Activation of the Cdc42p GTPase by cyclin‐dependent protein kinases in budding yeast

We combined an automated yeast genetics platform called synthetic genetic array (SGA) analysis with a yeast overexpression array, to enable a systematic approach to examine SDL interactions. Array‐based SDL screens permit systematic analysis of gain‐of‐function phenotypes and are based on the idea that deleterious effects of gene overexpression are often seen only in specific genetic backgrounds (Sopko et al, 2006). With this in mind, we screened a strain deleted for the PHO85 CDK and discovered 65 genes that cause lethality or slow growth specifically in the absence of the kinase (Sopko et al, 2006). We reasoned that since many known Pho85p substrates are negatively regulated by phosphorylation, an SDL interaction might reflect an accumulation of unmodified substrate. Consistent with this idea, the SDL data set was highly enriched for known in vivo substrates of the Pho85p kinase. To aid prioritization of other SDL hits for follow‐up studies, we compared the pho85Δ SDL profile to the spectrum of in vitro substrates for various forms of Pho85p identified using proteome chips coupled with in vitro kinase assays (Ptacek et al, 2005). Several proteins were identified in both screens, including the calcium‐responsive transcription factor, Crz1p, which we confirmed as a previously unappreciated in vivo target of Pho85p using a variety of assays (Sopko et al, 2006). The GAP, Rga2p, was another protein identified in both the SDL and proteome chip screens (Ptacek et al, 2005; Sopko et al, 2006). RGA2 is one of four genes in yeast encoding GAPs that act specifically to stimulate GTP hydrolysis by the Rho‐type GTPase Cdc42p (Smith et al, 2002). Cdc42p‐mediated GTP hydrolysis appears critical for the propagation and/or dissolution of an interaction between Cdc42p and downstream effectors, ultimately leading to the firing of specific pathways such as septin ring and actin filament assembly and organization. The results of our large‐scale screens suggest that Pho85p‐dependent phosphorylation of Rga2p may explain the clear role for Pho85p in cell polarity and morphogenesis, for which key substrates remain unidentified, an idea that we chose to explore.

We reasoned that if Pho85p were phosphorylating Rga2p in vivo, a specific Pho85p–cyclin complex might be involved. Since Pho85p has a clear role in regulating morphogenesis in G1 phase (Moffat and Andrews, 2004), we chose to focus on G1‐specific forms of Pho85p as potential Rga2p kinases. We predicted that, like the pho85Δ kinase mutant, RGA2 overexpression in a mutant lacking the relevant Pho85p cyclin(s) would result in a significant fitness defect. Pho85p is activated by three G1‐specific cyclins, Pcl1p, Pcl2p, and Pcl9p, and we found that overexpression of RGA2 produced a growth defect in strains deleted for various combinations of these G1 cyclins (Figure 1A, pcl1Δ, pcl2Δ, and pcl9Δdouble and triple mutants). The overexpression of RGA2 in other Pho85p cyclin mutant backgrounds failed to have any discernable effect (data not shown). Effects in single mutants were less dramatic, consistent with the documented genetic redundancy of these G1 cyclins (Measday et al, 1997; Huang et al, 2002). We saw a similar inhibition of growth when RGA2 was overexpressed in a strain lacking CLN1 and CLN2, the G1 cyclin counterparts for the CDK, Cdc28p (Figure 1A). This growth inhibition cannot be explained by an increased stability of Rga2p in these mutants, given that we saw no substantial differences in Rga2p levels in these mutants by Western blot (Supplementary Figure 1). A significant fraction of the double and triple clnΔ and pclΔmutants overexpressing RGA2 arrested with a large, round unbudded cell morphology (Figure 1B, pcl1Δpcl2Δ: 37%, pcl1Δpcl9Δ: 33%, and pcl1Δpcl2Δ pcl9Δ: 40%). This morphogenetic defect is similar to that seen in mutant strains bearing cdc24 loss‐of‐function alleles (Hartwell et al, 1974; Sloat et al, 1981; Zheng et al, 1994). Since Cdc24p, the GEF for Cdc42p, opposes the function of Cdc42p GAPs, our genetic results suggest that Rga2p is hyperactive when overexpressed in G1 CDK mutants.

Rga2p localizes to sites of polarized growth including the incipient bud site and bud tip, as well as the bud neck (Caviston et al, 2003). Coincident localization of Rga2 with relevant in vivo kinases should be detectable, and Pcl1p and Pcl2p are known to localize to the same regions of polarized growth (Figure 2A; Moffat and Andrews, 2004). Given the genetic redundancy we detected for RGA2 SDL phenotypes, we asked if, like Pcl1p and Pcl2p, a GFP‐tagged version of Pcl9p had overlapping localization with Rga2p (Figure 2A). We identified the bud neck and incipient bud site as the principal sites of Pcl9p localization (Figure 2A). Together with our genetic observations, these localization data suggest the G1‐specific Pcls are the relevant cyclins for targeting Pho85p‐dependent phosphorylation of Rga2p.

The physical interaction of kinases and their substrates can often be detected directly by copurification of the proteins or indirectly by association of kinase activity with the substrate. In particular, cyclins are known to function as substrate‐targeting subunits (Miller and Cross, 2001), and several interactions between cyclins and Pho85p targets have been reported (Huang et al, 1998; Friesen et al, 2003). We asked whether we could detect a physical interaction between Rga2p and any of the G1‐specific Pho85p cyclins. We used a FLAG‐tagged version of RGA2 to efficiently immunoprecipitate full‐length Rga2p‐FLAG from yeast extracts (Figure 2B, lanes 5–8). We detected copurification of Rga2p‐FLAG and Pcl1p, Pcl2p, or Pcl9p from extracts prepared from strains expressing 13xMYC‐PCL1, 13xMYC‐PCL2, or 13xMYC‐PCL9 (Figure 2B, lanes 6–8, respectively). These data suggest a direct physical interaction between Rga2p‐ and G1‐specific Pcl cyclins, consistent with a role for the Pcls in targeting Rga2p to the Pho85p kinase.

Rga2p possesses 18 potential CDK phosphorylation sites (S/TP; Figure 3A). Using five purified Rga2p protein fragments (Figure 3A, bottom) we performed in vitro kinase assays with recombinant Pho85p–Pcl and Cdc28p–Cln complexes to hone in on relevant sites of CDK phosphorylation. Fragments Rga2_2 and Rga2_4, both of which contain small clusters of S/TP sites, were excellent in vitro substrates for Pho85p–Pcl1p (data not shown), Pho85p–Pcl2p (Figure 3B), or Pho85p–Pcl9p (data not shown), while Cdc28p–Cln2p detectably phosphorylated all Rga2p peptides except Rga2_5 (Figure 3C). Interestingly, Rga2_5, the peptide encompassing the GAP domain failed to be efficiently phosphorylated by any of the kinases. Coomassie Blue‐stained gels corresponding to these autoradiographs are included in the Supplementary data for reference (Supplementary Figure 2). We subjected the in vitro phosphorylated peptides to mass spectrometry analysis and identified nine sites within the Rga2_2 and Rga2_4 peptides (T320, S330, S334, T561, S692, S763, S770, S772, and T779) that were phosphorylated by Pho85p–Pcl complexes in vitro (Table I). Five sites clustered in fragment Rga2_4 were also clearly phosphorylated in vitro by Cdc28p–Cln2p (S692, S763, S770, S772, and T779; Table I). Our failure to detect phosphorylated peptides by mass spectrometry on the other Rga2p fragments that were phosphorylated in vitro by Cdc28p (Figure 3C) may reflect variability in the kinase assay or a lack of enrichment for peptides encompassing these Rga2p fragments following protein digestion and derivatization.

Next, we sought to identify relevant in vivo sites of phosphorylation within Rga2p. We purified a C‐terminally FLAG‐tagged version of Rga2p from yeast extracts and identified eight in vivo sites of phosphorylation using mass spectrometry analysis (S334, S380, S692, S707, S733, S763, S770, and S772; Table I). Six of these sites were phosphorylated both in vivo and in vitro (Table I), while other phosphopeptides were detected only in one assay (in vitro: T320, S330, T561, and T779; in vivo: S380 and S707). In any case, the significant overlap between phosphorylation patterns seen in our CDK kinase assays and on Rga2p in vivo strongly supports the hypothesis that Rga2p is a substrate for G1 CDKs.

To further test this hypothesis, we purified Rga2p protein from pho85Δ and cln1Δcln2Δ mutants to assess alterations in Rga2p phosphorylation. We failed to detect phosphorylation at three sites (S334, S380, and S707; Table I) when Rga2p was purified from a pho85Δ strain, whereas phosphorylation of only S380 appeared dependent on CLN1 and CLN2 (Table I). Although we cannot exclude the possibility that an insufficient enrichment for derivatized peptides encompassing sites S380 and S707 is underrepresenting phosphorylation, we predict that a lack of phosphorylation detected at these sites is indicative of their CDK‐dependent phosphorylation. On the other hand, we recovered significant derivatization of non‐phosphorylated peptides encompassing S334 in pho85Δ, indicating that phosphorylation at this site is indeed dependent on Pho85p activity in vivo. The differential phosphorylation of Rga2p by Pho85p–Pcl and Cdc28p–Cln complexes suggests that these kinases may each contribute to the regulation of Rga2p via phosphorylation, both at unique and potentially overlapping sites (Table II).

To further substantiate the view that G1 CDKs phosphorylate Rga2p to regulate its activity, we analyzed the phenotype associated with overexpression of versions of Rga2p mutated at various S/TP sites. Rather than focusing solely on those residues implicated as in vivo sites by our biochemical analysis, we chose a broader approach that considered potential phosphorylation sites of Rga2p conserved among fungal species, as well as sites conserved among all Cdc42 GAPs (see Figure 4A). We expressed mutant alleles of Rga2p lacking one or more of 13 different S/TP sites (S/T‐A substitutions) in wild‐type cells. Although S380 was phosphorylated in our wild‐type Rga2p purification, we were unable to test the function of an S380A mutant as the mutant protein is unstable. Nonetheless, we predicted that expression of a hypo‐phosphorylated version of RGA2 might mimic the effect we saw when overexpressing wild‐type RGA2 in a pho85Δ strain or various CDK mutants. Overexpression of Rga2p mutants lacking any single site, or several double‐site combinations, had no obvious phenotypic consequence relative to wild‐type Rga2p (Figure 4A). Certain alanine substitution combinations, however, showed a pronounced phenotypic consequence and these effects were additive. An Rga2p mutant carrying eight substitutions (S160A, S330A, S334A, S707A, S751, S763, S770A, and S772A; hereafter referred to as Rga2^8Ap) had the most dramatic effect on growth when overexpressed in wild‐type cells (Figure 4A, RGA2^8A was expressed at levels comparable to wild‐type RGA2; Figure 4F). The residues altered in the Rga2^8Ap mutant include five serine residues that were phosphorylated both in vivo and in vitro (see above and Table I). Expressing RGA2^8A in wild‐type cells phenocopied the overexpression of wild‐type RGA2 in CDK mutants; a large proportion (∼47%) of cells displayed a large, round unbudded cell morphology after 6 h in galactose (Figure 4B). The morphological effects of expressing quadruple‐ and quintuple‐site mutants were similar to that of the 8 × mutant Rga2^8A, however more subtle. Notably, overexpression of RGA2^8A is still lethal in pho85Δ and cln1Δcln2Δ mutant strains, as we would predict if these kinases were responsible for phosphorylating Rga2p (data not shown) and overexpression of PCL1, PCL2, or PCL9 is unable to suppress RGA2^8A‐associated toxicity (Supplementary Figure 3). In addition, RGA2^8A exacerbated the temperature sensitivity and unbudded phenotype of a cdc24‐4 strain (Figure 4C). Together, our biochemical and genetic data suggest that a deficiency of CDK‐dependent phosphorylation of Rga2p leads to significant G1 defects, most consistent with a failure to negatively regulate Rga2p function.

To ask whether the phenotype displayed by wild‐type cells overexpressing RGA2^8A was due to unrestricted Rga2p GAP activity, we chose to mutate a conserved lysine residue (K872) within the GAP domain of Rga2p, previously shown to be essential for activation of the mammalian RhoA GTPase by the GAP p190 (Li et al, 1997). Furthermore, mutation of this residue to alanine resulted in elimination of GAP activity for Rga1p, another S. cerevisiae Cdc42p GAP, in vitro (Gladfelter et al, 2002), and diminished the interaction between the Rga1p GAP domain and Cdc42p. Substitution of lysine 872 with alanine in Rga2^8Ap resulted in the suppression of both the large, unbudded morphology (Figure 4B) and growth inhibition (Figure 4D) produced by RGA2^8A overexpression. Mutation of the GAP domain reduced the large cell size of RGA2^8A‐expressing cells from 67 to 42 fl, near the size of cells expressing vector or wild‐type RGA2 (Figure 4E). Suppression of RGA2^8A overexpression defects by mutation of the GAP domain implies that hypo‐phosphorylation of Rga2p leads to hyperactive GAP function.

Rga2p normally functions to counteract Cdc42p activation. We reasoned if CDK‐mediated phosphorylation of Rga2p acts to restrain its activity, then phosphorylation‐deficient Rga2p should be biochemically hyperactive. To test this idea, we assessed activated Cdc42p (Cdc42p‐GTP) levels in CDK mutants and cells expressing RGA2^8A. We predicted that the significant growth defects seen in pho85Δ or pclΔ cells overexpressing RGA2 or in wild‐type cells expressing a hypo‐phosphorylated derivative of Rga2p (RGA2^8A) reflect hyperactive GAP (Rga2p) activity, and would be evident as decreased levels of GTP‐bound Cdc42p in vivo. We used a CRIB domain from the human PAK1 kinase, a known Cdc42p effector, in a GST pull‐down assay to specifically recover Cdc42p‐GTP from yeast cell extracts (Caviston et al, 2002; Aguilar et al, 2006). We reproducibly recover less Cdc42‐GTP from CDK mutant extracts (Figure 5A, lanes 5–7), in particular from a pho85Δ extract (Figure 5A, lane 5; see also Supplementary Figure 4). Expression of RGA2^8A from the RGA2 chromosomal locus has little effect on growth, unless sensitized by reduction of function of CDC24 (Figure 4C). Consistent with this observation, we see only a slight decrease in the levels of Cdc42p‐GTP when RGA2^8A is expressed at endogenous levels, compared with wild type (Figure 5A, lane 1 versus lane 2). As well, we see relatively reduced levels of Cdc42‐GTP when Rga2^8A is the only endogenous GAP (RGA2^8Abem3Δrga1Δ versus rga2Δbem3Δrga1Δ; Figure 5A, lane 3 versus lane 4). A shift in the amount of Cdc42p‐GTP relative to total Cdc42p suggests either (1) a decrease in the activity of the Cdc42p GEF, Cdc24p, or (2) an increase in the activity of a Cdc42p GAP(s). Nonetheless, these data are consistent with our model that phosphorylation restricts Rga2p activity and that Rga2p is hyperactive in the absence of Pho85p function due to a lack of phosphorylation. This hypothesis is further strengthened by the fact deletion of PCL1 and PCL2 exacerbates the phenotype of a cdc24‐4 strain (Figure 5B).

To substantiate the kinase–substrate relationship between G1 CDKs and Rga2p in vivo, we used a TAP‐tagged version of Rga2p to examine Rga2p phosphoforms throughout the cell cycle. For these experiments, synchronized cells were released from a G1 arrest and samples were taken periodically for Western blotting. We found that Rga2p underwent an electrophoretic mobility shift 30–45 min after release from G1 arrest, coincident with DNA replication and bud emergence (Figure 6A). Treatment of extracts with phosphatase caused collapse of the Rga2p band, indicating that the reduced electrophoretic mobility is due to phosphorylation (Figure 6B). We assayed Rga2p phosphorylation in strains expressing an analogue‐sensitive allele of CDC28 (cdc28‐as) or PHO85 (pho85‐as) that can be specifically inhibited in vivo with the ATP analogue 1NM‐PP1 or 1Na‐PP1, respectively (Bishop et al, 2000; Carroll et al, 2001). We failed to detect a substantial change in the timing or level of Rga2p phosphorylation throughout the cell cycle when either PHO85 or CDC28 was singly genetically impaired, either using the as alleles or by removing specific cyclins (Figure 6C).

We reasoned that our failure to detect an obvious change in phosphorylation in single CDK mutants reflects the clear genetic redundancy of CDK function (Moffat and Andrews, 2004). Since cells lacking all G1‐CDK activity are inviable, we chose to assay cell cycle‐dependent phosphorylation of Rga2^8Ap, the version of Rga2p lacking confirmed CDK phosphorylation sites (Figure 6D). Cells expressing Rga2^8Ap failed to accumulate G1‐specific phosphoforms, suggesting that the mutant protein indeed lacks sites that are targeted for phosphorylation in vivo. No additional discernable shift in Rga2^8Ap mobility was detectable when the mutant protein was expressed in cln1Δcln2Δ cells (data not shown). These data support our hypothesis that both Pho85p and Cdc28p contribute to the phosphorylation of Rga2p in G1 phase.

As noted earlier, the localization of Rga2p is cell cycle‐dependent; Rga2p localizes to the incipient bud site in unbudded cells, whereas small‐budded cells display Rga2p at the bud tip. Rga2p fails to localize to any discrete region in medium‐budded cells (Caviston et al, 2003), and relocalizes to the bud neck during mitosis (see Figure 2A). To assess whether phosphorylation plays a role in proper localization of Rga2p, we examined localization of Rga2p‐GFP in various CDK and cyclin mutants. We were unable to detect any significant differences in Rga2p‐GFP localization in a pcl1Δ2Δ9Δ, pho85Δ, cln1Δcln2Δ, or aΔpho85as strain incubated with 25 μM analogue (data not shown). A GFP‐tagged version of Rga2^8Ap did not exhibit any observable differences in localization relative to wild‐type Rga2p‐GFP. Examination of Rga2^8Ap in a cln1Δcln2Δbackground however revealed a significantly altered localization: (1) more cells displayed signal, including medium‐budded cells that do not normally have any localized Rga2p‐GFP signal and (2) small and medium‐budded cells had more signal along the circumference of the bud (Figure 7A). Since Rga2^8Ap lacks abundant phosphorylation in late G1 phase, we conclude that phosphorylation of Rga2p plays a role in proper localization of this GAP.

Given that substitution of lysine 872 with alanine abrogated the effects of Rga2^8Ap overproduction (Figure 4B, D and E) and substitution of this conserved residue has been shown to abolish GAP–Cdc42p interactions (Gladfelter et al, 2002), we next examined localization of Cdc42p and Rga2^8Ap. Rga2^8A‐GFP and mcherry‐Cdc42p indeed have a largely overlapping and coincident localization along the circumference of small‐budded cln1Δcln2Δ cells (Figure 7B). This observation supports our prediction that Rga2^8A more readily interacts with Cdc42p. In fact, cells overexpressing RGA2^8A fail to localize Cdc42p to a discrete site as required for initiation of bud emergence (Supplementary Figure 5).

Standard yeast growth conditions were used. All yeast gene disruptions were achieved by homologous recombination by standard polymerase chain reaction (PCR)‐based methods, and verified by PCR and phenotypic analyses. Point mutants were made using PCR‐based site‐directed mutagenesis and confirmed by sequencing. Cell size analysis and synchronization were performed as described (Costanzo et al, 2004).

Cells were photographed with a CoolSNAP HQ high‐speed digital camera (Roeper Scientific) mounted on a Leica DM‐LB microscope. Images were captured and analyzed using MetaMorph software (Universal Imaging Media, PA). For confocal microscopy, proteins were visualized with a Leica DMI 6000B fluorescence microscope equipped with a spinning‐disc head and an argon laser (458, 488, and 514 nm) (Quorum Technologies), and an ImagEM‐CCD camera (Hamamatsu, Japan), and analyzed using Volocity software (distributed by Quorum Technologies Inc., Guelph, ON). Z‐stacks of 10 spinning‐disc confocal images separated by 0.2 μm were taken.

Recombinant Pcl–Pho85p and Cln2p–Cdc28p kinases were expressed and purified from insect cells as described (Huang et al, 1999). Substrates included 150 ng of five truncated versions of Rga2p: amino acids 1–240, 233–365, 357–658, 652–795, and 789–1009 (all N‐terminal GST fusions were expressed in Escherichia coli and purified using glutathione–Sepharose columns). Kinase reactions were performed as described (Costanzo et al, 2004). Details of the methods for purifying and processing phosphorylated proteins for mass spectrometry are available at the Andrews lab website (www.utoronto.ca/andrewslab/).

GTP‐bound Cdc42p in different lysates overexpressing CDC24 was detected using a GST‐PAK (a GST fusion of the CRIB domain from human Pak1) pull‐down assay adapted from Caviston et al (2002) and Aguilar et al (2006).

Western blotting was performed using monoclonal α‐myc 9E10 (produced by University of Toronto monoclonal antibody facility), polyclonal α‐TAP (Open Biosystems), polyclonal anti‐Cdc42 antibody (Santa Cruz Biotechnology, sc‐87), and monoclonal α‐Flag M2 (Sigma) antibodies. For immunoprecipitation, cells were disrupted in lysis buffer (50 mM Tris–HCl, pH 7.5, 100 mM NaCl, 1 mM EDTA, 5 mM NaF, and protease inhibitors) and clarified by centrifugation at 13k r.p.m. for 10 min. Extracts were diluted with IP buffer (50 mM Tris–HCl, pH 7.5, 100 mM NaCl, 1% Triton X‐100, 5 mM NaF, and protease inhibitors) and incubated with IgG sepharose (Amersham Biosciences). Resin was washed 3 × with IP buffer and resuspended in 2 × sample buffer. Extract and supernatant from resin were separated by electrophoresis on an SDS/8% polyacrylamide gel and transferred to PVDF membrane for Western blotting. Samples to be treated with lambda phosphatase were washed 2 × and resuspended in 200 μl buffer (2 mM MnCl₂, 100 mM NaCl, 2 mM DTT, 0.1 mM EGTA, 0.01% Brij 35, and 50 mM Tris–HCl, pH 8.0).

Supplementary data are available at The EMBO Journal Online (http://www.embojournal.org).