Enhancing bioreactor arrays for automated measurements and reactive control with ReacSight

Small-scale, low-cost bioreactors are becoming increasingly important in microbial systems and synthetic biology. They provide exceptional control over parameters like temperature, cell density, and media renewal, leading to improved experimental reproducibility and enabling sophisticated experiments. Studies have utilized these bioreactors to analyze antibiotic resistance evolution, characterize cell-cell communication, and perform genome-wide fitness characterization under dynamic conditions. However, a major limitation is the lack of automated measurement capabilities. Existing systems often rely on in situ optical density measurements, providing limited information on cell population characteristics such as gene expression, cellular stress, cell size, and morphology. Researchers typically resort to manual sample extraction, processing, and analysis using specialized instruments like cytometers or microscopes. This manual process is time-consuming, error-prone, limits temporal resolution, and hinders the implementation of reactive experiment control, a crucial aspect of systems and synthetic biology. Reactive control allows for dynamic adaptation of culture conditions based on real-time measurements, facilitating tasks such as maintaining specific population states, optimizing experiments, and accelerating model-based characterization of biological systems. While commercial robotic systems could address these limitations, their cost and integration complexity have limited their widespread adoption. Prior research demonstrates automated cytometry and reactive optogenetic control, but the setups are often limited to single continuous cultures or multiple cultures with restricted capabilities. This paper presents ReacSight, a flexible and generic strategy to overcome these limitations.

The existing literature highlights the growing need for automated, high-throughput methods in microbial systems biology and synthetic biology. Studies demonstrate the successful use of small-scale bioreactors for various experiments but emphasize the limitations of manual sample handling and analysis. Previous attempts at automation have utilized robotic systems, but their cost and complexity have hindered broader adoption. Research using continuous cultures and optogenetic control has shown promise, but scalability and accessibility remain challenges. The current study addresses these gaps by developing a more cost-effective and flexible automated system.

ReacSight combines hardware and software components to link bioreactor arrays with plate-based measurement devices. A pipetting robot (such as the Opentrons OT-2) serves as a physical interface, handling sample collection, treatment, and loading into the measurement device. This allows for automation of various steps before measurement. ReacSight's software architecture utilizes Python and the Flask web application framework, providing programmatic control of all instruments (bioreactors, pipetting robot, measurement device). A central computer orchestrates the experiment, communicating with instruments using HTTP requests, allowing user-friendly interaction and remote monitoring via platforms like Discord. Key software components include a generic object-oriented event system for reactive experiment control and an exhaustive logging system. The platform's versatility is demonstrated using both a custom-built bioreactor array and commercially available Chi.Bio bioreactors. For cytometry-based experiments, algorithms for automated gating and spectral deconvolution were developed. A simple ODE model was developed and fitted to experimental data to enable model-predictive control for real-time gene expression regulation. Competition assays were performed using histidine auxotrophy as a model for nutrient scarcity. Strain ratios were dynamically controlled by modulating the OD setpoint of turbidostat cultures. To further illustrate ReacSight's genericity, it was used to enhance a plate reader with pipetting capabilities for antibiotic treatment experiments on an E. coli clinical isolate. The ReacSight software and hardware designs are openly available.

ReacSight successfully enabled automated cytometry and reactive optogenetic control of yeast continuous cultures. Long-term turbidostat cultures showed stable, unimodal distributions of fluorophore levels. Optogenetic control experiments demonstrated dynamic regulation of gene expression across a wide range, with low cell-to-cell variability. Real-time control of gene expression was achieved using a model-predictive control algorithm, maintaining fluorophore levels at target setpoints in parallel reactors. Long-term stability experiments showed that the non-secreted inducible protein maintained high levels of expression for 5 days, while secreted protein expression declined, highlighting the impact of protein secretion on expression stability. High expression of secreted protein was achievable but triggered stress response. Competition assays revealed the impact of nutrient scarcity on fitness and cellular stress, showing a clear link between nutrient availability, growth rate, and UPR stress activation. Dynamic control of the composition of a two-strain consortium was achieved by leveraging OD-dependent and OD-independent growth phenotypes, although a steady-state error was observed, potentially due to model limitations or strain identification errors. Adapting ReacSight to a plate reader allowed automated maintenance of bacterial cell populations and provided insights into antibiotic treatment efficacy under various conditions and densities, revealing complex interactions between cell death, antibiotic degradation and filamentation.

ReacSight addresses a critical need for automating complex microbial experiments. The platform's modularity and open-source nature make it accessible and adaptable to various research settings. The successful integration of different bioreactor systems (custom and Chi.Bio) underscores its versatility. The results demonstrate the potential of ReacSight for various applications, including high-throughput screening, the study of complex interactions within microbial communities, and optimization of biotechnological processes. The observation of steady-state errors in the dynamic control of the two-strain consortium highlights the importance of accurate modeling and the need for further refinement of control algorithms. Similarly, experiments with repeated antibiotic treatments suggest the need for more sophisticated models of cell population dynamics that take into account interactions between tolerance mechanisms, antibiotic degradation, and cell density.

ReacSight provides a powerful, flexible, and accessible platform for automated measurements and reactive control in microbial systems biology. Its modular design enables customization and integration with various instruments, expanding the range of feasible experiments. Future work could focus on integrating more diverse measurement techniques, improving the accuracy of control algorithms, and applying the platform to a broader range of biological systems and biotechnological applications.

The study primarily focused on yeast and E. coli systems. The generalizability of ReacSight to other organisms requires further investigation. While the platform showed successful dynamic control, some limitations were observed, including steady-state errors in certain control experiments. These discrepancies might be due to model limitations, unanticipated biological interactions, or the influence of environmental factors not explicitly incorporated into the control algorithms. Further research is needed to fully understand and address these limitations.