Simulations reveal challenges to artificial community selection and possible strategies for success

Multi-species microbial communities often display functions - biochemical activities unattainable by member species alone, such as fighting pathogens. To improve community function, we can artificially select communities by growing \"Newborn\" communities over \"maturation time\" into \"Adult\" communities, and selecting highest-functioning Adults to \"reproduce\" by diluting each into multiple Newborns of the next cycle. Community selection has been attempted a few times on complex communities, often generating mixed results that are difficult to interpret. Here, we ask how costly community function may be improved via mutations and community selection. We simulate selection of two-species communities where Helpers digest Waste and generate Byproduct essential to Manufacturers; Manufacturers divert a fraction of their growth to make Product. Community function, the total Product in an \"Adult\", is sub-optimal even when both species have been pre-optimized as monocultures. If we dilute an Adult into Newborns by pipetting (a common experimental procedure), stochastic fluctuations in Newborn composition prevents community function from improving. Reducing fluctuations via cell sorting allows selection to work. Our conclusions hold regardless of whether H and M are commensal or mutualistic, or variations in model assumptions.