We successfully established the synthetic human gut bacterial community in an automated fermenter system in chemostat mode and recorded its variability as well as its responses to pH and feed interruptions in up to twelve replicate vessels. In each experiment, we measured pH, cell counts as well as metabolite concentrations. To count bacterial species in a high-throughput manner, we developed a new tool, CellScanner, that applies supervised classification to flow cytometry data to obtain species-specific counts from community data. In addition, we explored the interaction between BT and RI in depth in the absence and presence of mucin beads. Our experiments showed that the gut bacterial community reaches a stable composition dominated by BT and BH, which is highly reproducible across replicates in controlled conditions. We identified the technical variability of 16S rRNA gene sequencing as the main source of the remaining variation. Indeed, variability across replicates was significantly reduced when resolving community composition using flow cytometry counts obtained with CellScanner. The perturbation experiments revealed the existence of alternative stable states, i.e. the community changed its composition after the perturbation without returning to its original state. In addition, not only one, but several distinct compositions were attained after perturbation in replicate vessels, which were also reproduced in two independent experiments. Our in-depth exploration of the interaction between BT and RI, combined with viability staining, led to the discovery of a slow-growth mode in RI and highlighted the context-dependency of bacterial interactions. We investigated different modelling techniques to describe our observations and predict community dynamics in new scenarios. For this, we evaluated the performance of metabolic models of gut bacteria on previously published data, which emphasised the importance of model curation. We collected and analysed RNA-seq data for three gut bacterial species to better understand their metabolism in different conditions. However, given their greater flexibility, we opted for kinetic models to explore the dynamics of the BT/RI co-culture and to understand the mechanisms behind the alternative community states observed in vitro. In the context of this exploration, we developed the miaSim R package, which makes different microbial community modelling techniques accessible to the public.
