The p27Kip1 protein associates with G1-specific cyclin–CDK complexes and inhibits their catalytic activity. p27Kip1 is regulated at various levels, including translation, degradation by the ubiquitin/proteasome pathway and non-covalent sequestration. Here, we describe point mutants of p27 deficient in their interaction with either cyclins (p27c−), CDKs (p27k−) or both (p27ck−), and demonstrate that each contact is critical for kinase inhibition and induction of G1 arrest. Through its intact cyclin contact, p27k− associated with active cyclin E–CDK2 and, unlike wild type p27, p27c− or p27ck−, was efficiently phosphorylated by CDK2 on a conserved C-terminal CDK target site (TPKK). Retrovirally expressed p27k− was rapidly degraded through the proteasome in Rat1 cells, but was stabilized by secondary mutation of the TPKK site to VPKK. In this experimental setting, exogenous wild-type p27 formed inactive ternary complexes with cellular cyclin E–CDK2, was not degraded through the proteasome, and was not further stabilized by the VPKK mutation. p27ck−, which was not recruited to cyclin E–CDK2, also remained stable in vivo. Thus, selective degradation of p27k− depended upon association with active cyclin E–CDK2 and subsequent phosphorylation. Altogether, these data show that p27 must be phosphorylated by CDK2 on the TPKK site in order to be degraded by the proteasome. We propose that cellular p27 must also exist transiently in a cyclin-bound non-inhibitory conformation in vivo.
Progression through the various phases of the mitotic division cycle in all eukaryotic cells depends upon the activity of specific cyclin–cyclin-dependent kinase (CDK) complexes. CDK activity is regulated at multiple levels, including association with different cyclins, phosphorylation/dephosphorylation and association with a group of inhibitory proteins collectively called CKIs (reviewed by Morgan, 1995; Sherr and Roberts, 1995). Selective and time-controlled degradation of several of these regulatory proteins by the ubiquitin/proteasome pathway is also emerging as a major and universal mechanism controlling irreversible transitions during the cell cycle (reviewed by King et al., 1996; Nasmyth, 1996). Ubiquitination of cellular proteins requires the activity of three classes of enzymes: a ubiquitin activating enzyme (or E1), a ubiquitin conjugating enzyme (or E2) and a ubiquitin–protein ligase (or E3). Polyubiquitinated polypeptides are recognized as substrates by the proteasome, a large protein-degrading complex. This processing pathway degrades a large number of specific cellular proteins, and thus participates in many physiological processes (for reviews: Hershko and Ciechanover, 1992; Jentsch, 1992; Ciechanover, 1994; Isaksson et al., 1996; Pahl and Baeuerle, 1996).
Two major phases of the eukaryotic cell cycle, mitosis and the G1–S transition, are associated with selective protein degradation. During mitosis, chromosome separation at the metaphase–anaphase transition requires degradation of the yeast Pds1p protein (Cohen-Fix et al., 1996) and possibly of additional, yet unidentified proteins. At the end of mitosis, inactivation of CDK1 (Cdc28, CDC2) is brought about by degradation of mitotic cyclins. Ubiquitination of these mitotic targets is dependent upon the activity of a multiprotein E3 complex known as the cyclosome, or anaphase-promoting complex (APC; reviewed by King et al., 1996; Nasmyth, 1996). In both yeast and mammalian cells, degradation of mitotic cyclins by the APC is sustained until late G1, preventing premature accumulation of mitotic cyclins (Amon et al., 1994; Brandeis and Hunt, 1996). A second complex, different from the APC, directs distinct substrates to the ubiquitin/proteasome pathway in late G1. In yeast, this complex is required for S-phase entry and includes the products of the CDC34, CDC4, CDC53 and SKP1 genes (Bai et al., 1996; Mathias et al., 1996; Willems et al., 1996; reviewed by Jackson, 1996; King et al., 1996; Nasmyth, 1996). Genetic data imply that the essential substrate of this pathway for the G1–S transition is SIC1, an inhibitor of S-phase cyclin–CDK complexes (Nugroho and Mendenhall, 1994; Schwob et al., 1994). Other substrates include FAR1 (S.Henchoz, Y.Chi, B.Catarin, I.Herskowitz, R.Deshaies and M.Peter, submitted)_a CKI involved in pheromone-induced arrest (Peter and Herskowitz, 1994)_and G1 cyclins (CLN2 and 3) (Deshaies et al., 1995; Yaglom et al., 1995). In mammalian cells, both cyclin E and the CKI p27Kip1 (p27) are substrates of the ubiquitin/proteasome pathway (Pagano et al., 1995; Clurman et al., 1996; Won and Reed, 1996). Whether the mammalian counterpart of the CDC34 pathway mediates cyclin E or p27 degradation in vivo remains unclear, but the human CDC34 homologue Ubc3 can mediate p27 ubiquitination in vitro (Pagano et al., 1995).
At the outset of this work, we sought to understand the role of specific molecular interactions in p27 function and regulation. Our strategy was to design specific mutants of p27 affecting selected protein–protein contacts and to characterize their behaviour both in vitro and in vivo, by retroviral transduction in cultured cells (Vlach et al., 1996). Our results provide genetic and biochemical evidence that in order to be degraded by the proteasome, p27 must associate with cyclin E–CDK2 complexes in a non-inhibitory mode and be phosphorylated by CDK2 within these complexes on a specific threonine residue. Thus, phosphorylation-dependent degradation of cyclins and CKIs appears to be a universally conserved mechanism. Finally, we also discuss evidence that p27 sequestration induced by c-Myc and phosphorylation-induced degradation are two independent molecular events.
Cyclin- and CDK-specific interaction mutants of p27
The specificity of p27 mutations was addressed by co-incubating GST–p27 fusion proteins expressed in Escherichia coli with human cyclin E and CDK2 expressed in Sf9 cells; complexes were pulled down with glutathione beads and the associated cyclin and/or CDK was revealed by immunoblotting. GST–p27wt associated independently with cyclin E or CDK2 (Figure 2A, lane 2). GST–p27c− associated only with CDK2, and GST–p27k− only with cyclin E (lanes 3 and 4). When incubated with cyclin E–CDK2 complexes, GST–p27k− co-precipitated both proteins as efficiently as GST–p27wt, indicating formation of stable ternary complexes (Figure 2B, lanes 2 and 4). In contrast, the presence of cyclin E reduced the interaction between GST–p27c− and CDK2 (lane 3), suggesting that this mutant did not efficiently form ternary complexes. GST–p27ck− failed to co-precipitate cyclin E or CDK2, alone or in combination (Figure 2A and B, lane 5). These experiments were repeated with cyclin A–CDK2 and cyclin D1–CDK4 complexes. Only one notable difference was observed, in that GST–p27c− interacted with these complexes as efficiently as GST–p27wt (Figure 2C and D, compare lanes 2 and 3). However, the interaction pattern of individual cyclin D1 or CDK4 with the various GST–p27 proteins was the same as for cyclin E and CDK2, respectively (Figure 2A and D). Baculovirus-expressed cyclin A did not interact with GST–p27 (wt or mutants) in the absence of CDK2. In summary, p27c− and p27k− were deficient in associating with individual cyclins and CDKs, respectively. However, through their reciprocal intact interaction motifs, these mutants still formed stable ternary complexes with cyclin–CDK pairs (with the exception of p27c−–cyclin E–CDK2). The double mutant p27ck− lost the ability to form stable binary or ternary complexes.
Cyclin- and CDK-contacts are required for efficient kinase inhibition by p27
The in vitro kinase assays described below (Figures 3, 4, 5) were all performed with purified p27 proteins released from GST–p27 by thrombin cleavage. We assayed the CDK-inhibitory activities of p27 mutants by co-incubating cyclin E–CDK2 or cyclin A–CDK2 with serial dilutions of the p27 proteins and performing histone H1-kinase assays directly in solution. H1-kinase activity was strictly dependent upon the presence of cyclin E or A together with CDK2 in the Sf9 cell lysates (data not shown). p27c− and p27k− inhibited cyclin E–CDK2 markedly less effectively than p27wt, with half-inhibitory concentrations elevated by about one order of magnitude, whereas p27ck− was non-inhibitory (Figure 3A). p27k− antagonized cyclin A–CDK2 better than cyclin E–CDK2 (Figure 3A and B), although it failed to quantitatively suppress either activity at high concentration (Figure 3C and D, lanes 4). In contrast, p27c− quantitatively inhibited cyclin A–CDK2 at high concentration (Figure 3B and D, lane 3), correlating with the stable association between these proteins (Figure 2C). Upon dilution, however, p27c− reproducibly lost inhibitory activity faster than p27k− (Figure 3B, 10 ng), suggesting that its interaction with the complex was less effective. Results similar to those shown in Figure 3 (C and D) were obtained with kinase assays performed after immunoprecipitation of CDK2 or cyclins in the presence of GST–p27 proteins. Furthermore, the inhibitory pattern of GST–p27 mutants on cyclin D1–CDK4 was comparable with that on cyclin A–CDK2 (data not shown). In conclusion, both cyclin- and CDK-contacts were critical for the function of p27 as a CDK inhibitor.
Site-specific phosphorylation of p27k− by CDK2
In the above experiments, we noted that p27k− was selectively phosphorylated by cyclin E–CDK2 in histone H1-kinase assays (Figure 3C, lane 4). p27 contains three sites that conform with the minimal Ser/Thr-Pro (S/T-P) consensus CDK phosphorylation site. In murine p27, these are serine 10 (SPSL), serine 178 (SPNA) and threonine 187 (TPKK). Mutation of Thr187 to valine in p27k− (mutant p27k−V) reduced its phosphorylation to background levels (Figure 3C, lane 6). In contrast, mutation of both Ser10 and Ser178 (mutant p27k−AA) had no effect on phosphate incorporation (lane 7). The interaction patterns and H1-kinase-inhibitory activities of p27k−V and p27k−AA were identical to those of p27k− (Figures 2A–D and 3A–D; data not shown). Thus, p27k− was phosphorylated on Thr187 by cyclin E–CDK2.
The phosphorylation of p27 proteins was further investigated by co-incubation with cyclin–CDK2 complexes followed by kinase assays in the absence of histone H1. Neither p27wt nor p27c− were significantly phosphorylated (Figure 4A, lanes 2 and 3). At high p27 concentrations, both p27k− and p27ck− were phosphorylated by cyclin E–CDK2 (Figure 4A, panel i, lanes 4 and 5, and B, lanes 1 and 2). This suggested that the lack of p27ck− phosphorylation in the presence of excess (5 μg) histone H1 (Figure 3C, lane 5) was most likely due to substrate competition, and that the recruitment of p27k− through interaction with cyclin E allowed its phosphorylation under both conditions. Furthermore, when p27 proteins were diluted 10-fold in the H1-free kinase reaction, p27ck− phosphorylation by cyclin E–CDK2 was lost, whereas p27k− phosphorylation was unaltered (Figure 4B, lanes 3 and 4), showing that recruitment became rate-limiting for phosphorylation. The same experiments were repeated with cyclin A–CDK2, with one notable difference: while p27ck− was phosphorylated by this complex as efficiently as by cyclin E–CDK2, p27k− was a poor substate (Figure 4A, lanes 4 and 5, and B, lanes 5 and 6). As with cyclin E–CDK2, phosphorylation of p27ck− by cyclin A–CDK2 was lost upon dilution (Figure 4B, lanes 6 and 8), or in the presence of histone H1 (Figure 3D, lane 5). As control, none of the p27 proteins were phosphorylated in Sf9 lysates lacking cyclin–CDK complexes (Figure 4B, lane 9; data not shown). In summary, the non-interacting mutant p27ck− was phosphorylated by cyclin E–CDK2 or cyclin A–CDK2 in vitro in a concentration-dependent manner, analogous to histone H1. On the other hand, recruitment of p27k− through the cyclins suppressed its phosphorylation by cyclin A–CDK2, but favoured its phosphorylation by cyclin E–CDK2 at lower concentrations. Comparison of p27k− with the phosphosite mutants p27k−V and p27k−AA showed that p27k− phosphorylation occurred on Thr187 (Figure 4A, lanes 4, 6 and 7). Finally, cyclin D1–CDK4 phosphorylated all p27 proteins (wt, c−, k− or ck−) with an equal, albeit low efficiency, and on a site distinct from Thr187 (data not shown).
p27k− associates with active cyclin–CDK2 complexes
Altogether, the above data implied that p27k− should associate with kinase-active cyclin E–CDK2. To address this point directly, we immunoprecipitated ternary complexes with anti-p27 antibodies, and perfomed histone H1-kinase assays. Complexes formed with p27wt were devoid of any kinase activity (Figure 5, lane 2). In contrast, p27k− co-precipitated H1-kinase activity (lane 4). This activity was significantly above background levels (lane 1), was eliminated by peptide-blocking of the anti-p27 antibody (lane 8), and was dependent upon cyclin E and CDK2 in Sf9 lysates (data not shown). Furthermore, this activity was not due to release of the kinase from p27k− during incubation, since it was fully resistant to pre-incubation and washes in kinase-reaction buffer (data not shown). p27k−-associated activity was not dependent upon the presence of the phosphoacceptor site Thr187, since p27k−V also bound active complexes (Figure 5, lane 6) (the same was true for the Ser 10/178 mutant p27k−AA: lane 7). Neither p27c− nor p27ck− associated with active CDK2 (Figure 5, lanes 3 and 5). Thus, p27k− associated with active CDK2 through its direct contact with cyclin E.
In apparent conflict with this finding, we show in Figure 3A that p27k− had a residual inhibitory activity on cyclin E–CDK2. However, both sets of data were highly reproducible, either with purified p27 or with GST–p27 proteins. This implies that p27k−-associated cyclin E–CDK2 is only partially active in phosphorylating histone H1, in comparison with the free kinase complex. This may be due to substrate competition between p27k− and H1, residual inhibitory activity of p27k−, or both.
Cyclin A–CDK2 was unable to phosphorylate histone H1 when associated with p27k− (data not shown). This correlated with the inability of cyclin A–CDK2 to phosphorylate p27k− (Figures 3D and 4A) and with its more efficient inhibition by p27k− in solution (Figure 3B). Altogether, these observations implied that the effect of the p27k− mutation on CDK2 inhibition was less severe if cyclin A was present in the complex.
Cyclin- and CDK-contacts are critical for p27-induced G1 arrest
To assess the biological function of p27 mutants, we expressed these proteins in Rat1 cells from a retroviral vector. p27wt directly inhibited cyclin E–CDK2 and caused G1 arrest. Cyclin A mRNA and protein levels were suppressed in these cells, and thus also cyclin A–CDK2 activity (Vlach et al., 1996). In contrast, viruses expressing p27c−, p27k− or p27ck− induced no significant suppression of cellular cyclin E- and A-associated kinase activities (Figure 6A), cell cycle progression (Figure 6B) or long-term proliferation (data not shown).
p27k− is degraded at an enhanced rate in vivo
The expression of p27 mutants was assessed by immunoblotting (Figure 7A). We previously showed that retroviral p27wt was expressed at physiological levels (comparable with those of endogenous p27 in contact-inhibited cells; Vlach et al., 1996). In comparison with p27wt (lane 2), p27c− levels were unaltered and p27ck− levels were slightly enhanced (lanes 3 and 5). In contrast, p27k− accumulated in cells at low levels (lane 4). CDK2 levels were the same in all cell populations (Figure 7A). Decreased accumulation of p27k− was unlikely to result from differences in translational control (Hengst and Reed, 1996), since only the open reading frame of p27 was included in our constructs. Furthermore, retroviral mRNA levels in infected cells were the same for all mutants (data not shown). Pulse–chase analysis of 35S-labelled, retrovirally infected cells showed that p27k− was degraded at an enhanced rate in comparison with p27wt (Figure 8A, panels i and ii).
Degradation of p27k− requires the Thr187 phosphoacceptor site
Together with our in vitro data, the above observations suggested that phosphorylation of p27k− by cyclin E–CDK2 accounted for its instability in vivo. In this case, mutation of Thr187 to valine (as in p27k−V) should have restabilized the protein. This point was demonstrated. First, p27k−V accumulated at levels higher than those of p27k− and comparable with those of p27wt (Figure 7A, lanes 2, 4 and 6). Yet, like p27k−, p27k−V failed to suppress cellular cyclin A- and E-associated kinase activities or cell cycle progression (Figure 6). In contrast to Thr187, mutation of alanines 10, 178 or both in p27k− had no significant effects on the accumulation of the protein (data not shown). Second, p27k−V remained stable throughout the time course of our 35S-pulse–chase experiments (Figure 8A, panel iii).
To attempt a direct visualization of Thr187 phosphorylation in p27k−, we metabolically labelled infected cells with [32P]orthophosphate. However, no labelling of p27k− could be detected (Figure 8B, lane 2), most likely owing to the high degradation rate of this protein once phosphorylated. Furthermore, p27wt, p27c−, p27ck− and the phosphosite mutant p27k−V incorporated 32P at equal levels (Figure 8B, lanes 3–5; data not shown). Thus, p27 is phosphorylated in vivo on additional site(s) distinct from Thr187, and these phosphorylation events are independent of cyclin- and CDK-contacts. These circumstances precluded direct visualization of Thr187 phosphorylation in p27k−.
p27k− associates with active CDK2 in vivo
Since p27k− associated with active cyclin E–CDK2 in vitro (Figure 5), we investigated whether this could happen in vivo. To this end, retrovirally expressed p27 proteins were immunoprecipitated from cells, followed by H1-kinase assays. As expected from the in vitro results, p27wt, p27c− and p27ck− associated with no significant kinase activity (Figure 9A, lanes 2, 3 and 5). In contrast, p27k− co-precipitated active H1-kinase (lane 4). This was observed with two anti-p27 antibodies (C-19 and Neo Markers), and no kinase activity was recovered when the C-19 antibody was blocked with its cognate peptide (data not shown).
Cellular cyclin E and CDK2 associated with exogenous p27wt could be detected by immunoprecipitation followed by immunoblot analysis (Vlach et al., 1996). However, decreased levels of p27k− precluded detection of these interactions (data not shown). To identify the cellular kinase associated with p27k−, we treated kinase reactions with the CDK-inhibitory drugs Flavopiridol and Roscovitine. Both drugs eliminated kinase activity in p27k− immunoprecipitates (Figure 9B). At the concentrations used, Flavopiridol inhibits both CDK2 and CDK4, whereas Roscovitine inhibits only CDK2 (reviewed by Meijer, 1996). Together with our in vitro studies, these data imply that p27k− associates in cells with active CDK2 complexes, most likely through its recruitment by cyclin E.
p27k− degradation is mediated by the proteasome
p27 is a known substrate of the ubiquitin/proteasome pathway (Pagano et al., 1995). To test whether this pathway was responsible for degradation of p27k−, we treated infected cells with the proteasome inhibitor LLNL (Rock et al., 1994). This resulted in increased accumulation of p27k−, but not of p27wt, p27ck− or p27c− (Figure 10A and data not shown). Since LLNL also inhibits calpain, we tested the calpain-specific inhibitors LLM, E-64 and NCO-700 (see Materials and methods). Only LLNL had a significant stabilizing effect on p27k− (Figure 10B, panel i, lanes 6–10). Longer exposure of the immunoblot showed that endogenous p27 in control cells was specifically stabilized by LLNL (panel ii, lanes 1–5), in a manner analogous to p27k− and consistent with previous data (Pagano et al., 1995). That endogenous p27wt did not reach total levels comparable with those of p27k− in the presence of LLNL was expected, since the rate of synthesis of endogenous p27 was very low in growing cells, as demonstrated by our in vivo35S-labelling experiments (Figure 8A, panel i: compare ‘vector’ with p27k−, in the absence of chase). Altogether, our data show that the decreased expression of p27k− is due to enhanced degradation by the proteasome.
Retrovirally expressed wild-type p27 remains stable in G1-arrested cells
Like retrovirally expressed p27k−, endogenous p27wt in growing cells turned over through the proteasome (Figure 10B) (Pagano et al., 1995). However, retrovirally expressed p27wt was not degraded at a high rate, as shown by immunoblotting (Figure 7A, lane 2), pulse–chase experiments (Figure 8A, panel ii) and by the fact that LLNL did not increase its levels (Figure 10A). As already mentioned, exogenous p27wt associated with cyclin E–CDK2 in an inactive form (Figure 9A) (Vlach et al., 1996) and, under these circumstances, did not become phosphorylated on Thr187 (Figure 4). Consistent with these observations, mutation of Thr187 in the context of wild-type p27 (mutant p27V187) did not stabilize the protein in vivo (Figure 7B), and did not alter its CDK- and growth-inhibitory activities (data not shown).
In summary, the outcome of ectopic p27wt expression was the formation of inactive complexes with cyclin E–CDK2, with ensuing lack of phosphorylation on Thr187 and lack of targeting to the proteasome. In contrast, through its interaction with cyclin E, the CDK-contact mutant p27k− was recruited to active CDK2, phosphorylated on Thr187 and degraded by the proteasome. Altogether, our data imply that p27 degradation in cells requires its transitory association with active cyclin E–CDK2 and phosphorylation on Thr187, and suggest that this process might be regulated by additional signals.
Cyclin- and CDK-interaction mutants of p27Kip1: implications for cyclin–CDK2 function and specificity
We have constructed two mutants of murine p27Kip1 specifically deficient in interacting with cyclins (p27c−) and CDKs (p27k−), and the corresponding double mutant (p27ck−) defective in both interactions. Using these mutants, we demonstrated that contacts with both cyclin and CDK subunits are required for the optimal function of p27 as a kinase inhibitor. Two residues were mutated to alanines in each mutant: arginine 30/leucine 32 in p27c− and phenylalanines 62/64 in p27k− (Figure 1A). These residues lie within the previously mapped cyclin–CDK-binding domains of mammalian Kip/Cip family proteins, and are part of conserved sequence motifs in all CKIs of this family, including p21Cip1, p27Kip1, p57Kip2, Xenopus p27Xic1, two putative C.elegans gene products, as well as the Drosophila CKI Dacapo (Figure 1A, see references in first paragraph of Results). The crystal structure of p27–cyclin A–CDK2 ternary complexes showed that the targeted residues contact the cyclin and CDK subunits, as predicted (Russo et al., 1996). We propose in Figure 1A a possible alignment of the yeast CKI SIC1 with the Kip/Cip family. It will be of interest to address whether mutations equivalent to those in p27c− or p27k− affect interaction of SIC1 with its cyclin and CDK partners in yeast.
The cyclin-binding motif (or RXL motif) mutated in p27c− is found in proteins other than CKIs, including the retinoblastoma family members p107 and p130, as well as E2F-1, -2 and -3 (Figure 1B) (Zhu et al., 1995; Adams et al., 1996). Several observations suggest that these different RXL motifs indeed bind a common determinant on cyclins. Peptides spanning the RXL motif from representative members of all groups shown in Figure 1 can compete with recombinant GST–p21 or GST–E2F-1 for interaction with cyclins. Mutations of the conserved Arg or Leu to Ala in E2F-1-derived peptides, which correspond to the two substitutions introduced here in p27c−, eliminated their competing activity (Adams et al., 1996). Besides the RXL motif, p107, p130 or E2F proteins lack any obvious CDK-contact motif and can act as targets or tethers for active CDK2. For example, binding of cyclin A–CDK2 to E2F-1/DP-1 dimers results in their phosphorylation and suppression of DNA-binding activity, a mechanism thought to play an important role in downregulation of E2F function during S-phase (Dynlacht et al., 1994; Krek et al., 1994).
Consistent with the notion that the RXL motif can target substrates to active cyclin–CDK complexes, the CDK-contact mutant p27k− (in which the RXL motif was intact; Figure 1A) bound to active cyclin E–CDK2 and was efficiently phosphorylated on Thr187. Further mutation of the RXL motif, such as in the double mutant p27ck−, led to a loss of phosphorylation at limiting substrate concentrations. At high concentrations, p27ck− was phosphorylated by both cyclin E–CDK2 and cyclin A–CDK2, analogous to histone H1. Interestingly, p27k− also associated with cyclin A–CDK2, but was not significantly phosphorylated by this complex. Thus, recruitment of p27k− through the cyclin was inhibitory for Thr187 phosphorylation in the case of cyclin A, but stimulatory in the case of cyclin E. This stimulatory effect became evident at low p27k− concentrations or in the presence of excess histone H1, indicating that interaction with cyclin E became rate-limiting for phosphorylation under these conditions. In our experiments, neither cyclin E–CDK2 nor cyclin A–CDK2 phosphorylated p27wt, correlating with the formation of inactive ternary complexes. Cyclin D1–CDK4 phosphorylated p27wt, p27c−, p27k−, p27ck or the Ala187 derivative of p27k− (p27k−V) to identical, albeit very low levels, which may bear no physiological relevance.
While our work was in revision, Sheaff et al. (1997) reported that wild-type p27 (p27wt) could be phosphorylated by cyclin E–CDK2 in vitro on Thr187. Analysis of phosphorylation kinetics and of the effects of ATP concentration led these authors to propose that p27 bound to cyclin E–CDK2 in two conformations: first in a ‘loose’ state, under which CDK2 phosphorylated p27, and secondly in a ‘tight’ state, under which the kinase was inhibited. The loose state may be analogous to that adopted by our mutant p27k−, which does not undergo the transition to the inhibitory conformation. The tight state most likely is equivalent to the described structure of p27–cyclin A–CDK2 complexes, in which ATP binding to CDK2 is precluded (Russo et al., 1996). The residues mutated in p27k− (Figure 1A) play an important role in this structure. These observations altogether imply that p27 must transiently exist in a cyclin E-bound non-inhibitory conformation in order to become an effective CDK2 substrate.
The cyclin-contact mutant p27c− also formed stable ternary complexes with cyclin A–CDK2 and cyclin D1–CDK4, in this case through its intact CDK-contact domain. In contrast, the presence of cyclin E reduced interaction of p27c− with CDK2. This suggests either that CDK2 adopts slightly different conformations when bound to cyclin A or E, or that cyclin A can sterically hinder interactions of CDK2 with p27c−. Thus, interaction of CDK2 with other proteins (exemplified here by p27c−) may be differentially affected by cyclin E or A.
Recruitment of p27 to active cyclin E–CDK2 complexes triggers its degradation in vivo
When expressed in Rat1 cells with a retroviral vector, p27wt inactivated cyclin E–CDK2 and arrested cells in G1 (Vlach et al., 1996). All of the p27 mutants described here had lost these properties. While p27c− and p27ck− were expressed in vivo at levels equal to, or slightly higher than those of p27wt, the CDK-contact mutant p27k− accumulated to much lower levels in vivo. 35S-pulse–chase studies showed that this was due to enhanced degradation of p27k−. Treatment of cells with the proteasome inhibitor LLNL elevated p27k− levels, but had no effect on the other recombinant p27 proteins, showing that p27k− was degraded by the proteasome, as previously shown for cellular p27 (Pagano et al., 1995).
From our in vitro data, we predicted that association of p27k− with active cyclin E–CDK2 in cells would lead to phosphorylation on Thr187, and that this effect would trigger the enhanced turnover of the protein. We were able to prove the validity of this hypothesis. First, consistent with our in vitro data, we observed that p27k− recovered from cells was associated with catalytically active CDK2. No significant kinase activity was recovered with p27wt, p27c− or p27ck− from cells, even though those proteins were expressed at higher levels. Second, mutation of threonine 187 to valine in p27k− (mutant p27k−V) led to restabilization of the protein, without preventing its stable association with active cyclin–CDK2 complexes. Thus, although we could not biochemically detect phosphorylation of Thr187 in vivo (see Results), our data demonstrated that this phosphosite was required for degradation. Third, degradation of p27k− was strictly dependent upon interaction with cyclin E, since the double mutant p27ck− which did not bind cyclin E–CDK2 complexes was not degraded by the proteasome. In addition, we showed in vitro that although both cyclin E and cyclin A recruited p27k− onto CDK2, only cyclin E–CDK2 efficiently phosphorylated p27k−. Cyclin D1–CDK4 phosphorylated p27wt and all the mutants very inefficiently and with no discrimination. This implies that cyclin E, rather than cyclin A or D1, was responsible for the turnover of p27k−in vivo.
Our results with p27k− imply that cellular p27, in order to be degraded in vivo, must be phosphorylated by CDK2 on Thr187. This model predicts two possible, alternative outcomes upon expression of recombinant p27wt in cells. In the first, p27wt would form inactive ternary complexes with cyclin E–CDK2, would not become phosphorylated on Thr187 and as a consequence would not be targeted to degradation by the proteasome. Consistent with this scenario, retrovirally expressed p27wt associated with inactive cyclin E–CDK2, was stable in our pulse–chase analysis, and was not further stabilized by proteasome inhibitors or by mutation of Thr187. In this system, co-expression of cyclin E with p27wt at moderate levels was not sufficient to bypass CDK2 inhibition, destabilize p27wt or overcome G1 arrest (Vlach et al., 1996). In the second outcome, ectopically expressed p27wt would be phosphorylated by cyclin E–CDK2 on Thr187 and would be rapidly degraded. This was observed in the parallel work of Sheaff et al. (1997), who showed that p27wt expressed by transient transfection was eliminated from cells by co-expression of cyclin E. By generating a high and sudden burst of cyclin E activity, these authors were able to demonstrate that phosphorylation of p27wt on Thr187 preceded its elimination in vivo. As expected, cyclin E suppressed p27-induced G1 arrest in those experiments. Thus, the two complementary approaches used by ourselves and by Sheaff et al. (1997) verify the two major predictions of the proposed model, and establish that phosphorylation of p27 on its carboxy-terminal TPKK site is a prerequisite for its degradation through the proteasome. It is particularly noteworthy that the TPKK site lies within a conserved homology domain found at the carboxy-termini of p27 and p57 (Lee et al., 1995; Matsuoka et al., 1995), suggesting that the two proteins may be subject to similar control mechanisms.
The half-life of p27 increases upon contact-inhibition in HS68 cells and, conversely, decreases upon S-phase arrest in HeLa cells (Hengst et al., 1994). In vitro studies showed that proliferating cell extracts ubiquitinated p27 at higher rates than quiescent extracts (Pagano et al., 1995), and that S-phase extracts degraded p27 whereas mid-G1 extracts did not (Brandeis and Hunt, 1996). In yeast, G1-specific CLN–CDK1 activity is required for ubiquitination and degradation of SIC1, most likely through direct SIC1 phosphorylation (Schneider et al., 1996; Tyers, 1996). SIC1 inhibits S-phase-specific CLB–CDK1 complexes. Thus, through the degradation of SIC1, G1-cyclin–CDK complexes allow activation of S-phase complexes (reviewed by King et al., 1996; Nasmyth, 1996). Our data suggest that degradation of p27 induced by cyclin E–CDK2 may have in part a similar role in preventing inhibition of cyclin A–CDK2 by p27 during S-phase.
Similarly to p27, cyclin E is degraded by the ubiquitin/proteasome pathway following phosphorylation by CDK2 (Clurman et al., 1996; Won and Reed, 1996). A cyclin E mutant deficient in interaction with CDK2 was not phosphorylated, but remained free and was degraded (Clurman et al., 1996). In contrast, our p27ck− mutant, which did not bind cyclin E–CDK2 complexes, was not targeted to degradation. This suggests that the primary effect of phosphorylation is different for cyclin E and p27: to induce release from CDK2 in the case of cyclin E, and to generate a recognition site for components of the degradatory machinery in the case of p27. Alternatively, both mechanisms may be involved in p27 degradation. In yeast, G1 cyclins and both identified CKIs, SIC1 and FAR1, are targeted to the ubiquitin/proteasome pathway in a phosphorylation-dependent manner (see Introduction). Thus, phosphorylation-dependent degradation of G1-specific cyclins and CKIs is a fundamentally conserved process.
One question arising here concerns the relationship between the different mechanisms that regulate p27 function in cells. In particular, p27 can be sequestered away fron cyclin E–CDK2 complexes, either by D-type cyclins (Polyak et al., 1994a; Poon et al., 1995; Reynisdottir et al., 1995; Sherr and Roberts, 1995; Soos et al., 1996) or by an as yet uncharacterized Myc-induced activity (Vlach et al., 1996). Our own observations suggest that Myc-induced sequestration may be mechanistically unlinked from phosphorylation-induced degradation. First, Myc overrode cell cycle arrest by p27 without decreasing p27 levels or half-life in cells. Secondly, Myc could overcome G1-arrest induced by the phosphosite mutant p27V187 (Vlach et al., 1996 and unpublished data). Thus, rescue of p27-induced G1 arrest by Myc does not require Thr187 phosphorylation.
Mutant p27 cDNAs (see text) were constructed using PCR and were subcloned in the retroviral vector pBabePuro. Retroviruses were generated and used to infect Rat1 cells following a previously described protocol (Vlach et al., 1996).
Flavopiridol (NSC 649890) was obtained from the Drug Synthesis & Chemistry Branch, Developmental Therapeutics Program, Division of Cancer Treatment, National Cancer Institute (Bethesda, MD). Roscovitine was a gift from Dr Laurent Meijer (Roscoff). LLNL (N-acetyl-Leu-Leu-norleucinal) was purchased from Sigma (A-6185). LLM (N-acetyl-Leu-Leu-methioninal), E-64 and NCO-700 were purchased from Calbiochem (Nos 208721, 324890 and 479919, respectively).
Recombinant proteins, biochemical studies and antibodies
Mutant p27 cDNAs were subcloned in the vector pGEX-2T and used to express GST fusion proteins in E.coli by standard methods (Pharmacia Biotech protocols). After thrombin cleavage, the released p27 proteins were estimated to be over 90% pure. Non-cleaved GST–p27 represented <0.5% of the purified protein. Baculoviruses expressing cyclins and CDKs were used to infect Sf9 cells. Sf9 cells were lysed in the appropriate lysis buffers for CDK2 (Vlach et al., 1996) or CDK4 (Matsushime et al., 1994), at a final total protein concentration of 3 mg/ml. Inhibition of baculovirus-derived cyclin–CDK complexes by GST–p27 was described previously (Toyoshima and Hunter, 1994). Lysis of retrovirally infected Rat1 cells 2 days post-infection and subsequent biochemical experiments, including immunoprecipitations, immunoblotting and kinase assays, were as previously described (Vlach et al., 1996). The following antibodies were used: against p27, C-19 (Santa Cruz sc-528) or DCS72 (Neo Markers); CDK2, M2 (sc-163); rodent cyclin E, M20 (Santa Cruz sc–481); human cyclin E, C19 (sc-198); cyclin D1, 72-13G (sc-450); CDK4, C-22 (sc-260). For metabolic labelling, infected cells were labelled for 4 h with [35S]methionine/[35S]cysteine (0.11 mCi/ml) or with [32P]orthophosphate (0.5 mCi/ml).
We thank Willy Krek, Laurent Meijer, David Morgan, Edward Sausville and Charles Sherr for the gift of various materials and reagents. The Drug Synthesis & Chemistry Branch (NCI, Bethesda, MD) is acknowledged for the gift for Flavopiridol. We thank Michel Aguet, Kostis Alevizopoulos, Peter Greasley, Richard Iggo, Willy Krek, Kirsten Mundt and Matthias Peter for technical advice, helpful discussions and/or critical comments on the manuscript. We also thank Nathalie Lauper for help with the construction of mutant p27 cDNAs, and Sandra Henchoz, Matthias Peter, Martin Eilers and Jim Roberts for communicating unpublished data. J.V. was supported by a post-doctoral fellowship from the Swiss Cancer League. B.A. is a recipient of a START fellowship and of a research grant from the Swiss National Science Foundation.