Author response: Modular, robust, and extendible multicellular circuit design in yeast
Year: 2022
Type: peer-review
Abstract: Article Figures and data Abstract Editor's evaluation Introduction Results Discussion Materials and methods Appendix 1 Appendix 2 Appendix 3 Data availability References Decision letter Author response Article and author information Metrics Abstract Division of labor between cells is ubiquitous in biology but the use of multicellular consortia for engineering applications is only beginning to be explored. A significant advantage of multicellular circuits is their potential to be modular with respect to composition but this claim has not yet been extensively tested using experiments and quantitative modeling. Here, we construct a library of 24 yeast strains capable of sending, receiving or responding to three molecular signals, characterize them experimentally and build quantitative models of their input-output relationships. We then compose these strains into two- and three-strain cascades as well as a four-strain bistable switch and show that experimentally measured consortia dynamics can be predicted from the models of the constituent parts. To further explore the achievable range of behaviors, we perform a fully automated computational search over all two-, three-, and four-strain consortia to identify combinations that realize target behaviors including logic gates, band-pass filters, and time pulses. Strain combinations that are predicted to map onto a target behavior are further computationally optimized and then experimentally tested. Experiments closely track computational predictions. The high reliability of these model descriptions further strengthens the feasibility and highlights the potential for distributed computing in synthetic biology. Editor's evaluation This paper used multiple strains to build gene circuits and demonstrate the modular composition of strain circuits with an automated design strategy to achieve a target behavior from a large space of possible functional circuit architectures. This paper provides synthetic biologists with an alternative solution for the problems of scalability, robustness, and modularity. https://doi.org/10.7554/eLife.74540.sa0 Decision letter Reviews on Sciety eLife's review process Introduction The leading paradigm for genetic circuit design is to combine biological parts in a delicate balance within the same cell (Ellis et al., 2009; Kosuri et al., 2013; Ottoz et al., 2014). This approach has resulted in increasingly large genetic circuits that realize functions such as logic gates and circuits (Moon et al., 2012; Bonnet et al., 2013; Nielsen et al., 2016; Gander et al., 2017), time pulses (Gao et al., 2018; Guo and Murray, 2019), incoherent feed-forward loops (Entus et al., 2007; Ellis et al., 2009), bistable switches (Chen et al., 2012; Huang et al., 2012; Yang et al., 2019; Barbier et al., 2020; Grant et al., 2020), or oscillators (Elowitz and Leibler, 2000; Tigges et al., 2009; Tigges et al., 2010). Albeit very successful, single cell circuit engineering has limited scalability and robustness because parts cannot be reused, the genetic burden on the cell grows with circuit size (Nikolados et al., 2019), and retroactivity (Del Vecchio et al., 2008) and component crosstalk (Del Vecchio, 2015) interferes with expected behavior. Furthermore, competition over shared gene expression resources makes unrelated circuits unintentionally coupled (Zhang et al., 2021). An alternative approach is to generate complex behaviors using consortia of cells wherein different cell types perform distinct functions and communicate with each other through chemical signals to realize more complex behaviors. Such division of labor is common in nature and interest has recently emerged in engineering multi-cellular synthetic systems (Brenner et al., 2008). Although developed later, synthetic consortia have caught up with single-cell circuits in terms of the complexity of functions that have been realized, generating behaviors such as multicellular time pulses (Basu et al., 2004), oscillations (Chen et al., 2015; Danino et al., 2010), and logic gates (Regot et al., 2011; Tamsir et al., 2011), bioproduction (Egbert et al., 2016), or circuits that use quorum sensing to define social interactions (Kong et al., 2018; Balagaddé et al., 2008; Weber et al., 2007). Furthermore, the emergence of novel orthogonal cross-species signaling molecules (Billerbeck et al., 2018; Du et al., 2020) and genetic channel-selector devices to multiplex individual signals (Sexton and Tabor, 2020) has opened the path for engineering more complex and precise multicellular behaviors. To date, synthetic multicellular systems were largely designed with specific target behaviors in mind, rather than optimizing modularity of components to make them usable in a large number of contexts. Still, mathematical modeling has demonstrated that multi-cellular computation should easily access a larger space of behaviors: for instance, just three different cell populations can, in theory, generate up to 100 different logic functions (Regot et al., 2011) and even bimodality and cell-synchronization (Thurley et al., 2018). Furthermore, the intuitive modular nature of cell-to-cell communication should provide a useful tool to rationally design synthetic circuits with predetermined performance. Rational design significantly speeds up circuit assembly (Chen et al., 2012; Chen et al., 2020) and allows design of global functions from local behaviors (Salis et al., 2009; Carothers et al., 2011). As a matter of fact, mathematical models have been successful to bridge the gap between individual processes and collective behaviors when applied to ecological interactions between distinct populations of communicating cells (Shou et al., 2007; Momeni et al., 2013, Egbert et al., 2016). However, an example of a model-driven strain selection for multicellular circuit design is currently missing. Here, we propose a large vocabulary of yeast strains that use chemical signals for cell-to-cell communication and that can be modularly combined to realize a large number of functions (Figure 1A). Each strain senses one or two inputs and produces a single output. The transfer function relating inputs and outputs can be either activating or repressing. The output is a fluorescent protein which can be used to read out circuit behavior, another signaling molecule which can be used to connect strains or an enzyme that sequesters or degrades a signal thus disrupting communication. We experimentally characterize each strain, model their dynamics using differential equations and then use these models to predict the behavior of strain combinations (Figure 1B). Using a library of 24 strains, we rationally design complex multicellular behaviors, such as bandpass filters, negative and positive feedbacks, time pulses, logic gates, and bistable switches. These behaviors are common in nature and they underlie cell decisions on metabolism (for instance, responding to environmental signals, [Lee et al., 2002], utilization of carbon sources, [Brandman et al., 2005]) or cell differentiation (Kueh et al., 2016; Duddu et al., 2020), and have been repeatedly connected to multicellular organization (signal gradients in development) and signaling (auxin pulses in plants). The modularity of composition achieved by our multicellular circuits allow us to design these non-trivial behaviors based on model simulations alone. This qualitative and quantitative precision demonstrated in both time and steady-state experiments further extend circuit design automation (Nielsen et al., 2016; Chen et al., 2020). Figure 1 with 7 supplements see all Download asset Open asset Modular components for engineering multi-cellular signaling circuits. (A) Three input signals, two transfer functions and five outputs are used to assemble 24 distinct strains. (B) Differential equations are used to model and predict behavior of all strains and strain combinations in the paper. In this example, a model (Model 1) and symbolic representation for a fluorescent reporter repressed by high auxin concentration are shown. (C) Left: Time course fluorescence data for different auxin concentrations for the sensor strain shown in (B). Full lines are model simulations. Right: End point fluorescence data is shown as a function of auxin concentration color-matching the time series on the left. The steady-state simulation is shown in orange. (D) Symbolic representation, steady state data and model fit for all other single input reporter strains. (E) Modeling framework (Model 2) and fluorescence kinetics data for two different two-input reporter strains. Error bars represents the s.d. of three biological replicates. Figure 1—source data 1 The data used for plotting Figure 1 and supplements. https://cdn.elifesciences.org/articles/74540/elife-74540-fig1-data1-v3.xlsx Download elife-74540-fig1-data1-v3.xlsx Results Engineering yeast strains for signal sensing, synthesis and depletion As a first step toward the construction of a vocabulary of modular yeast strains, we selected a set of signals to enable cell-to-cell communication and then optimized chassis strains for signal sensing, synthesis, and depletion. We selected the plant hormone auxin (we will also refer to it as IAA, indole-3-acetic acid) and the yeast hormone α-factor as signaling molecules and additionally used the mammalian hormone β-estradiol as an inducer. The α-factor pathway has been subject of extensive studies and mechanisms for sensing through the surface receptor STE2, synthesis by the MFα1 gene and degradation through the BAR1 protease are well understood and have been engineered to generate a wide range of behaviors (Youk and Lim, 2014; Groves et al., 2016; Shaw et al., 2019). To have a baseline strain that does not interfere with signaling, we knocked out the native BAR1 gene as in Youk and Lim, 2014 to prevent α-factor degradation. Moreover, to account for growth-arrest induced by α-factor, which affects gene expression on a large scale, we knocked out FAR1 a protein that contributes to arresting the cell cycle at G1Chang and Chang and Herskowitz, 1990, and constitutively expressed POG1, a protein that promotes growth-arrest recovery (Leza and Elion, 1999). Unexpectedly, we detected no α-factor output from strains that both sense and secrete α-factor. We suspected that this was caused by the surface receptor STE2 internally binding to α-factor. It has been shown that, upon binding to α-factor, STE2 undergoes endocytosis and then shares the secretory pathway of α-factor itself (Schandel and Jenness, 1994). We solved this problem by overexpressing STE2 on a pGPD promoter Sun et al., 2012 in these strains, aiming to have some protein copies escaping this interaction. Next, we turned to the optimization of components for Auxin sensing, synthesis, and elimination. In prior work from our group, we (Khakhar et al., 2016; Pierre-Jerome et al., 2014) developed an auxin-responsive transcription factor using a chimeric dCas9-Aux/IAA protein regulation. In the presence of Auxin, the Aux/IAA degron part of the protein binds to the auxin signaling F-box protein (a modified TIR1 for this study) and acts as part of an E3 ubiquitin ligase to catalyze ubiquitination and degradation of the Aux/IAA-degron-containing protein. Similarly, we constructed a synthetic auxin synthesis pathway in yeast (Khakhar et al., 2016). We demonstrated conversion of the precursor molecule IAM (indole-3-acetamide) into IAA through expression of the IaaH gene in yeast. Here, we amplified Auxin secretion and thus effectively the strength of the signal produced, through integration of the auxin-efflux pump ABCB1/PGP1 from A. thaliana into our yeast strains, as previously reported in Geisler et al., 2005. We measured a significant increase in the auxin-synthesis yield as measured by a neighboring IAA-detecting cell when PGP1 was integrated (Figure 1—figure supplement 1). Unlike for α-factor, a depletion mechanism for Auxin had not yet been reported in yeast. To create an IAA depletion mechanism, we thus selected the plant protein GH3.3 that has been shown to conjugate IAA to aspartic acid, forming the signaling-inactive IAA-Asp. GH3.3 is part of a family of proteins that encode IAA-amino synthetases, which have been reported to control auxin homeostasis (Staswick et al., 2005). To test whether GH3.3 or related proteins could be used to inactivate Auxin in yeast, we first expressed codon-optimized versions of GH3.3 and GH3.6 from A. thaliana and C. papaya from a highly expressed pGPD promoter. We then tested the IAA to IAA-Asp conversion rate using mass spectrometry and found that GH3.3 from A. thaliana had the higher efficiency (Figure 1—figure supplement 1). Finally, we tested that IAA-Asp does not activate the IAA-mediated degradation pathway in yeast, by adding 10 µM of IAA-Asp in an auxin-sensor yeast culture and detecting no variation in fluorescence (Figure 1—figure supplement 1). Establishing a vocabulary of parts for cell to cell communication Having established conditions for efficient signal sensing, synthesis and inhibition, we combined signals with activating and repressive transfer functions to create a vocabulary of strains. Transfer functions that use α-factor as input are mediated by the transcription factor (STE2) that either directly induces expression of the gene of interest (activation) or induces expression of a repressor that inhibits the output (repression). Similarly, β-estradiol binds and activates a transcription factor (ZEV4, McIsaac et al., 2013) that either directly activates gene expression (activation) or induces expression of a repressor and inhibits output synthesis (repression). For both signaling molecules, we chose dCas9 fused with the repressor domain Mxi1 (Gander et al., 2017) as repressor (Figure 1—figure supplement 1 for the full pathways). To induce activation or repression with auxin, we build on the same auxin-mediated degradation pathway used for auxin sensing above. Specifically, activation results from degradation of a repressor, while repression results from degradation of an activator. We chose a dCas9-Mxi1-auxin degron (Khakhar et al., 2016) fusion as a repressor and used a dCas9-VP64-auxin degron fusion as the activator (Figure 1—figure supplement 1). We conducted extensive pathway optimization to increase the separation between high and low expression levels and IAA sensitivity, using a mechanistic model to explore the parameter space and guide the genetic engineering (Figure 1—figure supplement 1). We adopted the model proposed in Pierre-Jerome et al., 2014, where each parameter easily translates to a biological function, and performed parameter sensitivity analysis with respect to fold change between the baseline and the Aux/IAA-induced fully repressed state. We then selected the top six highest scoring parameter perturbations, designed circuit variants that reflected those changes, and tested them resulting in a good agreement with our predictions (Figure 1—figure supplement 1). Guided by the model, we combined five of the tested circuit variants to increase the fold change from 1.3-fold (sensor from Khakhar et al., 2016) to 3.1-fold: we used this final circuit for all the repressive strains, swapping the fluorescent reporter gene with the output gene for non-sensor strains. For the auxin activating strains, we used a similar model-driven approach to rationally design the activating pathway novel to this paper, obtaining a threefold-change activation. Combining the three different input signals, the activation/repression circuits and output secretion, we built and tested all the possible combinatorial designs presented in Figure 1A. The strains that sense β-estradiol and repress expression of BAR1 and GH3.3, the strain that senses α-factor and represses expression of GH3.3, and the strain that senses IAA and represses BAR1 expression are shown in Figure 1—figure supplement 1 and not used below, since their response was too slow to produce meaningful results. Of the remaining 24 strains, 6 sensor strains express GFP in response to the three input signals (three activators and three repressors), 12 strains (six activators and six repressors) synthesize a signaling molecule, and 6 strains act as signal attenuators (expressing BAR1 or GH3.3). Four of these 24 strains sense and secrete the same signaling molecule (α-factor or IAA, 'positive' or 'negative' feedback strains). Finally, two strains express repressors of their own input (α-factor expressing BAR1 and IAA expressing GH3), also describing a negative feedback topology (Figure 1—figure supplement 1). Sensor strain characterization For sensor strain characterization, we collected time series data for eight different input concentrations. Input concentrations were selected to fully cover the sensor dynamic range for model fitting. We normalized fluorescence data by cell size and took the mean of the histogram as in Groves et al., 2016 and then subtracted background fluorescence. Each measurement in the figure is an average of three experimental repeats (error bars representing the standard deviation). With a scalable and modular system in mind, we fitted a set of three ordinary differential equations (ODEs) with eight parameters for each strain to describe input sensing (x1), signal processing (x2) and fluorescence output synthesis (x3) (Figure 1B, Model 1). Signal processing (activation or repression) is modeled with a simple Hill function (Model 1b), which naturally incorporates signal saturation. Input sensing and output synthesis are modeled as linear ODEs (Model 1 a and c). A constant term in the last ODE accounts for background promoter activity. For instance, for the auxin sensor in Figure 1B, x represents the TIR1-auxin complex concentration, x2, the dCas9-VP64-auxin degron population, and x3, GFP concentration. These simple models capture the system dynamics, with the benefit of being easy to fit and analytically approachable. Parameters were fitted independently for each experimental repeat to obtain a mean and a standard deviation for the Hill coefficient. We also separately fitted a Hill curve to an average of the three experimental repeats and the resulting Hill coefficient (n) is reported in the figures and used for simulations (Figure 1B). The sensor strains range in sensitivity depending on the input. The β-estradiol sensors respond to inputs concentration ranging from 0.1 to 100 nM with an EC50 of 0.5 ± 0.0 nM for repression and 12.6 ± 0.9 nM for activation. α-factor sensors are sensitive to input concentrations ranging from 1 to 500 nM range with an EC50∼6.0±0.4 nM for activation and 89.0 ± 6.6 nM for repression. Finally the IAA sensor is least sensitive and responds to inputs ranging from 5 nM-10 uM withe an EC50∼964.8±93.8 nM for activation and 276.5 ± 8.8 nM for repression. The Hill coefficients for these sensors vary between 0.8 (3-repeat 0.8 ± α-factor repression) and ± α-factor ± β-estradiol repression) and ± β-estradiol ± auxin repression) and 0.8 auxin with previously reported for β-estradiol repression and more repression than dCas9 leading to Gander et al., 2017) and α-factor activation et al., 2016). Finally, the sensors achieve an that between a of 3 activating and a of activating We tested signal by adding combinations of at concentration to repressive sensors (Figure 1—figure supplement 1). the are within of the for signal concentrations within the range used for circuit In to we constructed two sensor strains that respond to both α-factor and In the first strain, α-factor induces fluorescence expression and IAA represses while in the strain, IAA induces fluorescence expression and α-factor represses it (Figure We collected data for eight different input combinations at five time up to saturation. The output signal is with respect to each input and the input functional range is similar to the one measured for the strains that respond to only one of the two To model these strains, we used a simple of our previously models with five ODEs (Figure Model two ODEs model input sensing two model input processing and one ODE the signals to their activating or nature and output These models fit the experimental data even when the output is over and circuits using cell-to-cell communication To the potential to build biological circuits using cell-to-cell we experimentally tested communication between strains that secrete an output and their sensor strains. For example, a strain that produces auxin in response to α-factor sensing was in with a sensor that switches GFP expression in response to auxin (Figure 1B). The two strain populations were at the same concentration and the fluorescent output was measured at steady state The experimental data is with a negative α-factor sensor as expected more α-factor, the the fluorescent as in Figure Figure 2 Download asset Open asset signaling (A) Symbolic representation of a that α-factor as the input and auxin as an To model models for individual strains are and only the last differential of the first model is the GFP output is with IAA fluorescence data and fit are shown for the example (B) the concentration of the strain, the strength of the signal by the strain can be This change in signal is modeled with a single parameter Right: data and model predictions for three experiments with the same strains but concentrations of the (C) Symbolic representation, model and data for a two strain wherein the strain a signal rather than (D) Symbolic representation, model and data for all two strain cascades where the input is activating the of the (E) Symbolic representation, model and data for all two strain cascades where the input is the of the Symbolic representation, model and data for signaling Error bars represents the s.d. of three biological replicates. Figure data 1 The data used for plotting Figure Download the genetic circuit of this strain is to the α-factor we tested the mathematical model previously fit to the sensor data We only the 3 output parameters to account for the that the output is auxin rather than The output of the cell model was used directly as input of the sensor cell To test the that strains that have common parts parameters and model we collected two on two different experiments of data we used one for first and the other for Figure A at input concentration was also used for in the we expected that higher cell concentration in an higher concentration of their output signal over the same time we to this output can be predicted by our models that we could use it to circuit behavior. We modeled this with a factor by the output where is the with respect to the standard we this using the same used in circuits (Figure predictions closely with data collected using 5 and 10 the concentration of cells and using the same strain as in the that the as we just it is not to the between the strains. in are not for model predictions that is no or competition for which is in these experiments strains are in the experiments and their concentrations are We further tested we could the concentration of signaling molecules by it from the system through BAR1 and strains (Figure We modeled signal degradation as a Hill function, where the output of the cell acts as a negative As only the three output parameters of the strain were fitted using cell fluorescence Finally, we tested all the in our vocabulary with the of those generating positive or negative activating strains function within the sensor range of their strains (Figure and the models fit or predict the the other the output of strains (Figure not fully cover the input range of their strains, as in the limited of the fluorescent that the strain to produce a more which we experimentally and Figure and the we also the of our model on two with different and different strain (Figure the simple model strategy we the dynamics even when different strain concentrations were These results the that multicellular circuits a of their individual parts are easy to through strain concentrations or signal and can be to response using positive feedback we to and design multicellular circuits with to obtain dynamic systems with a single To extend the range of behaviors and generate to the we used positive feedback strains that sense and secrete the same signaling molecule (Figure Figure 3 with 3 supplements see all Download asset Open asset strains and (A) Symbolic representation, model and data for two positive feedback circuits using factor and auxin as the activating (B) Symbolic representation, model and data for two positive feedback circuits that act through repression. (C) and model for a bistable switch from strains. (D) of as a function of strain where the represents one and predicted between and The of the are the concentrations for experimental (E) For the the show the IAA and concentration for the three data for the bistable Full is Model 3 and auxin and concentrations are reported as a function of over in the figure as and lines are normalized to their Error bars represents the s.d. of three biological replicates. Figure data 1 The data used for plotting Figure 3 and supplements. Download To the increase of on top of models to the positive feedback strains in Figure we also the sensor strains to a Hill coefficient as in Shaw et al., 2019). We then these fitted the data and the two Hill coefficients for both the feedback system and the response this the sensor In both the positive feedback the of the response 0.8 ± 0.0 to ± 0.1 and from ± to ± for the α-factor and the auxin We also positive feedback circuits that through repression (Figure In this we tested where either α-factor represses BAR1 synthesis, or auxin represses GH3.3 at low signaling molecule concentration, signal degradation through BAR1 or GH3.3 is
Source: eLife
Institutions: University of Washington, Virginia Tech
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: gold
Estimated APC paid: $2,500
OpenAlex ID: W4223606118