1. Central to the composition, structure and function of any microbial community is the complex species interaction web. But understanding the overwhelming complexity of ecological interaction webs has been challenging, owing at least partly to the lack of efficient tools for disentangling species interactions in natural or artificial microbial communities.

2. In this study, we developed a microbial experimental system which allows for rapidly generating microbial consortia with programmable ecological interactions. We engineered the model organism Escherichia coli to construct metabolism- and quorum sensing-based modules. The two engineered modules were used to create synthetic microbial consortia of synergy, competition and exploitation.

3. We showed that each of synthetic microbial consortia displayed the unique mode of population dynamics under certain initial inoculation conditions. We also demonstrated that the transitions between exploitation and the types of competition or synergy based on the same paired strains were plausible by tuning the two engineered modules. We lastly derived mathematical models to quantitatively capture the experimentally observed population dynamics of these synthetic microbial consortia.

4. This approach offers a fresh angle to engineering microbial systems for experimentally testing ecological questions with a much greater control and manipulation.

Mathematical modeling

Three different general frameworks were proposed to describe population dynamics within synthetic microbial consortia. Models Ⅰ, Ⅱ and Ⅲ are all built using ordinary differential equations. Specifically, these mathematical models consisted of two to five variables that were classified into the following compounds: 1) cell density of each strain N_i; 2) concentration of the common resource (e.g. glucose) and amino acids L_i ; 3) the signaling molecules Q. Custom-tailored codes were developed to simulate models in MATLAB (MATLAB R2020b, Mathworks, Inc., Natick, MA, USA). The ode45 subroutine was used to solve the ordinary differential equations and the lsqnonlin algorithm was performed to conduct the least square regression. The procedure of parameter estimation was based on multiple local optimizations of different parameter initial values (D'Alessandro et al., 2015). The “RandomStartPointSet” object was employed to randomly generate 10,000 initial point sets for each interaction in each model. The priori-knowledge strategy (Walter, 1987) was applied to constrain the obtained 10,000 parameter set estimations and the optimal one was selected with the minimum sum of squares for error (SSE).