Inhibitory GEF phosphorylation provides negative feedback in the yeast polarity circuit

Abstract

Cell polarity is critical for the form and function of many cell types. During polarity establishment, cells define a cortical “front” that behaves differently from the rest of the cortex. The front accumulates high levels of the active form of a polarity-determining Rho-family GTPase (Cdc42, Rac, or Rop), which then orients cytoskeletal elements through various effectors to generate the polarized morphology appropriate to the particular cell type [1, 2]. GTPase accumulation is thought to involve positive feedback, such that active GTPase promotes further delivery and/or activation of more GTPase in its vicinity [3]. Recent studies suggest that once a front forms, the concentration of polarity factors at the front can increase and decrease periodically, first clustering the factors at the cortex and then dispersing them back to the cytoplasm [4–7]. Such oscillatory behavior implies the presence of negative feedback in the polarity circuit [8], but the mechanism of negative feedback was not known. Here we show that, in the budding yeast Saccharomyces cerevisiae, the catalytic activity of the Cdc42-directed GEF is inhibited by Cdc42-stimulated effector kinases, thus providing negative feedback. We further show that replacing the GEF with a phosphosite-mutant GEF abolishes oscillations and leads to the accumulation of excess GTP-Cdc42 and other polarity factors at the front. These findings reveal a mechanism for negative feedback and suggest that the function of negative feedback via GEF inhibition is to buffer the level of Cdc42 at the polarity site.

Results and Discussion

Cdc24 phosphorylation reduces GEF activity in vitro

GTP-Cdc42 is known to activate p21-activated kinases (PAKs) that then phosphorylate the Cdc42-directed guanine nucleotide exchange factor (GEF), Cdc24 [9, 10]. To assess the effect of phosphorylation on Cdc24 GEF activity, we isolated HA-Cdc24 from yeast cells that overexpressed the protein, and measured its ability to stimulate GDP/GTP exchange on recombinant Cdc42 (Fig. 1A). HA-Cdc24 displayed robust GEF activity that was abolished by mutations in the catalytic domain, indicating that the activity was intrinsic to Cdc24 and not due to a co-precipitating factor (Fig. 1B). As judged by gel mobility, a large majority of the HA-Cdc24 was not phosphorylated. To generate a preparation of highly phosphorylated Cdc24, we co-expressed the activated Cdc42 mutant, Cdc42^Q61L, as well as a Bem1-Cla4 fusion protein [11], to promote Cla4 (PAK)-mediated hyperphosphorylation of HA-Cdc24 (Fig. 1C). Phosphorylated Cdc24 was less active than unphosphorylated Cdc24 (Fig. 1D). This was due to phosphorylation, because treatment with λ phosphatase caused an increase in GEF activity (Fig. 1E). We conclude that Cdc24 phosphorylation reduces GEF activity.

Figure 1. GEF phosphorylation inhibits activity.

A) Measurement of Cdc24 GEF activity in vitro. Immunoprecipitated (IP) HA-Cdc24 isolated from yeast is incubated with recombinant GDP-Cdc42 and γ-³⁵S-GTP. GTP-loading is assessed by a filtration assay. B) Cdc42 GTP-loading by wild-type Cdc24 or a catalytic domain mutant N452A/E453A (designated Cdc24-AA). Arbitrary units represent the amount of radioactive GTP loaded divided by the amount of HA-Cdc24 in the IP (inset). Activity is normalized to the WT Cdc24 sample. Mean +/− SEM (n=3). C) Western blot of Cdc24 isolated from wild-type cells (DLY15284) or cells overexpressing Cdc42^Q61L and Bem1-Cla4 (DLY15332). The latter (designated Cdc24^Phos) shows a mobility shift indicating hyperphosphorylation. Cdc11: loading control. D) Cdc24^Phos has reduced GEF activity. Assay as in B. Mean +/− SEM (n=4). E) Phosphatase treatment of Cdc24^Phos increases GEF activity. Cdc24^Phos was treated with buffer, λ phosphatase, or λ phosphatase plus phosphatase inhibitors, as indicated. Assay as in B. Mean +/− SEM (n=3). *** P<0.001; * P<0.02 (two-tailed t-test).

GEF activity was reduced approximately two-fold for phosphorylated versus unphosphorylated Cdc24 preparations (Fig. 1D,E). However, the phosphorylated preparation exhibited a heterogeneous mixture of phosphorylation states (insets, Fig. 1D,E), due to partial dephosphorylation during immunoprecipitation and washing. As 40–50% of the “phosphorylated” protein migrated at a comparable position to unphosphorylated Cdc24, it may be that all remaining activity stems from that less phosphorylated subset, and that the highly phosphorylated Cdc24 is completely inactive.

Assay for Cdc24 activity in vivo

To assess Cdc24 activity in vivo, we followed a strategy based on the expectation that if Cdc24 were targeted uniformly to the plasma membrane (rather than just to the polarity site), then it would generate GTP-Cdc42 all over the membrane (Fig. 2A). In turn, that should dominantly interfere with polarization mediated by the endogenous Cdc24, blocking proliferation [12]. The level of expression needed to block polarization would depend on the activity of the membrane-targeted Cdc24, providing a measure for Cdc24 activity in vivo. We used a 28-residue N-terminal peptide from Psr1 [13] to efficiently target Cdc24 to the plasma membrane. Levels of expression were controlled by an estrogen-inducible promoter system (Fig. 2A)[14]. High-level expression of membrane-targeted Cdc24 (MT-Cdc24) induced accumulation of GTP-Cdc42 all over the plasma membrane (Fig. 2B), and blocked both polarization (Fig. 2B) and proliferation (Fig. 2D).

Figure 2. Phosphosite mutant GEF is more active in vivo.

A) Assay for Cdc24 activity in vivo. An estradiol-regulated derivative of Gal4 allows graded induction of the GAL1 promoter, driving overexpression of wild-type (WT) or membrane-targeted (MT) GFP-Cdc24. WT Cdc24 is mostly cytoplasmic, and overexpression is well tolerated: cells polarize GTP-Cdc42 and form buds (top). A lipid-modified Psr1 peptide (blue) targets MT-Cdc24 uniformly around the plasma membrane, leading to uniform GTP-Cdc42 accumulation, loss of polarity, and failure to proliferate (bottom). B) Localization of the GTP-Cdc42 probe Gic2^1-208-tdTomato in cells overexpressing WT or MT GFP-Cdc24 derivatives. Cells overexpressing the indicated versions of Cdc24 were treated with 50 nM β-estradiol for 3–4 h, and single confocal planes were imaged (GFP, left; RFP, middle; merge, right). All membrane-targeted versions blocked polarization at this high level of expression, leading to accumulation of large unbudded cells. C) Quantification of the cortical-to-cytoplasmic fluorescence ratio for the GTP-Cdc42 probe in cells expressing MT-Cdc24 derivatives. Mean +/− SEM (n=5 fields of cells with ~40 cells/field). *** P<0.001, ** P<0.01 (two-tailed t-test). Cortical Cdc42 was reduced by catalytic mutations but increased by phosphosite mutations. D) MT-Cdc24 overexpression blocks proliferation by a combination of catalytic activity and Bem1 titration. Left: cells were spotted (10-fold serial dilutions) onto plates containing the indicated concentrations of β-estradiol to induce expression of Cdc24 or MT-Cdc24 derivatives, and incubated for 2 days at 24°C. MT-Cdc24-AA can titrate Bem1 but lacks catalytic activity, while MT-Cdc24-KR has catalytic activity but cannot titrate Bem1: both block proliferation, but less effectively than MT-Cdc24. Right: Blot shows levels of induction (MT-Cdc24 has slower mobility due to phosphorylation). Cdc11, loading control. E) MT-Cdc24 is highly phosphorylated. Mobility shift is reduced upon treatment with λ phosphatase, but not λ phosphatase plus phosphatase inhibitors. F) MT-Cdc24^33A is more potent at blocking proliferation than MT-Cdc24. Assay and blots as in D. G) Catalytically dead MT-Cdc24^33A is comparable at blocking proliferation to MT-Cdc24. Assay and blots as in D. H) MT-Cdc24^33A displays higher GEF activity than MT-Cdc24. Assays as in Fig. 1B except that the GEF was immunoprecipitated using anti-GFP. Mean +/− SEM (n=3). *** P<0.001 (two-tailed t-test). I) Cdc24 (which is largely unphosphorylated when overexpressed) has comparable GEF activity to Cdc24^33A. Mean +/− SEM (n=3).

In principle, the effect of membrane-targeted Cdc24 could stem from its catalytic activity (promoting uniform cortical GTP-Cdc42 accumulation), or from titrating Cdc24 interactors (e.g. Bem1 [15] and Rsr1-GTP [16]) away from the endogenous Cdc24, or both. We found that a membrane-targeted but catalytically dead Cdc24 (MT-Cdc24-AA) was no longer effective in promoting accumulation of GTP-Cdc42 all over the membrane (Fig. 2B,C) but was still able to block proliferation at high expression levels, presumably via titration (Fig. 2D). Membrane targeting of a Bem1-binding-deficient mutant (MT-Cdc24-KR: D824K, D831R [17]) was also able to block proliferation at high expression levels (Fig. 2D). At lower expression levels, only MT-Cdc24 (and not MT-Cdc24-AA or MT-Cdc24-KR) blocked proliferation (Fig. 2D). A double mutant lacking both catalytic activity and Bem1-binding ability did not block proliferation (Fig. 2D). Thus, the polarity-blocking effect of MT-Cdc24 reflects additive contributions from its catalytic activity and from Bem1 titration.

Nonphosphorylatable Cdc24 is hyperactive in vivo

MT-Cdc24 exhibited a gel mobility shift when compared to wild-type Cdc24, and phosphatase treatment reversed the shift (Fig. 2E), suggesting that MT-Cdc24 is highly phosphorylated. A previous study mapped 35 phosphorylation sites on Cdc24 using mass spectrometry, and generated a “nonphosphorylatable” Cdc24^38A mutant that lacked those sites and an additional 3 putative sites (although this mutant was still phosphorylated at a few additional sites, as judged by gel mobility)[18]. We generated a version of MT-Cdc24 containing 33 of those 38 phosphorylation-site mutations (the most N-terminal 4 mutations and most C-terminal mutation were omitted by the cloning strategy). We also generated variants harboring only 7 mutations (at designated “PAK consensus sites” [18]) or only 4 mutations (at designated “CDK consensus sites” [18]). Compared to MT-Cdc24, MT-Cdc24^33A exhibited a much smaller mobility shift, whereas MT-Cdc24^7A and MT-Cdc24^4A were still highly phosphorylated (Fig. 2F). MT-Cdc24^33A (but not MT-Cdc24^7A or MT-Cdc24^4A) was more effective at blocking proliferation than MT-Cdc24 (Fig. 2F). MT-Cdc24^33A was also more effective at promoting uniform GTP-Cdc42 accumulation (Fig. 2B,C), suggesting that it had greater catalytic activity. To assess whether the enhanced potency of MT-Cdc24^33A in blocking proliferation was due solely to enhanced catalytic activity, we generated a nonphosphorylatable but catalytically dead MT-Cdc24^33A-AA mutant. This construct was no more effective than MT-Cdc24-AA at inducing cortical GTP-Cdc42 accumulation (Fig. 2B,C) or blocking proliferation (Fig. 2G). Thus, mutation of most of the mapped Cdc24 phosphorylation sites increased the catalytic activity of membrane-targeted Cdc24 in vivo, leading to a more potent block in proliferation.

MT-Cdc24^33A also had more GEF activity than the highly phosphorylated MT-Cdc24 in vitro (Fig. 2H). In principle, the increased activity could be due either to the lack of phosphorylation or to some conformational effect of the mutations themselves. To distinguish between these possibilities, we generate a comparable GFP-Cdc24^33A that lacked the membrane-targeting Psr1 domain. In the absence of membrane targeting, overexpressed GFP-Cdc24 is almost all cytoplasmic (Fig. 2B) and unphosphorylated (Fig. 1B). In that context, Cdc24^33A had no more GEF activity than unphosphorylated Cdc24 (Fig. 2I). Thus, the increased activity of MT-Cdc24^33A is due to the absence of phosphorylation. We conclude that Cdc24 phosphorylation reduces GEF activity in vivo as well as in vitro.

Cdc24 phosphorylation was reported to abolish Cdc24-Bem1 interaction [9]. However, we found that immunoprecipitated Bem1-myc associated with both phosphorylated MT-Cdc24 and unphosphorylated Cdc24 (Fig. 3A). Similarly, phosphorylated MT-Cdc24 and unphosphorylated Cdc24 both associated with comparable amounts of Bem1-myc (Fig. 3B). Thus, we see no strong effect of Cdc24 phosphorylation on Bem1 binding.

Figure 3. GEF phosphorylation does not affect Bem1 binding; modeling negative feedback.

A) Immunoprecipitated Bem1-myc associates with both phosphorylated MT-Cdc24 and unphosphorylated Cdc24. B) Immunoprecipitated MT-Cdc24 and WT Cdc24 associate with comparable amounts of Bem1-myc. C) Schematic of models with and without negative feedback. In both models, the Bem1 complex can associate with GTP-Cdc42, and load GTP on neighboring GDP-Cdc42, providing positive feedback. Negative feedback is modeled as a transition of the Cdc42-bound Bem1 complex to an inactive, phosphorylated form (blue), which can dissociate from Cdc42 and be dephosphorylated in the cytoplasm. D) Simulations of polarity establishment using the two models. The amount of Cdc42 at the polarity site increases to a high plateau in the model with only positive feedback (red). Addition of negative feedback reduces Cdc42 accumulation and yields damped oscillations before reaching a low steady state (compared to the peak). E) Simulations of competition between two polarity clusters at opposite ends of the cell. Simulations were initiated with clusters containing Cdc42 at a 45:55 ratio, and the upper line shows growth of the winning cluster while the lower line shows disappearance of the losing cluster. Competition is resolved more rapidly with (blue) than without (red) negative feedback.

As the activity of isolated Cdc24 was increased by phosphatase treatment in vitro (under conditions where that could not be due to an increase in Bem1 binding: Fig. 1E), we conclude that Cdc24 phosphorylation directly affects catalytic activity, independent of Bem1 binding. Combined with previous studies [9, 10], our finding that phosphorylation of Cdc24 inhibits its catalytic activity reveals a negative feedback pathway acting in the yeast polarity circuit. The pathway involves the following steps: (i) GTP-Cdc42 activates PAKs; (ii) PAKs phosphorylate Cdc24; (iii) phosphorylation inhibits Cdc24 GEF activity; and (iv) reduced GEF activity leads to lower levels of GTP-Cdc42.

Computational modeling of negative feedback through Cdc24 phosphorylation

Because polarization can be initiated by stochastic fluctuations, which can occur at multiple locations, cells often begin to concentrate Cdc42 at more than one site [6, 19]. However, Cdc42 clusters then appear to compete with each other, so that one cluster grows to become the front while all of the other clusters disappear [6, 20]. Theoretical studies have shown that the known biochemical interactions and activities of Cdc42, the PAK-Bem1-Cdc24 complex, and a guanine nucleotide dissociation inhibitor (GDI) are sufficient in principle to explain both clustering of polarity factors and competition between clusters [20, 21]. To explain oscillatory clustering, an additional negative feedback loop was posited [6], although the mechanism of negative feedback was unknown.

Given our findings, we asked whether adding inhibitory multisite phosphorylation of Cdc24 to the previous positive-feedback-only model (Fig. 3C) could produce oscillatory clustering. As multisite phosphorylation can yield ultrasensitive behavior [22–24], we modeled phosphorylation as a Hill function of the concentration of Cdc42-bound Bem1 complexes (see Computational Methods). However, ultrasensitivity was not essential to produce the predictions discussed below.

Simulations incorporating negative feedback displayed dampened oscillations, as well as reduced Cdc42 accumulation at the polarity site (Fig. 3D). We also simulated cells in which two slightly unequal polarity clusters were formed at opposite ends of the cell, and followed the relative amount of Cdc42 in each cluster. Competition was resolved more rapidly in simulations containing negative feedback (Fig. 3E). Thus, if the model approximates the true situation in cells, then short-circuiting the negative feedback loop by rendering Cdc24 nonphosphorylatable should abolish oscillations, increase Cdc42 levels at the front, and slow down competition between clusters.

Cdc24 phosphorylation is required for oscillatory polarization but does not affect competition between polarity clusters

Previous work showed that nonphosphorylatable Cdc24^38A was functional, but the dynamics of polarization were not examined [18]. We generated yeast strains in which endogenous CDC24 was precisely replaced by CDC24^38A at its endogenous locus, and monitored polarization using Bem1-GFP [11], GFP-Cdc42, or the GTP-Cdc42 reporter, Gic2^1-208-tdTomato [25] as probes. We focused on diploid cells lacking the bud-site-selection protein Rsr1, where we had observed oscillatory cycles of clustering and dispersal of polarity factors [6]. Control experiments with HA-tagged constructs showed that Cdc24^38A was expressed at similar levels to Cdc24 (Fig. 4G), as reported [18].

Figure 4. Negative feedback via GEF phosphorylation promotes oscillatory clustering and lowers GTP-Cdc42 levels at the polarity site.

A) Dynamics of Bem1-GFP clustering in CDC24 and CDC24^38A cells. Cells were imaged side by side to allow direct comparison of Bem1-GFP intensities. Maximum projections of confocal stacks are shown, and intensity of the probe in the cluster is quantitated. Blue arrowhead: signal from old neck or neighboring cell (not polarity site). Arrows: clustering peak(s): red, CDC24 cells (control); green, CDC24^38A cells. More examples are shown in Fig. S1. B) Dynamics of GTP-Cdc42 clustering visualized with Gic2^1-208-tdTomato as above. More examples, Fig. S1. C) Dynamics of GFP-Cdc42 clustering, as above. More examples, Fig. S1. D) Rapid competition in CDC24 and CDC24^38A cells that initially form more than one cluster. E) Quantification of the time taken to resolve multi-cluster intermediates in CDC24 and CDC24^38A cells. F) Quantification of peak Bem1-GFP, Gic2^1-208-tdTomato, and GFP-Cdc42 intensities in CDC24 and CDC24^38A cells. In each case, mutant cells showed a statistically significant (P<0.001) increase in peak levels. For E and F, each dot is one cell taken from movies as in A-D; the line indicates average. G) Levels of HA-Cdc24 and HA-Cdc24^38A are comparable, as are total Bem1-GFP and GFP-Cdc42 probes in CDC24 and CDC24^38A cells. Cdc11, loading control.

During initial polarity establishment, CDC24 cells displayed oscillations in the concentration of polarity factors at the front (Fig. 4A-C, Fig. S1, and supplement movie 1), as reported [6]. In contrast, CDC24^38A cells kept polarity protein levels high at the front, without discernible oscillations (Fig. 4A-C, Fig. S1, and supplement movie 2). Unlike in the model with no negative feedback (Fig. 3D), however, levels of polarity factors did gradually decline from the peak in mutant cells (Fig. 4A-C, Fig. S1), perhaps reflecting additional sources of negative feedback. Two other negative feedback loops were recently proposed to act through the cytoskeletal targets of Cdc42, actin [26, 27] and septins [28]. These slower feedback loops cannot account for the rapid oscillations [6], but may contribute to the slower decline in polarity protein levels at the front observed in the CDC24^38A mutants.

As with CDC24 cells, several of the CDC24^38A cells we imaged initially developed more than one polarity cluster (Fig. 4D and supplement movie 2). To assess whether competition between clusters was altered in the mutant cells, we measured the co-existence time (the interval for which two or more clusters were present) for this subset of cells. Quantification indicated that competition timing was generally similar in CDC24 versus CDC24^38A cells (Fig. 4E). Thus, Cdc24 phosphorylation does not seem to be required to enforce rapid competition.

Cdc24 phosphorylation lowers polarity factor concentrations at the front

To compare the relative intensity of polarized signals, CDC24 and CDC24^38A cells were imaged side-by-side on the same slide (supplement video 3). Peak levels of Bem1, Cdc42, and GTP-Cdc42 were all higher in CDC24^38A cells than in CDC24 cells (Fig. 4F). This was not due to an increase in total cellular levels of Bem1 or Cdc42 (Fig. 4G). The effect of CDC24^38A was dominant to wild-type, as expected for a hyperactive protein (Fig. S1D). Thus, negative feedback through Cdc24 phosphorylation acts to reduce polarity factor accumulation at the cortex.

CDC24^38A cells did not display any major changes in polarization efficiency, proliferation rate, or cell morphology. Thus, it appears that cells can tolerate increased levels of polarized GTP-Cdc42 without severe ill effects, at least under lab growth conditions. Negative feedback may become more important when polarity-affecting stresses occur in more physiological environments. Alternatively, like cell-cycle checkpoints and DNA repair pathways, negative feedback may only be critical for a small minority of cells that experience polarity problems (e.g. because they contain potentially dangerous levels of polarity factors). Consistent with that hypothesis, preliminary observations suggest that negative feedback becomes important for proliferation when Cdc42 is overexpressed. It will be of great interest to determine whether polarity establishment in other systems also employs inhibitory GEF phosphorylation to regulate polarization.

Highlights.

• Cdc42 promotes inhibitory phosphorylation of Cdc42-directed GEF

• Negative feedback via GEF phosphorylation produces oscillatory dynamics

• GEF phosphorylation buffers the accumulation of Cdc42 at the cell’s front