P148 Longitudinal single-cell data informs deterministic modelling of inflammatory bowel disease

Abstract

Abstract Background Crucial contributions to our understanding of the inflammatory process in inflammatory bowel disease (IBD) have been made using single cell level approaches. Extracting mechanistic details of cell and gene dynamics would be critical to understanding the mechanisms mediating mucosal healing and predicting response to specific therapies. However, this remains challenging. Therefore, our aim was to use ordinary differential equations (ODEs) for mathematical modelling of cell abundance and prediction of biological processes in IBD. Methods We used mathematical modelling of published longitudinal single-cell mRNA sequencing (scRNA-seq) data from DSS colitis in mice to simulate cellular dynamics during IBD. We validated our predictions using pseudo-bulk gene expression analysis and flow cytometry. We also analysed a published microarray dataset from colonic biopsies of 51 IBD patients prior to infliximab treatment. Results Based on the scRNA-seq data, we used our mathematical model to predict the immune dynamics of a different mouse colitis protocol. We validated these predictions experimentally using flow cytometry and found good agreement between model simulation and experimental data (predicted vs. observed, Pearson's rho=0.92). We found that the estimated turnover rates predicted the protein level of the proliferation marker KI-67 (rho=0.88) in DSS colitis. Finally, we wanted to investigate whether our mathematical model reflected processes in human IBD. We reasoned that genes correlated with epithelial cell dynamics might be indicative of mucosal healing after DSS retrieval. We found more highly correlated genes when using simulated epithelial abundance compared to experimentally measured epithelial abundance. Many genes associated with tight junctions (Cldn8, Cftr, Crb3, Jam2) were among the highly positively correlated genes and genes associated with inflammation (Il1b, Jak2, Il10ra) were among the negatively correlated genes with epithelial cell abundance. We then looked for differentially expressed (DE) genes in pre-treatment colon biopsies between responders and non-responders to infliximab. We found a significant overlap of 25 positively correlated genes with upregulated DE genes in responders, i.e. associated with tight junctions, and 73 negatively correlated genes with upregulated genes in non-responders, i.e. associated with inflammatory pathways. Conclusion We demonstrate that mathematical modelling of colonic cell dynamics provides insight into IBD progression and treatment response. Furthermore, our results suggest that pre-treatment barrier integrity may predict and influence response to infliximab treatment.