IntroductionPaediatric ulcerative colitis (UC) is an immune-mediated disease of complex aetiology, with onset in children up to 17 years of age1, characterized by mucosal inflammation of the colon, extending from the rectum2. Paediatric disease is often more severe than its adult counterpart3,4 and is a major public health concern due to its increased mortality rates5 and cancer risks6 throughout lifetime. The incidence is increasing worldwide, with Western European countries presenting with one of the highest incidence rates7. Etiopathogenesis of UC is multi-factorial and complex, with paediatric UC having a stronger genetic component than adult disease8,9. Implicated susceptibility genes are involved in barrier function, adaptive immunity, and antigen presentation10, suggesting impaired immune response to commensal and pathogenic bacteria alike. Consequently, gut microbial signatures have been suggested in a contributing role, both for paediatric11,12 and adult13,14 disease, as well as having prognostic potential15 (reviewed in16 for paediatric disease).Children with UC are diagnosed after colonoscopy and biopsy, after which remission is induced with aminosalicylates (5-ASA) drugs and (usually) corticosteroids. A small number do not respond and will require ciclosporin, anti-TNF drugs, or colectomy. While remission is successfully achieved in most children, around half of them will relapse within 6 months17. A further remission is induced with a second course of corticosteroids with the addition of a thiopurine as an immunosuppressant18. Identification of children who are likely to relapse within 6 months of diagnosis would be advantageous, as they could be treated with immunosuppressants or biologics from the onset19.The available biomarkers for the aminosalicylates and corticosteroid treatment outcomes in paediatric UC are unfortunately limited. Various strategies have been adopted, including clinical features at diagnosis20,21, carriage of single-nucleotide polymorphisms22, and CD8 gene transcription in circulating CD8+ T-cells23,24. Other clinical features previously associated with sustained corticosteroid-free remission (52 weeks) include low baseline severity, baseline haemoglobin, week 4 clinical remission25, and decrease in faecal calprotectin by week 1226. While these approaches show some promise, they have significant problems. First, using clinical criteria, the best prognostic index is the disease activity 4–12 weeks after treatment and not a marker before it21,25; second, a large-scale gene-association study failed to link most clinical sub-phenotypes of IBD to known genetic loci27; and third, isolating CD8 + T-cells from children requires significant amounts of blood from venipuncture to obtain sufficient lymphocytes. We suggest that it is likely that the environment of the intestine, to which immune cells respond, will have a determining effect, and this forms the basis of a new approach to a long-standing goal.Biomarkers based on high-throughput sequencing technologies (“omics”) of rectal mucosa have also been proposed and include antimicrobial peptide gene signature and bacteria from the Ruminococcaceae family and Sutterella genus25. Here, we applied a multi-omics approach on both host and microbial features from intestinal biopsies using combinatory machine learning (ML) modelling. We found that the mucosa-related microbiota (including some of oral origin) at UC diagnosis and their interaction with the mucosal transcriptome, and to a lesser degree the epigenome, predict the therapeutic outcome (at 26 weeks post-diagnosis following treatment with corticosteroids and aminosalicylates).ResultsMucosal microbiome diversity and composition at diagnosis are associated with future relapseWe recruited 56 children between 2 and 16 years who were diagnosed as having UC through colonoscopy (see patient characteristics in Table 1). At diagnostic colonoscopy, 201 biopsies were collected from the terminal ileum (TI), ascending colon (AC), distal colon (DC), and rectum (R).Table 1 Cohort characteristics, including the full cohort as well as split by single omicsFull size tableWe sequenced the 16S rRNA V3V4 region to determine mucosal microbiota diversity and composition. After filtering and chimera-elimination, 156,565 reads per sample (range: 12,070–646,470, median: 137,686) generated 5196 ASVs, whereof 4464 ASVs were classified as bacteria (range: 29–368, median: 141). 16S rRNA amplicon sequencing was normalized to the total 16S rRNA gene copy number to better approximate abundance28. Principal coordinate analysis (PCoA) of Bray–Curtis distances, performed on a prevalence-filtered subset of 292 ASVs present in more than 10% samples, revealed significant shifts in both the first and third principal coordinates associated with future relapse (PCo1, PCo3, respectively; pooled samples, Wilcoxon test p 0.05; n = 8–10/group).Fig. 3: V. parvula induces pro-inflammatory response in epithelial cells and in IL10KO mice.A HT29 and B–D HCT116 Dual reporter cells cocultured with conditioned media of V. parvula (V. parv) and V. dispar (V. disp) and BHI-growth medium in the presence of IBD-drugs Tofacitinib (TOFA), methylprednisolone (MP), and Sulfasalazine (Sulfa), showing IL-8 secretion (A, D) and activation of NFkB (C) and IRF (D) pathways. Each dot represents an independent experiment. E–H IL10-/- mice were colonized with V. parvula and V. dispar, and body weight, caecum, and colon weight were measured, and levels of colon mKC. Each dot represents one IL10-/- mouse. ANOVA (one-way, two-sided) *p