Rapid prototyping and design of cybergenetic single-cell controllers

ARTICLE https://doi.org/10.1038/s41467-021-25754-6 OPEN Rapid prototyping and design of cybergenetic single-cell controllers 1234567890():,; Sant Kumar1, Marc Rullan1 & Mustafa Khammash 1✉ The design and implementation of synthetic circuits that operate robustly in the cellular context is fundamental for the advancement of synthetic biology. However, their practical implementation presents challenges due to low predictability of synthetic circuit design and time-intensive troubleshooting. Here, we present the Cyberloop, a testing framework to accelerate the design process and implementation of biomolecular controllers. Cellular fluorescence measurements are sent in real-time to a computer simulating candidate stochastic controllers, which in turn compute the control inputs and feed them back to the controlled cells via light stimulation. Applying this framework to yeast cells engineered with optogenetic tools, we examine and characterize different biomolecular controllers, test the impact of non-ideal circuit behaviors such as dilution on their operation, and qualitatively demonstrate improvements in controller function with certain network modifications. From this analysis, we derive conditions for desirable biomolecular controller performance, thereby avoiding pitfalls during its biological implementation. 1 Department of Biosystems Science and Engineering (D-BSSE), ETH Zürich, Mattenstrasse 26, 4058 Basel, Switzerland. ✉email: NATURE COMMUNICATIONS | (2021)12:5651 | https://doi.org/10.1038/s41467-021-25754-6 | www.nature.com/naturecommunications 1 ARTICLE P NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-25754-6 ropelled by advancements in DNA synthesis, laboratory automation, and a growing repository of characterized biological parts, synthetic biology is starting to bear fruit1–7. Recent years have seen a surge of synthetic circuits applied to biotechnology8 and medicine9. However, synthetic circuit construction still presents serious challenges due to unanticipated cross-talk between parts, loading and burden effects, operation in a cellular environment which is inherently stochastic, and long design-cycle time periods, among other reasons. Furthermore, synthetic circuits are usually developed in a context different to that of their end-application, and the circuit’s transplantation to new environments comes with complications: changes in culture conditions or host can significantly degrade circuit performance10. There are different strategies to mitigate the effect of host and environment context. Circuit reliability can be improved by means of network architecture11. A more general approach to engineer robustness is to control the circuit’s critical components with feedback regulation, a strategy commonly used in endogenous biological systems12–14. The properties of feedback regulation have previously been exploited in synthetic circuits to increase bioprocess yields15–17, or circuit robustness18. However, it is worth noting that previous implementations of in vivo feedback regulation were not capable of perfect adaptation, i.e., convergence to a constant activity level regardless of external disturbances. To achieve this feature, integral action is required19. Theoretical implementations of integral controllers solely employing chemical reactions (biomolecular controllers) have been proposed20–26, but experimental demonstrations remain challenging, with only a few examples of integral (or quasi-integral) implementations (in vivo implementations in27–30; in vitro implementation in31). The translation of circuit specifications to biomolecular realizations is non-trivial and depends on component availability, characterization, and cross-talk, among others10,32. Mathematical modeling is usually employed to identify and alleviate these issues. However, when the target system to be controlled is not quantitatively defined, this approach can lead to large discrepancies between predictions and experimental outcomes. An engineering strategy commonly used to develop complex, real-time embedded controllers is hardware-in-the-loop (HIL), where the controller being designed is interfaced with a realistic simulation of the system it should steer. HIL is widely used in industries where testing and optimizing the embedded controller in its final application setting is infeasible or very expensive, such as in the automotive or aerospace industries. For example, to ensure an airplane rudder functions suitably over the entire flight envelope, the full rudder and its controlling hardware are interfaced in closed-loop with an aerodynamic computer model of the rest of the airplane. In this way, the expected impact of the rudder dynamics on the airplane flight characteristics can be studied easily for a wide-range of flight conditions. In these cases, HIL vastly decreases development time and costs by shortening the design cycle and minimizing the number of test runs with the real system. To fulfill similar functionalities in a biological setting, we envisioned the Cyberloop (Fig. 1a), a hybrid framework to test and optimize synthetic circuits (biomolecular controllers in this work) under realistic conditions. In the Cyberloop, the targeted in vivo biological system is interfaced at the single-cell level with biomolecular controllers implemented in silico, thereby enabling rapid and cost-effective prototyping. Closing the loop with the true biological system instead of simply using simulations has clear advantages in the design process, as no assumptions need to be made regarding the system’s structure or parameters. The hybrid in vivo/in silico interaction in Cyberloop is achieved via fluorescence measurement and optogenetic 2 activation with light under the microscope33. Investigating/Measuring cellular behavior via fluorescent proteins (FP) is a wellknown and well-established method in the synthetic biology research community with thousands of FPs available now34,35 for different cell types. Moreover, optogenetics is a well-known biological technique that uses light to influence biological processes. Most notably, it facilitates a unique capability of controlling gene expression with excellent spatial as well as temporal resolution36. An interface of fluorescence measurement and optogenetic activation thus makes this Cyberloop framework applicable to different cell types, and empowers aiming at hundreds of cells individually in a parallel fashion. Using the Cyberloop with a genetically engineered strain of Saccharomyces cerevisiae (Fig. 1b), we first show how the behavior of a biomolecular controller (Autocatalytic Integral Control motif21) designed in a deterministic setting drastically changes when put into the stochastic cellular context and provide guidelines to reduce such effects. Secondly, we study the Antithetic Integral Control motif20, which received broad attention due to its robustness and good performance in stochastic settings. One key assumption requir (...truncated)