followed 78 pediatric IBD patients longitudinally, focusing on 20 newly diagnosed CD patients receiving EEN as first-line treatment. Our results corroborate previous observations that EEN is highly effective at inducing remission. We demonstrate that protective functions of EEN therapy are mediated through diet-induced changes in patient microbiomes despite considerable variations in individual microbiota over time. Dietary exposure of EEN components like MCFAs and EEN-like diets on individual patient-derived fecal microbiota modified the activity and function of selective taxa. EEN-mediated changes of fecal bacteria in the chemostat models convey anti-inflammatory phenotypes after FMT into IBD-relevant GF mice, strongly supporting a causal role of EEN-mediated protective changes in the microbiome. By contrast, FR diets antagonized these protective functions. Notably, and based on the model conditions, we identify patient-specific species and strain signatures in response to EEN therapy, confirming the highly personalized strain-level dynamics in patients and the absence of universal protective signatures.

Longitudinal sampling revealed an integrated network of taxonomic and metabolomic signatures across pediatric CD patients. In line with previous findings,²⁰ EEN therapy was associated with the genus Enterocloster (Lachnospiraceae; former genus Lachnoclostridium). In addition, several Enterocloster species responded to MCFA exposure; E. clostridioformis (zOUT5) was the most abundant species in inactive Il10^−/− mice and showed high dynamics in the CD-EEN cohort together with E. bolteae (zOTU5). Low levels of Lachnoclostridium have been associated with CD,²¹ supporting a protective function of Enterocloster in response to EEN therapy. Nevertheless, taxonomic shifts at the cessation of EEN were highly patient specific. For example, although in the majority of patients, E. bolteae (zOTU5) was at much higher relative abundance during EEN compared with PostEEN, the dominant strains in many patients changed over this transition, and most strains were detected in just one subject. This individual diversity and turnover contribute to challenges in predicting taxonomic responses to dietary changes globally. Antibiotics and pre-existing dysbiosis are both associated with increased strain engraftment in patients receiving fecal microbiota transplants.²² Likewise, the microbiome perturbation caused by EEN—and its cessation—may encourage strain turnover in some species, with potentially complex consequences for microbial function after therapy.

**Beyond taxonomic alterations, we highlighted six metabolites associated with EEN treatment and 16 metabolites associated with PostEEN samples. Independently, metagenomic analysis highlighted 583 bacterial gene functions with substantial changes in response to EEN, suggesting specific metabolic adaptations of the microbiota to host diets. Modulen IBD is characterized by easily digestible carbohydrates and high fat content, including medium-chain triglycerides.⁹ Our integrated network analysis highlighted caprylic (C8:0), decanoic (C10:0), and lauric (C12:0) acids as associated with EEN. MCFAs are part of the EEN formula and mediate anti-bacterial and anti-inflammatory functions.^23–25 The exposure of LA, DA, and OA to selected patient samples directly links nutritional components of EEN to selectively activated taxa in the microbiota of CD patients. Surprisingly, from the limited selection of donor samples, 50% (8 out of 16 zOTUs) of the EEN-associated bacteria significantly responded in the BONCAT experiments. Diets high in fat trigger the release of BAs,²⁶ providing a rationale for C. innocuum and Hungatella hathewayi to be associated with EEN.^27,28 Concordantly, our untargeted metabolomics analysis identified several BAs as differentially abundant, highlighting a potential microbial conversion of primary to secondary BAs in non-inflamed Il10^−/− mice, potentially promoting wound healing and epithelial regeneration.²⁶ In a recent study (FARMM), EEN specifically affected amino acid pathways in the microbiome, creating a metabolite milieu low in bacteria-derived indoles,¹⁷ which was also low in abundance in the EEN-like mouse diet. In addition, EEN-related protective effects in Il10^−/− mice are partially mediated by Isobutyrate production,¹⁶ which was highest in mice receiving EEN-conditioned microbiota on an EEN-like diet, supporting the hypothesis that microbiota and metabolite changes contribute to the EEN therapeutic efficacy.

In healthy humans, dietary fiber intake supports beneficial functions of the gastrointestinal tract.²⁹ In animal models, low fiber diets induce adverse microbiota changes associated with impaired intestinal barrier integrity and bacteria-related mucus degradation,^30,31 highlighting the fiber paradox in understanding EEN-mediated anti-inflammatory mechanisms in IBD. Dietary fiber consists of complex carbohydrates, which are resistant to digestion and absorption by the host. However, dietary fiber degradation and availability for the host depends on the fermenting potential of host-specific microbial communities, which are modified by intestinal inflammation.^29,32–34 Using the combined gut chemostat-to-mouse transfer approach, we showed that a fiber-free EEN-like medium maintains the anti-inflammatory and remitting phenotype associated with EEN therapy. Concordantly, the protective effect of EEN-conditioned microbiota is reversible upon exposure to FR diet, as a proxy for regular diet PreEEN and PostEEN, and potentially resistant to repetitive EEN as shown before.¹² A potentially adverse effect of fiber in IBD was elucidated by Armstrong and colleagues, demonstrating the incapability and loss of function of bacteria to degrade fibers promoting intestinal inflammation.³⁴ Furthermore, several additional studies link the presence of dietary fiber with increased inflammation.^35,36 In addition, fiber-free diets prevent and diminish the colonization of identified pathobionts in different CD-like mouse models,^37,38 supporting our hypothesis that EEN therapy directly creates protective niches in the gut microbiome. Preserving a beneficial microbiome after cessation of EEN may be a new approach to guide mucosal healing and remission in pediatric CD.³⁹ In our IBD cohort, 40% (8/20) of pediatric CD patients relapsed within the first year after EEN treatment, highlighting the necessity to maintain remitting conditions. Our data clearly demonstrate that EEN therapy mediates protective changes in individually variant microbiome profiles, leading toward personalized patient-to-patient species and strain signatures. These findings support the idea of developing patient-tailored maintenance strategies in pediatric CD involving combinations of personalized nutritional and bacterial interventions.

RESOURCE AVAILABILITY

Lead contact

Further information and requests should be directed to and will be fulfilled by the lead contact, Prof. Dr. Dirk Haller (dirk.haller@tum.de).

Materials availability

This study did not generate new, unique reagents.

Data and code availability

• Some data that support the findings of this study will be available on reasonable request from the corresponding author (T.S.). The data are not publicly available due to the pediatric age of the research participants. Mouse untargeted metabolomics data are available at Massive (https://massive.ucsd.edu) under the number MSV000094047. 16S rRNA gene sequencing results of the cohort, mouse, and fermenter experiments will be available at the time of publication upon request from the lead contact.

• Mutliomic data integration of 16S rRNA and metabolomics code is available at https://doi.org/10.5281/zenodo.10808412. Metagenomics analysis code is available at https://doi.org/10.5281/zenodo.13777347.

• Any additional information required to reanalyze the data reported in this work paper is available from the lead contact upon request.

STAR★METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Patient cohort, recruitment and study design

Patients (aged 3–18 years) with suspected or established IBD were recruited to a monocentric pediatric IBD cohort study (ethics approval Ludwig-Maximilians University of Munich, approval No. 17–801 and German Clinical Trials Register Accession No. DRKS00013306, date registration 19.03.2018). All patients and parents/legal guardians provided written informed consent/assent. Patients without confirmed IBD after full diagnostic work-up according to revised Porto criteria were excluded.⁵³ All other patients were followed prospectively for one year and treated according to current ESPGHAN/ECCO guidelines for CD or ulcerative colitis (UC)/IBD unclassified (IBDU).^6,54,55 All patients were recruited consecutively to our study. No additional selection criteria have been applied. Patients are representative in terms of phenotype (Paris classification) and disease activity. Disease activity, as well as endoscopic, histologic, radiologic and laboratory findings were collected prospectively. Disease activity indices included the Physician Global Assessment (PGA) and the weighted pediatric Crohn’s disease activity index (wPCDAI) for CD, as well as the pediatric ulcerative colitis activity index (PUCAI) for UC/IBDU. For EEN-treated CD patients, remission was defined as wPCDAI of <12.5 points and response as reduction in wPCDAI of >17.5 points, but not reaching remission. Relapse was defined as increase of wPCDAI to ≥12.5 points and/or fecal calprotectin of >250 mg/L requiring changes in medical treatment or abdominal surgery or reintroduction of EEN. Patients received detailed instructions for fecal specimen collections and time points, as well as stool collection kits. Each sample was accompanied with a questionnaire on date and time of specimen collection, problems during sampling, Bristol stool scale and recent antibiotic use.

EEN therapy was prescribed for 6–8 weeks, followed by gradual re-introduction of solid food and concomitant start of maintenance treatment. EEN was performed with Modulen^® IBD (Nestlé Health Science) or in case of intolerance (e.g. cow’s milk protein allergy) with Neocate Junior^® (Nutricia, Erlangen, Germany). The amount of prescribed formula was based on the Daily Recommendations Intake (DRI) using the online calculator: https://www.nal.usda.gov/human-nutrition-and-food-safety/dri-calculator. All EEN-treated patients received dietary advice and dietetic support to comply with complete liquid formula feeding.

Animal ethics statement

All animal experiments were conducted at Technical University of Munich and approved by the Committee on Animal Health Care and Use of the State of Upper Bavaria (TVA ROB-55.2–2532.Vet_02–19-190) and performed in strict compliance with the EEC recommendations for the care and use of laboratory animals (European Communities Council Directive of November 24, 1986 (86/609/EEC).

Animals and housing conditions

GF wild-type (WT) and Il10-deficient (Il10^−/−) mice on 129Sv/Ev background were bred and kept at the Core Facility Gnotobiology of the ZIEL Institute for Food & Health, Technical University of Munich, Freising, Germany. Mice were housed as described elsewhere,¹⁹ but using single gnoto-cage units (Isocage P System, Tecniplast, Italy). Cages are ventilated with HEPA-filtered air at 22 ± 1°C with a 12-h light/dark cycle. Female and male mice were assigned randomly to the treatment groups with a ratio of 1:1 and six mice per genotype. Mice received standard chow (V1124–300; Ssniff, Soest, Germany; autoclaved) or an EEN-like purified diet (EEN-like, SNIFF S5745-E902, Figure S13D) and sterile water ad libitum. EEN-like diet was introduced one week prior to fecal microbial transfer (FMT) to allow for adaptation. Mouse intervention (FMT) started at the age of eight weeks, EEN-like diet at the age of 7 weeks. Handling of mice in isocages was performed under biological safety cabinets in sterile conditions.

METHOD DETAILS

Fecal specimen collection

Samples were collected at home or in the hospital. Patients provided samples at baseline (before bowel preparation for endoscopy) and every four weeks in (1) 4 mL DNA stabilizer (magiX PBI Microbiome Preservation Buffer, microBIOMix GmbH, Regensburg),⁵⁶ (2) 8 mL 20% glycerol in reduced phosphate-buffered saline (PBS, 0.05% L-cysteine) and (3) naĸïve without stabilizing solution. Patients on EEN collected stool in (1) weekly for the first 12 weeks. Samples in (1) were shipped at ambient temperature by mail to the study center. Fecal specimens in (2) and (3) were immediately frozen at home (approx. −18 to −20°C) and brought in on wet ice packs at the next visit. Upon receipt, samples were aliquoted, barcoded, linked to clinical data with CentraXX Bio (Kairos GmbH) and stored immediately at −80°C until further analysis.

Selection of donor samples for BONCAT and transfer experiments into germ-free mice

We selected Stool samples for direct exposure of MCFAs and FMT experiments into germ-free (GF) mice to show proof-of-concept examples. Therefore, we selected samples based on sampling timepoint (include various states along EEN therapy: active disease sample from treatment-naive patient before the start of EEN (PreEEN, pE), EEN-induced remission (EEN, EE) and a relapse sample after reintroduction of a regular diet (PostEEN, PE)); clinical activity (pga, calprotectin), drug exposure (shorter time under co-medication preferred) and histopathology score in Il10^−/− mice of FMT1 (direct transfer). For MP1 (model patient 1) the pE was selected as EE and PE samples induced inflammation in the mouse model, so lowered disease activity in the patient seemed to be influenced by Anit-TNF. PreEEN sample form MP2 had very bad quality due to its bloody and diarrhetic consistency, therefore EE and relapse sample were chosen. PE sample in remission was not chosen as a direct impact of EEN therapy was no longer expected. EE sample from MP3 was chosen to investigate a sample where mucosal healing was not fully achieved during EEN therapy. Details are in Table S3.

Preparing gavage for humanizing germ-free mice

Preparation of FMT with human fecal microbiota was performed according to Metwaly et al.¹⁹ For post ex vivo FMT, frozen samples in glycerol (20% end concentration) were thawed on ice. All steps were performed in an anaerobic chamber (90% N₂ and 10% H₂). Ex vivo suspensions (2 mL) were centrifuged at 100×g for 3 min at 4°C to pellet debris. Supernatant was collected and centrifuged at 8000×g for 10 min at 4°C to pellet the bacteria. The bacterial pellet was resuspended in 1.5 mL reduced PBS (0.05% L-cysteine-HCl). The clear supernatant was transferred into an anaerobic crimped tube, which was transferred to the gnotobiotic facility. All mice were gavaged on three consecutive days and sacrificed after four weeks of colonization (12 weeks of age).

Tissue Staining and scoring

Colonic and cecal Swiss-roll tissues were fixed, stained and scored as described before.¹⁹ In brief, tissue was fixed in 4% formal-dehyde/PBS for 48 h at room temperature, subsequently dehydrated (Leica TP1020), and embedded in paraffin (McCormick; Leica EG1150C). Formalin-fixed paraffin-embedded (FFPE) tissue sections were stained with hematoxylin and eosin (H&E) and scored 0 (not inflamed) to 12 (highly inflamed). Inflammation cut off was defined as a histopathology score greater than three based on scoring in WT mice.

Gene expression

Tissue sections from colon were collected in RNAlater (Sigma Aldrich) for 24 h at 4°C and long-term at −80 °C until processing. Total RNA from tissue was isolated by using the NucleoSpin RNA II kit (Macherey-Nagel GmbH) according to manufacturer’s instructions. Complementary DNA was synthetized from 500 ng total RNA by using random hexamers and moloney murine leukemia virus (M-MLV) reverse transcriptase (RT) Point Mutant Synthesis System (Promega). Quantification was performed by using the LightCycler 480 Universal Probe Library System (Roche Primers and probes used were: Gapdh (glyceraldehyde 3-phosphate dehydrogenase) (5’-tccactcatggcaaattcaa-3’; 5’ tttgatgttagtggggtctcg-3’; Probe #9) and Tnf (Tumor necrosis factor) (5’-tgcctatgtct-cagcctcttc 3’; 5’-gaggccatttgggaacttct-3’; Probe #49). Relative induction of gene mRNA expression was calculated based the 2-ΔΔCt method⁵⁷ using the expression of Gapdh for normalization.

Ex vivo gut chemostat model

The ex vivo continuous fermentation was performed in 1.4-L bioreactors (Multifors 2) and monitored with Eve^®software (Infors AG, Basel, Switzerland). Parameters were set as described before.⁵⁸ Briefly, vessels were maintained at 37°C, pH 6.8 with a constant influx of N₂ to maintain an anaerobic atmosphere. Inoculation material was prepared in an anaerobic chamber (90% N₂ and 10% H₂). Approximately 2.5 g of frozen fecal samples was suspended in reduced PBS supplemented with 0.03 g/mL L-cysteine-HCl, filtered (70 mm Cell Strainer, Fisher Scientific) and evenly distributed between two vessels, which led to about 1.0×10⁸ to 8.16×10⁸ CFU/vessel. Two consecutive batch fermentations (24 h in 400 mL, another 24 h with fresh 400 mL medium added) allowed fecal microbial communities to establish. Continuous fermentation included complete medium exchange every 36 h (retention time) with a working volume of 400 mL. After 10 days of continuous fermentation, where stable community formation was achieved, samples for transfer into GF mice were collected and medium was switched afterwards. After 1.5 days of complete medium exchange from medium 1 to medium 2, samples for transfer into mice were collected on day 9 of continuous fermentation in the second medium. Medium composition was either fiber-rich (FR) or EEN-like, mimicking patient conditions at sampling (PreEEN and PostEEN, respectively; Figures S7A–S7E).

FR medium was prepared according to Macfarlane et al.,⁵⁹ but supplemented with additional vitamins (pantothenate 10 mg/L, nicotinamide 5 mg/L, thiamine 4 mg/L, biotin 2 mg/L, vitamin B12 0.5 mg/L, menadione 1 mg/L, and p-aminobenzoic acid 5 mg/L) as described in Gibson and Wang.⁶⁰ EEN-like medium was based on the Macfarlane medium, but with modifications simulating colonic-luminal content during an EEN diet. To determine the carbohydrate:protein ratio of digested Modulen IBD^® in the colon, similar calculations as described previously were conducted.^61,62 Since Modulen IBD^® lacks fibers, these ingredients present in the FR medium (i.e., pectin from citrus, guar gum, xylan from oat spelt, inulin from Dahlia tubers, and arabinogalactan from larch wood) were replaced by soluble rice starch and simple sugars. A total carbohydrate concentration of 13 g/L was chosen in alignment with Cinquin et al..⁶¹ As the formulation of Modulen IBD^® is not publicly available, we decided to include some Modulen IBD^® not to miss crucial nutrients in our model-colon medium. Towards this end, 10% of the 1kcal/mL standard preparation of Modulen IBD^® was added to 1L of EEN-like medium. Finally, the carbohydrate:protein ratio was 75:25 in the model medium. To further mimic Modulen IBD^®, casein was used as main protein source with the exclusion of tryptone and peptone (Figures S13A–S13C).

Samples were aseptically collected daily at the same daytime and stored at −80°C until further analyses. Samples for transfer into the GF mice were collected and mixed with glycerol (20% end concentration) and stored at −80°C until gavage preparation. In order to diminish potential vessel effects in the technical duplicates of each fermenter colonization, samples from both vessels were mixed 50:50 for gavage.

High-throughput 16S-ribosomal RNA (rRNA) gene amplicon sequencing analysis

Total DNA from human stool, colonic mouse content and the continuous ex vivo fermentation was isolated using bead beating and sequenced for 16S rRNA gene amplicons as described.⁴⁰ Downstream analysis was performed in R v4.2.1 using Rhea (https://lagkouvardos.github.io/Rhea/).⁴² zOTU tables were normalized to 10.000 reads, beta-diversity was computed using generalized UniFrac distances.⁶³ Alpha-diversity was assessed on the basis of taxa richness and Shannon effective number of zOTUs.⁶⁴ Trees were visualized and annotated using EvolView (http://www.evolgenius.info/evolview/).⁴³ Differential analyses of bacteria in different groups was done using LDA Effect Size (LEfSe).⁴⁴ zOTUs were re-checked using the 16S-based ID tool of EzBioCloud⁴¹ using the database update 2023.08.23, reflecting all taxa published and accepted by IJSEM until April 2023.

Shotgun Metagenomic Sequencing and Analysis

Metagenomic sequencing was performed for 96 human stool samples and 66 samples from transfer experiments (54 mouse, 12 ex vivo) using the genomic DNA already isolated for 16S rRNA gene amplicon sequencing. DNA was randomly sheared into short fragments. The obtained fragments were end repaired, A-tailed, and further ligated with Illumina adapters. The fragments with adapters were size-selected, PCR amplified, and purified. The library was checked with Qubit and real-time PCR for quantification and BioAnalyzer for size distribution detection as required by Illumina (TruSeq^® DNA Library Prep Kit). Quantified libraries were pooled and sequenced on a NovaSeq platform (Illumina) in PE150 mode. Raw data were aimed at 15 Gb per demultiplexed sample. From raw shotgun metagenomic sequencing data, adapter sequences were removed using Trimmomatic v0.39⁶⁵ without considering any mismatch and parameters LEADING:3, TRAILING:3, and SLIDINGWINDOW:4:15. We removed human reads by mapping against human reference genome hg38 by using Bowtie2 v2.4.5⁶⁶ with the default parameters. Cleaned fastq data files were used in the following steps.

Strain tracking and strain gene content analyses were performed based on the StrainPGC workflow.⁴⁸ Briefly, preprocessed meta-genomes were metagenotyped using GT-Pro v1.0.1.⁴⁷ Across all samples, 369 species - as defined in the Unified Human Gut Genome (UHGG)⁴⁹ - were robustly detected and only these were included in downstream pangenome profiling. Single nucleotide polymorphism (SNP) profiles were filtered, subsampled, and strain relative abundances were estimated with StrainFacts v0.4.0.¹⁸ Within individual species, estimated strain fractions add up to 100%. By contrast, other analyses considered the overall strain composition: the relative abundance of each strain across all species.

Gene profiling was performed against a subset of the MIDASDB-UHGG pangenome database v.15⁶⁷ dereplicated at a 99% ANI threshold performed and mapping reads using Bowtie2 v2.4.5.⁵⁰ Species abundances were estimated based on the average depth of core genes. Global functional gene family profiling was performed by aggregating gene depths into Clusters of Orthologous Genes (COGs)⁴⁵ across all species and then normalizing based on 25 ubiquitous, single-copy genes to estimate ‘‘gene copies per genome’’.

Ex vivo anaerobic incubations with medium chain fatty acids (MCFAs)

Fecal samples (MP1 PreEEN, MP2 EEN, MP2 relapse, and MP3 EEN) previously collected in 20% glycerol in reduced phosphate-buffered saline (PBS, 0.05% L-cysteine) were thawed and introduced into an anaerobic chamber (10% H₂, 90% N₂). All reagents and materials were introduced in the anaerobic chamber 48 hours before sample processing to ensure anaerobic state by the start of the experiment.

Stool samples (1–1,5 g) were homogenized with 1X PBS (1:10 dilution) and filtered using a 40 μm size filter (Corning, Germany) to remove large particles, which was followed by three washing steps in 1X sterile PBS to remove residual of glycerol and leftover nutrients in the stool. Afterwards, samples were added to sterile falcon tubes containing a final concentration of 50μM of the cellular activity marker L-azidohomoalanine (AHA), (Baseclick GmbH, Germany) and 10μM of MCFAs (lauric acid, decanoid acid and octanoic acid) dissolved in ethanol with a final concentration of 0.5%. 0.5% ethanol was used as negative control and 2 mg/mL of glucose as positive control for each experiment. Samples were incubated for 24 hours under anaerobic conditions. After 24 hours of incubation, part of the samples were frozen for DNA extraction and other aliquots were washed twice in PBS and then fixed in 1:1 ethanol:PBS for further analysis with fluorescence activated cell sorting (FACS). Both ethanol fixed and frozen samples were collected in triplicates.^68,69

Bioorthogonal non-canonical amino acid tagging (BONCAT) and images analysis of translationally active bacteria

To identify utilization capability of MCFAs by the gut microbiota, we employed BONCAT, a fluorescence-based single cell labeling technique. BONCAT involves incorporation of the non-canonical amino acid AHA to replace L-methionine during protein synthesis, followed by fluorescent labeling of AHA-containing proteins using azide-alkyne click chemistry⁷⁰ (Figure 2A). Cu(I)-catalyzed click chemistry reaction was performed on microscopy slides as previously described⁷⁰ and counter-stained with 4′, 6-diamidino-2-phenylindole (DAPI).

20 pictures for each sample were collected using a confocal microscope (Olympus, fluoview, FV10i, Germany) and processed using Fiji.⁵² Translationally active cells were quantified using the software digital image analysis in microbial ecology (Daime).⁵¹ Translationally active cells were measured as biovolume fraction comparing the Cy5-labeled cells (BONCAT positive) to the total biomass stained with DAPI.

BONCAT-FACS and DNA extraction

A combination of BONCAT and FACS was used to determine which bacteria are capable of MCFAs utilization. Briefly, ethanol fixed samples were washed in 1X sterile PBS, resuspended in 96% ethanol and centrifuged again. Afterwards, the BONCAT dye solution was added to the samples and incubated for 30 min in the dark at room temperature.⁷⁰ After incubation, samples were washed three time in 1X sterile PBS and filtered with a 35 mm nylon mesh using BD tubes 12×75 mm (Corning, Germany) right before sorting. Cy5-positive bacteria were sorted, collected into sterile tubes and store at−80°C until DNA extraction. A representation of the gating strategy is shown in Figure S1.

Bacterial DNA from sorted and unsorted samples were extracted using the QIAmp DNA mini kit (Qiagen, Germany) following the protocols for bacteria according to the manufacturer’s instructions with an additional lysozyme treatment.⁶⁹ Every sample was sorted in triplicates. FACS data were further analyzed with FlowJo^™ v10.10.0 Software (BD Life Sciences; Reference: FlowJo^™ Software (for Windows) Version v10.10.0. Ashland, OR: Becton, Dickinson and Company; 2023).

BONCAT data analysis

Statistical analysis was performed using R statistical software (https://www.r-project.org/). 16S rRNA amplicon sequencing analysis was carried out with Rhea, beta diversity was assessed using UniFrac distances as previously described^42,63 and the statistical significance of factors was evaluated using permutational multivariate analysis of variance (perMANOVA). Differences in relative abundance between sorted and unsorted fractions and between sorted MCFAs and ethanol control, were analyzed using ANOVA and Tukey’s test for multiple comparisons. For quantifications of the percentage of active cells, Kruskal-Wallis and Dunn’s test were applied. Significantly enriched and depleted zOTUs sequences were identified using the 16S-based ID tool of EzBioCloud.⁴¹

Metabolomics

Sample preparation / Metabolites extraction

Sample preparation of mouse colon content:

Approximately 20 mg of mouse colon content was weighed in a 2 mL bead beater tube (FastPrep-Tubes, Matrix D, MP Biomedicals Germany GmbH, Eschwege, Germany). Next, 1 mL of methanol-based dehydrocholic acid extraction solvent (c = 1.3 μmol/L) was added as an internal standard for work-up losses. The samples were extracted with a bead beater FastPep-24TM 5G (MP Biomedicals Germany GmbH, Eschwege, Germany) supplied with a CoolPrepTM (MP Biomedicals Germany, cooled with dry ice) 3-times each for 20 s of beating at a speed of 6 m/s and followed by a break of 30 s each.

Sample preparation of fermenter or human samples:

Either 200 mg of the fermenter samples or 100 mg of human fecal samples was weighed in a 15 mL bead beater tube (FastPrep-Tubes, Matrix D, MP Biomedicals Germany GmbH, Eschwege, Germany). Next, 5 mL of methanol-based dehydrocholic acid extraction solvent (c = 1.3 μmol/L) was added as an internal standard for work-up losses. The samples were extracted with the same FastPrep protocol described above.

Targeted bile acid measurement:

A standard of 20 μL of isotopically labeled bile acids (ca. 7 μM each) were added to 100 μL of sample extract. Targeted bile acid measurement was performed using a QTRAP 5500 triple quadrupole mass spectrometer (Sciex, Darmstadt, Germany) coupled to an ExionLC AD (Sciex, Darmstadt, Germany) ultrahigh performance liquid chromatography system. A multiple reaction monitoring (MRM) method was used for the detection and quantification of the bile acids according to Reiter et al.⁷¹ Data acquisition and instrumental control were performed with Analyst v1.7 software (Sciex, Darmstadt, Germany).

Untargeted mass spectrometric measurement:

The untargeted analysis was performed using a Nexera UHPLC system (Shimadzu, Duisburg, Germany) coupled to a Q-TOF mass spectrometer (TripleTOF 6600, AB Sciex, Darmstadt, Germany). Separation of the fecal samples was performed either using a UPLC BEH Amide 2.1 × 100 mm, 1.7 μm analytic column (Waters, Eschborn, Germany) with a 400 μL/min flow rate or with a Kinetex XB18 2.1 × 100 mm, 1.7 μm (Phenomenex, Aschaffenburg, Germany) with a 300 μL/min flow rate. For the HILIC-separation the settings were as follows: The mobile phase was 5 mM ammonium acetate in water (eluent A) and 5 mM ammonium acetate in acetonitrile/water (95/5, v/v) (eluent B). The gradient profile was 100% B from 0 to 1.5 min, 60% B at 8 min and 20% B at 10 min to 11.5 min and 100% B at 12 to 15 min. For the reversed-phase separation, eluent A was 0.1% formic acid and eluent B was 0.1% formic acid in acetonitrile. The gradient profile started with 0.2% B which was held for 0.5 min. Afterwards, the concentration of eluent B was increased to 100% until 10 min which was held for 3.25 min. Subsequently, the column was equilibrated at starting conditions. A volume of 5 μL per sample was injected. The autosampler was cooled to 10 °C and the column oven heated to 40 °C. Every tenth run, a quality control (QC) sample, which was pooled from all samples, was injected. The samples were measured in a randomized order and in the Information Dependent Acquisition (IDA) mode. MS settings in the positive mode were as follows: Gas 1 55, Gas 2 65, Curtain gas 35, Temperature 500 °C, Ion Spray Voltage 5500, declustering potential 80. The mass range of the TOF MS and MS/MS scans were 50–2000 m/z and the collision energy was ramped from 15–55 V. MS settings in the negative mode were as follows: Gas 1 55, Gas 2 65, Cur 35, Temperature 500 °C, Ion Spray Voltage −4500, declustering potential −80. The mass range of the TOF MS and MS/MS scans were 50–2000 m/z and the collision energy was ramped from −15–55 V.

Data processing

The “msconvert” from ProteoWizard⁷² was used to convert raw files to mzXML (de-noised by centroid peaks). The bioconductor/R package xcms⁷³ was used for data processing and feature identification. More specifically, the matchedFilter algorithm was used to identify peaks (full width at half-maximum set to 7.5 s). Then the peaks were grouped into features using the “peak density” method.⁷³ The area under the peak was integrated to represent the abundance of features. The retention time was adjusted based on the peak groups presented in most samples. To annotate features with names of metabolites, the exact mass and MS2 fragmentation pattern of the measured features were compared to the records in HMBD⁷⁴ and the public MS/MS spectra in MSDIAL,⁷⁵ referred to as MS1 and MS2 annotation, respectively. Missing values were imputed with half of the limit of detection (LOD) methods, i.e., for every feature, the missing values were replaced with half of the minimal measured value of that feature in all measurements. To confirm a MS2 spectra is well annotated, we manually reviewed our MS2 fragmentation pattern and compared it with records in the public database, previously measured reference standards and SIRIUS⁷⁶ to evaluate the correctness of the annotation. Significances of well-annotated metabolic features were computed with analysis of variance (ANOVA) per donor or Wilcoxon signed-rank tests and FDR-corrected using the Benjamin-Hochberg method for three donors. Subsequent hierarchical clustering using only differentially abundant metabolites (q < 0.05) was conducted using complete linkage based on Euclidean distances. Calculations, as well as heatmaps, were computed in R v4.2.1. For each principal component analysis (PCA), we included negative and positive features from RP and HILIC with at least 50% non-missing values per condition (inflamed/non-inflamed). The remaining missing values were imputed as described above. The prcomp function from R v4.3.2 computed the principal components after centering and scaling each feature to zero mean and unit variance. Sparse Partial Least Squares Discriminant Analysis (sPLS-DA) was performed on centered and scaled features using the mixOmics bioconductor/R package v6.26.0^36,77. Mean balanced error rate, was used to select the number of components and features in M-fold cross-validation where M corresponds to the minimum number of samples per condition. Figure S11 shows raw values for features with the highest Variable Importance in the Projection (VIP), i.e., features contributing the most to explaining the conditions.

Data integration of 16s rRNA amplicon data with metabolomics and metagenomics

We matched 60 samples from 15 patients of the 16S-rRNA gene sequencing to 60 samples of the untargeted metabolomics analysis according to patient ID and same time points. The aim was to select features that jointly discriminate the 23 EEN samples (clinical phenotype mild and remission) and 37 PostEEN samples (>2 weeks after EEN, clinical remission). For the integrated selection of metabolome and microbiota features, sPLS-DA (mixOmics v6.20) was used^36,77. For feature selection concerning metabolomics, only annotated metabolites were considered. As described above, input metabolites were mean-centered and scaled to unit variance. zOTUs (zero-radius operational taxonomic units) were centered log-ratio transformed. To estimate the degree of overfitting, the model was trained on 70% of the data, with a k-fold cross-validation for feature selection during training. The remaining 30% of the data were used to evaluate the generalization power. The model performance was measured with an Area Under the Receiver Operating Characteristic (ROC-AUC) on the samples in the test set. Subsequently, Spearman’s rank correlation coefficients between all selected metabolites and zOTUs were computed and their respective p-values were corrected with the Benjamini-Hochberg procedure.⁷⁸ All coefficients with an FDR >0.05 were set to 0. Using all non-zero correlations, a network was generated and the Louvain method for community detection⁷⁹ was applied to extract highly connected subgraphs from the correlation network. For each community, a sPLS-DA model was trained with the same train-test scheme explained previously for feature selection. Using the mean loadings scores over all k iterations, the most relevant metabolites and zOTUs were identified. More precisely, for metabolites and zOTUs separately, the absolute mean loadings were plotted in ascending order and all features on the right of the knee point (20 zOTUs and 22 metabolites) were selected. Analyses were run in R v4.2.1 and python v3.8.11. For network analysis the igraph v1.3.4⁸⁰ and networkx v3.1⁸¹ packages were used.

Profiles of 16S rRNA amplicon relative abundance were further integrated with metagenomic species profiles in order to identify those which represent the same, underlying taxa. This was done by cross-decomposition of zOTU relative abundance and species relative abundances estimated from the metagenomic workflow, using the PLSSVD model implemented in SciKit-Learn,⁵⁰ multiplying the matrices of weights for each data-type provides a matrix of coefficients mapping zOTUs to metagenomic species. We scaled these coefficients by the square-root of the mean relative abundance of that zOTU and species, respectively, producing a matrix of matching scores for each comparison, while accounting for all other taxa. Based on this score, we ranked zOTUs for each species and species for each zOTU. We selected as putative matches all zOTU-species pairs that were either reciprocal best hits or a combination of one best hit and one second-best hit, and where at least one of the genus or family classifications matched. In this way, we accounted for the possibility of multiple zOTUs representing a species or multiple species being represented by a zOTU.

QUANTIFICATION AND STATISTICAL ANALYSIS

MDS and PCA latent dimensions in Figures 1 and S1 were tested for significance using linear mixed models of the form

$$\mathrm{a}\mathrm{x}\mathrm{i}\mathrm{s}\sim \mathrm{i}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}\mathrm{c}\mathrm{e}\mathrm{p}\mathrm{t}+\mathrm{d}\mathrm{i}\mathrm{e}\mathrm{t}+\mathrm{d}\mathrm{i}\mathrm{s}\mathrm{e}\mathrm{a}\mathrm{s}\mathrm{e}+\mathrm{d}\mathrm{i}\mathrm{e}\mathrm{t}\mathrm{*}\mathrm{d}\mathrm{i}\mathrm{s}\mathrm{e}\mathrm{a}\mathrm{s}\mathrm{e}+\left(\frac{1}{\text{patient}}\right)$$

where diet, disease and their interaction are included as fixed effects and patients are included as random effects. All coefficients are tested with a t-test for Figure 1 and an ANOVA for Figure S1. Reported p-values are Bonferroni-corrected. To estimate the degrees of freedom, we used Satterthwaite’s method as implemented in the lmerTest package (version 3.1–3).

Further statistical analyses and plots were generated using GraphPad Prism v8.0 (GraphPad, La Jolla, CA) using Mann-Whitney test, ANOVA followed by pairwise comparison testing (Tukey, Bonferroni correction), by Kruskal-Wallis test followed by Dunn’s multiple comparisons. Data is presented as mean ± SD. P-values below 0.05 were considered significant (p < 0.05, *; p < 0.01, **; p < 0.001, ***; and p < 0.0001, ****).

Taxonomic turnover in both zOTU and strain compositions were calculated based on the Bray-Curtis dissimilarity. Pairs where one sample was collected during EEN and the other during PostEEN were classified as ‘‘transition’’. All within-subject pairwise comparisons were computed and fit using cubic spline regression – implemented in the statsmodels python

library v1.3.2⁴⁶ – of the time between sample collections, with a regression term for each subject, as well as for each pair class: EEN, transition, or PostEEN. The null hypothesis of no difference between pair classes was tested by permutation: 999 random permutations of samples within subjects. Changes in mean, normalized, COG depths during EEN and PostEEN time points were tested using the Wilcoxon signed-rank test across subjects. False discovery rates were calculated using the Benjamini-Hochberg procedure. Mean fold change for each COG was calculated as the ratio between these two estimates.

Statistical details of each experiment can be found in the respective figure legends.

ADDITIONAL RESOURCES

Clinical Trial Registry

The study is registered at the German Clinical Trials Registry under the Accession No. DRKS00013306, date of registration 19.03.2018.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Biological samples
Pediatric fecal samples	This study	N/A
Chemicals, peptides, and recombinant proteins
RNAlater®	Sigma	Cat#MFCD03453003
DNA stabilizer - magiX PBI Microbiome	microBIOMix GmbH	N/A
Preservation Buffer
Modulen IBD®	Nestle Health Science	N/A
Sodium chloride	Sigma	Cat#31434-M
Potassium chloride	Sigma	Cat#P9333
Magnesium sulfate anhydrous	Sigma	Cat#746452
Calcium chloride anhydrous	Sigma	Cat#C1016
Sodium hydrogen carbonate	Sigma	Cat#S6014
Potassium dihydrogen phosphate	Sigma	Cat#P5655
Iron-(II)-sulfate heptahydrate	Fisher	Cat#201392500
Bile salts (no 3)	Fluka	Cat#48305
Tryptone	Sigma	Cat#T9410
Peptone	Sigma	Cat#91249
Yeast Extract	Sigma	Cat#Y1625
Starch (from wheat)	Sigma	Cat#S5127
Pectin (from citrus)	Alfa Aesar	Cat#J61021
Guar gum	Sigma	Cat#G4129
Xylan (from beech wood)	Roth	Cat#4414.4
Inulin (from dahlia tubers)	Sigma	Cat#I3754
Arabinogalactan (from larch wood)	SCBT	Cat#sc-210833
Anti-foam solution B	Sigma	Cat#A5757
Tween 80	Sigma	Cat#P8074
Porcine gastric mucin (type II)	Sigma	Cat#M2378
Casein	Fisher	Cat#A13707
L-Cysteine HCl	Fisher Acros	Cat#111781000
Pantothenate	Applichem	Cat#A2088
Nicotinamide	VWR	Cat#A15970
Thiamine	Applichem	Cat#A2088
Biotin	Fisher	Cat#23009
Vitamin B12	Sigma	Cat#V2876
Menadione	Fisher	Cat#127180250
4-Aminobenoic acid (PABA)	Sigma	Cat#A9878
Hemin	Sigma	Cat#51280
Phosphoric acid, 85.5 %	Fisher	Cat#11478443
Sodium hydroxide pellets	Sigma	Cat#S5881
Brain-Heart Infusion	Sigma	Cat#53286
WCA broth	Th.Geyer	Cat#CM0643B
Agar agar	FisherSci	Cat#10572775
Phenosafranin	Sigma	Cat#199648
L-Cysteine	Roth	Cat#3467.x
DL-Dithiothreitol (DTT)	Sigma	Cat#43819
Phosphate Buffered Saline	Invitrogen	Cat#18912-014
Glycerol, 100 %	Fisher	Cat#BP229-4
Sodium hydroxide solution, 1 M, 10 L	Fisher	Cat#J762024
NH4OH, ≥ 25 %	Sigma	Cat#30501-M
KCl, 3 M, 1 L	VWR	Cat#83605.290
Starch from rice	Sigma	Cat#9005-25-8
Sucrose	Sigma	Cat#57-50-1
Fructose	Sigma	Cat#57-48-7
Glucose	Sigma	Cat#50-99-7
4-azido-L-homoalanine hydrochloride	baseclick	Cat#BCAA-005-10
THPTA ligand (BONCAT dye solution)	baseclick	Cat#BCMI-006
Aminoguanidine hydrochloride (BONCAT dye solution)	ThermoFisher Scientific	Cat#368910250
L-Ascorbic acid sodium salt (BONCAT dye solution)	ThermoFisher Scientific	Cat#352681000
Sulfo-Cy5-Alkyne (BONCAT dye solution)	Jena Bioscience	Cat#CLK-TA116-1
Copper (II) sulfate pentahydrate (BONCAT dye solution)	Sigma	Cat#C8027
Lauric acid	ThermoFisher Scientific	Cat#16728
Decanoic acid	ThermoFisher Scientific	Cat#16727
Octanoic acid	Sigma	Cat#C2875
4′, 6-diamidino-2-phenylindole (DAPI)	Sigma	Cat#D9542
Critical commercial assays
NucleoSpin® RNAII Kit	Macherey-Nagel	Cat#740955.250
NucleoSpin® gDNA columns	Macherey-Nagel	Cat#740230
M-MLV Reverse Transcriptase	Promega	Cat#M1705; Cat#M5313
QIAamp DNA Mini Kit	Qiagen	Cat#51306
Deposited data
untargeted metabolomics data (mouse, fermenter)	BayBioMS	Massive MSV000094047
Experimental models: Organisms/strains
Mouse: Il10−/− (129S6/SvEv) (GF)	National Gnotobiotic Rodent Resource Center (NGRRC), Chapel Hill	https://www.med.unc.edu/ngrrc/products-services/
Mouse: 129S6/SvEv (GF)	National Gnotobiotic Rodent Resource Center (NGRRC), Chapel Hill	https://www.med.unc.edu/ngrrc/products-services/
Oligonucleotides
16S rRNA gene Illumina sequencing primers	Reitmeier et al.40	341F-ovh and 785r-ovh
Tnf Forward	Sigma	TGCCTATGTCTCAGCCTCTTC
Tnf Reverse	Sigma	GAGGCCATTTGGGAACTTCT
UPL probe #49	Roche	N/A
Gapdh Forward	Sigma	TCCACTCATGGCAAATTCAA
Gapdh Reverse	Sigma	GAGGCCATTTGGGAACTTCT
UPL probe #9	Roche	N/A
Software and algorithms
CentraXX Bio	Kairos GmbH	N/A
EzBioCloud database	Yoon et al.41	https://www.ezbiocloud.net/
IMNGS	Lagkouvardos et al.42	https://www.imngs.org/
Rhea	Lagkouvardos et al.42	https://github.com/Lagkouvardos/Rhea
EvolView	Evolview43	https://www.evolgenius.info/
GraphPad Prism, v.9.4.1	GraphPad Software	https://www.graphpad.com/; RRID: SCR_002798
Light Cycler® 480 Software	Roche Diagnostics	N/A
ViewPoint Light software	PreciPoint	https://precipoint.com/en/products/viewpoint-viewer-software
RStudio	RStudio	https://posit.co/products/open-source/rstudio/; RRID: SCR_000432
Metagenomics analysis code	This study	https://doi.org/10.5281/zenodo.13777347
Multiomic data integration of 16S rRNA and metabolomics code	This study	https://doi.org/10.5281/zenodo.10808412
LEfSe	Segataet al.44	https://huttenhower.sph.harvard.edu/lefse/
Eve® software	Infors AG	2023 H1 Bioprocess software Version 1.125
EggNOG DB	Galperin et al.45	https://doi.org/10.1093/nar/gkaa1018
Statsmodels v1.3.2	Seabold and Perktold46	https://doi.org/10.25080/majora-92bf1922-011
GT-Pro v1.0.1	Shi et al.47	https://doi.org/10.1093/nar/gkaa1018
MIDAS v3 and MIDASDB-UHGG	Smith et al.48	https://doi.org/10.1101/2024.04.10.588779
StrainPGC v0.1.0	Smith et al.48	https://doi.org/10.1101/2024.04.10.588779
StrainFacts v.0.4.0	Smith et al.18	https://doi.org/10.3389/fbinf.2022.867386
UHGG v2.0.2	Almeida et al.49	https://doi.org/10.1038/s41587-020-0603-3
Scikit-learn v1.3.2	Pedregosa et al.50	https://jmlr.csail.mit.edu/papers/v12/pedregosa11a.html
Digital image analysis in microbial ecology (Daime)	Daims et al.51	https://doi.org/10.111VJ.1462-2920.2005.00880.X.
FlowJo™ v10.10.0	FlowJo	FlowJo™ Software (for Windows) Version v10.10.0. Ashland, OR: Becton, Dickinson and Company; 2023 https://www.flowjo.com/solutions/flowjo
Fiji	Schindelin et al.52	https://doi.org/10.1038/nmeth.2019
Other
Empty fecal collection tubes	Susse Labortechnik	Cat#H8555T
Bioreactor	Infors AG	Multifors 2
Gnoto-cages	Tecniplast	Isocage P System
Mouse diet: Chow	SNIFF	V1124-300
Mouse diet: EEN-like	SNIFF	S5745-E902
Anaerobic chamber	Don Whitley Scientific Limited	Whitley A35 Workstation
Falcon 5ml round bottom polystyrene	Corning	Cat#352235
test tube with cell strainer snap cap

Highlights.

• EEN induces remission in patients with pediatric Crohn’s disease

• EEN creates temporally and individually variable microbiome profiles

• Medium-chain fatty acids link EEN to changes in the microbiota of CD patients

• Protective microbiota functions are validated in chemostat and gnotobiotic models