Scalable dynamic characterization of synthetic gene circuits

Abstract

The dynamic behavior of synthetic gene circuits plays a key role in ensuring their correct function. However, our ability to accurately predict this dynamic behavior is limited by our quantitative understanding of the circuits being constructed. This understanding can be represented as a mathematical model, which can be used to optimize circuit performance and inform the design of future circuits. Previous inference methods have used fluorescent reporters to quantify average behaviors over an extended time window, resulting in a static characterization which is a poor predictor of dynamics. Here we present a method for characterizing the dynamic behavior of synthetic gene circuits. The method relies on parameter inference techniques applied to timeseries measurements of cell cultures growing in microtiter plates. We use our method to design and characterize gene circuits in E. coli that provide core functionality for engineering cell behavior at the population level. We arrange 23 biological parts into 9 devices and combine them to construct and measure 9 gene circuits including relays, receivers and a degrader. We demonstrate that the behaviors of simple devices can be modeled dynamically and used to predict the behaviors of more complex circuits. Furthermore, our method allows incremental inference of models as new circuits are constructed, and lays the foundation for iteratively learning dynamic models from data in a scalable manner.

Interpretation of morphogen gradients by a synthetic bistable circuit

“…The model is based on one derived for the Receiver circuit 24 , but incorporates the repressor proteins, TetR and LacI, and their regulation of LuxR/ LasR expression (see Supplementary Methods for a complete derivation). We identified parameter values that enabled the model to reproduce timecourse fluorescence data using a previously established inference methodology in which a sequence of parameter inference tasks are applied to models and data for circuits of increasing complexity 26 (Supplementary Methods). This enabled us to simplify the identification of parameter values of the Exclusive Receiver model by reusing values of the subset of parameters that also appear in the Receiver model.…”

Efficient Amortised Bayesian Inference for Hierarchical and Nonlinear Dynamical Systems

“…Markov-Chain Monte Carlo (MCMC) methods have long been the gold standard for inference in ODEs (e.g., Xun et al 2013). Dalchau et al (2019) applied MCMC inference to the synthetic biological problem described in Section 4, however reports convergence times of 24-48 hours. Due to numerical integration at each time step, MCMC is not scalable across large time series models, and lacks the flexibility, interpretability, and compositionality of our method.…”

“…Our general approach for constructing prescribed (whitebox) models of biological circuits combines a populationlevel model for cell culture growth with more detailed models for the concentrations of intra-and intercellular molecules, and resembles the approach commonly used in the synthetic biology literature (Balagaddé et al, 2008;Daniel et al, 2013;Chen et al, 2015;Dalchau et al, 2019). Cell growth models are generally described by the product of the current cell density c(t) and the specific growth rate γ(c(t)), which describes both the per capita growth rate and the decrease in intracellular concentrations due to an increased volume.…”

“…As such, model reductions are commonly applied to reduce the number of dependent variables, but result in more complex nonlinearities. Following this approach, the white-box model we consider here (Section 4) was derived in detail previously (Grant et al, 2016;Dalchau et al, 2019). It describes the time-evolution of the response of double receiver devices to HSL signals C 6 and C 12 , vector u in (2).…”

The Synthetic Biology Open Language (SBOL) Version 3: Simplified Data Exchange for Bioengineering

“…Since its initial publication in 2011, SBOL has become the recommended format for engineered nucleic acid constructs in ACS Synthetic Biology (Hillson et al, 2016 ), and is supported by many biological design tools. For instance, Eugene (Bilitchenko et al, 2011 ; Oberortner et al, 2014 ; Oberortner and Densmore, 2015 ), GEC (Pedersen and Phillips, 2009 ; Dalchau et al, 2019 ), Cello (Vaidyanathan et al, 2015 ; Nielsen et al, 2016 ), GenoCAD (Czar et al, 2009 ), ShortBOL (Crowther et al, 2020 ), and GeneTech (Baig and Madsen, 2017 ) provide computational frameworks for combinatorial design space exploration, where users can specify structural, functional, and performance constraints. The outputs generated by these tools in SBOL can then be directly used by DNA assembly planning software tools such as BOOST (Oberortner et al, 2017 ), Raven (Appleton et al, 2014 ), j5 (Hillson et al, 2012 ), and DeviceEditor (Chen et al, 2012 ) to automate the process of physically building DNA constructs.…”