key: cord-0264926-24n1riwk
authors: Palme, Julius; Wang, Jue; Springer, Michael
title: Variation in the modality of a yeast signaling pathway is mediated by a single regulator
date: 2020-08-06
journal: bioRxiv
DOI: 10.1101/131938
sha: a57f9b63fd783cf473196f02fb62c50d1418de70
doc_id: 264926
cord_uid: 24n1riwk

Bimodal gene expression by genetically identical cells is a pervasive feature of signaling networks. In the galactose-utilization (GAL) pathway of Saccharomyces cerevisiae, induction can be unimodal or bimodal depending on natural genetic variation and pre-induction conditions. Here, we find that this variation of modality is regulated by an interplay between two features of the pathway response, the fraction of cells that are in the induced subpopulation and their expression level. Combined, the variations in these features are sufficient to explain the observed effects of natural variation and pre-induction conditions on the modality of induction in both mechanistic and phenomenological models. Both natural variation and pre-induction conditions act by modulating the expression and function of the galactose sensor GAL3. The ability to alter modality may allow organisms to adapt their level of “bet hedging” to the conditions they experience, and thus help optimize fitness in complex, fluctuating natural environments.

Non-genetic heterogeneity is a pervasive feature of gene expression and cellular signaling (1-3). Bimodal 26 responses, where cells in an isogenic population adopt one of two distinct states, are particularly 27 important for microbes coping with fluctuating environments (4, 5) and cells of multicellular organisms 28 differentiating into discrete types (6, 7). 29

The galactose-utilization (GAL) pathway in Saccharomyces cerevisiae is a well-characterized bimodal 30 response and a classic model of microbial decision-making (8, 9) . Bimodality of GAL gene expression 31 has been attributed to bistability arising from positive feedback through the Gal1p kinase and the Gal3p 32 transducer (10, 11). Perturbing many of the components of the GAL pathway such as the Gal2p permease, 33 the Gal4p activator, and the Gal80p repressor have been found to affect quantitative features of the GAL 34 response (11-14) and in principle could affect the feedback in the system. However, only changes in 35

Gal1p and Gal3p (10, 11) have been shown to affect modality, i.e. whether the response is bimodal or 36 unimodal. 37

Our existing insight into modality in the GAL system comes almost entirely from measuring one pathway 38 phenotype, the induced fraction, under one environmental perturbation, galactose titration (10, 13, 15-39 17). The few studies that have deviated from this experimental approach have resulted in observations 40 that raise new questions. For example, the GAL response was found to be unimodal or bimodal 41 depending on the carbon source prior to encountering galactose (18); the molecular basis of this behavior 42 is unknown. In our own previous work, we found that natural yeast isolates differ in the inducibility of 43 GAL genes in mixtures of glucose and galactose (17, 19) , and that some strains showed a bimodal 44 response while others had a unimodal response (17). This natural variation provided an opportunity to 45 dissect the genetic variants modulating bimodality in nature and gain insight into the quantitative 46 behaviors organisms have evolved to respond to their environments. 47

In this work, we confirm and expand the observation that the pattern of GAL pathway induction can be 48 either unimodal or bimodal depending on genetic background and pre-induction conditions. A 49 computational model of GAL induction led us to a conceptual framework for variation in modality that 50 identified the induced fraction and the threshold for glucose inhibition of expression as the two critical 51 factors. Using this simple framework, we can explain the variations in modality we observed, and predict 52 how new perturbations affect modality. Finally, we have determined that both natural variation and pre-53 induction conditions achieve changes in modality by tuning the expression and activity of a single 54 signaling protein in the GAL pathway. These results reveal a simple evolutionary mechanism by which 55 organisms can shape their responses to the environment, and suggest that modality is a highly adaptable 56 feature of a signaling response. 57

Results 58 Genetic and environmental factors affect the modality of the GAL response 59 To study natural variation in the modality of the GAL response, we measured the expression of a GAL1 Figure S2 ). We also found that 67 growth history affected modality. For example, the laboratory strain S288C showed unimodal GAL 68 induction when grown with raffinose as a carbon source prior to encountering mixtures of glucose and 69

galactose, but gave a bimodal response when mannose was used as the initial carbon source instead 70 ( Figure 1D, (18) ). These observations suggest that the modality of the GAL pathway is variable if glucose 5 is present as well as galactose. However, it is unclear how adding a second input to a bimodal system can 72 cause this qualitative change in pathway induction. Bimodality in the GAL pathway is commonly attributed to bistability arising from positive feedback. To 92 investigate the mechanism by which glucose regulates bistability (and thus the modality) of GAL 93 induction, we adapted a previous ODE model that described the regulation of the GAL pathway by 94 glucose and galactose (15). In this model, Gal80p binds the transcription factor Gal4p and keeps it in an 95 inactive state. In the presence of galactose, Gal3p binds Gal80p, freeing Gal4p from the Gal80p-Gal4p 96 complex. Bistability can arise in this system as a result of transcriptional feedback loops where free Gal4p 97 leads to the production of Gal3p and Gal80p (10). Conversely, glucose inhibits GAL activation as it 98 reduces the intracellular galactose concentration through competition for binding to hexose transporters 99 (22) and activates Mig1p which in turn decreases the production of Gal3p and Gal4p (Figure 2A) . 100

To analyze how glucose affects bistability in this model, we simulated the ODE system in different 101 concentrations of glucose and galactose. Since Gal4p drives the expression of GAL genes, we used the 102 steady-state concentration of free Gal4p as a read-out for pathway activity. To determine bistability in the 103 system, we started all simulations from two extreme initial conditions, the steady-state concentrations of 104 components in pure galactose or pure glucose. We then simulated switching to a mixture of sugars, and 105 analyzed whether these initial conditions led to different steady-state levels of components in the new 106 environment. As expected, the system displays bistability when galactose is titrated in the absence of 107 glucose ( Figure 2B ). Next, we analyzed the effect of titrating glucose in the presence of a constant 108 concentration of galactose, mirroring our experimental approach ( Figure 1C, 1D ). In the model, the 109 addition of glucose converts the system from bistable to monostable ( Figure 2C ), indicating that the ODE 110 system recapitulates the experimentally observed effect of glucose addition. To investigate the molecular 111 basis for the switch, we then analyzed the effects of glucose in the simulations. Glucose has two effects: 112 1) it directly increases the activity of the transcriptional repressor Mig1p and 2) it indirectly decreases the 7 increasing glucose decreases both the total transcriptional output of GAL3 and GAL4 and the fraction of 115

Gal3p that is in the active form ( Figure 2C ). The indirect effect of glucose on Gal3p activity is equivalent 116 to the direct effect of galactose on Gal3p activity. As the system is bistable in pure galactose, this indirect 117 effect cannot be the molecular cause of the change in bimodality; therefore, glucose must eliminate 118 bistability through Mig1p. 119

Molecularly, bistability occurs when the system is near the activation threshold of Gal3p (10). Noise in 120 the level of active Gal3p presumably leads to some cells to be above the activation threshold while others 121 are below it. This difference is amplified by positive feedback, leading to a bimodal distribution. A 122 second requirement is that Gal4p must be expressed at sufficient levels near the Gal3p activation 123 threshold, so that Gal4p mediated positive feedback can occur. Figure 2C shows that when Gal4p 124 abundance is low near this threshold, the system is monostable. To further test this hypothesis, we varied 125 the Michaelis constants of parameters that affect glucose-dependent inhibition of Gal4p production (see 126 Methods). These simulations confirmed that the system is bistable only when the concentration of Gal4p 127 is sufficient to drive positive feedback upon activation and the system is near the Gal3p activation 128 threshold ( Figure 2D ). In contrast, varying a parameter that controls the Mig1p inhibition of GAL3 had no 129 effect on whether the simulations were monostable or bistable ( Figure S3 ). 130 Figure S4 ). We hypothesized that a simpler phenomenological model focusing only on two 142 free parameters should be able to recapitulate the variation in modality we experimentally observed. We 143 therefore constructed a model in which the fraction of active Gal3p determines the fraction of cells that 144 are in the induced state and the amount of free Gal4p determines the GAL1 expression level of the 145 induced subpopulation ( Figure 3A ). Mathematically, we described these two behaviors as Hill functions 146 that decrease with increasing glucose concentration. To simulate induction profiles, we generated 147 population distributions at different glucose concentrations by using the induced fraction to set the 148 relative size of the two populations, the expression level to set the mean level of the induced 149 subpopulation, and a normal distribution based on experimental noise to set the spread in expression. This 150 simplified model focuses on how modality is affected by differences in the pathway's responses to sugar 151 ratio and glucose repression. 152

To confirm that the simplified model provides the same qualitative results as our molecular model, we 153 varied the glucose threshold for the expression level of induced cells while keeping the glucose threshold 154 for the induced fraction constant. In agreement with the simulations from our molecular model, the 155 induction profile is unimodal when the induced fraction changes at concentrations of glucose that still 156 inhibit expression level ( Figure 3B , top). Conversely, the induction profile is bimodal when the induced 157 fraction is in a regime of greatest change while the expression level is not inhibited by glucose ( Figure  158 3B, bottom). Next, we varied the glucose threshold for the induced fraction while keeping the glucose 159 threshold for the expression level constant. Again, these changes were sufficient to change modality 160 ( Figure 3C ). Together, these results show that varying the relative position of induced fraction and 161 expression level regulation in a phenomenological model is sufficient to recapitulate changes in modality. 162 factors that might explain these erroneous predictions. First, to measure the F 90 , there must be measurable 193 gene induction. Therefore, the calculated F 90 is a lower bound for the actual value of F 90 in unimodal 194 strains; in many cases the calculated F 90 will be higher than the actual F 90 . If the calculated F 90 for 195 simulations of unimodal strains is increased by even a factor of 2, the predicted bimodal area disappears 196 ( Figure S8 ). Second, the slopes of the induced fraction curves could also be subject to variation and affect 197 modality. Increasing the steepness of the induced fraction curve or induced level curves by increasing the isolates appear to have steeper induction curves ( Figure S10 ). These two explanations are not mutually 200 exclusive, and we believe that these minor errors do not substantially detract from the value of our model. 201

We next scanned the parameters for our phenomenological model using a wide range of summary metrics 202 and slopes to delimit a phase diagram of GAL induction modality. We then compared this phase diagram 203 to the experimental data from natural isolates ( Figure 4E ). The modality of the natural isolates agrees well 204 with their predicted modality in the phase diagram, supporting our hypothesis that the E 10 and F 90 205 measurements capture the important biological features that determine the modality of a strain. 206

Our ODE simulations gave us a molecular explanation for unimodality: Mig1p activity leads to 207 unimodality by inhibiting GAL4 expression in the regime where Gal3p activation is increasing. Therefore, 208 an important prediction of the model is that removing Mig1p regulation should restore bimodality in 209 unimodal strains. To test this prediction, we analyzed the induction profile of a mig1Δ strain. Deleting 210 MIG1 removes the glucose dependent regulation of GAL4 expression level and thus the E 10 of the deletion 211 strain is increased compared to the wild-type strain ( Figure 4F ). As predicted, it also converts the strain 212 from unimodal to bimodal. In addition, the observed F 90 value for the bimodal mig1Δ strain is higher than 213 that for the unimodal wild-type. This supports our hypothesis that Mig1p-dependent repression conceals 214 the actual F 90 value of unimodal strains and that the observed F 90 value of unimodal strains is only a lower 215 bound for the actual F 90 value. 216 History-dependence of modality can be explained by changes in the induced fraction 234 It has previously been reported that pre-induction growth conditions can affect the modality of GAL 235 induction (18), which offers another opportunity to test the predictions from our modeling framework. To 236 see how metabolic history affects F 90 and E 10 , we grew 13 natural isolate strains in mannose, raffinose, 237 acetate, or glycerol prior to transferring them into mixtures of glucose and galactose ( Figure S11 ). There 238 are three broad categories of strain responses to pre-induction conditions. The first category contains 239 strains that are unimodal when pre-grown in raffinose that become bimodal when pre-grown in mannose 240 ( Figure 5A , 5B). The second is strains that are bimodal when pre-grown in raffinose but become 241 unimodal when pre-grown in glycerol or acetate ( Figure 5C, 5D) . The last is strains that are always 242 bimodal or always unimodal in all pre-induction carbon sources tested here ( Figure 5E , 5F). Analyzing 243 the effect of pre-induction conditions on our summary metrics, we found that the average fold change 244 between the lowest and highest F 90 values of each strain is 8.61 while the average fold change for the E 10 245 is 1.78 ( Figure S12 ). Our heuristic model predicts that the magnitude of the changes in To determine how pre-induction carbon source modulates F 90 , we measured the expression of GAL genes 260 in pre-induction conditions using transcriptional reporters ( Figure 6A ). We found that GAL genes are 261 down-regulated in carbon sources that lead to bimodal induction (mannose) and up-regulated in carbon 262 sources that lead to unimodal induction (acetate, glycerol). Amongst all GAL genes, the expression levels 263 of GAL3 and GAL4 show the strongest fold change between the carbon sources tested ( Figure 6A ). 264

We hypothesized that GAL3 was more likely than GAL4 to be the dominant factor due to its high dynamic 265 range of expression ( Figure 6A ) and prior evidence for GAL3 concentration affecting the GAL decision 266 (13). We therefore analyzed the regulation of GAL3 by a range of pre-induction carbon sources in 9 267 natural isolates using a transcriptional reporter for the GAL3 S288C promoter. We found that in each strain 268 the GAL3 expression level in a pre-induction conditions generally correlates with the modality observed 269 later ( Figure 6B ). 270

To test if GAL3 expression prior to induction is the key determinant of modality, we used a tetracycline-271 inducible promoter to control the expression of GAL3 directly. We predicted that forcing a change in 272 GAL3 expression while keeping the pre-induction carbon the same should change modality. Conversely, 273 changing the pre-induction carbon without changing GAL3 expression should not change modality. 274

For the laboratory strain S288C, pre-induction growth in mannose leads to low GAL3 expression and a 275 bimodal induction profile, while pre-induction growth in raffinose leads to higher GAL3 expression and a in mannose using tetracyclin induction, we saw an increase in the induced fraction and a loss of 278 bimodality ( Figure 6C) . Thus, the GAL3 concentration pre-induction is sufficient to set the modality in 279 this strain background. Artificially setting the GAL3 level of mannose pre-induction cultures to that of a 280 raffinose pre-induction culture converted the induction profiles to one similar to a raffinose pre-induction 281 culture ( Figure 6C, II and IV) . 282

In addition to showing unimodal induction after raffinose pre-induction, this strain also shows unimodal 283 behavior after pre-induction in acetate or glycerol (Fig. 5A) . We deleted the endogenous GAL3 gene 284 from our tetracycline induction strain, so that we could force pre-induction GAL3 expression to remain 285 low in these sugars. As predicted, when pre-induction GAL3 expression in these sugars was low we saw 286 bimodal induction almost identical to that seen with mannose pre-induction (Fig. 6D ). Pre-induction 287 GAL3 concentrations also set the induced fraction with almost no dependence on the pre-induction carbon 288 source ( Figure 6D ). We conclude that regulation of GAL3 expression in pre-induction conditions is the 289 major driver of history-dependence in the modality of GAL induction. 290 The central role of GAL3 expression in setting the modality of induction suggests that natural variation in 308 GAL3 alleles could be responsible for the observed differences in modality between isolates ( Figure 1C) . 309

Previously, we showed that polymorphisms in the GAL3 gene explain most of the natural variation in the 310 decision to induce the GAL pathway (17), i.e. the F 90 (17), suggesting that allele swaps of the GAL3 ORF 311 should alter the F 90 of the strain, which in some cases would be enough to switch the modality of 312 induction. To test this prediction, we determined the modality of a set of 30 allele swap strains comprised 313 of 10 GAL3 alleles in 3 genetic backgrounds. The experimentally determined modality of all allele swap 314 strains agrees with the expected modality in each case ( Figure 7A , Figure S13 ). For example, replacing 315 the GAL3 allele of two unimodal strains, BC187 and S288C, with the GAL3 allele of any of our bimodal 316 strains is sufficient to change the induction profiles from unimodal to bimodal ( Figure 7B, 7C) . In 317 agreement with our model, the F 90 in these strains decreases sufficiently to convert strains from unimodal 318 to bimodal. Replacing the GAL3 gene of the bimodal strain YJM978 with the alleles of unimodal strains 319 does not change the modality ( Figure 7D ). This is also in agreement with our model which predicts that 320 the magnitude of the increase in F 90 caused by allele swaps in the YJM978 strain background are 321 insufficient to change modality ( Figure 7D ). 322

To further explore the variations in modality amongst the natural isolates, we analyzed the contribution of 323 promoter and coding sequence variation. We found that SNPs in either promoter or coding regions (CDS) 324 are sufficient to change modality ( Figure 7E-F) . For example, S288C with GAL3 YPS606 is unimodal, but 325

replacing the YPS606 promoter in this strain with the YJM421 promoter leads to bimodal induction with the YJM978 coding sequence leads to bimodal induction ( Figure 7F ). The mechanisms by which 328 promoter and CDS changes are able to change the modality will be the subject of future work. conditions. Here, we show that the modality of GAL induction in different strains depends on the relative 342 position of the sugar concentration thresholds at which induced fraction and expression level are 343 regulated. Glucose inhibits the expression of GAL genes, preventing the positive feedback that is crucial 344 for bistability. In general, any input to a pathway that can create or eliminate feedback has the potential to 345 modulate bistability. Since many signaling responses are controlled by multiple inputs, our findings imply 346 that other unimodal responses could be bimodal in different conditions and vice versa. 347

Bimodality in the GAL response is considered a bet-hedging strategy where a fraction of the population 348 prepares for glucose depletion by inducing the GAL pathway while other cells maximize their current 349 growth and do not induce the pathway (15, 19) . This heterogeneity helps populations deal with uncertain, 350 fluctuating environments. Bet-hedging is advantageous in the GAL system when the switching rate 351 between glucose and galactose environments is high (24); indeed, cells evolve bimodality in MAL gene 352 expression when they are continuously switched between glucose and maltose (25). Because cells can 353 sometimes be in environments with a high switching rate and other times in environments with a low 354 switching rate, a strategy that allows the extent of bet-hedging to be tuned could be the most optimal. In 355 this work, we show strains can tune the amount of bimodality both physiologically, based on their 356 metabolic history, and genetically, presumably based on the environmental statistics that different natural 357

isolates have faced in their evolutionary history. Further work will be needed to determine the 358 evolutionary consequences of tunable bimodal responses such as the ones we characterize here.

Previous work on cell-to-cell heterogeneity has typically emphasized the complex genetic architecture of 360 the pathways involved (26, 27). In contrast, the physiological and genetic variation in modality in the 361 GAL pathway can be explained by changes in the behavior of a single protein. Modulating levels of 362 Gal3p, the intracellular galactose sensor, is sufficient for changing modality. Swapping the GAL3 alleles 363 of natural isolates can turn a unimodal strain into a bimodal strain. We show that the environment tunes 364 the expression level of GAL3, and this tuning is sufficient to change the modality of GAL induction 365 (Figure 7 and Figure S13 ). Circuit designs such as these, where a single gene controls modality, may have 366 been selected in evolution, since they allow cells to easily adapt their behavior on both physiological and 367 evolutionary timescales. 368

The control of GAL pathway modality by mannose, raffinose, glycerol, and acetate suggests an addition 369 layer of metabolic regulation that has been largely missed in previous analyses of this pathway. These 370 findings show that factors other than canonical glucose catabolite repression can be important in 371 determining the inducibility of GAL genes, consistent with our findings that many mutants outside the 372 GAL pathway can have a significant effect on GAL response (28). The fact that pre-induction carbon 373 sources mostly affect F 90 , just as GAL3 allele swaps do, suggests that the GAL3 positive feedback loop 374 may be a nexus of regulation of GAL genes by multiple signals in the cell. In future studies, 375 understanding the metabolic regulation of this well-studied system could give insight into the connections 376 between metabolism and metabolic signaling in a variety of systems. 377 Methods 378 Strains and media 379 Strains were obtained as described in (17, 19 ). An initial set of 36 strains were assayed in a glucose 380 gradient (1% to 0.0039%) with a constant background of 0.25% galactose. Strains DBVPG6765, 381 CLIB324, L-1528, M22, W303, YIIC17-E5 were excluded from downstream analysis due to poor growth 382 in our media conditions. Strain 378604X was also excluded due to a high basal expression phenotype that 383 was an outlier in our collection. The genetic basis of this behavior is likely an interesting topic for follow-384 up studies. All experiments were performed in synthetic minimal medium ("S"), which contains 1. Then, cells were harvested and fixed by washing twice in TE and resuspended in TE before transferring to 401 microtiter plate for measurement. Flow cytometry was performed using a Stratedigm S1000EX with 402 A700 automated plate handling system. 403 GAL3 titration in pre-induction conditions 404 To titrate GAL3 levels in the presence of the native GAL3 gene, the AGA1 gene was replaced with a 405

MYO2pr-rtTA-TetO7pr-GAL3 construct in a hoΔ:GAL1pr-YFP strain. Cells were grown for 16 hours in 406 S + 2% mannose as described above, but the medium was supplemented with doxycycline 407 (MilliporeSigma) concentrations ranging from 38.9 μ g/ml to 0.0176 μ g/ml in 1.5x dilutions steps. To measure the total GAL3 expression level after pre-induction growth, the AGA1 gene was replaced with a 409

MYO2pr-rtTA-TetO7pr-YFP construct in a hoΔ:GAL3pr-YFP reporter strain. After pre-induction growth 410 in the same dilution doxycycline concentrations, cells were harvested and YFP levels were determined 411 using flow cytometry as described above. 412

To titrate GAL3 levels in the absence of the native GAL3 gene, the AGA1 gene was replaced with a 413

MYO2pr-rtTA-TetO7pr-GAL3-mScarlet construct in a gal3Δ hoΔ:GAL1pr-YFP strain. Cells were grown 414 for 16 hours in S + 2% pre-induction carbon source as described above, but the medium was 415 supplemented with doxycycline concentrations ranging from 38.9 To determine the modality of GAL induction experiments, a Gaussian function was fitted to the 424 population distribution for each of the 9 sugar combinations. If the degree-of-freedom adjusted R 2 of the 425 fit was less than 0.99, two Gaussian functions were fitted to the data. Distributions were then determined 426 to be bimodal if the distance between the means of the Gaussians was more than twice of the highest 427 standard deviation of the Gaussian (as in (10)) and the fraction of the smaller Gaussian was higher than 428 0.15. GAL induction experiments or simulations that had a bimodal distribution in at least one 429 combination of glucose and galactose in all replicates were called bimodal. 430 ODE model 431

The ODE system from (15) was implemented using the G1 term to describe Gal3p and the R term to 432 describe Mig1p with the parameters described in Supplementary Table 1. To determine the intracellular  433 galactose concentration (α gal ), the extracellular galactose concentration was divided by the extracellular 434 glucose concentration and multiplied with a conversion factor μ . Steady-state species levels were obtained 435 by simulating the system using the ode function of the R package deSolve in different levels of galactose 436 (logarithmically spaced between 10 0 to 10 4 ) and glucose (logarithmically spaced between 10 -2 to 10 1.5 437 with a constant galactose level of 500 The function was scaled from -3 to -0.5 to match the range of the experimental data. To obtain realistic 446 versions for the n constant, this function was fitted to the induced level curves of natural isolates, the 447 mean fitted n value was extracted for every natural isolate, and the mean of these values was used for 448 simulations (induced level curve: 1.15, induced fraction curve: 1.69, see Figure S11 ). 449

For the induced fraction, the following function was used: 450 Figure S1 . GAL induction of natural isolates in different galactose concentrations. Each plot represents 9 histograms with color intensities corresponding to the density of cells with a given fluorescence value (normalized by side-scatter (SSC)). Galactose concentration is titrated in two-fold steps from 1% to 0.0039%.

s 9 ue to Figure S2 . GAL induction of natural isolates in different glucose and galactose concentrations. Each plot represents 9 histograms with color intensities corresponding to the density of cells with a given fluorescence value (normalized by side-scatter (SSC)). Glucose concentration is titrated in two-fold steps from 1% to 0.0039%, galactose concentration is constant at 0.25%. Orange isolate names indicate unimodal induction, blue isolate names indicate bimodal induction. Figure S3 . Bistability was assessed as described in Figure 2 . The color indicates the fraction of glucose concentrations that showed bistability. re se he Figure S5 . Identification of the induced subpopulation. A reference distribution from 2% glucose (black histogram) is subtracted from the population distribution (orange histogram) to yield the induced subpopulation (orange shading).

ck ed Figure S6 . Experimental and simulated GAL induction profiles of bimodal strains. Simulations were performed as shown in Figure 3 . Glucose concentration is titrated in two-fold steps from 1% to 0.0039%, galactose concentration is constant at 0.25%. re %, Figure S7 . Experimental and simulated GAL induction profiles of unimodal strains. Simulations were performed as shown in Figure 3 . Glucose concentration is titrated in two-fold steps from 1% to 0.0039%, galactose concentration is constant at 0.25%. re %, Figure S8 . Experimental and simulated GAL induction profiles of unimodal strains with higher F 90 .

Simulations were identical to those in Figure S6 , but the experimentally determined F90 was increased two-fold for simulation purposes.

. ed Figure S9 . Experimental and simulated GAL induction profiles of unimodal strains with varying steepness. Simulations were identical to those in Figure S6 , but the n parameter of the functions (see Figure 3 and Methods) was increased to the highest value that was observed when the functions were fitted to experimental data (induced level curve: 1.72, induced fraction curve: 3.10, see Figure S10 ). Figure S11 . GAL induction of natural isolates in different glucose and galactose concentrations after growth in different carbon sources. Each plot represents 9 histograms with color intensities corresponding to the density of cells with a given fluorescence value (normalized by side-scatter (SSC)). Glucose concentration is titrated in two-fold steps from 1% to 0.0039%, galactose concentration is constant at 0.25%. Orange titles indicate unimodal induction, blue titles indicate bimodal induction. Figure S12 . Fold change between the highest and lowest E10 and F90 values after growth in different preinduction conditions for all isolates shown in Figure S11 . Gray bars indicate the means of the maximal fold changes.

al Figure S13 . GAL induction of allele swap strains in different glucose and galactose concentrations. Each plot represents 9 histograms with color intensities corresponding to the density of cells with a given fluorescence value (normalized by side-scatter (SSC)). Glucose concentration is titrated in two-fold steps from 1% to 0.0039%, galactose concentration is constant at 0.25%. Plot titles indicate the source of the GAL3 allele. Orange titles indicate unimodal induction, blue titles indicate bimodal induction. ch en ps he

Stochasticity in gene expression: from theories 475 to phenotypes

Cellular decision making and biological noise: 477 from microbes to mammals

Nature, nurture, or chance: stochastic gene expression and its 479 consequences

Microbial bet-hedging: the power of 481 being different

Bistability, epigenetics, and bet-hedging in bacteria

A positive-feedback-based bistable "memory module" that governs 485 a cell fate decision

Systems biology of stem cell fate and cellular 487 reprogramming

A model fungal gene regulatory mechanism: the GAL genes of Saccharomyces 489 cerevisiae

Synergistic dual positive feedback loops 492 established by molecular sequestration generate robust bimodal response

Enhancement of cellular memory by reducing 495 stochastic transitions

The regulatory roles of the galactose permease and kinase in the 497 induction response of the GAL network in Saccharomyces cerevisiae

A general mechanism for 500 network-dosage compensation in gene circuits

Dual feedback loops in the GAL regulon suppress cellular heterogeneity 502 in yeast

Population diversification in a yeast 504 metabolic program promotes anticipation of environmental shifts

Evolution of gene network activity by tuning the strength 506 of negative-feedback regulation

Polymorphisms in the yeast galactose sensor underlie a natural continuum of 508 nutrient-decision phenotypes

Cell signaling can direct either binary or graded transcriptional 510 responses

Natural Variation in Preparation for Nutrient Depletion Reveals a Cost-512

Population genomics of domestic and wild yeasts

Genomic sequence diversity and population structure of Saccharomyces 515 cerevisiae assessed by RAD-seq

Galactose metabolic genes in yeast respond to a ratio of galactose 517 and glucose

Decoupling transcription factor expression and activity enables 519 dimmer-switch gene regulation

Stochastic switching as a survival strategy in 521 fluctuating environments

Different levels of catabolite repression optimize growth in stable and 523 variable environments

Cell-to-Cell Stochastic Variation in Gene Expression Is a Complex Genetic 525

Natural sequence variants of yeast environmental sensors confer cell-to-527 cell expression variability

Widespread Cumulative Influence of Small Effect Size Mutations on 529

U-Net: deep learning for cell counting, detection, and morphometry

 This function was fitted to the  452   induced level curves of bimodal natural isolates, the mean fitted n value extracted for every natural  453 isolate, and the mean of these values was used for simulations ( Figure S11 ). 454 To delineate possible unimodal and bimodal regimes, GAL induction was simulated using all possible 463 Figure S11 ). The E 10 metric, the F 90 467 metrics, and the modality of the induction profile were determined from these simulations as described 468above. In the F 90 -E 10 space, unimodal and bimodal regimes were delineated by the bounding line with a 469 slope of 1 that would capture all the unimodal or bimodal simulations respectively on one side of the line. 470Data availability