Dynamic cybergenetic control of bacterial co-culture composition via optogenetic feedback

Abstract

Communities of microbes play important roles in natural environments and hold great potential for deploying division-of-labor strategies in synthetic biology and bioproduction. However, the difficulty of controlling the composition of microbial consortia over time hinders their optimal use in many applications. Here, we present a fully automated, high-throughput platform that combines real-time measurements and computer-controlled optogenetic modulation of bacterial growth to implement precise and robust compositional control of a two-strain E. coli community. In addition, we develop a general framework for dynamic modeling of synthetic genetic circuits in the physiological context of E. coli and use a host-aware model to determine the optimal control parameters of our closed-loop compositional control system. Our platform succeeds in stabilizing the strain ratio of multiple parallel co-cultures at arbitrary levels and in changing these targets over time, opening the door for the implementation of dynamic compositional programs in synthetic bacterial communities.

Fig. 1. Compositional control of a bacterial co-culture via optogenetic feedback.

a Because of the competitive exclusion principle, two non-interacting strains that compete for common space and resources cannot stably coexist in a co-culture (left). In this study, we show that the precise composition of such a co-culture can be modulated and stabilized at arbitrary strain ratios using external optogenetic feedback (right). b Our co-cultures contain a constitutive strain, which grows at a fixed rate independent of light, and a photophilic strain, whose growth is stimulated by blue-light. The fate of the co-culture can be controlled through the choice of external light inputs, with the constitutive strain taking over the culture in the dark and the photophilic strain taking over under strong illumination. c We implement external optogenetic feedback in a fully automated platform that includes a continuous culture with an LED for the delivery of light inputs, automated sampling coupled to a flow cytometer to allow us to monitor the strain ratio with high temporal resolution and a controller algorithm, running on a computer, that updates the intensity of the input light based on the current state of the co-culture and the control objective.

Fig. 2. Optogenetic control of cellular growth rate.

a Schematic of the genetic circuits used in this study. Control over the growth rate is achieved by growing E. coli in the presence of a fixed concentration of chloramphenicol and titrating the expression of chloramphenicol acetyltransferase (CAT), an enzyme that inactivates the antibiotic. The photophilic strain (top) carries a split T7 polymerase fused to light-inducible heterodimerization domains and the CAT gene, placed under control of a T7-promoter. Blue-light illumination leads to reconstitution of active T7-polymerase units and production of CAT, which in turn results in faster growth. In the constitutive strain (bottom), the CAT gene is expressed from a constitutive promoter, so that the growth rate is independent of light. b Dose-response of the growth rate of the photophilic strain to blue-light intensity in the presence of sub-lethal concentrations of chloramphenicol (10.5 μM). Data are presented as mean values +/− SEM, with the median steady-state growth rate of individual biological replicates shown as data points with different transparency (n = 3 biologically independent samples). The raw time-course data used to determine the steady-state growth is presented in Supplementary Fig. 1. c Dynamic response of the photophilic strain to a step changes in blue-light intensity. The transient phases of growth downshifts and upshifts have different duration. d Comparison between the controllable range of growth rates of the photophilic strain (minimum and maximum) and the growth rates of several constitutive strains. Data are presented as mean values +/− SEM, with the median steady-state growth rate of individual biological replicates shown as data points (n = 3 biologically independent samples). For the composition of a photophilic-constitutive co-culture to be controllable, the growth rate of the constitutive strain must lie in between the extremes of the photophilic strain. Therefore, strain bJAG236 cannot be used together with the photophilic strain and was only measured in duplicates (n = 2 biologically independent samples). Source data are provided as a Source Data file.

Fig. 3. evotron—automated high-throughput culture, sampling, and light-stimulation platform.

We used a modified eVOLVER platform for maintaining and stimulating our target cell culture, and developed an Opentrons OT-2 Robot-based generic and modular setup to facilitate automated periodic sampling and measurement in our experiments. a Left: Modified eVOLVER smart sleeve. We re-designed the glass vial cap and the tube-holder for stable and consistent OD (optical density) sensor measurements (Supplementary Fig. 3). We also integrated one blue LED per sleeve in the framework for dynamic light-stimulation of the target culture during an optogenetic experiment. Center: Turbidostat-mode operation. The modified eVOLVER platform was used in turbidostat mode to maintain cell culture density within a desired range during the course of an experiment via a controlled dilution and cell-culture removal process. Right: OD measurements during an experiment. Cell density was maintained within a 0.1–0.15 OD range in all of our experiments. b Opentrons OT-2 Robot-based automated sampling platform. We placed the modified eVOLVER platform on the OT-2 deck, ensuring that all 16 sleeves stayed within the accessible region of the OT-2 pipette head. The pipette head was fitted with a custom-designed adapter (3D printed) holding a sampling-needle that can be lowered into the cell culture in individual vials for sampling. We also placed cleaning solutions on the OT-2 deck to clean the sampling-needle and tubing after each sampling in order to avoid cross-contamination. At every sampling instance, the sampling-needle is moved to the desired culture vial and lowered into it. A sampling pump then extracts around 0.5 ml of cell culture into a sampling tubing, and draws it through the tubing into a flow-cytometer sample vial. Once the cytometry measurement is done, a separate waste-pump removes the left-over sample from the flow-cytometer sample vial. The sampling-needle is then moved and lowered into the cleaning solutions one-by-one, with sampling-pump and waste-pump running sequentially to clean the entire culture sample path. We developed a primary and secondary routines running on a control computer and an embedded controller respectively to execute sampling steps, run feedback-control algorithms over the measurement, and set the stimulating LED intensity accordingly. Communication channels between different elements are shown with dotted green lines in the figure. c Dynamic responses from Fig. 2c are shown with the automated fluorescence (mCherry) measurements obtained using evotron platform. Source data are provided as a Source Data file.

Fig. 4. Host-aware modeling framework applied to the photophilic strain.

a In proteome-partition models, empirical correlations between growth rate and ribosomal mass fraction arise from a balance between co-regulated sectors of the proteome: a fixed, house-keeping fraction (Φ_Q) and two flexible sectors, the ribosomal fraction (Φ_R) and the catabolic fraction (Φ_P). The give-and-take regulation of the latter two sectors determines the cellular growth rate, mediating adaptation to environmental conditions characterized by nutrient quality and translational capacity. Figure adapted from refs. and . b In our approach to host-aware modeling, the dynamics of arbitrary synthetic genetic circuits, described by a system of ODEs, are embedded into a proteome-partition framework that captures the physiological response of the cell. In contrast to a traditional ODE model, which does not consider the physiological adaptation of the host's growth rate, gene-expression burden caused by the circuit or limitations in the host's gene-expression resources, the host-aware framework seamlessly incorporates all of these host-circuit interactions without the need for extra free parameters. c Host-aware modeling framework applied to the photophilic strain. Blue-light intensity enters as an external parameter and the growth-modulating effect of expressing CAT in the presence of chloramphenicol introduces a direct interaction between the circuit and the host's growth rate. d Dynamic upshift and downshift experiments used to determine the best parameter values for the host-aware model of the photophilic strain. The parameterized model simultaneously recapitulates the dynamics of both cellular growth rate and resistance expression. Source data are provided as a Source Data file.

Fig. 5. Open-loop dynamics of the photophilic-consitutive co-culture.

a The composition of the photophilic-constitutive co-culture can be characterized by the photophilic strain fraction φ_p, whose dynamics obey the depicted ODE. Phase portraits illustrate three qualitatively distinct scenarios. Full and empty circles denote stable and unstable fixed points, respectively, and trajectories evolve in time following the flow denoted by the arrows. In all cases, the sign and magnitude of the growth rate difference, λ_p − λ_c, determine the fate of φ_p and the speed of convergence to equilibrium. If λ_p > λ_c, the photophilic strain inevitably dominates: φ_p = 1 (Left). If λ_p < λ_c, the constitutive strain dominates: φ_p = 0 (Center). If the strains grow at equal pace (Right), any value of φ_p can be a fixed point of the dynamics. b (Left) Schematic of open-loop case. A constant light intensity is delivered throughout the experiment and samples are collected periodically to monitor the co-culture composition. (Right) Experimental behavior of the co-culture in an open-loop setting, illustrating the three qualitative cases from (a). Dotted lines represent computational simulations of the equation in (a), where ${\lambda }_p={\lambda }_p\left(L\left(t\right)\right)$ is determined dynamically from the delivered input profile L(t) with a host-aware model of the photophilic strain. Source data are provided as a Source Data file.

Fig. 6. Computational screening for optimal controller parameters and closed-loop control of co-culture composition.

a Schematic of closed-loop control experiment with a PID controller. b Computational screening for optimal PID gains. Sets of randomly sampled gains (K_p, K_i, K_d) are used to simulate the expected trajectory of the co-culture in a closed-loop setting with a defined target setpoint. Trajectories are scored according to their total deviation from the target strain ratio in the relevant time window. c–e Closed-loop control of co-cultures with the same initial strain ratio and different target setpoints. Both model predictions (Right) and experiments (Left) are shown. c Target photophilic fraction: ${\phi }_p^{\mathrm{set}}=0.7$. d Target photophilic fraction: ${\phi }_p^{\mathrm{set}}=0.3$. e The strain ratio is forced to track setpoints that change from a target photophilic fraction of ${\phi }_p^{\mathrm{set}}=0.2$ to ${\phi }_p^{\mathrm{set}}=0.8$ (t = 10 h) and to ${\phi }_p^{\mathrm{set}}=0.4$ (t = 30 h). Source data.