Predictable Design in Biological Engineering: Debugging of Synthetic Circuits By in Vivo and in silico Approaches

The bottom-up design process of biological systems is one of the hallmarks of synthetic biology. This approach is adopted in all the fields of engineering and, similarly, it will significantly boost the potential of living systems engineering. However, our ability to predictably engineer genetic systems from sets of quantitatively characterized parts is still limited due to unpredicted interactions among circuit components. The evaluation of the bottom-up approach in a wide number of model systems can elucidate the actual predictability boundaries where the designed systems can behave as intended. The subsequent study of non-functional or unpredictable systems offers the opportunity to determine specific features, parts and conditions that significantly affect predictability. In this work, a set of increasingly complex model systems was considered. All of them were bottom-up designed from sets of pre-characterized parts, but the quantitative characterization of the final systems resulted to be unpredictable, thus requiring a debugging process, herein reported, based on both experimental procedures and mathematical modelling.

The model systems included: 1) a library of novel synthetic repressible devices, 2) transcriptional regulators cascades, 3) a feedback-controlled circuit, 4) enzyme production systems, and 5) circuits controlled by small RNAs. Mathematical models describing the quantitative behaviour of the systems were developed during the design step. Such models were used to investigate the unpredictable experimental output of the systems, measured via population-based, single-cell fluorescence measurements, or enzyme assays. Since the bottom-up design relies on parameter estimates, obtained during biological parts characterization, sensitivity analysis was coupled with Monte Carlo simulations to study the propagation of the uncertainty of parameter values towards the final output of the system. The contribution of biological noise in interconnected circuits was also studied in silico, via stochastic models, and in vivo, via single-cell measurements, to evaluate its importance in output predictability. The re-characterization of parts in a multi-faceted fashion was finally carried out to deepen the knowledge of parts functioning and identify possible conditions where specific modules do not have predictable behaviour. Crosstalk among parts like transcription factor-promoter pairs, identified via ad-hoc experiments, have also been crucial to elucidate previously unpredicted and non-modelled molecular interactions.

Taken together, the studied systems spanned a wide range of design architectures, parts and strains, and their debugging process enabled to decouple the contributions of context-dependent variability of biological parts, cell-to-cell variability, parameter estimation uncertainty, circuit mutations due to genetic instability, crosstalk, and metabolic burden.