Synthetic Bistability and Differentiation in Yeast

Abstract

Engineered systems that control cellular differentiation and pattern formation are essential for applications like tissue engineering, biomaterial fabrication, and synthetic ecosystems. Synthetic circuits that can take on multiple states have been made to engineer multicellular systems. However, how to use these states to drive interesting cellular behavior remains challenging. Here, we present a cellular differentiation program involving a novel synthetic bistable switch coupled to an antibiotic resistance gene that affects growth in yeast (S. cerevisiae). The switch is composed of a positive feedback loop involving a novel transcription factor and can be switched ON and OFF via two different transient inducer inputs. By further coupling the bistable switch with an antibiotic resistance gene, we obtained a growth differentiation circuit, where yeast cells can be switched to stable HIGH or LOW growth rate states via transient inducer inputs. This work demonstrates a rationally designed and experimentally validated cellular differentiation behavior in yeast.

Synthetic biology may one day allow us to routinely engineer multicellular systems, which could revolutionize tissue engineering,¹ crop improvement,² biomaterial fabrication,³ and the design of synthetic ecosystems.⁴

Synthetic biologists have developed genetic circuits coupled with cell-to-cell communication to demonstrate the rudimentary capabilities of engineering multicellular systems. A pulsegenerator circuit coupled with cell-to-cell communication using acyl-homoserine lactone (AHL) realized spatiotemporal control of gene expression in a multicellular bacterial system.⁵ A band-detector circuit coupled with cell-to-cell communication in E. coli was engineered to form various spatial color patterns as a synthetic multicellular bacterial system.⁶ Light sensors, cell-to-cell communication, and genetic logic gates were used to implement an edge detection program as a coordinated multicellular behavior.⁷ Multicellular behavior such as quorum sensing and population level bistability were achieved via engineering haploid budding yeast cell to secrete and sense the mating pheromone a-factor.⁸ Engineered multicellular behavior of cell-to-cell communication was also achieved in yeast using auxin biosynthesis and CRISPR based transcription factors that respond to auxin.⁹

In nature, for example, in the human immune system, both B and T cells differentiate from hematopoietic stem cells.¹⁰ Cell-to-cell communications via cytokines between B and T cells coordinate defense against foreign invaders.¹¹ Both B and T cells sense specific antigens, and their proliferation is tightly controlled.¹² In synthetic biology, computational simulations of multicellular systems that involve cell differentiation, cell-to-cell communication, signal detection, and growth control have generated a variety of multicelled behaviors such as microcolony edge detection and morphogenesis.¹³

Here we demonstrate a cellular differentiation program built with a novel bistable switch coupled to an antibiotic resistance gene that affects growth in yeast (S. cerevisiae). The synthetic bistable switch is built with a positive feedback loop involving a fluorescently tagged synthetic transcriptional activator we call ZAVNY. The switch can be turned ON and OFF via two different inducers, β-estradiol (β-e) and the plant hormone auxin, respectively. The ON and OFF states are represented by high and low expression of ZAVNY. Both the ON and OFF states are stable without the continuing presence of either inducer. By coupling the bistable switch with an antibiotic resistance module, we built a simple growth differentiation program, where yeast cells can be forced to either a HIGH or LOW growth rate state in antibiotic media with the transient induction of β-e or auxin. Crucially, our goal was to have each growth state be stable even without the continuing presence of either inducer.

RESULTS

A Functional Bistable Switch Was Constructed Using a Positive Feedback Loop on a Novel Transcriptional Activator.

The core of the switch is a positive feedback loop using the novel transcriptional activator ZAVNY (Figure 1a), a chimeric, fluorescently tagged protein that is degraded in the presence of auxin. ZAVNY consists of five domains: A zinc finger DNA binding domain (ZDBD),^14,15 an auxin degron,¹⁶ a VP16 activation domain,^17,18 a nuclear localization signal (NLS),¹⁹ and an enhanced yellow fluorescent protein (EYFP)²⁰ (Supporting Information Figure S1). The ZDBD binds the pGALZ promoter, a variant of the native yeast promoter pGAL1 with its binding sites replaced with zinc finger recognition site to reduce interference with the native yeast system.¹⁴ VP16 acts as an activator and has been used in similar chimeric activators.^17,18 The NLS (2xSV40)^15,19 enables localization to the nucleus. The auxin degron, a protein subdomain from the plant Aux/IAA family, makes ZAVNY degradable by an auxin receptor (TIR1 or AFB2, both in FBox family) in the presence of auxin.¹⁶ The arrangement of these domains was inspired by the β-e inducible zinc finger transcriptional activator.¹⁴ We swapped out the Estrogen Receptor (ER) domain with auxin degron, added the NLS to make up the nuclear localization function provided by the ER, and attached EYFP in the C-terminus to allow direct assay of ZAVNY. A positive feedback loop using ZAVNY introduces bistability, and was inspired by previous theoretical work.^21,22 In principle, turning on the feedback loop would produce a stable ON state and turning off the feedback loop would produce a stable OFF state, assuming the parameters of the circuit were appropriately tuned.

Figure 1.

Circuit diagram and experimental data for the bistable switch. (a) Circuit diagram. Both TIR1^DM and ZEV are driven by constitutive yeast promoters: pACTl and pGPD. ZAVNY is degraded in the presence of auxin by TIR1^DM. (b) Control circuit without positive feedback. ZAVNY is replaced with EYFP. (c) Control circuit without auxin receptor. Auxin should not degrade ZAVNY in this circuit. (d,e,f) Time-course cytometry data under a variety of inputs. Vertical lines indicate the dilution events. Model fit resulted from the model in Figure 2. Fluorescence (a.u.) was normalized by the untransformed nonfluorescent base yeast strain MATa W303–1A. (d) Experiment data for the bistable switch under cycled inputs 10, 00, 01, 00 showing that the switch can be switched on and off repeatedly. (e) Data for the switch and control circuit without positive feedback with inputs 10 and 00, demonstrating long-term ON state stability. (f) Data for the switch and the control circuit without auxin receptor under inputs 10, 00, 01, 00, demonstrating long-term OFF state stability. (g,h,i) Non-normalized cytometry data corresponding to d, e, f presented in density plots with kernel smoothing.

We designed the switch to use two chemical inducers, β-e and auxin, so that the positive feedback loop could be turned on and off. We introduced the two components ZEV and TIR1^DM, to control the positive feedback loop (Figure 1a). ZEV is a widely used β-e inducible transcriptional activator.¹⁴ β-e induces ZEV to localize to the nucleus, activating the pGALZ promoter, thus turning the positive feedback loop on. TIR1^DM is a mutant variant of the auxin receptor TIR1 with a faster degradation rate.²³ Auxin signals TIR1^DM to degrade ZAVNY, thus turning the positive feedback loop off. Both ZEV and TIR1^DM are driven by constitutive yeast promoters. Inducing the circuit with β-e or auxin directly controls the state of the switch. Time course cytometry experiments below are consistent with our design expectations.

To test the switch, we developed time-course cytometry assays in which yeast were grown in one of each of three types of media representing the possible inputs to the switch. We use a two bit binary number to denote the input, similar to how input is represented in digital electronics.²⁴ The leftmost bit stands for the presence or absence of 100 nM β-e, whereas the rightmost bit stands for the presence or absence of 20 μM auxin. In detail, the input 00 is plain synthetic complete (SC) media, the input 10 is SC media containing 100 nM β-e, and the input 01 is SC media containing 20 μM auxin. Cells were diluted to 1 event/μL with the input media when applying an input. Short applications of an input (6–9 h) corresponded to simple batch growth in the media, whereas in longer applications, cells were periodically diluted (every 6–9 h) to 1 event/μL with the same media to keep them growing at or near log phase.

From the OFF state at 0 h (Figure 1e, h), the switch (Figure 1a) was switched into the ON state, indicated by high fluorescence, via 9 h of growth with input 10. Upon application of input 00, the switch remained in the ON state for at least 47 h, whereas the control circuit without positive feedback (Figure 1b) returned to basal levels of fluorescence within 10 h under the same conditions. The switch has a higher steady state with input 10 than with input 00, likely due to extra activation of ZAVNY by ZEV with input 10, consistent with our mathematical model prediction.

From the ON state at 16 h (Figure 1f), the switch was switched into the OFF state, indicated by low fluorescence, via application of input 01. The switch returned to basal level within 6 h. Upon application of input 00 after 12 h, the switch cells remained OFF for at least 32 h (Figure 1f,i). In contrast, the control circuit without the auxin receptor TIR1^DM (Figure 1c) could not be turned OFF (Figure 1f,i) via the application of input 01.

To test whether the switch could be cycled repeatedly, it was cycled between the ON and OFF states twice, in an 85 h experiment, by sequentially applying inputs 10, 00, 01, and 00 twice. We observed essentially the same behavior in each cycle (Figure 1d,g), with the switch ON for the first half of each cycle, and OFF for the second half of each cycle.

A Model Describes the Behavior of the Switch and Our Controls.

The observed behaviors were predicted by simple mathematical models of the switch (Figure 2a-i), the control circuit without positive feedback (Figure 2a-ii), and the control circuit without auxin receptor (Figure 2a-iii). Each model was derived from a full mechanistic model of the reactions involved, and reduced to a one-dimensional model using mass conservation and quasi-steady state assumptions (Supporting Information, Section S2). Parameters were estimated using the data shown in Figure 1, and are listed in Figure 2c. The switch model predicts two stable equilibria for the concentration of ZAVNY, separated by an unstable equilibrium (Figure 2b-i). When the concentration of β-e is increased, the equilibrium structure bifurcates to produce a single, stable, and nonzero equilibrium that is higher than the high state without the β-e, consistent with the experimental results that the switch has a higher steady state with input 10 than with input 00. When the concentration of auxin is increased, the equilibrium structure bifurcates to produce a single stable equilibrium at zero. Thus, the model predicts that pulsing the switch with β-e or auxin can push the concentration of ZAVNY between the low state and the high state, because these signals push the switch into a single equilibrium state that is either low or high. Hysteresis leads to the switch persisting in either the ON or OFF state, consistent with experimental results where β-e or auxin is diluted away. In contrast, the control without positive feedback does not exhibit a bifurcation, and having only a single stable equilibrium that merely monotonically increases with the concentration of β-e, which means that removing β-e will return the system to low state, consistent with the experimental results that this control returned to low state after β-e diluted away. In addition, the control circuit without the auxin receptor cannot be switched off once turned on because the addition of auxin has no effect on the equilibrium structure (Figure 2b-iii).

Figure 2.

Mathematical model and analysis. (a) Simplified circuit diagram and mathematical models of three experimental yeast strains: (i) the full bistable switch, (ii) the control circuit without positive feedback, and (iii) the control circuit without auxin receptor. In the ODEs, A is the concentration of ZAVNY, Y is the concentration of EYFP, v is the concentration of β-e, and u is the concentration of auxin. Parameters in these equations are explained in Figure 2c. (b) Bifurcation diagrams of the three experimental yeast strains under varying concentrations of β-e and auxin appeared in the same order as shown in (a). (c) Parameter description, estimate, and unit. The estimated values of the parameters are from the model fit of experimental data shown in Figure 1d,e,f. (d) Bifurcation diagrams for the bistable switch system for a Hill coefficient n. The dots and circles indicate the stable and unstable equilibriums given the estimated parameter values.

The Switch Was Used To Drive Antibiotic Resistance To Construct a Growth Differentiation Circuit.

A growth differentiation circuit was constructed to demonstrate that the switch could be used to construct a controllable cell differentiation circuit wherein the cell has obvious and distinct behaviors in each state. In the growth differentiation circuit, ZAVNY additionally drove the expression of the Zeocin resistance gene BleoMX²⁵ (Figure 3a). The hypothesis is that when the switch is OFF, the concentration of ZAVNY is low and BleoMX is not expressed so that the cell is not resistant to Zeocin and does not grow, noting that Zeocin concentration can be tuned to prevent growth but not completely kill cells,²⁶ whereas when the switch is ON, the concentration of ZAVNY is high, driving the expression of BleoMX so that the cell is resistant to Zeocin and it grows. Since the bistable switch can be switched between ON and OFF states repeatedly via β-e and auxin, the growth differentiation circuit could be switched repeatedly as well. A control circuit without the bistable switch circuitry was also constructed to test whether the switch drives the effect of growth differentiation or not.(Figure 3b).

Figure 3.

A growth differentiation circuit in yeast. (a) Circuit diagram of the growth differentiation circuit in yeast. (b) Circuit diagram of the control circuit without the bistable switch circuitry. (c,d,e,f) Growth rate and growth curve of the growth differentiation circuit and the control circuit without the bistable switch circuitry under series of inputs. Vertical lines indicate the dilution events. (c) Experimental data showed the switchability of the resistance memory. Cells were grown in SC media containing 400 μg/mL Zeocin plus respective inducers of β-e or auxin determined by the inputs. Growth rate was high with input 10, remained high with input 00, changed to low with input 01, remained low with input 00, and changed to back to high with input 10. (d) Experimental data showed ON state stability of the resistance memory. Growth rate remained high after input 10 with at least four steps of 00. (e) Experimental data demonstrating the OFF state stability of the resistance memory, noting that the strains were induced with 20 μM auxin to switch to OFF state before 0 h. (f) Experimental data showed that the control circuit without the bistable switch circuitry did not hold the ON state. The growth rate dropped from high to low after input 10 was changed to input 00 at 20 h.

To test the growth differentiation circuit, we developed time-course plate reader assays to measure growth rates under different inputs. SC media additionally containing 400 μg/mL Zeocin was used as base media for all inputs. Similar to the bistable switch input notation, input 00 contained only this base media, input 10 contained 100 nM β-e, and input 01 contained 20 μM auxin. The growth differentiation circuit cells were cycled through the inputs 10, 00, 01, 00, and 10, as shown in Figure 3c. The observed growth rate was high when the switch was supposed to be ON, while the observed growth rate was low to zero when the switch was supposed to be OFF. This behavior is consistent with the hypothesis that once BleoMX reaches a low enough level, Zeocin functions to prevent cell growth. The growth differentiation circuit cells were able to “remember” a pulse of β-e for at least four subsequent dilution steps using input 00 (Figure 3d). In contrast, our control circuit without bistable switch, did not “remember” a pulse of β-e, and showed no growth within two steps after the removal of β-e (Figure 3f). We did not observe any growth differentiation circuit cells able to escape the OFF state after five steps (Figure 3e).

Minimal Changes to the Bistable Switch Model Describe the Growth Differentiation Circuit.

Minimal changes to the model described the growth differentiation circuit. In particular, we developed a model of the growth differentiation circuit that included the bistable switch model above, and also included equations for the rates of change of the BleoMX protein, the population size, and the nutrient concentration (Figure 4a). The rate of change of the population was assumed to increase as nutrients and BleoMX increased. This model was fit to the data and used to estimate the concentrations of ZAVNY and BleoMX (Figure 4c), and the results were consistent with our expectation that ZAVNY drives BleoMX and BleoMX promotes growth. A model of the control circuit with no positive feedback also produced the expected result in which the concentration of BleoMX dilutes away quickly after activation by ZAVNY is removed, crashing the population.

Figure 4.

Mathematical model and growth curve fitting of the growth differentiation circuit. (a) Simplified circuit diagram and mathematical models of two experimental yeast strains: (i) growth differentiation circuit, and (ii) control circuit without bistable switch. (b) Description, estimate, and unit of parameters shown in the mathematical models in addition to the parameters of the bistable switch in Figure 2c. The estimated values of the parameters are from the model fit of experimental data shown in Figure 3c–f. (c,d,e,f) Model prediction and growth curve fitting for the time course plate reader assay. The red line in each figure shows the model fit with parameters estimates shown in b. Simulated ZAVNY and BloeMX concentration dynamics are shown in the upper portion of each subfigure. Vertical lines indicate the dilution events.

DISCUSSION

A synthetic circuit that allows persistent, predictable control of cell states would be a key enabler for engineering multicellular systems. This study presented a cellular differentiation program built with a bistable switch connected with antibiotic resistance in yeast. The bistable switch consists of a single positive feedback loop on a novel transcriptional activator and two additional components for turning it ON and OFF with two external signals β-e and auxin. When the switch was connected to drive an antibiotic resistance gene, the cells differentiated from a LOW growth rate state to a HIGH growth rate state when pulsed with β-e and from the HIGH growth rate state to the LOW growth rate state when pulsed with auxin.

Many synthetic bistable switches are formed from the mutual inhibition of two transcriptional repressors.^27,28 In contrast, the single positive feedback motif is ubiquitous in cells²⁹ and in some cases supplies the cell with a memory to maintain gene expression. Single positive feedback regulation has been used previously to generate a bimodal distribution of cell phenotype,³⁰ cellular memory,³¹ and a hysteretic switch,³² where various methods of switching on the positive feedback loop with an external signal were developed. However, no clear method existed for switching off previously published positive feedback loops. In our switch, the auxin degron provides the ability to switch to the OFF state by signaling the degradation of the transcription factor itself.

Our mathematical model suggested a Hill coefficient greater than 1.88 is needed for the switch to be bistable. The model fit estimated a Hill coefficient of 2, which suggested there might be cooperativity of 2 for ZAVNY interaction of pGALZ. Previous literature suggests that single zinc finger binding domain and pGALZ with single zinc finger binding site has a Hill coefficient around 1.¹⁴ We did not investigate the molecular mechanism of how cooperativity arises in the ZAVNY interaction with pGALZ. There are additional domains such as the auxin degron and EYFP added in ZAVNY compared with previous transcriptional activators in the literature,^14,15 so these additional domains and/or interactions with others could contribute to the cooperativity. We think this may be of interest for a follow-up study.

We explored several means of connecting the switch to genes in yeast that control growth. One example was a mutant FAR1 gene, which resulted in growth arrest state when the switch is in ON state. However, cells expressing the mutant FAR1 were not able to recover when the bistable switch turned to OFF state. Our hypothesis was that the mutant FAR1 level still remained in the arrested cells due to little protein degradation activity and no cell growth even though the switch is turned off. In contrast, Zeocin concentration in our switch coupled with antibiotic resistance gene design gave us a crucial tuning knob to achieve desired recoverable growth differentiation behavior. The concentration of Zeocin was chosen based on a dose–response experiment shown in Supporting Information Figure S4 where the ON and OFF growth rates differ the most while in the OFF state Zeocin did not kill the cells completely but prevents growth.²⁶

Most synthetic circuits in engineered multicellular systems only involve a few genes, far fewer even than the early stages of Drosophila embryo development, which involves 46 genes in a 3 h window.³³ To scale up synthetic circuits, one could construct n bistable switches each with a positive feedback loop with n transcriptional activators, each of which could also regulate a downstream gene responsible for a certain role in cell development. Since each bistable switch can code two states, the number of states that could theoretically be obtained is thus 2ⁿ. Constructing n transcriptional activators is realizable with current technologies such as CRISPR³⁴ or TALE.³⁵ Constructing orthogonal or multiplexed signaling systems that control a large number of states remains a difficult but seemingly surmountable challenge, for example, one may be able to use RNA to program orthogonal pairs of signals.³⁶

METHODS

Plasmid Construction.

All plasmids used are listed in Supporting Information Table S1 with publicly available sequences linked in Benchling.³⁷ They were constructed on ampicillin resistant backbone vector via Gibson assembly,³⁸ amplified in E. coli DH5α, extracted using QIAGEN miniprep kits, and sequence verified with Genewiz. All plasmids were constructed through the plasmid construction process in Aquarium,³⁹ a software-enabled human-in-the-loop automation platform developed in the Klavins lab. Technicians in the University of Washington Biofabrication Center (UW BIOFAB),⁴⁰ an Aquarium enabled cost center for providing molecular cloning services, carried out the lab procedures.

Strain Construction.

All yeast strains used are listed in Supporting Information Table S2 by their reference name along with the plasmids that were integrated into their genome. The plasmids listed in each strain were digested by PmeI restriction enzyme from NEB to get linearized fragments and then sequentially integrated into the yeast genome using a standard lithium acetate yeast transformation protocol.⁴¹ The base yeast strain used for the yeast transformation is MATa W303–1A, a gift from the Gottschling laboratory. All yeast strains were constructed through the yeast cloning process in Aquarium, and carried out by technicians in the UW BIOFAB.

Media Preparation.

Synthetic complete (SC) media was used as the base yeast media across the time-course cytometry assays. Synthetic complete (SC) media with 400 μg/mL were used as the base yeast media across the plate reader assays. For both cytometry and plate reader assays, the input 00 was base yeast media, the input 10 was base media containing 100 nM β-e, and the input 01 was base media containing 20 μM auxin. β-e was made from β-Estradiol dry stock, and auxin was made from 3-Indoleacetic acid dry stock, both ordered from Sigma-Aldrich.

Time-Course Cytometry Assays.

To prepare each assay, for each experimental strain, three freshly grown colonies on YPD plate were inoculated in SC media in a 96 deepwell plate, and then grew in 30 °C shaker incubator for 12-h. For the bistable switch strain, 20 μM auxin were added in the media after inoculation, while other procedures kept the same. This was to prepare the bistable switch strain to be at OFF state at 0 h in the assay. After a 12-h incubation, cytometry measurements were performed for each culture. Cell density was estimated using cytometry data gated for yeast singlets population by a custom Python package named Cowfish.⁴² Each culture was then diluted to 1 event per μL in its respective input media. Immediately after dilution, yeast cell measurements were performed over flow cytometer to obtain the 0 h data. Subsequent measurements were acquired every 6–12 h according to the experimental charts (Figure 1d,e,f). Each culture was diluted to a constant cell density of 1 event per μL in the respective input media when applying each input step.

Flow Cytometry.

For each measurement, 100 μL of each sample in the 96 deepwell plate was transferred to a Costar 96 well assay plate and placed on the BD Accuri C6 flow cytometer plate adapter. Yeast cell measurements including fluorescence, size, and count were obtained using excitation wavelengths of 488 and 640 nm and an emission detection filter at 533/30 nm (FL1 channel). A total of 10 000 events above a 400 000 FSC-H threshold (to exclude debris) were measured for each sample at a flow rate of 66 mL/min and core size of 22 mm using the Accuri C6 software. Data were exported as FCS 3.0 files and processed by the Python package Cowfish.

Time-Course Plate Reader Assays.

The preparation step is similar to the time-course cytometry assays. Three freshly grown colonies from YPD plate for each strain were inoculated in SC media in a 96 deepwell plate, and then grew in 30 °C shaker incubator for 12 h. For the growth differentiation circuit strain, 20 μM auxin were added in the media after inoculation to make sure the switch was switched to OFF state. After 12-h incubation, cell density was estimated using cytometry data gated for yeast singlets population by the Python package Cowfish. Yeast culture was then diluted to 1 event per μL in 1 mL of respective input media in a VWR 24-well tissue culture plate. The plate was placed in a Biotek Synergy HT plate reader at 30 °C with shaking and measured every 15 min. Yeast culture was diluted every 20–24 h to a constant cell density of 1 event per μL in the respective input media according to the experimental chart (Figure 3c,d,e,f).

Plate Reader Measurements.

Yeast cells were cultured in a VWR 24-well tissue culture plate with lid covered in a Biotek Synergy HT plate reader at 30 °C with shaking. Cell culture optical densities (OD, 600 nm) were taken every 15 min through scheduling from the Gen5 software. Data were exported as tab-delimited text file and processed by custom Python scripts to fit exponential growth curve and calculate growth rate.