Decoupling transcription factor expression and activity enables dimmer switch gene regulation

Abstract

Gene-regulatory networks achieve complex mappings of inputs to outputs through mechanisms that are poorly understood. We found that in the galactose-responsive pathway in Saccharomyces cerevisiae, the decision to activate the transcription of genes encoding pathway components is controlled independently from the expression level, resulting in behavior resembling that of a mechanical dimmer switch. This was not a direct result of chromatin regulation or combinatorial control at galactose-responsive promoters; rather, this behavior was achieved by hierarchical regulation of the expression and activity of a single transcription factor. Hierarchical regulation is ubiquitous, and thus dimmer switch regulation is likely a key feature of many biological systems. Dimmer switch gene regulation may allow cells to fine-tune their responses to multi-input environments on both physiological and evolutionary time scales.

To respond appropriately to varying circumstances, cells use transcriptional programs to integrate multiple inputs from their environment and determine the appropriate output. The yeast galactose-responsive (GAL) pathway, which controls the decision to metabolize galactose in the presence of other sugars, is a model system for multi-input responses (1, 2). We found that the output of this pathway has two independently controlled features: (i) the fraction of cells that express genes in the pathway and (ii) their level of expression. We set out to determine the molecular mechanism underlying this behavior.

The decision to express GAL genes is triggered by galactose but inhibited by glucose, a sugar that is easier to metabolize. The two features of this response are a switch-like decision to activate the pathway and a rheostat-like (i.e., graded) control of the expression level (Fig. 1 and fig. S1). We measured the steady-state GAL response in 77 different combinations of glucose and galactose using a yellow fluorescent protein (YFP) fused to the promoter of Gal1p, the first enzyme in the GAL pathway (Fig. 1A, figs. S1 and S3, and materials and methods) (3). The fraction of cells that activate the pathway (ON fraction, defined by comparing the YFP signal with an uninduced reference sample) was a one-dimensional switch-like function of the ratio of the galactose:glucose concentrations in the medium (1) (Fig. 1, B and J, and fig. S1, C and E). However, we found that when most of the cells were activated, the mean expression level of cells in the ON subpopulation (ON expression level) depended solely on the glucose concentration (Fig. 1, C and K; fig. S1, D and F; and materials and methods). This behavior is analogous to that of a mechanical dimmer switch often used for home lights. The galactose:glucose ratio, like the on-off light switch, determines whether cells turn on the pathway. The glucose concentration, like a dimmer knob or rheostat, controls the expression level of ON cells. Inside a narrow region near where the ON fraction is transitioning from mostly off to mostly on, the ON expression level is well described by the product of the glucose dependence of ON expression level and the ratiometric dependence of the ON fraction (supplementary text and fig. S2, F to K).

Fig. 1. Switch-like activation and rheostat-like control of expression level are genetically decoupled in the yeast GAL pathway.

(A) Schematic of the yeast GAL pathway, including the hypothesized competitive transport mechanism (1) by which the internal galactose level is a function of the ratio of external concentrations of the sugars, leading the GAL branch of the pathway (orange) to be responsive to the galactose:glucose ratio. (B to I) ON fraction and ON expression level heatmaps of the wild-type, mig1Δ, gal80Δ, and mig1Δgal80Δ strains in a glucose-galactose double gradient (two biological replicates are shown for each strain and glucose-galactose combination). Black lines delimit the region of ON fraction threshold (range of galactose:glucose ratios where 0.2 < ON fraction < 0.8). (J) Plot of the ON fraction versus galactose:glucose ratio for the wild-type, mig1Δ, gal80Δ, and mig1Δgal80Δ strains for experiments in (B) to (I). Scatterplot X-values are randomly jittered for visualization purposes. Deletion of GAL80 eliminates the switch. (K) Plot of ON expression level in the ON region (ON fraction > 0.8) versus the glucose concentration for the wild-type, mig1Δ, gal80Δ, and mig1Δgal80Δ strains for experiments in (B) to (I). Scatterplot X-values are randomly jittered for visualization purposes. Deletion of MIG1 eliminates the rheostat. See fit parameters in table S4.

We found that the switch and rheostat were controlled by separate genetic elements (Fig. 1 and fig. S2). We measured the steady-state GAL response of mutant yeast strains lacking key GAL pathway regulators (gal80Δ, mig1Δ, and mig1Δgal80Δ) in the same 77 combinations of glucose and galactose (fig. S2, A to E, and fig. S3). The gal80Δ ON fraction was always 100% independently of the concentrations of glucose and galactose (Fig. 1, D and J), whereas the gal80Δ rheostat was normal; the ON expression level responded to glucose concentration similarly to wild-type cells (Fig. 1, C, E, and K). By contrast, the mig1Δ strain had a normal switch, where the ON fraction responded to the galactose:glucose ratio similarly to wild-type cells (Fig. 1, F and J, and supplementary text), but its rheostat was always at its maximum outside the switch threshold region regardless of glucose concentration (Fig. 1, G and K, and fig. S2J). Thus, Gal80p is necessary for the switch and Mig1p is necessary for the rheostat. Consistent with this, the mig1Δgal80Δ strain was constitutively ON at its maximal level (Fig. 1, H to K) regardless of glucose or galactose concentration.

The first mechanism that we considered for the switch-and-rheostat response was a “chromatin-decoupled regulation” model inspired by pioneer factors (4) and observations in the yeast phosphate pathway (5). This model proposes that the switch is controlled by Gal4p-mediated chromatin remodeling and the rheostat by Mig1p-mediated transcriptional regulation (6) in fully accessible GAL1 promoters (fig. S4A). To test this idea, we measured chromatin accessibility and gene expression in many glucose and galactose combinations with a modified version of the assay for transposase-accessible chromatin sequencing (ATAC-seq) (7, 8) (figs. S4 and S8 and materials and methods). Consistent with the model, the chromatin accessibility of the OFF subpopulation was almost identical to that of maximally repressed wild-type cells (Fig. 2A and fig. S4, D to F). However, whereas the model predicts that all ON cells have fully accessible chromatin, we found a graded range of chromatin accessibility at the GAL1 promoter that correlated with GAL reporter expression (Fig. 2A and fig. S4, H to K).

Fig. 2. Decoupling occurs through regulation of the Gal4p transcription factor rather than directly at the GAL1 promoter through chromatin.

(A) Scatterplot of YFP expression level (x-axis) versus the percentage chromatin accessibility (y-axis) at the GAL1 promoter across all samples from wild-type, mig1Δ, gal80Δ, and mig1Δgal80Δ, and TetO7pr-MIG1 mig1Δgal80Δ strain backgrounds. Data plotted include two biological replicates (circles indicate replicate 1, squares indicate replicate 2) for each strain and glucose-galactose and doxycycline condition. (B and C) ON fraction as a function of the galactose:glucose ratio (B) and ON expression level as a function of glucose concentration (C) in a glucose-galactose double gradient assay (two biological replicates) across the wild-type, mig1Δ, GAL1pr^mig1bsΔ, and GAL4pr^mig1bsΔ strains. Scatterplot X-values are randomly jittered for visualization purposes. See pairwise comparisons from (B) in fig. S5, E to G. (D) Plot of GAL4pr-mScarlet expression versus sugar concentration in a glucose titration series across the wild-type, mig1Δ, gal80Δ, and GAL4pr^mig1bsΔ strains. Thin lines represent individual biological replicate data (two replicates for wild-type and four replicates for all other strains), and thick lines represent the average across replicate measurements. Scatterplot X-values are randomly jittered for visualization purposes. (E) RFP versus YFP fluorescence in a gal4Δ TetO7pr-mScarlet-GAL4 GAL1pr-YFP strain from microscopy of a doxycycline (DOX) titration series (see fig. S6D for a flow cytometry YFP versus DOX plot). Scatter represents single-cell measurements, with colors corresponding to discrete DOX concentrations; black circles are the average YFP and RFP values at each given DOX concentration (two biological replicates).

The chromatin-decoupled regulation model also failed to explain the behavior of the rheostat-only gal80Δ strain, which showed a range of glucose-dependent chromatin accessibility (Fig. 2A, bottom right) despite always having an ON fraction close to 1. Furthermore, the switch-only mig1Δ strain’s chromatin was either fully open or fully closed (Fig. 2A, top right), indicating that not only the switch but also the rheostat modulates chromatin accessibility. We created a doxycycline-titratable Mig1p strain to test this hypothesis (mig1Δgal80Δ TetO7pr-MIG1 GAL1pr-YFP; SLYM03 in table S1, also see the materials and methods). Increasing Mig1p expression by increasing doxycycline lowered both the expression of GAL1pr-YFP (fig. S4L) and the chromatin accessibility in a sugar-independent manner. The relationship between chromatin accessibility and transcription was the same as that of a wild-type strain (Fig. 2A). These data show that chromatin accessibility is not the primary mediator of decoupling between the switch and the rheostat (fig. S4A).

Our observation of a Mig1p-dependent correlation between chromatin accessibility and expression level conflicts with a report that nucleosome and Mig1p binding are mutually exclusive at the GAL1pr (6). To resolve this, we deleted the Mig1p-binding sites in the GAL1pr-YFP reporter (fig. S5A) and measured the response to varying combinations of glucose and galactose concentrations. Unexpectedly, the phenotype of the GAL1pr^mig1bsΔ reporter strain was nearly identical to that of the wild type (Fig. 2, B and C, and fig. S5C) despite the fact that the Mig1p-binding sites were highly conserved and presumed to be functionally important (9). This indicated that the direct effect of Mig1p at GAL1pr has a relatively minor contribution to the steady-state GAL1 expression level. Therefore, the Mig1p rheostat must regulate GAL1 upstream of the GAL1 promoter.

In addition to regulating GAL1 expression, Mig1p also inhibits GAL4 expression (10, 11) (Fig. 1A), and changing Gal4p concentrations can affect the GAL response (12, 13). We therefore investigated whether modulation of GAL4 expression might be involved in the rheostat. Deleting the Mig1p-binding site in the GAL4 promoter (fig. S5B) was sufficient to phenocopy the mig1Δ strain (Fig. 2, B and C, and fig. S5D), suggesting that glucose control of Gal1p levels is achieved solely through the regulation of Gal4p abundance. Supporting this hypothesis, YFP reporters for other Gal4p-regulated promoters, including two synthetic promoters, responded to glucose in the same way as GAL1, even if the promoter did not contain a Mig1p-binding site (fig. S5, H to M). As further support for this hypothesis, when we introduced a transcriptional reporter for GAL4 expression (GAL4pr-mScarlet-I) into the wild-type and mutant backgrounds, we found similar glucose-dependent titration of GAL4 expression level in the wild-type and gal80Δ backgrounds, which is in contrast to the elevated and less variable GAL4 expression for the GAL4pr^mig1bsΔ and mig1Δ backgrounds. (Fig. 2D and fig. S6, A and B). To directly test whether Gal4p abundance regulates GAL1 expression and thus mediates the Mig1p rheostat, we built a strain with a doxycycline-titratable mScarlet-I-Gal4p fusion (14) (gal4Δ TetO7pr-mScarlet-I-GAL4 GAL1pr-YFP; SLYM08 in table S1; also see the materials and methods) (fig. S6, C and D). By measuring fluorescence from both mScarlet-I-Gal4p and GAL1pr-YFP, we observed a direct correlation between the fluorescence of red fluorescent protein (RFP) and YFP in the ON subpopulation (Fig. 2E and fig. S6, E to H), confirming that the GAL pathway responds to Gal4p abundance (see also fig. S5, N and O).

Our results support a “hierarchically decoupled regulation” model in which the abundance and activity of a single transcription factor, Gal4p, are regulated independently (Fig. 3). In this model, transcriptional regulation of Gal4p abundance by the upstream transcription factor Mig1p mediates the response to glucose, whereas protein binding of Gal80p to Gal4p (15, 16) regulates Gal4p activity in response to the galactose:glucose ratio. Unlike our initial chromatin-decoupled regulation model (fig. S4A), a single transcription factor, Gal4p, controlled both the switch and the rheostat at the final step of the pathway. In both models, decoupling was achieved by regulation working through two distinct mechanisms; this is reminiscent of other cases, such as the frequency versus amplitude modulation of the Msn2p-Msn4p stress responses in yeast (17). Our model is agnostic to mechanistic details of how Gal4p activates downstream GAL promoters and is thus compatible with recent observations that different Gal4p-binding sites have different functional roles (18).

Fig. 3. Mechanistic model: switch-and-rheostat works through decoupled regulation of Gal4p activity and abundance.

(A) Switch-and-rheostat regulation in the GAL pathway is achieved through a hierarchical design. In response to glucose, the rheostat works through Mig1p transcriptional regulation of GAL4, where the activity of Gal4p is then regulated in response to the galactose:glucose ratio. (B) Molecular mechanism of the independent control of Gal4p abundance and activity. Internal galactose concentrations (which depend on the external ratio of glucose:galactose) controls the activity of Gal3p (x-axis). Gal80p sequesters Gal4p; active Gal3p sequesters Gal80p. Thus, when the active Gal3p concentration exceeds the total Gal80p concentration, Gal4p will convert sharply in the galactose:glucose ratio space from inactive (gray circles) to active (orange circles). However, the total amount of active Gal4p controls the amount of transcriptional output from the GAL1 promoter (green shading). Glucose controls Mig1p activity, thereby setting the total level of Gal4p (y-axis). Thus, glucose controls the level of Gal1p by controlling the total level of Gal4p and thereby the amount of active Gal4p.

What physiological function could be served by decoupling the on-off switch of pathway activation from the expression level of the pathway? When faced with mixtures of sugars, yeast first use glucose, then less-preferred carbon sources (19), a phenomenon called diauxic growth (20). Yeast prepare by expressing GAL genes before glucose is depleted; the earlier a strain expresses GAL genes, the higher the fitness advantage it has once glucose is depleted. However, preparation comes at a fitness cost in the period before the glucose runs out (21). One possible function of the decoupled switch- and-rheostat design is to allow early activation of the pathway but with a reduced cost.

To test this, we set up a competition between a wild-type strain and a rheostat-lacking GAL4pr^mig1bsΔ strain during diauxic growth conditions (fig. S7A). GAL1pr-YFP expression quickly reached its maximum in the GAL4pr^mig1bsΔ strain but was delayed in a wild-type strain until glucose was depleted (fig. S7, B and D). During this time (t = 0 to 7 hours), the wild-type strain had a fitness advantage over the GAL4pr^mig1bsΔ strain (Fig. 4 and materials and methods), presumably due to the saving of resources required to maximally activate the GAL pathway (21, 22). After glucose was exhausted (fig. S7B), expression of GAL genes increased in the wild-type strain (fig. S7D). During this period (t = 7 to 10 hours), the GAL4pr^mig1bsΔ strain had a competitive advantage (Fig. 4) that lasted until the wild-type strain also reached maximal expression of GAL genes (21). However, this temporary benefit for the GAL4pr^mig1bsΔ strain was insufficient to offset the initial cost of over-expressing GAL genes (Fig. 4). The observed fitness differences were not due to differences in growth on glucose or galactose alone (Fig. 4 and fig. S7, F and G). In addition, the fitness advantage of the wild-type strain was absent when both strains were switched directly to galactose (Fig. 4 and fig. S7E). We conclude that the Mig1p rheostat reduces the fitness cost of preactivating GAL genes during the gradual depletion of glucose.

Fig. 4. Physiological benefit of switch-and-rheostat design.

Log₂ ratio of GAL4pr^mig1bsΔ cell counts over wild-type cell counts versus time during diauxic growth (black, four biological replicate cocultures), an “instantaneous” shift to pure galactose (blue, four biological replicate cocultures), steady-state growth on 2% glucose (yellow, four biological replicate cocultures), or steady-state growth on 2% galactose (green, four biological replicate cocultures). Thick lines are the average of the four measurements (thin lines). A positive value on the y-axis indicates that GAL4pr^mig1bsΔ has a net fitness advantage since the initial time point, and a negative value indicates that the wild type has a net fitness advantage since the initial time point.

Decoupled control is a useful property because it allows response features to be independently controlled physiologically and evolutionarily (23–25). Decoupling has often been proposed to involve independent transcription factor-binding sites on promoters and to be aided by chromatin (5). We show here that the same result can be accomplished by hierarchical regulation of the abundance and activity of a transcription factor. Because regulation of this kind is common, it is likely that decoupling of responses is also achieved by this mechanism in other systems (26, 27).