Dynamically reshaping signaling networks to program cell fate via genetic controllers

Abstract

Engineering of cell fate through synthetic gene circuits requires methods to precisely implement control around native decision-making pathways, and offers the potential to direct cell processes. We demonstrate a class of genetic control systems, molecular network diverters, that interface with a native signaling pathway to route cells to divergent fates in response to environmental signals without modification of native genetic material. A method for identifying control points within natural networks is described that enables the construction of synthetic control systems that activate or attenuate native pathways to direct cell fate. We integrate opposing genetic programs by developing network architectures for reduced antagonism and demonstrate rational tuning of performance. Extension of these control strategies to mammalian systems should facilitate the engineering of complex cellular signaling systems.

Organisms orchestrate complex, coordinated tasks by dynamically programming the extracellular space with distributed molecular signals that are processed by individual cells into concerted responses (1-3). Programmed cells can harness sophisticated and complex biological processes to direct developmental programs and redirect aberrantly activated cell processes (4-10). The engineering of biological systems to regulate cell fate requires precise control over gene circuit performance (11, 12) and the ability to interface with key decision-making pathways (13). Advances in synthetic biology may facilitate the design of sophisticated genetic circuits capable of sensing and actuating changes in native signaling networks (14).

Mitogen-activated protein-kinase (MAPK) pathways are a class of signaling pathways that control such key cellular processes as differentiation, mitosis, and apoptosis (1). Many diseases, including one third of human cancers, result from aberrant signaling through MAPK pathways (15, 16). Conservation of the form and function of MAPK pathways has facilitated the translation of principles identified in yeast to higher eukaryotes (17, 18). In the model organism Saccharomyces cerevisiae, pheromone stimulates cells to activate a three-tiered MAPK cascade that increases transcription of mating genes, induces cell cycle arrest, and initiates polarized cell growth. Synthetic circuits can be constructed with pathway-responsive promoters to form feedback control systems that directly prescribe dynamic pathway activation profiles (19, 20). Chimeric protein scaffolds can also be used to route cells to alternative pathway responses (21). Although successful in modulating pathway activity or fate or both, these strategies primarily rely on genetic knockouts of endogenous genes, which can alter wild-type behavior and are difficult to implement in higher eukaryotes such as humans.

Interfacing native networks with purely synthetic exogenous circuits that route cell fate through precisely controlled ectopic expression of pathway regulators provides a less invasive scheme for directing cell fate. Such control systems require little or no manipulation of the host's native genetic material, preserve access to wild-type behaviors, and minimize difficulty in transfer to higher eukaryotes. Modulation of pathway components at key control points, or pathway regulators, can alter a network response and redirect cellular fate. Introducing feedback loops at these control points reshapes network topology, alters dynamic signaling profiles, and may enhance robustness of phenotypic selection (22, 23). Synthetic RNA-based transducers can be used to link diverse environmental signals to exogenous control systems, conditionally reshaping network topology and thus redirecting cell fate (4, 5, 24). These synthetic control systems, or molecular network diverters, are composed of a promoter, which acts as a modulator, a pathway regulator, and an RNA-based transducer (Fig. 1, A and B). The modulator and transducer determine the strength, mode, and signal responsiveness of a diverter. Molecular network diverters conditionally divert the native network and confer orthogonal control of cell fate within a genetically homogenous cell population through specified environmental signals. Orthogonal control through diverters provides an added degree of freedom in specifying cell fate, preserving existing mechanical, chemical, and biochemical channels. We set out to develop molecular network diverters to activate or attenuate signaling through a native MAPK pathway to conditionally redirect, or route, cells to one of three distinct fates.

Fig. 1.

Molecular network diverters and key pathway control points. (A) General architecture of a molecular network diverter. Transducer: RNA-based controller responsive to an environmental signal; modulator: promoter that controls level and mode of expression (feedback, non-feedback); pathway regulator: protein that modulates pathway activity. (B) Diagram of molecular network diverters reshaping native signaling networks in response to environmental signals. (C) Schematic of the yeast mating pathway. Pheromone (α-factor) binds to a transmembrane receptor (Ste2), where the signal is relayed through G-proteins (Ste4), an adaptor protein (Ste50), and a scaffold (Ste5)-bound three-tiered MAPK cascade. Phosphorylated Fus3 (dephosphorylated by Msg5) is translocated to the nucleus, resulting in activation of mating genes. Pathway activity and mating response are monitored through a pFUS1-GFP reporter and pheromone-induced cell cycle arrest, respectively.

Results

Identifying Control Points within Natural Networks

To build molecular network diverters for the yeast mating pathway, we identified titratable positive and negative regulators of pathway activity that result in divergent cell fate decisions when ectopically expressed (Fig. 1C). We measured pathway activity by monitoring fluorescence from a transcriptional reporter (pFUS1-GFP). Cell fate determination was measured through halo assays, where cells are plated around a filter disc saturated in pheromone, which creates a pheromone concentration gradient, and mating-associated cell cycle arrest above a certain concentration of pheromone from wild-type cells results in a “halo” or cleared region around the disc. We used an engineered galactose-inducible promoter system to titrate amounts of various pathway signaling proteins (25). The set of tested pathway proteins was selected for those that were more likely to dominate pathway activity and exhibit minimal toxicity when overexpressed (26, 27). The majority of examined proteins failed to alter pathway activation when overexpressed, supporting the hypothesis that within native pathways regulatory architectures filter out perturbations in the amounts of various components (28-34). Despite the endogenous control schemes, we identified two control points where ectopic overexpression of associated signaling proteins dominated pathway activity (Fig. 2, A and B and fig. S1, A and B). Ste4 activated the mating pathway and induced cell cycle arrest in the absence of pheromone (fig. S2A), whereas overexpression of Msg5 reduced pathway activation and eliminated pheromone-induced cell cycle arrest (fig. S2B).

Fig 2.

A synthetic biology toolbox for constructing molecular network diverters. (A) Identification of activating regulators. Normalized pathway activity is reported as the geometric mean GFP value from cells overexpressing (+gal) or not overexpressing (−gal) the indicated proteins of the mating pathway. GFP values are normalized to that of the control (cells harboring a blank plasmid in the presence of saturating pheromone at the corresponding galactose concentration; −/+gal: 0/1% galactose; −/+α: 0/100 nM α mating factor). For n = 3, * p < 0.01 (B) Identification of attenuating regulators. Normalized pathway activity from attenuating constructs is reported as described in (A) except cells were assayed in the presence of saturating concentration of pheromone. (C) Relative promoter activities for modulating concentrations of pathway regulators. Three constitutive promoters (pADH1, pCYC1, pTEF1mutant7/pC) and one mating-responsive promoter (pFUS1/pFB) were characterized. pFB activities were measured in the absence (pFB−) and presence (pFB+) of α factor. Normalized GFP expression is reported as the geometric mean GFP value from cells harboring the specified construct normalized to that of the control (cells harboring the pC construct). (D, E) Activities of theophylline-responsive (SXth, D) and tetracycline-responsive (SXtc, E) RNA transducer sets (−/+th: 0/5 mM theophylline; −/+ tc: 0/1 mM tetracycline). Normalized GFP expression is reported as the geometric mean GFP value from cells harboring the specified construct normalized to that of the control (cells harboring the ON state control construct at the corresponding signal concentration). The TAR (transducer activation ratio) is reported as the ratio of normalized GFP expression for a given transducer in the presence and absence of signal. For n ≥ 3, p < 0.001 for all non-control transducers. (F) Pathway response as a function of predicted Ste4 expression for constitutive and feedback architectures (circles: pFB; squares: pC). Predicted relative Ste4/Msg5 expression is reported as the calculated expression normalized to that of the control (construct pairing pC and ON RNA transducer). Normalized pathway activity from activating constructs is reported as the geometric mean GFP value from cells harboring the specified construct in the absence of pheromone at the indicated signal concentration normalized to that of the control (cells harboring a blank plasmid in the presence of saturating pheromone). Each data set is fit to a hill function (line). Dash indicates extrapolated fit. Numbers indicate corresponding pathway activity and cell cycle arrest data for identical constructs. (G) Pathway response as a function of predicted Msg5 expression for constitutive and feedback architectures (squares: pC; circles: pFB). Normalized pathway activity from attenuating constructs is reported as described in (F) except cells are assayed in the presence of saturating pheromone. Values reported represent mean ± s.d. for n ≥ 3.

We examined the relations between pathway response and the predicted abundance of the regulator (Ste4, Msg5). Amounts of each pathway regulator were varied by pairing the regulator with promoters and RNA-based transducers of various activities (Fig. 2, C, D, and E). By compiling pathway activity data, we determined the pathway response curve as a function of predicted Ste4 concentrations, calculated from the combined promoter and transducer activities (Fig. 2F, fig. S3, and SOM text S1). By combining the pathway response curve with data on cell fate, we identified a narrow transitory range of ectopic Ste4 expression at which cell fate diverged from wild-type to an alternative “mating” fate. The “mating” fate is characterized by constitutive activation of the pathway and cell cycle arrest in the absence of pheromone, resulting in undetectable cell growth outside the halo region. We also examined a positive feedback architecture by replacing the constitutive promoter (pC) with the pathway-responsive promoter (pFB), which increased expression of the regulator in response to pathway activity. To analyze the quantitative effect of the various modes of expression (pC, pFB) on the pathway response to Ste4, we fit each response curve to a Hill function (table S1). The Hill function model indicates that the positive feedback architecture shifted the concentration of Ste4 at which half-maximal pathway activation was achieved (EC₅₀) to predicted concentrations of Ste4 almost one tenth those for the nonfeedback system (EC_50-Ste4,FB: 0.02, EC_50-Ste4,C: 0.21) (Fig. 2F). The results support the hypothesis that positive feedback increased the pathway sensitivity to Ste4 expression, shifting the threshold of pathway response to lower amounts of Ste4 and sharpening the divergence of the wild-type and mating fates.

Using similar strategies we generated pathway response curves as a function of predicted concentrations of the negative regulator, Msg5, under constitutive (pC) and feedback (pFB) expression. Quantification of the pathway response characteristics was performed by fitting the data to a Hill function (table S2). In contrast to the Ste4 system, the Msg5 systems demonstrated different efficacies (V_max-Msg5,C: 0.86, V_max-Msg5,FB: 0.53), resulting in different maximal pathway attenuation for each expression mode (pC, pFB) (Fig. 2G). Constitutive expression of Msg5 resulted in a sharper decrease in pathway activity and greater maximal pathway attenuation compared to the negative feedback architecture, which showed a more graded response that leveled out to higher values of pathway activity. For both modes of Msg5 expression, the pathway response curves combined with the phenotypic assays demonstrated a low and narrow range of ectopic Msg5 expression (<0.1) across which cell fate diverged from wild-type to an alternative “non-mating” fate. The “non-mating” fate is characterized by constitutive inhibition of the pathway and cell cycle arrest in the presence of pheromone, resulting in cell growth into the halo region. The data indicate that Msg5 is a very potent regulator of pathway activity.

Developing Positive and Negative Diverters Based on Identified Control Points

We constructed a positive diverter with a positive feedback architecture to activate the pathway in the absence of pheromone in response to an environmental signal. By using the pathway-responsive promoter (pFB) and various tetracycline-responsive RNA-based transducers (SXtc) to control Ste4 expression, we constructed positive feedback diverters of varying strength. Within the positive feedback architecture (pFB-Ste4-SXtc) we identified transducers that demonstrated robust small-molecule pathway activation and routing of cells to the mating fate (Fig. 3A and fig. S1C). As transducer activity increased, thus increasing the strength of the positive feedback diverter, cells exhibited increasing pathway activity. Positive feedback diverters composed of tetracycline-responsive transducers S3tc and S4tc activated the mating pathway and induced switching to the mating fate in the presence of tetracycline while maintaining the wild-type fate in its absence as measured by halo assays. We also constructed positive diverters without feedback (pC-Ste4-SXtc) with the same components as the positive feedback diverters except with a constitutive promoter (pC) replacing the pathway-responsive promoter. As expected from the pathway response curves (Fig. 2F), the positive diverters without feedback, or booster diverters, exhibited weaker effects than the positive feedback diverters (fig. S4). We built positive diverters with and without feedback (pFB-Ste4-SXth, pC-Ste4-SXth) with theophylline-responsive RNA-based transducers (SXth), which allowed similar control of cell fate with a different small-molecule input (fig. S5). Transducers responsive to different inputs with similar basal activities and dynamic ranges induced similar pathway activities and similarly routed cells to the mating fate, indicating fate routing is independent of the input signal identity. The data support the ability to rationally tune network diverters through the exchange of modular, well-defined parts.

Fig. 3.

Optimized design of independent positive and negative diverters. (A) A positive feedback diverter supports robust pathway activation in the absence of pheromone (−α) and cell fate routing to the mating fate (−/+tc: 0/1 mM tetracycline). Normalized pathway activity from activating constructs is reported as described in Fig. 2F. Dagger: see fig. S3 and SOM text S1. (B) A resistance diverter supports robust pathway attenuation in the presence of pheromone (+α) and cell fate routing to the non-mating fate (−/+th: 0/5 mM theophylline). Normalized pathway activity from attenuating constructs is reported as described in Fig. 2G. For n ≥ 5, * p < 0.05, ** p < 0.005. Values reported represent mean ± s.d.

We constructed negative diverters to attenuate the pathway in the presence of pheromone in response to an environmental signal using similar design strategies. Due to the efficacy and potency of Msg5, negative diverters that maintained wild-type fate in the absence of input signal required stringent transducers (those with low basal activities) (35). We evaluated pathway attenuation and fate routing by negative diverters with the constitutive architecture and theophylline-responsive transducers (pC-Msg5-SXth; Fig. 3B and fig. S1D). Pathway activity decreased with increasing transducer activity and maintenance of the wild-type halo in the absence of theophylline required stringent transducers (S1th, S2th; Fig. 2E). The negative diverter without feedback, or resistance diverter, composed of theophylline-responsive transducer S2th reduced mating pathway activity and induced switching to the non-mating fate in the presence of theophylline while maintaining a slightly diminished wild-type fate in its absence. Fate routing was not observed with any single negative diverter constructed with the feedback architecture (pFB-Msg5-SXth), although cells harboring individual constructs within the negative feedback set adopted either wild-type or non-mating fates (fig. S6). The Hill function model of the pathway response to negative feedback indicates this architecture has reduced efficacy (Fig. 2G). The data indicate that the pathway is less sensitive to negative feedback (36). Thus, the transducers do not span the requisite expression range separating the wild-type and non-mating fates in this architecture. Reduced pathway sensitivity to changes in Msg5 concentrations would be expected to result in smaller changes in pathway activity, potentially preventing fate routing.

Dual Diverter Highlights Need to Precisely Balance Opposing Activities

A dual diverter is a genetic control system that will support routing of an entire cell population to one of three cell fates in response to specified environmental signals. We designed a dual-diverter architecture that integrated positive and negative diverters optimized for independent cell fate routing to either the mating or non-mating fates. We paired several tetracycline-responsive positive feedback diverters (pFB-Ste4-SXtc) with resistance diverters that incorporated a set of theophylline-responsive RNA transducers (pC-Msg5-SXth) to examine effects of pairing diverters encoding opposing functions. When integrating opposing diverters that route signals through a native network, basal activity from the non-triggered diverter may antagonize routing from the triggered diverter (Fig. 4A), even if independently neither diverter affects cell fate in the absence of their cognate signal. Our results indicate that integrating the independently optimized positive feedback and resistance diverters leads to antagonism, such that independent routing to the alternative fates cannot be achieved in cells harboring these opposing functions (fig. S7).

Fig. 4.

Integration of opposing diverters optimized for independent cell fate routing results in antagonism. (A) Diagram of the integration of independently optimized opposing diverters resulting in antagonism, which inhibits the diverters from routing cells to alternate fates. (B) Schematic of black box model of system with native pheromone input and synthetic Ste4 and Msg5 input to active Fus3 with pFUS1 activity as the model output. (C-E) Model of pathway activation in the absence of pheromone (left, −α) and attenuation in the presence of pheromone (right, +α) as a function of varying strengths of the positive feedback and resistance diverters. To represent the presence or absence of cognate signals for each diverter, simulations were performed with booster strength set to 0 (C), S4tc (+tc: 0.41, left, D and E), and S4tc (−tc: 0.08, right, D and E); negative feedback strength set to 0 (C and D), S3th (−th: 0.06, left, E), and S3th (+th: 0.34, right, E). White star: positive feedback strength: S3tc (+tc: 0.21); resistance strength: S2th (−th: 0.04). Black star: positive feedback strength: S3tc (−tc: 0.04); resistance strength: S2th (+th: 0.13).

We developed a simple black box model of our system to explore the pathway response to integration of multiple modules (i.e., individual regulator expression cassettes; positive feedback, booster, negative feedback, and resistance modules) of varying strength set by transducer activity (Fig. 4B and table S3). We analyzed the dual-diverter architecture that integrated positive feedback and resistance diverters by simulating pathway output for varying positive feedback and resistance strengths in the absence and presence of pheromone (Fig. 4C). The model predicts that pathway activation across the range of positive feedback strengths was significantly diminished even in the presence of a low strength resistance diverter and completely inhibited for resistance strengths (amounts of Msg5) above K_M,Msg5 (the concentration of Msg5 at which half-maximal pathway attenuation was achieved). The modeling results suggest that even at a resistance strength corresponding to the lowest basal transducer activity the resistance module will inhibit pathway activation from the positive feedback diverter. Conversely, positive feedback strength limited the maximal attenuation from the resistance diverter, leading to reduced maximal attenuation as positive feedback strength increased. These results indicate that the basal expression from the integrated opposing diverters may significantly antagonize the other diverter.

Given the limitations of performance for a dual diverter based on integrating the positive feedback and resistance diverters, we examined alternative designs to limit antagonism and enhance the cell fate routing functions. The modeling results indicate that strategies based on modifying the strength of the positive feedback diverter for increased activity may limit the performance of the negative diverter (Fig. 4C). Thus, we used the model to examine the impact of adding a second positive module, or a booster module encoding constitutive expression of Ste4, to the positive feedback diverter (Fig. 4D). The combination of the booster and positive feedback modules enhanced system tolerance to Msg5 expression and allowed for pathway activation at amounts of Msg5 produced at basal strength (in the absence of cognate signal) of the resistance diverter. In the absence of its cognate signal, the basal strength of the booster module was predicted to have minimal effect on attenuation from the resistance diverter.

Based on the enhancement to the performance of the positive diverter achieved by combining two modules, we examined the impact of adding a second negative module, or a negative feedback module, to the resistance diverter (Fig 4E). Negative feedback was predicted to globally enhance attenuation, making it robust to minor variations in Ste4 and Msg5 expression from the other modules. The addition of negative feedback to the negative diverter was predicted to have minimal impact on pathway activation through the positive diverter. Thus, the model predicts that restructuring the positive and negative diverters to include two differentially regulated modules of the same pathway regulator will enhance the function of the individual diverters, while minimizing the impact on the opposing diverter.

Genetic Strategies for Minimizing Pathway Antagonism

We developed and incorporated a number of strategies to reduce antagonism in the dual diverter and allow for multiple fate routing. We constructed an alternative RNA transducer architecture that minimized basal expression from the positive feedback diverter in response to the cognate signal from the negative diverter to maximize pathway attenuation. The composite transducer S3tc - comprising S4tc (tetracycline-responsive “ON” switch) and S1OFFth (theophylline-responsive “OFF” switch) – was designed to reduce expression of Ste4 in response to theophylline and increase Ste4 expression in response to tetracycline (Fig. 5A). When this composite transducer was implemented in the positive feedback diverter, pathway activity was between those of S2tc and S4tc for basal and tetracycline-triggered activities. In the presence of theophylline, activity for S3tc was similar to S2tc basal activity (Fig. 5B). Thus, S3tc yielded a larger fold difference in pathway activity (8.0; activity_tc:activity_th) than did S2tc (3.8) or S4tc (2.5). Several integrated dual-diverter configurations utilizing the positive feedback diverter incorporating S3tc (pFB-Ste4-S3tc) achieved strong pathway attenuation when paired with resistance diverters. As predicted from the model, all configurations displayed weak pathway activation (fig. S8), suggesting that basal expression of Msg5 from the resistance diverter prevented the requisite pathway activation to trigger positive feedback-induced amplification. Improved performance of a dual diverter may be achieved with the use of RNA transducers exhibiting even lower basal activities and larger dynamic ranges. However, computational models of RNA transducer activities indicate a trade-off between stringency and input sensitivity of these genetic devices (37). Given these constraints, the dual-diverter architecture is inherently biased to allow basal expression from the opposing diverters to promote antagonism, which cannot be mitigated solely by changing transducer or promoter activities.

Fig. 5.

Optimizing the dual diverter architecture and components for the integration of opposing diverters. (A) Schematic of the composite RNA transducer S3tc designed to reduce expression of the linked regulator in response to theophylline and increase expression in response to tetracycline. (B) Pathway activity of cells with composite transducer S3tc in response to tetracycline or theophylline (−/+tc: 0/1 mM tetracycline; −/+th: 0/20 mM theophylline; −/+α: 0/100 nM α mating factor). For n=6, ** p < 0.02. (C) Schematic of the amplifying diverter comprising a positive feedback module and a booster module. (D) Schematic of the attenuating diverter comprising a resistance module and a negative feedback module. (E) Impact of resistance, booster, and negative feedback modules on the resolution and variation in population-level pathway activation and routing to mating fate for cells harboring the positive feedback module (white: 0 mM tetracycline; blue: 1 mM tetracycline). GFP histograms for triggered and non−triggered cell populations and routing to the mating fate are shown for cells harboring: (left) the positive feedback module (pFB-Ste4-S3tc); (second from left) the positive feedback and resistance (pC-Msg5-S2th) modules; (second from right) the positive feedback, resistance, and booster (pC-Ste4-S3tc) modules; (right) the positive feedback, resistance, booster, and negative feedback (pFB-Msg5-S3th) modules. For n=3, * p < 0.0005. (F) Pathway activity of the dual diverter as a function of varying resistance strength. R² represented for linear fit of data. (G) Predicted pathway activation from the dual diverter as a function of booster strength in the absence (left) and presence (right) of low resistance. Lines represent predicted relationships under the indicated positive feedback strengths, as set by transducer activity. (H) Predicted pathway activation and buffer capacity from the dual diverter as a function of varying resistance strength relative to K_M,Msg5. Solid: pathway activity; dashed: buffer capacity. Values reported represent mean ± s.d for n ≥ 3.

We developed an alternate positive diverter architecture to support increased pathway activation with minimal impact on the negative diverter (Fig. 5C). The modeling results indicate that strategies based on modifying the strength of the positive feedback module for increased activity limit the performance of the negative diverter (Fig. 4C). Thus, we added a second module, or a booster module, encoding constitutive expression of Ste4 and a strong tetracycline-responsive RNA-based transducer (pC-Ste4-S4tc), to the positive feedback diverter. The resulting enhanced positive diverter, or amplifying diverter, combined a positive feedback module (pFB-Ste4-S3tc) and booster module (pC-Ste4-S4tc) to increase pathway activation while minimizing antagonism of the attenuating function of the negative diverter. As predicted by the model, this dual-diverter architecture restores pathway activation (Fig. 5E). However, this architecture did not allow for routing to the non-mating fate (fig. S9). The data indicate that amplification of pathway activation from the positive feedback module in the amplifying diverter in the presence of high concentrations of pheromone overwhelmed attenuation from the resistance diverter.

To counteract amplification from the amplifying diverter at high concentrations of pheromone, we constructed an enhanced negative diverter by adding a negative feedback module, linking a pathway-responsive promoter and a theophylline-responsive RNA-based transducer to Msg5 (pFB-Msg5-SXth), to the resistance diverter (Fig 5D). The resulting attenuating diverter combined a negative feedback module (pFB-Msg5-S3th) and resistance module (pC-Msg5-S2th) (Fig. 5D). As predicted by the model, the attenuating diverter had minimal impact on pathway activation through the amplifying diverter, which retained tetracycline-induced routing to the mating fate (Fig. 5E).

We observed that the robustness of pathway activation varied with the different dual-diverter architectures. Although effective in routing cells to the mating fate, the high basal pathway activity associated with the positive feedback diverter provided little resolution between triggered and non-triggered populations (Fig. 5E and fig. S10A). Addition of the resistance diverter inhibited routing to the mating fate. By combining the positive feedback and booster modules in the amplifying diverter, cells were routed to the mating fate in the presence of the resistance diverter. Addition of the negative feedback module significantly increased resolution by reducing population heterogeneity (fig. S10B).

We probed the performance of the dual diverter constructed with amplifying and attenuating diverters under varying strengths of the resistance module by varying the transducer activity (pC-Msg5-SXth). The data demonstrated a strong correlation between resistance strength and pathway activity (Fig. 5F). Pathway activation diminished and attenuation increased with increasing resistance strength (fig. S11, A and B). Thus, resistance strength must be precisely tuned for the dual diverter to achieve routing to multiple alternative and divergent cell fates.

We used our model to analyze the impact of the resistance module on system robustness and output. Basal concentration of Msg5 from the resistance module acts as a buffer to pathway activity, filtering out perturbations in Ste4 concentrations from the amplifying diverter (Fig. 5G). In the presence of the resistance module, the amplifying diverter maintains low pathway activity until Ste4 concentrations cross a threshold and pathway activity is amplified. We evaluated the buffer capacity of the system by identifying the Ste4 expression from the booster module that resulted in pathway activation (above basal pathway activity ~0.07) as a function of resistance strength (Fig 5H). Buffer capacity increased for increasing resistance strength and was inversely related to pathway activation with an optimum for buffer capacity and pathway activation when resistance strength is set at 80% of K_M,Msg5. Our results suggest that there exists a trade-off in tuning our system for both signal output and robustness.

Optimized Dual Diverter Routes Cells to Three Distinct Fates

Our optimized dual diverter integrated a number of strategies to minimize antagonism, including improved transducer architecture, amplifying and attenuating diverters, and a precisely tuned amount of resistance (Fig. 6, A and B). Although the dual diverter (booster: pC-Ste4-S4tc; positive feedback: pFB-Ste4-S3tc; resistance: pC-Msg5-S2th; negative feedback: pFB-Msg5-S3th) exhibited lower pathway activation than did the positive feedback diverter alone or ON control (dual diverter without the negative feedback module), the dual diverter exhibited a stronger fold pathway activation (7.2) than the positive feedback diverter (1.6), approaching that of the ON control (10.9) (Fig. 6C). The dual diverter attenuated pathway activity as well as the resistance diverter alone, approaching that of the OFF control (positive feedback: pFB-Ste4-OFF; resistance: pC-Msg5-S3th; negative feedback: pFB-Msg5-S3th) (Fig. 6E). The dual diverter routed cells to the alternative mating fate as well as did the positive feedback diverter (Fig. 6D). Additionally, the cells harboring the dual diverter maintained a more robust wild-type halo compared to those harboring the resistance diverter, while routing nearly as well to the non-mating fate. Finally, cell routing could be modulated through the concentration of environmental signals (fig. S11C). Although the ON control and dual diverter exhibited similar pathway attenuation, only the latter routed cells to the non-mating fate, indicating negative feedback as a critical element in regulating longer timescale events such as cell cycle arrest. Thus, the optimized dual diverter achieved conditional routing of genetically identical cells to diverse phenotypes in response to distinct molecular signals.

Fig. 6.

Conditional routing of genetically identical cells containing a dual diverter to diverse fates in response to distinct environmental signals. (A) Diagram of integrated dual diverter with amplifying diverter and attenuating diverter to modulate the mating pathway. (B) Dual-diverter circuit diagram of behavior of the attenuating and amplifying diverters under different environmental signal conditions (+th: theophylline; +tc: tetracycline). Dashed lines indicate post−transcriptional interactions through the RNA-based transducers; solid lines indicate protein-mediated interactions. (C, E) Pathway activation and attenuation levels for the dual diverter, single diverters, and controls (−/+tc: 0/1 mM tetracycline; −/+th: 0/20 mM theophylline; −/+α: 0/100 nM α mating factor). Normalized pathway activity from activating and attenuating constructs is reported as described in Fig. 2. (D) Routing to non-mating and mating fates for the dual diverter, single diverters, and controls (red: 20 mM theophylline; blue: 2 mM tetracycline). (F) Mating efficiencies of cells harboring the dual diverter, single diverters, and controls. Mating efficiency is reported as the number of diploid colonies (determined by mating the experimental strain with a tester strain) normalized to the number of colonies from diploids and the experimental strain. (−/+tc: 0/2 mM tetracycline; −/+th: 0/20 mM theophylline). For (C) and (E), n=6 or more and for (F), n=3, ** p < 0.005. Values reported represent mean ± s.d.

Beyond controlling transcriptional activation of the pathway and cell cycle arrest, the diverters also controlled mating. The positive feedback diverter and the ON control allowed a tetracycline-induced increase in mating efficiency and increased basal mating efficiency (Fig. 6F). The resistance diverter and the OFF control exhibited a theophylline-triggered reduction in mating efficiency. The dual diverter increased mating efficiency by ~70% in response to tetracycline, similar to the effect of the positive feedback diverter (~66%), and reduced mating efficiency to an amount similar to that from cells harboring the resistance diverter in the presence of theophylline. The mating data indicated that the dual diverter is optimized to route cells to the divergent fates as effectively as did each single diverter.

Discussion

Our studies describe a class of genetic control systems, molecular network diverters, that route cells to divergent fates in response to specified environmental signals. The system integrates complex positive and negative routing functions through stringent RNA-based transducers to limit antagonism, amplify activation, and induce attenuation, allowing genetically identical cells to be conditionally routed to one of three fates in response to specified environmental signals. We focused our analysis of the effects of the molecular network diverters on steady-state pathway activation, as measured by transcriptional activation of the pathway, and cell fate determination, as measured by cell cycle arrest and mating efficiency. Other properties of the network, including temporal response of pathway activation and pathway sensitivity to pheromone concentrations, can also be measured and analyzed for changes in the engineered systems and the relation of these network properties to fine-tuning cell fate decisions (33, 38). Positive or negative feedback loops as well as constitutive overexpression of signaling components can tune MAPK pathways for prescribed temporal pathway activation responses (19). In addition, alteration of the relative amounts of pathway kinases can tune the sensitivity of the pathway response to pheromone concentrations (33). Thus, the molecular network diverters can be adapted to the regulation of other pathway signaling components to engineer a wider range of network properties.

The modular architecture of the diverters, availability of tuned RNA-based transducers, and incorporation of feedback allows for precise balancing of positive and negative regulator activities even in the presence of external pathway stimulation. Our work identified design principles for networks that induce differentiation of cells in response to environmental signals and that enhance the robust performance of integrated mutually-antagonistic genetic programs. As an example, integrated negative regulators can act to buffer a system to noise amplification mediated through positive feedback loops by providing a resistance to amplification. Additionally, pathway output can be tuned through resistance, which may be a key design consideration when adopting the network diverter architecture to new signaling pathways with different thresholds of pathway activity corresponding to alternative fates. Further, negative feedback can play an important role by reducing population heterogeneity and mediating robust, long-term cell fate decisions. The dual-diverter configuration enables routing to alternative fates and minimizes impact on the opposing diverter by integrating differential regulatory strategies on functionally redundant genes. Differential regulation of functionally redundant parts is observed in natural biological networks, such as those regulating bone formation, osteoblast differentiation, plant defense, and metabolism (39, 40), and may represent a common strategy for amplifying and attenuating pathway responses to environmental signals.

Our studies focused on the development and elucidation of design principles for this class of genetic control systems in the model organism S. cerevisiae. Conservation of the form and function of MAPK pathways through higher eukaryotes highlights the possible applications of these controllers in mammalian systems. For example, a 3-fold overexpression of MPK (homolog to yeast Msg5) can significantly reduce MAPK phosphorylation and block persistent pathway activation (41). In addition, overexpression of MPK-1 blocks apoptosis by inhibiting ERK1/2 and JNK signaling (42-44). As another example, overexpression of a neuro-endocrine-associated phosphatase can suppress NGF-stimulated neuron differentiation in PC12 cells (45). However, differences in the regulation of these pathways in higher organisms compared with yeast may prohibit a direct translation of the strategies described here, and extension into higher organism may require further engineering optimization efforts.

Molecular network diverters provide a foundation for robustly programming spatial and temporal control over cell fate. The extension of similar control strategies to mammalian systems may facilitate construction of sophisticated genetic programs encoding complex cellular functions such as patterning of cell fate, tissue homeostasis, and autonomous immune surveillance (14). The connection of synthetic circuits to downstream effector processes, like drug-delivery and apoptosis, require reliable signal processing to target therapeutic effects. Strategies that enhance the robustness of signal processing in synthetic circuits will enable such smart therapeutics. Our ability to construct synthetic circuits that interface with native pathways, like molecular network diverters, will continue to expand as more pathway-responsive genetic elements and regulatory RNAs are identified. Advances in our capabilities to extract information from and direct responses in native pathways will facilitate the translation of synthetic biology tools to applications in medicine and biotechnology.

Materials and Methods

Plasmid and strain construction

Details are available in SOM Materials and Methods.

Measuring promoter and RNA transducer activities

Details are available in SOM Materials and Methods.

Measuring mating pathway activity through a transcriptional reporter

For the galactose titration studies, plasmids harboring the galactose-inducible promoter (pGAL1) controlling expression of various pathway regulators (STE4, STE50, STE11, STE7, FUS3, MSG5) and appropriate controls were transformed into the reporter yeast strain CSY532. Cells were inoculated into the appropriate dropout media, grown overnight at 30°C, and back diluted into fresh media in the presence of varying concentrations of galactose (0, 0.25, 0.5, 1, 2 and 3%) to an OD₆₀₀ of <0.1. To identify negative pathway regulators, after growing for 3 h at 30°C, cells were stimulated with saturating pheromone levels (100 nM final concentration of α mating factor (Sigma-Aldrich, St. Louis, MO)) to activate the mating pathway. The pathway was not stimulated when evaluating the potential of positive regulators. Following 6 h of growth post-back dilution, GFP fluorescence levels from the pFUS1-yEGFP3 reporter were evaluated via flow cytometry using a Cell Lab Quanta SC flow cytometer (Beckman Coulter, Fullerton, CA). GFP was exited from a 488 nm laser and emission was measured from 520/40 nm bandpass filter with a PMT setting of 5.0. Cells were gated for viability based on side scatter and EV. For each sample, the GFP values of 10,000 viable cells were measured. Cells were then gated for GFP levels above background. Flow cytometry data was analyzed using the FlowJo v.10 (Tree Star, Inc.) software package. Normalized pathway activity is calculated as the geometric mean GFP value of cells harboring the indicated construct normalized to that of cells harboring the blank plasmid control (not bearing a gene) stimulated with saturating α mating factor at the corresponding galactose concentration for three biological replicates. Error bars represent the standard deviation of at least three replicates.

The molecular network diverter plasmids and appropriate controls were transformed into the reporter yeast strain CSY840. Cells were inoculated into the appropriate dropout media, grown overnight at 30°C, and back diluted into fresh media in the presence or absence of the appropriate trigger molecule (theophylline or tetracycline) at the specified concentration to an OD₆₀₀ of <0.1. For negative diverters, after growing for 3 h at 30°C, cells were stimulated with saturating pheromone levels (100 nM final concentration α mating factor) to activate the mating pathway, while positive diverters were tested in the absence of pheromone. Following 6 h of growth post-back dilution, GFP fluorescence levels from the pFUS1-yEGFP3 reporter were evaluated via flow cytometry using a Cell Lab Quanta SC flow cytometer as described above. For cells stimulated with tetracycline, GFP values were corrected by subtracting the calculated tetracycline autofluorescence. Tetracycline autofluorescence was calculated as the difference in GFP values between the 0 mM and 1 mM tetracycline samples for the wild-type control. GFP values for the wild-type control were not altered in the presence of theophylline. Normalized pathway activity is calculated as the geometric mean GFP value of cells harboring the indicated construct normalized to that of cells harboring the blank plasmid control stimulated with saturating α mating factor in the absence of either small molecule. Error bars represent the standard deviation of at least three replicates. P-values were determined using a two-tailed t-test of the compared data sets.

Measuring mating pathway activity via halo assays

Mating associated cell-cycle arrest was evaluated via halo assays as previously described (46). Briefly, halo assays were performed on cultures harboring the indicated constructs grown overnight in the appropriate dropout media, back diluted into fresh media to an OD₆₀₀ of <0.1, and grown to an OD₆₀₀ between 0.2-0.4. Each replicate was plated at a volume of 300 μl on the appropriate dropout plates. For galactose-titration halo assays, cells were plated on non-inducing, non-repressing solid media with varying concentrations of galactose (0, 0.25, 0.5, 1, 2, and 3%). For characterization of the molecular network diverters, cells were plated on solid dropout media containing the indicated concentration of signal molecule (theophylline or tetracycline). After plating the cells, a gradient of α mating factor was established by saturating a filter disk (2 mm diameter) of Whatman paper with 9 μl of 0.1 mg/mL α mating factor and placing the disk on the center of the plate. Cells were grown for 18 h at 30°C and imaged via epi-white illumination with a GelDoc XR+ System (Bio-Rad). Representative images are presented from experiments performed in at least triplicate.

Images from halo assays were analyzed using ImageJ (NIH, Bethesda, MD) software analysis to extract pixel intensity at each position along a horizontal slice of the halo image using the rectangular tool. Standard slices were centered on the filter disc. Heat maps were generated using the “HeatMap” function from MATLAB (Mathworks, Natick, MA) for the halo assays at a standard window size. To ensure the heat map captured the changes in growth and was standardized across all maps, we excluded high intensity data (e.g., the filter disc) and fixed the range of intensities (pixels=120-200).

Calculating relative regulator expression

Details are available in SOM Materials and Methods.

Evaluating selection of mating resistant population

Details are available in SOM Materials and Methods.

Determining resolution and variation in positive diverter networks

Details are available in SOM Materials and Methods.

Measuring mating efficiencies

Details are available in SOM Materials and Methods.

Modeling of molecular networks diverter interactions with the mating pathway

We built a simple phenomenological model to describe the dynamic behavior of key components of our pathway: Ste4, Fus3, Msg5, and Fus1 (output). The model consists of four deterministic ordinary differential equations (ODEs, 1-4 below) and is consistent with previous models described in the literature (47-49). A complete, detailed description of this signaling cascade would include additional phosphorylation intermediates and scaffold proteins (19). However, detailed mechanistic ODE models of the MAPK pathway in Xenopus show that the resulting input/output behavior of this multi-layered signaling cascade is equivalent to a highly cooperative activation enzymatic process (47). Thus, we reduced the model to a four-component system, where the unmodeled phosphorylation and dephosphorylation kinetics between Ste4, Msg5, and Fus3 are qualitatively captured by Hill functions with high Hill coefficients.

Each synthetic regulatory element in our pathway is individually modeled. We used zeroth-order production rates to capture the effects of constitutive promoters and switches on the native pathway. These production rates are modulated according to the strength of the switch employed. We used Hill functions to model the pathway-responsive promoter (pFUS1/pFB), which generate the positive (Ste4 activation) and negative (Msg5 activation) feedback loops.

The model was fit to normalized pFUS1-GFP steady-state data. We used the results of several experimental assays testing the effects of each individual synthetic regulatory module on the native pathway output. The steady-state data sets simultaneously used in the fitting algorithm include the response of the pathway (i) at different concentrations of pheromone (α-factor), (ii) in the presence of pC-Ste4 with a range of transducers, (iii) in the presence of pC-Msg5 with a range of transducers, (iv) in the presence of pFB-Ste4 with range of transducers, (v) in the presence of pFB-Msg5 with a range of transducers. Fits were generated using MATLAB ‘fmincon’ routine, minimizing the squared error between the numerically generated steady-state of Fus1 and the normalized measured data. ODEs were integrated using a forward Euler method. Upper and lower bounds were chosen for each parameter to guarantee positivity and boundedness of the solutions.

The parameters resulting from our fit are in general agreement with expected relative values of reaction rates for promoter activity and for phosphorylation/dephosphorylation pathways (table S3). The Mgs5-mediated maximal dephosphorylation rate of Fus3 is high relative to other rates; the efficacy of Msg5 as a strong phosphatase for Fus3 is well known (50, 51), but few experimental kinetic measurements are available in the literature (52). Our parameter for Msg5 dephosphorylation is consistent with previous work (52). The half-maximal concentration of Msg5 (K_M,Msg5) is low, in agreement with our experimental observations.

$$\frac{d}{dt}\left[Ste4\right]={\beta }_{Ste4}-{\delta }_{Ste4}\left[Ste4\right]+k_{C,Ste4}SCtc+k_{FB,Ste4}SFtc\frac{{\left[Fus1\right]}^{n_{pf}}}{K_{M,Fus1,pf}^{n_{pf}}+{\left[Fus1\right]}^{n_{pf}}}$$ (1)

$$\frac{d}{dt}\left[Fus3\right]={\beta }_{Fus3}+k_{Ste4}\frac{{\left[Ste4\right]}^m}{K_{M,Ste4}^m+{\left[Ste4\right]}^m}+k_{\alpha }\frac{{\left[\alpha \right]}^n}{K_{M,\alpha }^n+{\left[\alpha \right]}^n}-{\delta }_{Fus3}\left[Fus3\right]-k_{Msg5}\left[Fus3\right]\frac{{\left[Msg5\right]}^q}{K_{M,Msg5}^q+{\left[Msg5\right]}^q}$$ (2)

$$\frac{d}{dt}\left[Msg5\right]={\beta }_{Msg5}-{\delta }_{Msg5}\left[Msg5\right]+k_{C,Msg5}SCth+k_{FB,Msg5}SFth\frac{{\left[Fus1\right]}^{n_{nf}}}{K_{M,Fus1,nf}^{n_{nf}}+{\left[Fus1\right]}^{n_{nf}}}$$ (3)

$$\frac{d}{dt}\left[Fus1\right]={\beta }_{Fus1}-{\delta }_{Fus1}\left[Fus1\right]+k_{Fus3}\frac{{\left[Fus3\right]}^p}{K_{M,Fus3}^p+{\left[Fus3\right]}^p}$$ (4)

Summary.

We demonstrate a class of genetic control systems that interface with native signaling pathways to route cells to divergent fates in response to environmental signals.