DOP31 Gut metabolomic and compositional signatures predict response to treatment with exclusive enteral nutrition in children with active Crohn’s disease

Abstract

Abstract Background Predicting response to treatment with exclusive enteral nutrition (EEN) in paediatric Crohn’s disease (CD) can lead to more personalised, cost-effective, and efficacious therapy. We explored the ability of pre-treatment clinical and multi-omics parameters to predict faecal calprotectin (FCal) levels at EEN completion. Methods In 37 children [median (IQR), 12.4y (10.1, 15.0)] with active CD, disease parameters, dietary intake, 19 inflammatory cytokines, and 92 plasma inflammation-related proteomic markers in plasma, diet-related bacterial metabolites and 1H NMR metabolomics in faeces, as well as the faecal, duodenal, ileal, ascending and descending colon microbiome (16S rRNA sequencing), and dietary intake were measured prior to EEN initiation. Fifteen children responded to EEN (RS) and 12 not (non-RS) using a FCal cut-off <300. Results Disease and host immunological parameters did not predict FCal levels at EEN completion. Responders had lower fibre intake, half the concentration of butyrate, acetate, phenylacetate and higher bacterial richness, in faeces, than non-RS. A model trained with the discriminant duodenal operational taxonomic units (OTUs) demonstrated a sensitivity of 67%, specificity of 90% and positive predictive value of 80% to differentiate RS from non-RS. The predictive ability of the faecal and the microbiomes of other mucosal sites was inferior to that of the duodenum. In multicomponent prediction including all datasets and faecal microbiome, higher levels of phenylacetate and relative abundance of OTU of Bacteroides, were predictive of non-RS with a sensitivity of 73%, specificity of 81% and positive predictive value of 73%. Replacing the faecal with the duodenal microbiome in a subset of participants (n=14), produced a model with 100% accuracy to predict RS from non-RS. The single most important variable in this model was a Lachnospira OTU which was absent in all RS but highly abundant in non-RS. Conclusion We identified pre-treatment microbial signals and diet-related metabolites which may comprise targets for pre-treatment optimisation and personalised nutritional therapy in CD.