Fecal microbial and metabolic signatures in children with very early onset inflammatory bowel disease Authors: Kine Eide Kvitne, Simone Zuffa, Vincent Charron-Lamoureux, Ipsita Mohanty, Abubaker Patan, Helena Mannochio-Russo, Jasmine Zemlin, Lindsey A Burnett, Lisa S Zhang, Mia C Cecala, Ceylan Ersoz, James A Connelly, Natasha Halasa, Maribeth Nicholson, Pieter C Dorrestein, Shirley M Tsunoda, Janet Markle Published: 2025 ABSTRACT AbstractVery early onset inflammatory bowel disease (VEO-IBD) is a clinically distinct form of IBD manifesting in children before the age of six years. Disease in these children is especially severe and often refractory to treatment. While previous studies have investigated changes in the fecal microbiome and metabolome in adult and pediatric IBD, insights in VEO-IBD remain limited. This multi-omics analysis reveals changes in the fecal microbiome and metabolome in children diagnosed with VEO-IBD compared with age- and sexmatched healthy controls. Untargeted metabolomics analysis identified a depletion of short-chain N-acyl lipids and an enrichment of dipeptides, tripeptides, and oxo bile acids in children with VEO-IBD. Differential abundance analysis of 16S rRNA sequencing data showed lower abundance of beneficial bacteria such as Bifidobacterium and Blautia, and higher abundance of Lachnospira, Veillonella, and Bacteroides in VEO-IBD. Multi-omics integration revealed associations between the altered gut microbiome composition and metabolic dysregulation, specifically for the N-acyl lipids. This study offers unique insight into fecal microbial and metabolic signatures in VEO-IBD, paving the way for a better understanding of disease patterns and thereby more effective treatment strategies.Subject terms: Diseases, Gastroenterology, Microbiology Introduction Inflammatory bowel disease (IBD) comprises a group of chronic conditions characterized by inflammation of the gastrointestinal (GI) tract, with ulcerative colitis (UC) and Crohn’s disease (CD) as the two main subtypes1. Very early onset inflammatory bowel disease (VEO-IBD) is a form of IBD manifesting in children younger than six years of age2,3. Compared to IBD diagnosed at later ages, VEO-IBD is generally associated with a more severe disease course4,5, delayed time to accurate diagnosis6,7, and lower initial response rate to conventional treatments8. The prolonged exposure to chronic inflammation is of significant concern for key developmental processes such as growth, bone health, and immune function in children with VEO-IBD9. Additionally, monogenic etiologies, mutations in a single gene, play a larger role in the pathogenesis of VEO-IBD, especially in infantile onset cases (<2 years)7,10. The incidence of VEO-IBD is rapidly increasing worldwide11,12, with the majority of children with VEO-IBD (70–80%) not having an identified genetic etiology13. This underscores the need to better understand its multifactorial etiology, driven by complex interactions between host, microbial, and environmental factors14,15. The gut microbiome, a key modulator of many physiological processes, including host digestion and vitamin synthesis16, drug metabolism17, immune function18, and metabolic homeostasis19, is increasingly recognized for its important role in the pathophysiology of IBD. Its role may be particularly relevant in children with VEO-IBD, as the gut microbiome is immature, less diverse, and undergoes rapid changes during the first years of life20. It was recently demonstrated that there is a lower abundance of beneficial bacteria belonging to Bifidobacterium, Collinsella, and Akkermansia, and a higher abundance of potential pathogens such as Escherichia, Ruminococcus, Clostridium, and Veillonella in children with VEO-IBD compared to age-matched healthy controls21. Alterations in the gut microbial profiles can disrupt the production and modulation of microbial metabolites that act as signaling molecules, thereby influencing key processes such as immune function and homeostasis. Indeed, the levels of several metabolites belonging to the classes of bile acids (BAs)22,23, short-chain fatty acids (SCFA)24,25, sphingolipids26, and amino acids (AA)25,26 have been shown to be altered in IBD. To the best of our knowledge, no multi-omics study has been previously conducted in VEO-IBD, limiting our understanding of how the gut microbiome and metabolome may contribute to this disease. In this study, we therefore combined untargeted metabolomics and 16S rRNA gene amplicon sequencing to investigate characteristics of the fecal metabolome and microbiome in children with VEO-IBD compared with healthy age- and sex-matched controls. We demonstrate an altered gut microbiome composition in VEO-IBD, combined with altered levels of important signaling molecules belonging to the classes of N-acyl lipids, small peptides, and BAs. We also show, through a joint analysis, a link between the changes in microbiome composition and metabolic alterations in VEO-IBD. These findings will help advance our understanding of the distinct biology of VEO-IBD and support the development of targeted diagnostic and therapeutic strategies. Results Study population characteristics The study recruited 28 children initially diagnosed with isolated VEO-IBD and 28 healthy age- and sex-matched controls. Two individuals in the VEO-IBD group were subsequently excluded due to their clinical phenotype, i.e., disease evolved to systemic autoimmunity and/or autoinflammation, leaving 26 samples from children with VEO-IBD for subsequent analyses. One stool sample per subject was collected and used for 16S rRNA gene amplicon sequencing, to profile the fecal microbial communities, and untargeted liquid chromatography coupled with tandem mass spectrometry (LC-MS/MS) to profile both known and unknown metabolites (Fig. 1a). Mean age ± SD of study participants were 6.7 ± 3.4 and 6.4 ± 3.5 years in the VEO-IBD group and control group (Wilcoxon rank-sum test, p = 0.71), respectively, with 54% being male in both groups (Chi-squared test: p = 1). Demographic characteristics of the study participants are listed in Table 1. Briefly, most patients with VEO-IBD received pharmacological treatment, including tumor necrosis factor-α (TNF)-α inhibitors (54%), aminosalicylates (38%), antibiotics (12%), or a combination of these therapies; and 4 subjects were diagnosed with monogenic VEO-IBD. Mean time from VEO-IBD diagnosis to sample collection was 3.3 ± 3.2 years, with 46% of samples being collected within the first year. Additional details of this patient cohort are available in Supplementary Table 1. N-acyl lipids, small peptides, oxo bile acids, and aminosalicylate drug metabolites separate children with VEO-IBD from healthy controls. a Overview of the study design. b PCA of fecal metabolic profiles showed separation based on cohort (PERMANOVA: R2 = 0.039, F-statistic = 1.9, p = 0.001) and age (PERMANOVA: R2 = 0.031, F-statistic = 1.5, p = 0.016). c A PLS-DA model was constructed to extract features separating groups (VEO-IBD and healthy controls). Boxplot highlights the natural log ratio of differential features from the PLS-DA model (numerator = healthy controls, denominator = VEO-IBD). Statistical significance assessed by Wilcoxon rank-sum test. d Volcano plot of N-acyl lipids, with the dotted lines showing the significance threshold (p adjusted (Benjamini-Hochberg (BH)) < 0.05 and absolute log2FC > 2). e Boxplot of the relative abundance of dipeptides and tripeptides with a VIP score > 1 from the PLS-DA model, respectively. Relative abundance was calculated by dividing the sum of peak areas for relevant dipeptides and tripeptides by the total ion current. Statistical significance assessed by Wilcoxon rank-sum test. f Boxplot of the natural log ratio of BAs with a VIP score > 1 from the PLS-DA model (numerator = VEO-IBD, denominator = healthy controls). Statistical significance assessed by Wilcoxon rank-sum test. g Boxplot of putative 3α-7α-12-oxo BA (Cliff’s delta, δ = –0.51 [95% CI: –0.74, –0.19]) and 3α-7-oxo-12-oxo BA (Cliff’s delta, δ = –0.43 [95% CI: –0.68, –0.09]). Statistical significance assessed by Wilcoxon rank-sum test with BH corrected p values. h Barplot of other differential features between children with VEO-IBD and healthy controls (VIP score > 1). i Molecular network of 5-ASA, the major metabolite of sulfasalazine and balsalazide, and its drug metabolites. Orange nodes represent features with a VIP > 2, with larger nodes indicating higher VIP scores. All boxplots show the first (lower), median, and third (upper) quartiles, with whiskers 1.5 times the interquartile range. Annotations are based on MS/MS spectral matches with the GNPS spectral libraries (level 2–3). ‡Candidate dihydroxylated BA (∆ 135.033). †MS/MS spectral match to 2-acetamidophenol, a stereoisomer of acetaminophen, which is indistinguishable from acetaminophen via mass spectrometry. Table 1Characteristics of the study populationCharacteristicsChildren with VEO-IBD (n = 26)Healthy Controls (n = 28)Age at enrollment (years)6.7 (1–14)6.4 (1–14)Sex (female/male)12/1413/15Monogenic IBD, n (%)4 (15%)NAPharmacological treatment, n (%)^NA - Aminosalicylate (sulfasalazine, mesalamine, balsalazide)10 (38%) - TNF-α inhibitor (infliximab, adalimumab)14 (54%) - Antibiotics (amoxicillin, azithromycin, sulfamethoxazole-trimetoprim) 3 (12%)Numbers are given as mean (absolute range) or count (%).NA not applicable, TNF tumor necrosis factor.^Medication use was not available for one patient in the VEO-IBD group. Characteristics of the study population Numbers are given as mean (absolute range) or count (%). NA not applicable, TNF tumor necrosis factor. ^Medication use was not available for one patient in the VEO-IBD group. Children with VEO-IBD display distinct metabolic profiles Using untargeted LC-MS/MS metabolomics, we observed a significant separation in the fecal metabolome between the children with VEO-IBD and the healthy controls (Fig. 1b) (PERMANOVA: R2 = 0.039, F-statistic = 1.9, p = 0.001). Age also had a significant impact on the fecal metabolic profiles (PERMANOVA: R2 = 0.031, F-statistic = 1.5, p = 0.016). To identify discriminant metabolites between the two groups, a supervised Partial Least Squares - Discriminant Analysis (PLS-DA) model was generated (Supplementary. Fig. 1a, classification error rate (CER) = 0.21) and 1,678 known and unknown metabolic features with variable importance in projection (VIP) > 1 were extracted (Supplementary. Data. 1). The natural log ratio of the extracted features significantly separated the controls from the VEO-IBD group (Wilcoxon rank-sum test, p = 9.0e-08; Cliff’s delta, δ = 0.83 [95% CI: 0.60, 0.93]) (Fig. 1c). One group of metabolites that stood out as key discriminant features between children with VEO-IBD and healthy controls were the N-acyl lipids, signaling molecules containing an amine group (head) and a fatty acid (tail), linked by an amide bond27 (Fig. 1d). Several putatively annotated short-chain (C2-C6) N-acyl lipids with methionine (Met), phenylalanine (Phe), leucine (Leu)/isoleucine (Ile), and tyrosine (Tyr) as head groups were significantly depleted in children with VEO-IBD (Fig. 1d, Supplementary. Fig. 1b). In contrast, we observed a higher relative abundance of dipeptides (Wilcoxon rank-sum test, p = 0.018; Cliff’s delta, δ = –0.40 [95% CI: –0.66, –0.06]) and tripeptides (Wilcoxon rank-sum test, p = 0.0046; Cliff’s delta, δ = –0.47 [95% CI: –0.71, –0.14]) in children with VEO-IBD than in healthy controls (Fig. 1e). Putatively annotated dipeptides that were enriched in children with VEO-IBD included asparagine (Asn)-Phe, glutamic acid (Glu)-Phe, Phe-Met, Ile-Leu, Trp-Phe, Phe-Leu, Tyr-Phe, valine (Val)-Ile, threonine (Thr)-Phe, Met-Leu, Ile-Met, aspartic acid (Asp)-Phe, and Ala-Phe (Supplementary. Data 1). Annotated tripeptides included putative Ile-Proline(Pro)-Ile and Thr-Val-Leu. Given the limited number of annotated tripeptides in our dataset, we also included unknown features predicted to be tripeptides by SIRIUS (probability > 0.7) in the analysis. A total of 43 tripeptides with a VIP > 1 were identified, and a group comparison of the relative abundance supported higher levels of tripeptides in children with VEO-IBD (Wilcoxon rank-sum test, p = 0.00011; Cliff’s delta, δ = –0.63 [95% CI: –0.83, –0.29]; Supplementary. Fig. 1c). Several putatively annotated BAs were discriminant features (Wilcoxon rank-sum test, p = 0.0035; Cliff’s delta, δ = –0.49 [95% CI: –0.71, –0.17]) (Fig. 1f). Notably, children with VEO-IBD displayed a higher abundance of oxo BAs compared with healthy controls, specifically putative 3α-7α-12-oxo and 3α-7-oxo-12-oxo (Fig. 1g). Among the other annotated oxo BAs, there was a trend towards a higher abundance of three different 3α-7α-12-oxo isomers and 3α-7-oxo-12-oxo in VEO-IBD, while 3α-12-oxo did not differ between the groups (Supplementary. Fig. 1d). Microbial AA conjugated BAs such as putative Glu-cholic acid (CA) was also enriched in VEO-IBD (Fig. 1h). We did not observe any difference in putative host-derived CA, microbially derived deoxycholic acid (DCA), or taurine or glycine conjugated BAs (Supplementary. Fig. 1e). Additional discriminant metabolic features enriched in VEO-IBD included spectral matches to drug metabolites of the aminosalicylates, acetaminophen/2-acetamidophenol (stereoisomers and thereby indistinguishable via mass spectrometry), a candidate dihydroxylated BA (refers to a BA with uncharacterized modification)28, and 5’-methylthioadenosine. On the other hand, putative N-acetyl-Met, N-acetyl-Ile/Leu, Arg-Phe, Val-Trp, and stercobilin were enriched in healthy controls (Fig. 1h). It should be noted that we observed poor alignment (38%) between acetaminophen detection and reported use, possibly due to limited MS/MS fragmentation affecting annotation or inaccuracies in self-reported use of acetaminophen. Around 40% of the children with VEO-IBD reported use of an aminosalicylate. Using the GNPS drug library29, a resource encompassing MS/MS data of drugs and their metabolites/analogs, combined with a molecular network of the aminosalicylates, we were able to annotate N-acetyl 5-ASA and N-propionyl 5-ASA as key discriminant features between the VEO-IBD and healthy control groups (Fig. 1i). These gut microbially derived metabolites30 were detected in all patients reported to receive treatment with an aminosalicylate. In contrast, 5-aminosalicylate (5-ASA) eluted in the dead volume, sulfasalazine was not a discriminant feature (Wilcoxon rank-sum test: p = 0.25), and no spectral matches to balsalazide were observed, highlighting the importance of considering drug metabolites in addition to parent compounds. To account for potential confounding effects of medication use on the fecal metabolome, we included medications previously shown to alter the microbiome and/or metabolome31–37 as covariates in the PERMANOVA model (“dist_metabolites ~ Cohort + Age + Aminosalicylate + ABX + TNF_alpha_i”). Medications were grouped as aminosalicylates (yes/no), antibiotics (yes/no), and TNF-α inhibitors (yes/no). Use of aminosalicylate showed a trend towards significance (PERMANOVA: F = 1.3, p = 0.069), while no statistically significant effect was observed for the other categories (Supplementary Table 2). Using the GNPS drug library, we also identified and annotated sulfasalazine, mesalamine, azithromycin, trimethoprim, and their corresponding metabolites. Exclusion of these features from the down-stream analysis did not alter group separation (Supplementary. Fig. 1f), suggesting that the distinct fecal metabolic profiles observed in children with VEO-IBD in our study are mainly driven by disease-associated mechanisms rather than medication use. Several other unannotated metabolic features were also driving the separation between children with VEO-IBD and healthy controls. To predict these metabolites’ chemical classes, their MS/MS spectra were processed with SIRIUS. Features with a VIP > 2.5 were further investigated. Of the metabolites enriched in VEO-IBD, many were predicted to be dipeptides and tripeptides, while depleted metabolites included piperidine, purine, and simple indole alkaloids (Supplementary. Fig. 2a). To investigate their possible microbial origin, we also searched the MS/MS spectra via microbeMASST (see methods). The output showed that out of the 20 most discriminant features (top 10 in each group), 13 have been previously observed in microbial monocultures. Gut microbiome composition is altered in children with VEO-IBD versus healthy controls We next looked at the microbial alpha diversity by measuring the Shannon Diversity Index and found no statistically significant difference between children with VEO-IBD and healthy controls (Wilcoxon rank-sum test, p = 0.48) (Fig. 2a). However, when investigating age, we observed a moderate positive correlation between age and the Shannon Diversity Index in the VEO-IBD group (Pearson correlation, r = 0.41, p = 0.041), but not in the healthy control group (p = 0.33) (Fig. 2b). There was also a significant dissimilarity in beta diversity based on PCA of the robust center log ratio (rclr) transformed data (PERMANOVA: R2 = 0.087, F-statistic = 4.5, p < 0.001) (Fig. 2c). A PLS-DA model was generated to identify discriminant taxa at genus level (Supplementary. Fig. 3a, CER = 0.11). We identified 32 genera with VIP > 1; 17 that were enriched in the healthy control group and 15 that were enriched in the VEO-IBD group (Supplementary. Data. 2). Natural log ratios of the selected taxa confirmed a significant difference in the fecal microbiome composition between the two groups (Wilcoxon rank-sum test, p = 3.1e-09) (Fig. 2d). Key discriminant genera, with a higher abundance in VEO-IBD, included Lachnospira, Sutterella, Clostridium, Bacteroides, Alistipes, and unknown genera of the Enterobacteriaceae family (Fig. 2e, Supplementary. Fig. 3b). On the other hand, genera including Coprococcus, Bifidobacterium, Collinsella, Blautia, and Dorea were depleted in VEO-IBD. To further investigate differences in fecal microbiome composition, differential abundance analysis was also conducted using ALDEx238. The output showed similar findings as from the PLS-DA model, with higher abundance of Veillonella, Sutterella, Lachnospira, Clostridium, and Bacteroides and lower abundance of Coprococcus, Collinsella, and Turibacter in VEO-IBD compared with healthy controls (Fig. 2f). Fecal microbiome composition varies considerably between children with VEO-IBD and healthy controls. a Boxplot of alpha diversity measured as Shannon’s Diversity Index (Wilcoxon rank-sum test, p = 0.48). b Scatter plot showing the correlation between age and Shannon’s Diversity Index (Pearson correlation, VEO-IBD group: r = 0.41, p = 0.041; healthy control group: r = 0.21, p = 0.33). c PCA of rclr transformed data showing a significant difference in beta diversity between the two cohorts (PERMANOVA: R2 = 0.087, F-statistic = 4.5, p < 0.001) with no effect of age. d Boxplot of the natural log ratio of differential genera extracted from the PLS-DA model (numerator = healthy control, denominator = children with VEO-IBD). Statistical significance assessed by Wilcoxon rank-sum test. e Boxplot of selected differential genera obtained from the PLS-DA model. Statistical significance assessed by Wilcoxon rank-sum test with BH-adjusted p-values. f Balloon plot of differential abundance taxa collapsed at genus level using ALDEx2. Only taxa with BH adjusted p value < 0.05 are reported. Taxa names reported as Family_Genera. All boxplots show the first (lower), median, and third (upper) quartiles, with whiskers 1.5 times the interquartile range. Joint analysis identifies link between N-acyl lipid dysregulation and the gut microbiome To investigate if any of the metabolic changes we detected could be associated with changes in the composition of the gut microbiome, a multi-omics analysis was performed using DIABLO39. The output of the first component (correlation cut-off > 0.5) showed a correlation between the putative short-chain N-acyl lipids Met-C3:0, Met-C4:0, Phe-C3:0, Phe-C4:0, and Ile/Leu-C4:0, and the genera Coprococcus, Lachnospira, Collinsella, Bifidobacterium, Blautia, Dorea, Alistipes, and Bacteroides (Fig. 3a). More specifically, N-acyl lipids were positively correlated with Coprococcus, Collinsella, Bifidobacterium, Blautia, and Dorea, genera more abundant in the healthy controls, and negatively correlated with Lachnospira, Bacteroides, and Alistipes, enriched in children with VEO-IBD. Similarly, N-acetyl-Ile/Leu and N-acetyl-Met, both more abundant in healthy controls, were positively correlated with Coprococcus, Collinsella, and Blautia and inversely correlated with Lachnospira and Alistipes. Additionally, we observed correlations between specific genera and unannotated metabolic features (Fig. 3a). Constructed sub-molecular networks did not contain any annotated features. SIRIUS predictions indicated that these unknown metabolic features belonged to the pathways or classes of alkaloids (m/z 243.1128, m/z 197.1284, and m/z 188.128), dipeptides (m/z 395.1811), and tripeptides (m/z 346.1975) (Fig. 3b). Next, we searched them via microbeMASST and found that three of them have been previously observed in microbial monocultures. Of particular interest was the unknown feature with m/z 346.1975 enriched in VEO-IBD, which SIRIUS-predicted to be a tripeptide, matching to genera such as Bacteroides and Akkermansia (Supplementary. Fig. 3c). We also searched this feature (m/z of 346.1975) via tissueMASST to investigate its presence in other publicly available metabolomics IBD datasets. The output showed that this metabolic feature is detected with higher frequency in feces of individuals with IBD compared with healthy (Chi-squared test: χ2 = 149, p = 2.2e–16), and also in other chronic inflammatory diseases such as rheumatoid arthritis compared with healthy (Chi-squared test: χ2 = 29.7, p = 5.1e–08) (Fig. 3b). Multi-omics analysis of the microbiome and metabolome data reveals significant correlation between specific microbial genera, N-acyl lipids, and unknown metabolic features. a Integration of the metabolomics and microbiome data using DIABLO. The correlation cut-off was set to > 0.5. Red lines show positive correlation, blue lines show negative correlation. b Sub-molecular networks of unknown metabolic features highlighted by DIABLO. MS/MS spectra of the unknown metabolites were processed using SIRIUS to predict their chemical pathways or classes. The MS/MS spectra of one of the unknown features enriched in VEO-IBD (m/z 346.1975), predicted to be a tripeptide, was searched across publicly available metabolomics datasets using tissueMASST. The output showed a significantly higher detection rate of this feature in fecal samples from individuals with IBD compared to healthy (Chi-squared test: p < 2.2e–16). Pie charts show the proportion of spectral matches found in deposited datasets, with blue representing a match and yellow representing a non-match. c Other key discriminant features belonging to the classes of di- and tripeptides were therefore also searched via tissueMASST. Barplots showing a higher detection frequency in human fecal samples from individuals with IBD than healthy individuals. Statistical significance assessed with Chi-squared test. The Y-axis shows the percentage of samples where the feature was detected vs not detected. d To investigate if they are likely of microbial origin, these features were also searched against microbeMASST. The output for Thr-Val-Leu is illustrated, but they all matched to microbial monocultures. Pie charts show the proportion of spectral matches found in deposited datasets, with blue representing a match and yellow representing a non-match. Significance: ****p < 0.0001. Metabolomic annotations are based on MS/MS spectral matches with the GNPS spectral libraries (level 2–3). This observation motivated us to explore the detection frequency of other discriminant di- and tripeptides from the metabolomics PLS-DA model in other publicly available metabolomics IBD datasets. The MS/MS spectra of Glu-Phe, Asn-Phe, Thr-Val-Leu, Ile-Pro-Ile, and an unknown feature SIRIUS-predicted to be a tripeptide (m/z 360.2129), were therefore searched via tissueMASST. We focused on the output from datasets comprising human stool samples. The output confirmed a higher detection rate of all these features in human fecal samples from individuals with IBD compared with healthy individuals (Chi-squared test: all p ≤ 8.98e–04) (Fig. 3c). We also searched the MS/MS spectra of these features via microbeMASST. The output indicated that they might be produced by bacteria. The spectra of these metabolites were most frequently observed in cultures of Bacteroides, Veillonella, and Akkermansia, although they matched to several bacterial cultures, as illustrated for Thr-Val-Leu (Fig. 3d). Discussion A more thorough understanding of gut microbiome composition and function in IBD may set the stage for more effective and personalized treatments. For most patients with IBD, the peak age at onset is between 20–40 years40. However, a small but growing percentage of IBD patients are diagnosed with VEO-IBD at age 6 or younger, and disease in these children is especially severe and often treatment-refractory3. The timely selection of an individualized treatment plan is critical for these children, since VEO-IBD is associated with poor long-term outcomes, including surgical resection, increased risk of colorectal cancer, and even death7. The gut microbiome represents a promising therapeutic target in IBD, and significant research efforts have been dedicated to investigate changes in the gut microbiome and metabolome in pediatric26,41,42 and adult IBD43,44. However, studies specifically looking at VEO-IBD remain scarce21, limiting our understanding of whether gut microbial and metabolic changes may contribute to the distinct clinical phenotype of VEO-IBD. Given that the gut microbiome and immune system evolve rapidly in early childhood20, VEO-IBD may involve unique perturbations that differ, not only from healthy peers, but also from those seen in IBD diagnosed at later ages. Untargeted LC-MS/MS metabolomics allows the identification of thousands of known and unknown metabolites in a given sample. When combined with 16S rRNA gene amplicon sequencing, which characterizes the gut microbiome composition, it can provide valuable insights into the intricate relationship between host, microbes, and metabolites. To our knowledge, this is the first multi-omics analysis investigating changes in the fecal microbiome and metabolome in children diagnosed with VEO-IBD compared with healthy age- and sex-matched controls. N-acyl lipids, an understudied class of molecules, appear to play an important role in several biological functions, such as immune regulation27,45. In this study, short-chain N-acyl lipids were among the top discriminant features between VEO-IBD and healthy participants. Several putatively annotated N-acyl lipids had a lower abundance in children with VEO-IBD compared with healthy controls. This has, to our knowledge, not been previously described in the literature. Whether this represents a distinct characteristic of VEO-IBD or is a shared trait across IBD subtypes is not yet known. Our study also demonstrated correlation between the abundance of these short-chain N-acyl lipids and the abundance of specific bacterial genera in a multi-omics analysis of the fecal metabolome and the microbiome. Putatively annotated Met-C3:0, Met-C4:0, Phe-C3:0, Phe-C4:0, and Ile/Leu-C4:0 were negatively correlated with bacterial genera enriched in children with VEO-IBD, and positively correlated with genera more abundant in healthy controls. This pattern suggests that the dysregulation of N-acyl lipids in VEO-IBD may be linked to underlying gut microbiome dysbiosis. We recently demonstrated that many of these N-acyl lipids are microbially produced when both the amine group and lipid chain are present27. We speculate that a lower abundance of specific SCFA-producing bacteria in VEO-IBD, such as Coprococcus, Blautia, Dorea, and Bifidobacterium46, may alter SCFAs availability, thereby limiting their ability to form N-acyl lipids with an amine group. Recent studies have demonstrated a correlation between higher dipeptide abundance and increased disease activity in UC47. The elevated dipeptide levels have been linked to an overabundance of microbial proteases, particularly serine proteases and dipeptidases produced by Bacteroides, specifically B. vulgatus47. Consistent with this finding, we observed a higher relative abundance of dipeptides and tripeptides alongside a higher abundance of Bacteroides in children with VEO-IBD compared with healthy controls. Furthermore, a microbeMASST search confirmed a higher detection frequency of several di- and tripeptides in cultures of Bacteroides and Veillonella, both of which were enriched in VEO-IBD. These findings suggest that the elevated levels of small peptides may be secondary to increased proteolytic activity associated with the altered gut microbiome in VEO-IBD. While it remains unknown whether these peptides directly contribute to VEO-IBD pathophysiology or are simply byproducts of the altered gut microbiome composition, there is evidence pointing to a potential role in disease. For instance, protease inhibitors have been shown to prevent colitis in B. vulgatus monocolonized, IL10-deficient mice47, suggesting that microbial protease activity could be a viable therapeutic target in IBD. Additionally, a tissueMASST search found that key discriminant di- and tripeptides were more frequently detected in fecal samples from individuals with IBD than from healthy subjects. This suggests the possibility that certain di- or tripeptides may serve as biomarkers for IBD or disease activity and/or severity. Previous studies have shown dysregulation of BA metabolism in IBD22,23. BAs are cholesterol-derived signaling molecules synthesized in the liver that activate receptors such as farnesoid X receptor (FXR) and pregnane X receptor (PXR)48. After being conjugated with taurine and glycine, BAs are secreted into the small intestine via the biliary system, where they undergo extensive modifications via microbial metabolism48. Consistent with previous IBD literature23,49, we observed a shift in the bile acid pool in children with VEO-IBD, characterized by a higher abundance of microbial conjugated BAs such as putative Glu-CA. Bile salt hydrolase (BSH), an enzyme widely expressed across most bacterial phyla50, plays a key role in the modification of BAs, including AA conjugation of BA51,52. Notably, Glu-CA has been shown to be more resistant to BSH-mediated hydrolysis than other microbial conjugated BAs in an in vitro model with 17 gut bacterial isolates exposed to AA conjugated BAs53. We also observed elevated levels of some, but not all, 12-oxo BAs in VEO-IBD. These oxo BAs, which contain a ketone in the carbon 12 position in the steroid ring, have been implicated in activating the pyrin inflammasome and promoting secretion of pro-inflammatory cytokines, such as interleukin (IL)-18 and IL-1β54. This suggests a potential role for the oxo BAs in immune regulation and disease pathophysiology in VEO-IBD. Interestingly, unlike previous reports in pediatric IBD41,55, we did not observe differences in levels of host-derived CA or microbially derived DCA. These findings suggest that BA metabolism in VEO-IBD may follow a distinct pattern compared to IBD diagnosed later in childhood or in adulthood. In line with previous findings in pediatric IBD21,56, we observed significant differences in gut microbiome composition between children with VEO-IBD compared with age- and sex-matched controls. This was evidenced by a significant difference in microbial community dissimilarity between the groups. Specifically, we found a significantly lower abundance of beneficial commensal gut bacteria including Bifidobacterium, Blautia, and Coprococcus, microorganisms associated with host health57,58. Supporting this, Conrad et al. recently reported increased microbial dissimilarity in VEO-IBD compared with IBD diagnosed in older children (> 6 years)21, highlighting that VEO-IBD may involve distinct perturbations that differ from IBD in older children. Consistent with their findings, we also observed a higher abundance of OTUs of the genus Veillonella and Clostridium. Interestingly, alpha diversity did not differ between VEO-IBD and healthy controls, which contrasts earlier reports in VEO-IBD21 and pediatric IBD59. However, within the VEO-IBD group, we found a positive correlation between age and microbial diversity, a pattern not observed in healthy controls. This suggests that microbial diversification in VEO-IBD may be delayed, potentially due to early life dysbiosis combined with prolonged exposure to chronic intestinal inflammation and/or medications prescribed to control the symptoms of IBD, disrupting normal microbiome development. Ecological theory suggests that the microbiota is not solely composed of microbes and their genes, but also includes the environment60–62. Here, ecological causes of dysbiosis can include loss of colonization resistance, and changes in oxygen availability. Mouse studies have shown that the neonatal microbiome is unable to effectively provide colonization resistance, i.e., a healthy microbiome’s ability to prevent the invasion and overgrowth of pathogens63, due to low abundance of bacteria from the class Clostridia64. This finding dovetails with the ‘oxygen hypothesis’ of microbiome ecology. At homeostasis, SCFA, including butyrate are abundantly produced by Clostridia such as Lachnospiraceae and Ruminococcaceae65,66. Butyrate regulates mitochondrial oxidative phosphorylation in colonocytes67, and this bacterial-host crosstalk maintains the colonic surface in a state of physiological hypoxia68,69. Without adequate butyrate, colonic epithelial cells switch to a more glycolytic metabolic program68, and increased oxygen availability in the lumen permits the growth of bacteria that can thrive on aerobic respiration (facultative anaerobes such as Salmonella70). This vulnerability may be pronounced in early life, since relative to adults, infants’ intestinal microbiota are less diverse and contain large numbers of facultative anaerobes63. Overall, dysbiosis and chronic inflammation in VEO-IBD may create an altered ecological niche that favors the proliferation of specific opportunistic and inflammation-tolerant bacteria. We observed a lower abundance of putative N-acetyl-Met alongside a higher abundance of 5’-methylthioadenosine (MTA) in VEO-IBD, indicating a potential disruption in the methionine cycle. Additionally, microbially derived metabolites of aminosalicylate drugs30 were among the key discriminant features of the VEO-IBD group. Our multi-omics analysis also identified several unknown metabolic features that strongly correlated with microbial genera either enriched or depleted in VEO-IBD. Notably, using tissueMASST, we found that one of these metabolites, enriched in VEO-IBD, was also more frequently detected in the fecal samples from individuals with IBD compared to healthy subjects when searching its spectra against publicly available metabolomics datasets within the GNPS/MassIVE ecosystem. To explore the microbial origin of this feature, we searched microbeMASST and identified matches across multiple bacterial phyla, including Bacteroides, that were more abundant in VEO-IBD. Based on SIRIUS predictions and manual inspection of the MS/MS spectra, this metabolite is likely a modified tripeptide. However, further investigation is needed to confirm its structure and understand its putative role in the pathophysiology of VEO-IBD. We note several possible limitations of this study. First, this study was cross-sectional, with only one sample collected per participant, and included a relatively small sample size. The findings should therefore be validated in larger, independent cohorts. In addition, samples were collected after initiation of treatment. Although the post hoc analysis suggested only a minor impact of pharmacologic treatment on the fecal metabolome, potential confounding effects of medications cannot be entirely ruled out. However, it should be noted that recruiting patients with VEO-IBD, especially that are treatment naive, remains exceedingly challenging. Second, factors that can influence the gut microbiome composition and functionality, such as probiotics intake and dietary modifications were not reported in detail. Finally, metabolite annotations are based on MS/MS spectral matches with the GNPS spectral libraries, representing a level 2-3 annotation according to the Metabolomics Standard Initiative (MSI)71. Generally, MS/MS data cannot differentiate between stereoisomers, such as acetaminophen and 2-acetamidophenol. The poor alignment between acetaminophen detection and reported use (38%) may be explained by the parameters used for the FBMN job, which required at least 5 matching fragments for annotation, or inaccuracies in self-reported use of acetaminophen. In summary, this integrative multi-omics analysis revealed a strong connection between gut microbial and metabolic alterations in VEO-IBD. Importantly, several of the identified molecules have not been previously reported as altered in IBD diagnosed in older children or adults, and may therefore represent features distinct to VEO-IBD. Future studies directly comparing children with VEO-IBD with pediatric and adult IBD will be important to confirm these findings. Targeted manipulation of gut microbiota composition and function via pre/probiotics, fecal microbiota transplantation, or other modulators may provide an opportunity to target processes that contribute to VEO-IBD development and/or severity. Methods Participants Potential participants included children initially diagnosed with isolated VEO-IBD and followed at Monroe Carell Jr. Children’s Hospital at Vanderbilt in the Pediatric Gastroenterology, Hepatology, and Nutrition clinic as well as healthy children, previously enrolled in a study at Vanderbilt University Medical Center (VUMC) focusing on pediatric vaccine-preventable diseases. The studies were approved by the Institutional Review Board at VUMC (studies #170067, #202017, #200412, and #111296) and performed in accordance with the ethical standards of the institution and with the Declaration of Helsinki. Written informed consent was obtained from all parents/legal guardians and assent of pediatric participants prior to any study procedures. The healthy controls were not immunocompromised and were free of any parental-reported vomiting or diarrhea for at least 14 days prior to providing a stool specimen. Exact age at the time of sample collection was not available for the healthy controls, however each control was matched to a child with VEO-IBD of similar age (±1 year). Fecal specimen collection Fecal material from each child with VEO-IBD was collected at home by the participants’ parents following a single bowel movement. Parents received oral and written instructions on how to collect fecal material into a sterile plastic container, which was then stored in a home freezer until transport to VUMC. For healthy controls, stool samples were collected either at VUMC during the visit, or if not possible, a courier was sent to the home to collect the sample. Each sample was transported to VUMC within 24 hours of collection, aliquoted into 1.5 mL cryovials, and stored at –80 °C until processing. Untargeted metabolomics sample preparation 50% methanol (MeOH:H2O) was used to extract metabolites from the fecal samples. Briefly, 800 µL of cold 50% MeOH was added to each sample, before they were homogenized at 25 Hz for 5 min on a tissue lyzer, incubated for 30 min at 4 °C, and centrifuged at max speed (15,000 x g) for 10 min at 4 °C. Supernatant (440 µL) was then collected, dried overnight in a CentriVap, and stored at -80 °C for later acquisition. Prior to instrumental analysis, the samples were reconstituted in 200 µL 50% methanol with 1 μM sulfadimethoxine as the internal standard. Untargeted metabolomics data acquisition Samples (5 μL) were injected into a Vanquish ultra-high-performance liquid chromatography (UHPLC) system coupled to a QExactive quadrupole orbitrap (Thermo Scientific) mass spectrometer. The chromatographic separation was done on a polar C18 column (Kinetex Polar C18, 100 mm × 2.1 mm, 2.6 µm particle size, 100 A pore size; Phenomenex) with a matching Guard cartridge (2.1 mm) at 40 °C column temperature. The mobile phase consisted of solvent A H2O + 0.1% formic acid (FA), and solvent B acetonitrile (ACN) + 0.1% FA. Samples were eluted at a flow rate of 0.5 mL/min using the following gradient: 0–0.5 min, 5% B, 0.5–1.1 min, 5–25% B, 1.1–7.5 min, 25-60% B, 7.5–8.5 min, 60-99% B, 8.5–9.5 min, 99% B, 9.5–10 min, 99-5% B, 10.0–10.5 min 5% B, 10.5–10.75, 5-99% B, 10.75–11.25 min, 99% B, 11.25–11.5 min, 99–5% B, 11.5–12 min, 5% B. All solvents used were LC-MS grade. Data dependent acquisition (DDA) of tandem mass spectrometry (MS/MS) spectra was performed in positive mode with the following parameters: sheath gas flow 53 L/min, aux gas flow rate 14 L/min, sweep gas flow 3 L/min, spray voltage 3.5 kV, inlet capillary 269 °C, aux gas heater 438 °C, and 50 V S-lens level. The MS scan range was set to 150–1000 m/z with a resolution of 35,000 at m/z 200; maximum ion injection time, 100 ms; automatic gain control (AGC) target, 5.0E4. MS/MS spectra were collected with a resolution of 17,500 and an AGC target of 5E5 with a maximum injection time of 150 ms, and fragmented the top 5 most abundant ions per cycle with a 1 m/z isolation window, isolation offset set to 0 m/z, stepped collision energies of 25, 40, 60, and a dynamic exclusion window of 10 s. Untargeted metabolomics data processing The acquired raw files were converted into mzML format using MSConvert72. Feature detection and extraction were performed using MZmine 4.273. For mass detection, MS1 and MS2 noise levels were set to 5.0E4 and 1.0E3, respectively. Parameters for the chromatogram builder were set to minimum 5 consecutive scans, 1.0E5 minimum absolute height, and m/z tolerance 10 ppm. Parameters for the local minimum resolver function were set to 85% for the chromatographic threshold, 0.2 min for minimum search range RT, and 1.7 for minimum ratio of peak top/edge. 13 C isotope filter and finder were applied. Join aligner was used to align features with weight for m/z set to 80 and RT tolerance set to 0.2 min. Before using the peak finder function, features not detected in at least 2 samples were removed. metaCorrelate and ion identity networking were performed. The mgf spectra file and associated feature table were then exported and used to generate a feature based molecular network (FBMN)74 on GNPS275. Briefly, fragment tolerances for both parent and fragment ions were set at 0.02, while networking and annotation parameters were set to minimum 5 matching peaks and cosine similarity > 0.7. The GNPS2 FBMN job is available at https://gnps2.org/status?task=8f1989e83449460a9a5748cd8f32df30. The network.graphml file from the FBMN was imported and manipulated in Cytoscape76 (version 3.10.3). MS/MS spectra of unknown features were imported into SIRIUS77 (version 6.0.7) and CANOPUS78 was used to predict their chemical pathways or classes. Pathways or classes were retained if probability scores were > 0.7. Known and unknown features of interest were also searched using two domain-specific MASSTs79, microbeMASST80 and tissueMASST81. MicrobeMASST is a taxonomically informed mass spectrometry search tool within the GNPS ecosystem where MS/MS spectra can be linked to their respective microbial producers by querying against a curated reference database of >60,000 microbial monocultures. TissueMASST, a similar MS/MS search tool, enables investigation of molecules of interest across diverse biological contexts by the search of MS/MS spectra across publicly available metabolomics datasets acquired from preclinical animal models (mouse and rat) and humans and with associated metadata on tissue localization and disease status. For microbeMASST and tissueMASST searches, matches were defined by a minimum of 5 matching peaks and cosine similarity > 0.7. Untargeted metabolomics data analysis Metabolomics data was then imported in R 4.4.2 (R Foundation for Statistical Computing, Vienna, Austria) for downstream data analysis. Quality control (QC) samples were used to evaluate data quality. Features with a retention time (RT) < 0.70 or > 9.5 min were excluded from the analysis. Because polymers were detected in the LC-MS/MS run, the package ‘homologueDiscoverer v 0.0.0.9000‘82 was used to remove them. Six samples (VEO-IBD; n = 5, healthy; n = 1) were excluded from the metabolomics analysis because of low intensity in the total ion chromatogram (TIC; n = 3) or high contamination (n = 3). Blank subtraction was performed by removing features detected in blank samples where the mean peak areas were less than five times that observed in the pooled QC samples. Features with near zero variance were removed using the package ‘caret v 6.0‘. A validation of features annotated as BA candidates was performed using massQL27,28,83. Multivariate analysis was conducted using the package ‘mixOmics v 6.30’. Principal component analysis (PCA) and partial least square discriminant analysis (PLS-DA) were performed on peak areas after robust center log ratio transformation (rclr) using the package ‘vegan v 2.6.10’. PERMANOVA was used to evaluate group centroid separation. The performance of the PLS-DA models was evaluated using a 4-fold cross-validation. Variable importance in projection (VIP) scores were calculated per feature and features with VIP > 1 were considered significant. Natural log ratios of features of interest were generated by summing the peak areas. Wilcoxon rank-sum test, followed by Benjamini-Hochberg (BH) correction to adjust for multiple comparisons, was used to investigate differences between the children with VEO-IBD and the healthy control group. Effect sizes were reported as Cliff’s delta. Chi-squared test was used to compare categorical variables between the two groups. P < 0.05 was considered statistically significant. Pearson correlation was used to investigate correlation between age and alpha diversity. 16S rRNA gene amplicon sequencing and processing The UC San Diego Microbiome Core performed DNA extraction and sequencing following the previously developed standard protocols from the Earth Microbiome Project (http://earthmicrobiome.org/protocols-and-standards/16s/)84. Briefly, extraction was conducted using the MagMAX Microbiome Ultra Nucleic Acid Isolation Kit (Thermo Fisher Scientific, USA) and KingFisher Flex robots (Thermo Fisher Scientific, USA). Negative (blanks) and positive (Zymo mock communities) controls were included in the analysis and used for quality controls. Sequencing was conducted on the V4 region of 16S rRNA gene using the 515 F forward and the 806 R reverse primers on a MiSeq System (Illumina, USA). Generated demultiplexed fastq files were then imported into Qiita85 (ID 15748) for processing. Briefly, the Qiita default workflow for 16S rRNA data was applied, which included trimming to 150 base pairs and Deblur for the generation of the final OTU (operational taxonomic unit) table. Taxonomic classification was also performed in Qiita using the Greengenes 13.8 phylogeny database86. The obtained biom file was then converted into a tsv table using QIIME 287 and imported in R 4.4.2 for downstream analysis. Microbiome data analysis The sequencing depth of the microbiome data was inspected and five samples (VEO-IBD; n = 1, healthy; n = 4) were excluded from the downstream analysis due to low read counts (<9000 reads). The ‘phyloseq v 1.50‘88 package was used to manipulate the microbiome data. Alpha diversity analysis was conducted by rarefying the data to the lowest number of reads observed in the cohort and calculated using the Shannon Diversity Index. Before ordination and differential abundance analysis, OTUs were collapsed at genus level. Beta diversity analysis was conducted via PCA of the rclr-transformed data. Significant differences in community profiles were tested via PERMANOVA89. A PLS-DA model was also generated to extract features driving group separation. The performance was evaluated using a 4-fold cross-validation, and features with VIP > 1 were considered significant. Finally, differential abundance analysis was performed using ALDEx238 and features with BH adjusted p value < 0.05 were considered significant. Multi-omics analysis The joint analysis between the microbiome and the metabolomics data was performed using DIABLO39, a supervised method using multi-block PLS-DA to identify feature correlations between datasets in relation to a categorical outcome. The model was first tuned to retain top features for each omics block maximizing discriminatory covariance between groups. Model performance was evaluated via leave-one-out cross validation. A circos plot was finally generated to display intra feature correlation with a cut off set to 0.5. Supplementary information npj_Supplementary Supplementary_Data_1. Supplementary_Data_2. Acknowledgements The authors would like to acknowledge support from Laura Stewart, Rendie McHenry, Dan Payne and Jim Chappell (VUMC) with healthy donor sample collection. Author contributions J.M. conceptualized the study. M.N., J.A.C., and N.H. recruited patients and oversaw sample collection. Samples were processed by L.S.Z., M.C.C., and C.E. K.E.K., J.Z., and V.C. performed sample extraction and LC-MS/MS analysis. K.E.K. conducted untargeted metabolomics analysis. K.E.K. and S.Z. conducted microbiome analysis. K.E.K. drafted the manuscript. P.C.D., S.M.T., and J.M. acquired funding and supervised this project. All authors provided feedback, reviewed, and approved the manuscript. Data availability Untargeted LC-MS/MS data generated in this study are publicly available at GNPS/MassIVE (https://massive.ucsd.edu/)under the accession code MSV000097610. The associated FBMN job is publicly available at GNPS2: https://gnps2.org/status?task=8f1989e83449460a9a5748cd8f32df30. 16S sequencing data is available through ENA(https://www.ebi.ac.uk/ena/browser/home) via the following accession code: PRJEB90214. Additional information and data sequencing pipelines for the microbiome data are available in Qiita (https://qiita.ucsd.edu/) under the Study ID 15748. Code availability The code used for data analysis and to generate the figures can be found in the GitHub repository: https://github.com/kinekvitne/manuscript_VEO-IBD. Competing interests S.M.T. receives research funding from Veloxis Pharmaceuticals. P.C.D. is an advisor and holds equity in Cybele, Sirenas, and BileOmix, and he is a scientific co-founder, advisor, holds equity and/or receives income from Ometa, Enveda, and Arome with prior approval by UC San Diego. P.C.D. consulted for DSM Animal Health in 2023. J.A.C. is an advisor for Pharming Pharmaceuticals. All other authors declare no conflicts of interest. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at 10.1038/s41522-025-00899-0. References Associated Data Supplementary Materials npj_Supplementary Supplementary_Data_1. Supplementary_Data_2. Data Availability Statement Untargeted LC-MS/MS data generated in this study are publicly available at GNPS/MassIVE (https://massive.ucsd.edu/)under the accession code MSV000097610. The associated FBMN job is publicly available at GNPS2: https://gnps2.org/status?task=8f1989e83449460a9a5748cd8f32df30. 16S sequencing data is available through ENA(https://www.ebi.ac.uk/ena/browser/home) via the following accession code: PRJEB90214. Additional information and data sequencing pipelines for the microbiome data are available in Qiita (https://qiita.ucsd.edu/) under the Study ID 15748. The code used for data analysis and to generate the figures can be found in the GitHub repository: https://github.com/kinekvitne/manuscript_VEO-IBD.