Incomplete Folding upon Binding Mediates Cdk4/Cyclin D Complex Activation by Tyrosine Phosphorylation of Inhibitor p27 Protein*

p27Kip1 (p27), an intrinsically disordered protein, regulates the various Cdk/cyclin complexes that control cell cycle progression. The kinase inhibitory domain of p27 contains a cyclin-binding subdomain (D1), a Cdk-binding subdomain (D2), and a linker helix subdomain that connects D1 and D2. Here, we report that, despite extensive sequence conservation between Cdk4/cyclin D1 (hereafter Cdk4/cyclin D) and Cdk2/cyclin A, the thermodynamic details describing how the individual p27 subdomains contribute to equally high affinity binding to these two Cdk/cyclin complexes are strikingly different. Differences in enthalpy/entropy compensation revealed that the D2 subdomain of p27 folds incompletely when binding Cdk4/cyclin D versus Cdk2/cyclin A. Incomplete binding-induced folding exposes tyrosine 88 of p27 for phosphorylation by the nonreceptor tyrosine kinase Abl. Importantly, tyrosine phosphorylation (of p27) relieves Cdk inhibition by p27, enabling cell cycle entry. Furthermore, the interaction between a conserved hydrophobic patch on cyclin D and subdomain D1 is much weaker than that with cyclin A; consequently, a construct containing subdomains D1 and LH (p27-D1LH) does not inhibit substrate binding to Cdk4/cyclin D as it does to Cdk2/cyclin A. Our results provide a mechanism by which Cdk4 (within the p27/Cdk4/cyclin D complex) is poised to be activated by extrinsic mitogenic signals that impinge upon p27 at the earliest stage of cell division. More broadly, our results further illustrate the regulatory versatility of intrinsically disordered proteins.

Progression from the G 1 to S phase of cell division is promoted by sequential activation of several Cdk/cyclin complexes, including Cdk4/cyclin D (and Cdk6/cyclin D), Cdk2/ cyclin E, and Cdk2/cyclin A. In early G 1 phase, however, p27 is present at high levels, which restrains cell cycle progression by potently inhibiting Cdk4/cyclin D and Cdk6/cyclin D and consequently limiting expression of cyclins E and A and activation of their complexes with Cdk2 (17). Recently, it was shown that this restraint on cell cycle entry is relieved by tyrosine phosphorylation of p27 (18). Phosphorylation of tyrosine 88 (Tyr-88) ejects an inhibitory segment of p27 from the active site of Cdk2, partially restores kinase activity (18), and ultimately leads to p27 ubiquitination and degradation at the G 1 -to-S phase boundary (18). This mechanism, which couples nonreceptor tyrosine kinase signaling and cell cycle regulation (14), generates fully active Cdk2/cyclin E and Cdk2/cyclin A complexes by mediating p27 elimination and drives cells into S phase.
The discovery that Cdk/cyclin complexes containing tyrosine-phosphorylated p27 are partially catalytically active (18) provided an explanation for past findings (19) showing that ternary complexes of p21 and p27 with Cdk4/cyclin D were catalytically active. In support of this, Blain and co-workers (20) recently showed that p27-bound Cdk4/cyclin D complexes containing tyrosine-phosphorylated p27 are catalytically active, whereas those containing unmodified p27 are inactive. Thus, a critical regulatory step for cell cycle entry is partial activation of Cdk4 through tyrosine phosphorylation of p27 within Cdk4/ cyclin D complexes, which enables subsequent hypophosphor-ylation of the retinoblastoma protein (Rb) (21,22). The structure of the kinase inhibitory domain of p27 (p27-KID) bound to Cdk2/cyclin A was determined more than 10 years ago (23) and provided detailed insights in the molecular basis of Cdk2 inhibition by p27 (Fig. 1A). Details of the binding mechanism were revealed through kinetic and thermodynamic analyses using ITC and surface plasmon resonance (15). These studies showed that p27 subdomain D1 rapidly binds to a conserved binding pocket on the surface of cyclin A, followed by folding of other portions of p27 upon binding to and remodeling of Cdk2. During the first step of this sequential binding mechanism, the binding of p27 subdomain D1 causes only subtle structural changes in cyclin A. After p27-D1 binding to cyclin A, the LH subdomain of p27 fully folds into an ␣-helix that bridges over to Cdk2. Three secondary structures, a ␤-hairpin, ␤-strand, and 3 10 -helix, form as p27 sequentially folds upon binding to Cdk2. Cdk2 is dramatically remolded during p27 binding, including displacement of a ␤-strand by that formed by p27. Although these results provide detailed insights into how p27 binds Cdk2/cyclin A, they are insufficient for understanding the molecular details regarding how p27 interacts with the other Cdk/cyclin complexes that regulate the G 1 to S cell cycle transition (e.g. Cdk4/cyclin D).
Here, we report the use of circular dichroism (CD), isothermal titration calorimetry (ITC), computational docking, and biochemical assays to characterize and compare the interactions of Cdk4/cyclin D and Cdk2/cyclin A with p27-KID and various p27-KID deletion constructs. The biochemical assays confirmed that full-length p27 and p27-KID were equally potent inhibitors of Cdk4/cyclin D and Cdk2/cyclin A, as shown previously (24). Consistent with these results, ITC results indicated that the K d values for p27-KID binding to these two Cdk/ cyclin complexes were the same. Strikingly, however, the enthalpy/entropy balance in the Gibbs free energy of p27-KID binding to these two complexes was dramatically different. Furthermore, the region of p27 known to bind tightly to cyclin A, termed subdomain D1 (15) and known to contain a linear motif termed the RXL motif (25), bound relatively weakly to cyclin D. In contrast, another region, termed subdomain D2, exhibited similar ⌬G values for binding to Cdk4 and Cdk2, but the enthalpy/entropy balance for the two was dramatically different. Thus, although p27-KID bound and inhibited Cdk4/cyclin D and Cdk2/cyclin A with similar K d and IC 50 values, respectively, the thermodynamic forces governing binding and inhibition for the two complexes were different. We propose that p27 has evolved to utilize different combinations of residues within its conserved D1 and D2 subdomains to bind and inhibit different Cdk/cyclin complexes, giving rise to the different binding enthalpy/entropy signatures we observed. We further propose that these thermodynamic differences mediate differential exposure of residues within these domains of p27 (within Cdk/cyclin complexes) to post-translational modifications. Supporting this, we observed that Tyr-88 of p27 was more readily phosphorylated by Abl kinase domain (ABL-KD) when bound to Cdk4/cyclin D than to Cdk2/cyclin A. This observation has implications as to how mitogenic signals, which activate tyrosine kinases, impinge on p27 to drive cell cycle entry. ments of the KID, including p27-LHD2 (residues 38 -105), p27-D1LH (residues 22-59), and p27-KID-⌬3 10 (residues 22-85), were expressed in Escherichia coli BL21 (DE3) cells in His-tagged form and purified using established procedures (15). Briefly, the p27 proteins were purified using Ni 2ϩ -affinity chromatography (chelating-Sepharose, GE Healthcare) followed by cleavage of the His tag with thrombin, followed finally by further purification using high performance liquid chromatography (HPLC) with a C 4 column (Vydac) and a water, acetonitrile, 0.1% trifluoroacetic acid solvent system. p27 Y89F, T187A was expressed in E. coli BL21 (DE3) cells and purified using the procedures described above without cleavage of the His tag.
The C-terminal domains of Rb (Rb C , residues 773-928) and Rb C -5A (in which Arg residues 830 and 857 and Lys residues 870, 873, and 889 were mutated to Ala) were expressed in E. coli BL21 (DE3) cells fused at the N terminus to His 6 and a solubility enhancement tag (SET) (26) and expressed in E. coli BL21 (DE3) cells. Rb C and Rb C -5A were purified using Ni 2ϩ affinity, anion exchange (Q-Sepharose, GE Healthcare), and size-exclusion (Superdex-200, GE Healthcare) chromatography.
Truncated human cyclin A2 (residues 173-432; termed cyclin A here), full-length human Cdk2, and Cdk2 phosphorylated on Thr-160 were expressed and purified using established procedures (15). Briefly, His-tagged proteins were expressed in soluble form in E. coli BL21 (DE3) and purified using Ni 2ϩaffinity chromatography followed by cleavage of the His tag with thrombin. Individual proteins were further purified using anion-exchange and size-exclusion chromatography. The Cdk2/cyclin A complex was purified using gel filtration chromatography after incubating the purified components with a 1:1 mole ratio at 4°C for 30 min.
The Cdk4/cyclin D complex was expressed as a fusion protein, fus-Cdk4/cyclin D, in Sf9 insect cells using a baculovirus originally provided by Dr. Steve Reed (Scripps Research Institute). The expressed protein corresponded to cyclin D fused at the C terminus to a linker containing a PreScission protease site (linker sequence, ASKGGGGSLEVLFQPSR) fused further to Cdk4. fus-Cdk4/cyclin D was purified from Sf9 cells at 4°C using Ni 2ϩ -affinity, anion-exchange, and size-exclusion chromatography. fus-Cdk4/cyclin D was cleaved to give the noncovalent Cdk4/cyclin D complex using GST-tagged PreScission protease (Amersham Biosciences) followed by removal of the protease by GST affinity chromatography (glutathione-Sepharose, GE Healthcare). Cdk4/cyclin D was further purified using an additional round of sizeexclusion chromatography, as described above.
The identity of all proteins was confirmed using mass spectrometry, and protein concentrations were generally determined using UV absorbance at 280 nm. The molar extinction coefficients used were 35,560, 31,860, and 66,476 M Ϫ1 cm Ϫ1 for Cdk2, cyclin A, and the Cdk2/cyclin A complex, respectively; 49,810 M Ϫ1 cm Ϫ1 for the Cdk4/cyclin D complex; and 15,700, 15,580, 15,460, and 12,900 M Ϫ1 cm Ϫ1 for p27, p27-KID, p27-LHD2, and p27-KID-⌬3 10 , respectively, as determined using the software VectorNTI (Invitrogen). Because of its limited UV absorbance, p27-D1LH was quantified after HPLC purification and solvent removal by lyophilization using gravimetric methods.
CD Spectropolarimetry-CD measurements were performed using an AVIV model 62A DS circular dichroism spectropola-rimeter equipped with a thermoelectrically controlled cell holder using a 1-cm path quartz cell. For thermal denaturation experiments, ellipticity at 222 nm was measured at 2°C intervals in the temperature range from 15 to 99°C with a heating rate of 1°C per min. Samples were incubated in 1.0 mM HEPES buffer, pH 7.4, and 50 mM NaCl for 1 min at each temperature.
ITC Studies-Protein samples for ITC were further purified and exchanged into ITC buffer (20 mM HEPES, pH 7.4, 300 mM NaCl, and 5 mM DTT) using size-exclusion chromatography (Superdex-200). p27 samples were incubated with 20 mM DTT for 30 min at room temperature immediately prior to this step. ITC experiments were performed using an ITC-200 (Microcal) calorimeter. Titrations consisted of a preliminary 0.5-l injection of p27-KID, p27-LHD2, or p27-D1LH followed by multiple injections of 2 l into a solution of Cdk4/cyclin D. Experiments were performed at 25°C in ITC buffer. The time interval between injections was 180 s. The protein concentrations for these experiments were as follows: 60 M p27-KID, 75 M p27-D1LH, and 55 M p27-LHD2 for titration into 5 M Cdk4/ cyclin D; and 50 M p27-KID, 60 M p27-D1LH, and 75 M p27-LHD2 for titration into 5 M Cdk2/cyclin A. Titrations of Rb C and Rb C -5A into the two Cdk/cyclin complexes were performed as described above using 10 M Cdk2/cyclin A or Cdk4/ cyclin D and 100 M Rb C or Rb C -5A. Thermodynamic parameters were obtained by fitting the raw data using Origin software (OriginLab) according to the manufacturer's instructions using a 1:1 binding model.
In Vitro Cdk4 and Cdk2 Activity Assays-A solution of 1 nM Cdk4/cyclin D or 80 pM Cdk2/cyclin A containing 4 M substrate, Rb C , and varying concentrations of the various p27 proteins or peptides were incubated at 4°C for 1 h in kinase buffer (50 mM HEPES, pH 7.0, 15 mM MgCl 2 , 1 mM EGTA, 0.1 mM Na 3 VO 4 , 1 mM NaF, 10 mM ␤-glycerophosphate, 5 mM DTT, and 10% glycerol). 6 Ci of [␥-32 P]ATP (PerkinElmer Life Sciences) and unlabeled ATP up to 200 M were added to each reaction followed by incubation at 30°C for 30 min. Reactions were terminated by addition of 0.25 volume of 4ϫ SDS loading buffer, and 32 P-labeled Rb C was resolved using 10% SDS-PAGE and analyzed using a phosphorimager (Typhoon 9200; GE Healthcare). IC 50 values were determined after fitting values of normalized percentage of kinase activity versus log ([inhibitor]) using the dose-response model using Prism software (Graphpad Software). Measurements were performed in triplicate, and the error was reported as 95% confidence interval of the IC 50 value.
Docking Calculations-AutoDock 4.2 (27) was used to investigate interactions between p27 subdomain D1 and the Cdk2/ cyclin A and Cdk4/cyclin D complexes. Docking studies were performed using a peptide containing the RXL motif with the sequence RNLF as the ligand (28) and two Cdk/cyclin complexes as the receptor, Cdk2/cyclin A (PDB 1JST) (29) and Cdk4/cyclin D (PDB 2W96) (30). AutoDock treated the ligand as conformationally flexible (including both backbone and side chain atoms) and employed a Lamarckian-type algorithm using a scoring function based on the AMBER force field to estimate binding free energies. Initial blind docking studies were carried out using a grid with spacing of 0.35 Å containing 256 ϫ 256 ϫ 256 points in each Cartesian dimension centered on each of the Cdk/cyclin complexes. The results were clustered using a 5 Å root mean squared deviation criterion and revealed a single binding site for Cdk2/cyclin A corresponding to a hydrophobic patch on cyclin A mainly formed by a helix with the sequence MRAIL, referred to as the MRAIL patch. In the case of Cdk4/ cyclin D, two distinct binding sites were identified from the clustering analysis. The first of these corresponded to the MRAIL patch of cyclin D. An additional potential binding site, however, was located at the interface between Cdk4 and cyclin D and was judged to be structurally inaccessible to the D1 subdomain of p27 when the D2 subdomain is bound to Cdk4 within the Cdk4/cyclin D complex (supplemental Fig. 1A). Therefore, only the former cluster was deemed biochemically relevant to p27 function and considered during the second round of docking calculations. The grids used for subsequent targeted docking calculations were therefore centered on the MRAIL pockets of the two Cdk/cyclin complexes.
Targeted docking was performed on the sites described above, employing a 0.30 Å grid with 256 ϫ 256 ϫ 256 points and an initial population size of 200. The parameters used allowed 10 8 energy evaluations per run over 60,000 generations with a mutation rate of 0.10 and a crossover rate of 0.90. Results from these initial calculations were clustered at 5.0 Å and further refined with the same parameter set using a grid spacing of 0.10 Å. These final docking calculations revealed a single cluster (generated as described above) containing 37 structures for the Cdk2/cyclin A complex. For the Cdk4/cyclin D complex, three clusters were identified with populations of 40, 14, and 2 structures, respectively. Clusters for which the structure with the lowest binding energy exhibited a binding energy of more than 1.0 kcal/mol above that of the lowest energy cluster were not considered further. Locally modified versions of AutoGrid 4.2 and AutoDock 4.2 were required to generate grid boxes of sufficient size at 0.10 Å resolution to allow fully extended conformations of the model peptide sequence. The binding constant for Cdk4/cyclin D reported in Table 3 was based on a Boltzmann-weighted average of the three lowest energy clusters as defined above. The calculated free energies of binding for the tightest binding representative from each cluster as determined by AutoDock were used to generate the partition coefficients.
In Vitro Tyrosine Phosphorylation Assays-Six His-tagged, full-length p27 proteins with tyrosine 89 mutated to phenylalanine and threonine 187 mutated to alanine (p27 Y89F, T187A ) were used in these assays to prevent the following: (i) nonspecific phosphorylation of Tyr-89 by ABL-KD and (ii) phosphorylation of Thr-187 by the two Cdks. p27 Y89F, T187A was incubated with equimolar amounts of either Cdk2/cyclin A or Cdk4/cyclin D for 1 h at 4°C followed by Ni 2ϩ -affinity and gel filtration chromatography to isolate ternary complexes free of unbound p27 Y89F, T187A . Cdk4/cyclin D was prepared using the procedures described above followed by de-phosphorylation by protein phosphatase 1 (PP1; New England Biolabs) to reduce nonspecific activity toward p27 Y89F, T187A . Phosphorylation of Tyr-88 by ABL-KD (18) was determined as follows. Reactions included 500 nM ABL-KD, 5 M p27, and 1 mM ATP including 5 Ci of [␥-32 P]ATP and were terminated after different time intervals through addition of 8 M urea in 20 mM Tris-HCl solution. 32 PO 4 -p27 Y89F, T187A was resolved using 10% SDS-PAGE and analyzed using a PhosphorImager (Typhoon 9200; GE Healthcare). The extent of autophosphoryla-tion of ABL-KD observed in the reaction with p27 Y89F, T187A bound to Cdk2/cyclin A (supplemental Fig. 2) was smaller than that observed in the reactions with free p27 Y89F, T187A and p27 Y89F, T187A bound to Cdk4/cyclin D. To eliminate this difference as a factor in quantitative analysis of Tyr-88 phosphorylation, the raw data reported in supplemental Fig. 2 were normalized by multiplication by the following ratios: data for free p27 Y89F, T187A , 1.0; data for p27 Y89F, T187A /Cdk2/cyclin A, 3.0; and data for p27 Y89F, T187A /Cdk4/cyclin D, 1.2. The normalized raw data are illustrated in Fig. 6B and the relative initial rates in Fig. 6C. Relative initial reaction rates were obtained by linear fitting of the first three data points, after normalization, using Prism software (Graphpad Software).

Characterization of Recombinant Cdk4/Cyclin D-The
binary Cdk4/cyclin D complex was expressed as a single polypeptide chain (termed fus-Cdk4/cyclin D) followed by proteolytic cleavage of a linker between the Cdk and cyclin subunits and purification using anion-exchange and gel filtration chromatography ( Fig. 2A). The single chain and noncovalent Cdk4/ cyclin D complexes exhibited nearly identical secondary structure (supplemental Fig. 3) and similar apparent thermal  Table 1). Gel filtration chromatography analysis showed that Cdk4/cyclin D bound to the N-terminal kinase inhibitory domain of p27 (p27-KID, Fig. 2B). These results demonstrate that Cdk4/cyclin D is folded and capable of binding to p27-KID with 1:1:1 (Cdk4/ cyclin D/p27-KID) stoichiometry.
Inhibition of Cdk4/Cyclin D Activity by p27-KID and p27-KID Deletion Constructs-The N-terminal domain of p27 (p27-KID) is necessary and sufficient for inhibition of Cdk2/cyclin A, Cdk2/cyclin E, and Cdk4/cyclin D (32). The crystal structure of p27-KID bound to Cdk2/cyclin A showed that a segment of p27-KID termed subdomain D1 (residues 22-37; Fig. 1A) bound to cyclin A and another segment termed subdomain D2 (residues 60 -105) bound to Cdk2. Subdomains D1 and D2 are linked by subdomain LH, which formed an elongated ␣-helix that spans the ϳ35-Å distance between cyclin A and Cdk2, respectively. A previous report described the thermodynamic, kinetic, and functional contributions of the different subdomains of p27-KID to interactions with Cdk2/cyclin A (15), but data are not available to describe interactions with Cdk4/ cyclin D. To address this knowledge gap, we investigated the contributions of the different subdomains of p27-KID to Cdk4/ cyclin D inhibition using a library of p27 constructs, including full-length p27 (p27), p27-KID, and several p27-KID deletion constructs (Fig. 3, supplemental Fig. 6, and Table 1). p27-KID and p27 were equally active in inhibiting Cdk4/cyclin D, exhibiting IC 50 values of 0.9 and 1.1 nM, respectively (Fig. 3, supplemental Fig. 6, and Table 1). The C terminus of subdomain D2 forms a single turn of 3 10 helix when bound to Cdk2/cyclin A, and Tyr-88 within this segment binds within the ATP binding pocket of the kinase active site (Fig. 1A) (23). Deletion of this segment from p27-KID (p27-KID-⌬3 10 ; Fig. 1B) did not significantly affect the IC 50 value for kinase inhibition (Fig. 3, supplemental Fig. 6, and Table 1), but this construct was unable to completely inhibit Cdk4, as was observed previously with Cdk2/cyclin A (32,33). A construct in which subdomain D1, containing the RXL motif, was deleted (p27-LHD2) exhibited a significantly increased IC 50 value (128.7 nM) and was unable to completely inhibit Cdk4 (Fig. 3, supplemental Fig. 6, and Table  1). These results are reminiscent of previous results with the related Cdk inhibitor, p21, in which deletion of the RXL motif led to a 100-fold increase in the IC 50 value and incomplete inhibition of Cdk4/cyclin D (34). A construct corresponding to deletion of subdomain D2 (p27-D1LH) failed to inhibit Cdk4 at concentrations up to 33 M (supplemental Fig. 6). Thus, subdomains D1, LH, and D2 of p27 are required for potent and complete inhibition of Cdk4 within the Cdk4/cyclin D complex. The 3 10 helix segment within subdomain D2 is dispensable for high potency but is required for complete kinase inhibition. These results further demonstrated that the p27 subdomain requirements for inhibition of Cdk4/cyclin D are similar to those observed previously for Cdk2/cyclin A (15).
Thermal Stability of Complex Cdk4/Cyclin D-We performed thermal denaturation experiments to determine the extent to which binding of p27-KID stabilized Cdk4/cyclin D. The apparent thermal denaturation temperature (T m app ) value for fus-Cdk4/cyclin D was 51°C, and this value decreased slightly when the fusion linker was cleaved (T m app ϭ 47°C; supplemental Fig. 4). Notably, Cdk4/cyclin D was less thermally stable than Cdk2/cyclin A (T m app ϭ 55°C) (35). Consistent with this observation, the crystal structure of Cdk4/cyclin D (30) revealed that the Cdk/cyclin interface was ϳ60% smaller than that observed for Cdk2/cyclin A (36). Furthermore, isolated Cdk4 and cyclin D were previously shown to assemble inefficiently in vitro (37). The p27-KID/Cdk4/cyclin D ternary complex was significantly more stable (T m app ϭ 57°C) than the binary Cdk4/cyclin D complex. p27-KID is a "molecular staple" that holds Cdk4 and cyclin D together, as observed for Cdk2/ cyclin A (35). However, the magnitude of the stabilization for Cdk4/cyclin D (⌬T m app ϭ ϩ10°C) was smaller than that observed for Cdk2/cyclin A (⌬T m app ϭ ϩ24°C) (35), suggesting that p27-KID interacts differently to the two Cdk/cyclin complexes.
Thermodynamic Analysis of p27-KID Binding to Cdk/Cyclin Complexes Using ITC-ITC was used to analyze the thermodynamics of interactions between p27-KID and Cdk4/cyclin D. ITC directly measures the enthalpy change (⌬H) associated with a binding event, and through quantitative fitting of binding isotherms, the binding affinity (from which the Gibbs free energy change (⌬G) can be derived) and stoichiometry of binding The data were normalized such that Cdk4/cyclin D activity in the absence of a p27 construct corresponded to 100% activity. The error bars correspond to one S.D. of the mean based on triplicate measurements.

TABLE 1 Inhibition of Cdk4/cyclin D and Cdk2/cyclin A by p27 and the p27-KID constructs
Cdk inhibition assays were performed by monitoring phosphorylation of Rb C in the presence of a range of p27 construct concentrations. IC 50 values were determined through fitting of inhibition curves; the experiments were performed in triplicate and the error is reported as the 95% confidence interval of the IC 50 value. (n) can also be determined. With ⌬H and ⌬G known, the entropy change associated with binding can be determined from ϪT⌬S ϭ ⌬G Ϫ ⌬H. Consistent with the low IC 50 value (0.9 nM) discussed above, ITC showed that the binding of p27-KID to Cdk4/cyclin D was highly favorable (K d ϭ 2.5 Ϯ 0.5 nM; Fig. 4A and Table 2). This K d value was very similar to that determined previously for p27-KID binding to Cdk2/cyclin A (15) and analyzed again here (K d ϭ 5.1 Ϯ 0.5 nM; Fig. 4B and Table 2). However, the ⌬H values for p27-KID binding to Cdk4/cyclin D and Cdk2/cyclin A were strikingly different (Ϫ34.4 Ϯ 0.7 and Ϫ49.1 Ϯ 0.2 kcal mol Ϫ1 , respectively). This difference in favorable ⌬H values (͉⌬⌬H͉ ϳ15 kcal mol Ϫ1 ) was mirrored by a difference of similar magnitude in the unfavorable ϪT⌬S values (͉T⌬⌬S͉ ϳ15 kcal mol Ϫ1 ; Table 2). These differences in the binding thermodynamics suggested that, whereas binding affinity was approximately equal (͉⌬⌬G͉ ϭ 0.4 kcal mol Ϫ1 ), the extent of interaction between p27-KID and the two Cdk/cyclin complexes was different (e.g. the number of residues involved in the interactions was different) as evidenced by the large ͉⌬⌬H͉ and ͉⌬⌬S͉ values.
A different picture emerged when subdomain D2 was deleted from p27-KID (p27-D1LH). p27-D1LH bound weakly to Cdk4/ cyclin D (⌬G ϭ Ϫ8.5 Ϯ 0.1 kcal mol Ϫ1 ; ⌬⌬G ϭ Ϫ3.2 Ϯ 0.2 kcal mol Ϫ1 ) but, as shown previously, bound with high affinity to Cdk2/cyclin A (⌬G ϭ Ϫ10.5 Ϯ 0.2 kcal mol Ϫ1 ; ⌬⌬G ϭ Ϫ0.8 Ϯ 0.3 kcal mol Ϫ1 ) (15). The corresponding changes in other thermodynamic parameters upon deletion of subdomain D2 for binding to Cdk4/cyclin D were ⌬⌬H ϭ Ϫ21.8 Ϯ 1.1 kcal mol Ϫ1 and ϪT⌬⌬S ϭ ϩ18.6 Ϯ 0.8 kcal mol Ϫ1 (Table 2). These values for binding to Cdk2/cyclin A were ⌬⌬H ϭ Ϫ30.3 Ϯ 0.7 kcal mol Ϫ1 and ϪT⌬⌬S ϭ ϩ29.5 Ϯ 0.8 kcal mol Ϫ1 (Table 2), illustrating that deletion of subdomain D2 more significantly affected interactions with Cdk2/cyclin A than with Cdk4/cyclin D. These results for the two p27-KID deletion constructs showed that subdomain D2 played an important role in binding to both Cdk4/cyclin D and Cdk2/cyclin A but that subdomain D1, although important for binding to Cdk2/cyclin A, contributed much less favorably in interactions with Cdk4/cyclin D. Furthermore, the smaller absolute magnitude of the ⌬H and ϪT⌬S values for p27-LHD2 binding to Cdk4/cyclin D, in comparison with these values for binding to Cdk2/cyclin A, suggested that a smaller number of residues within this construct contact the former Cdk/cyclin complex. This latter observation further suggested that the extent of folding upon binding, which occurred when p27-KID bound to Cdk4/cyclin D, was smaller than that which occurred upon binding to Cdk2/cyclin A.

Characterization of Interactions between the RXL Motif and Cdk/Cyclin Complexes Using Computational Docking Methods-
The sequence of cyclin D is 23% identical and 36% similar to that of the cyclin box of cyclin A (residues 173-432) (supplemental Fig. 1B), and the structures of the two cyclins are very similar (the C␣ atom root mean squared deviation between cyclin D and cyclin A is 0.9 Å). Within cyclin A, the MRAIL patch mediates substrate recognition (28). Within Cdk2 substrates and inhibitors, one or more copies of the RXL motif, with the sequence (Arg/Lys)-Xaa-Leu (with Xaa being any residue), have been identified as mediators of binding to the MRAIL patch on cyclin A (23,25,28,38). The RXL motif of p27 is within the D1 domain, and the finding of differential binding affinity of D1-containing p27 constructs to Cdk2/cyclin A versus Cdk4/cyclin D, detailed above, suggests that there may be differential affinity for the RXL motif. In the absence of structural data for p27 bound to Cdk4/cyclin D, we performed computational docking experiments to investigate the physical basis for the large difference in the affinity of the p27 RXL motif for Cdk2/cyclin A versus Cdk4/cyclin D. In these experiments, a peptide with the sequence RNLF (corresponding to the p27 RXL motif) was docked onto the surfaces of the two Cdk/cyclin complexes. These experiments complement those performed earlier by Liu et al. (39), in which the structural similarity between cyclin A and cyclin D was used to model a subdomain D1-containing segment of p27 bound to cyclin D.
The RNLF peptide docked to a single site (Cluster A1, Fig.  5A) within the MRAIL patch on the surface of cyclin A within the Cdk2/cyclin A complex. Among the 37 structures within this cluster, the lowest calculated binding energy (⌬G low (dock)) was Ϫ9.81 kcal mol Ϫ1 (Fig. 5A and Table 3). It should be noted that the docked conformation of the Arg residue is different from that observed in crystals (Fig. 5A) due to binding within an alternative acidic pocket. This result reflects the use of the unbound (to p27) Cdk2/cyclin A crystal structure for docking experiments. For Cdk4/cyclin D, the peptide docked to two sites within the MRAIL patch (Clusters D1 and D3 in Fig. 5B) and one nearby site (Cluster D2, Fig. 5B). The Boltzmannweighted average of the lowest binding energies for these three sites (⌬G ave (dock)) was Ϫ8.11 kcal mol Ϫ1 (Fig. 5B and Table 3). The difference of the computed energies for the RXL peptide binding to the two Cdk/cyclin complexes (⌬⌬G(dock) ϭ ⌬G avg (dock) Cdk4/cyclin D Ϫ ⌬G low (dock) Cdk2/cyclin A ) was ϩ1.70 kcal mol Ϫ1 , which is similar to the ⌬⌬G value derived from ITC measurements (⌬⌬G(ITC)) for p27-D1LH binding to the two Cdk/cyclin complexes (⌬⌬G(ITC) ϭ ⌬G(ITC) Cdk4/cyclin D Ϫ ⌬G(ITC) Cdk2/cyclin A ϭ ϩ2.0 kcal mol Ϫ1 ; Table 2). The similarity of the values of ⌬⌬G(dock) and ⌬⌬G(ITC) suggested that the docking experiments yielded a reasonable, approximate physical model of how the RXL motif binds to the surface of cyclin D. For both cyclin A and cyclin D, the Arg residue of the RXL motif bound within an acidic pocket (Fig. 5). Although the docking algorithm FIGURE 5. Comparison of docking results for p27 RXL motif peptide (RNLF) binding to Cdk2/cyclin A and Cdk4/cyclin D. A, a single cluster (labeled Cluster A1) was observed with Cdk2/cyclin A; the lowest energy representative is illustrated in stick format (orange). The structure of the RNLF segment from the D1 subdomain of p27 from the crystal structure of p27-KID/ Cdk2/cyclin A (PDB code 1JSU) is shown as light gray sticks for reference. B, three clusters (labeled Clusters D1, D2, and D3) were observed with Cdk4/ cyclin D; the low energy representative for each of these three clusters is illustrated in stick format in orange, yellow, and green, respectively. The Boltzmann-weighted average computed binding energy (⌬G ave (dock)) for these three clusters is indicated. In both panels, the surfaces of cyclin A and cyclin D within 5 Å of atoms of the docked RXL motif peptide are colored by electrostatic potential. Surfaces having negative electrostatic potential are shown in red, and those having positive electrostatic potential are shown in blue. The molecular images were prepared using the PyMOL Molecular Graphics System (DeLano Scientific, San Carlos, CA).

TABLE 3 Summary of results for computational docking of the RXL peptide (RNLF) to Cdk2/cyclin A and Cdk4/cyclin D
Docking studies were performed as described under "Experimental Procedures." A single cluster of docked molecules was observed with Cdk2/cyclin A, and three clusters were observed with Cdk4/cyclin D. For Cdk4/cyclin D, the average computed binding energy was calculated using Boltzmann-weighted averaging. did not accurately (as compared with the crystal structure; Fig. 5A) position the Arg residue of the peptide within the MRAIL pocket of cyclin A, the docking results nonetheless point out that cyclin A exhibits a more extensive constellation of acidic residues than cyclin D that contributes more favorably to ⌬⌬G(dock) (supplemental Fig. 1, C-H). In addition, the Leu and Phe residues of the RNLF peptide bound within a hydrophobic pocket on both cyclins. These interactions, however, were more favorable with cyclin A than cyclin D (supplemental Fig. 1, C-H). The results of these docking experiments provide a rationale for the reduced affinity of the RXL motif-containing D1 subdomain of p27 for Cdk4/cyclin D versus Cdk2/cyclin A. Phosphorylation of p27 within Cdk/Cyclin Complexes by the Nonreceptor Tyrosine Kinase Abl-The results of the structural and thermodynamic investigations described above revealed that, although both the D1 and D2 subdomains of p27 participate in binding to Cdk2/cyclin A and Cdk4/cyclin D, these two subdomains interact with the Cdk and cyclin subunits of these two complexes in distinct ways. In particular, the D2 subdomain appears to interact less extensively with Cdk4 than with Cdk2 based on the large difference in ⌬H values for p27-LHD2 binding to Cdk4/cyclin D and Cdk2/cyclin A ( Table 2). Past studies (18) showed that, despite tight binding within the ATP pocket of Cdk2, Tyr-88 of p27 (when bound to Cdk2/cyclin A) could be phosphorylated by the nonreceptor tyrosine kinase (NRTK) Abl as well as the fusion oncoprotein Bcr-Abl. p27 is also phosphorylated on Tyr-88 in the context of its complex with Cdk4/cyclin D (20). Based on our thermodynamic observations, we hypothesized that Tyr-88 of p27 is more accessible to phosphorylation when bound to Cdk4/cyclin D than Cdk2/ cyclin A due to less extensive interactions of subdomain D2 (of p27) with the N-terminal lobe of Cdk4 than with Cdk2. To test this hypothesis, we probed Tyr-88 accessibility to phosphorylation within two ternary complexes (bound to Cdk4/cyclin D and Cdk2/cyclin A; Fig. 6A) using the kinase domain of Abl (ABL-KD). Unbound p27 was used as a control for fully accessible Tyr-88. In addition, Tyr-89 of p27 was mutated to phenylalanine to prevent nonspecific phosphorylation by ABL-KD, and Tyr-187 was mutated to alanine to prevent nonspecific phosphorylation by Cdk2 (termed p27 Y89F, T187A ). Free p27 Y89F, T187A was highly phosphorylated by ABL-KD (normalized initial rate defined as 100%), and tyrosine phosphorylation was reduced by complexation with Cdk4/cyclin D (normalized initial rate, 41%) and Cdk2/cyclin A (normalized initial rate, 12%; Fig. 6, B and C and supplemental Fig. 2). In accord with our structural hypothesis, the initial rate of Tyr-88 phosphorylation in the presence of Cdk4/cyclin D was 3.4-fold greater than that with Cdk2/cyclin A. We observed that the extent of autophosphorylation of ABL-KD was similar for free p27 Y89F, T187A and p27 Y89F, T187A bound to Cdk4/cyclin D but that this reaction was partially inhibited in the presence of p27 Y89F, T187A bound to Cdk2/cyclin A (supplemental Fig. 2). As a consequence, the extent of p27 phosphorylation reported above (Fig. 6, B and C) was normalized to reflect the apparent ABL-KD activity based on the extent of ABL-KD autophosphorylation in the three reactions. These results suggested that, due to differences in interactions between the D2 subdomain and Cdks, Tyr-88 of p27 is more accessible for phosphorylation by ABL-KD when bound to Cdk4/cyclin D, and possibly other NRTKs, than when bound to Cdk2/cyclin A.

Cluster
Clarifying the Differential Substrate Recruitment Roles of Cyclin D and Cyclin A-A second significant difference in how p27 bound to Cdk4/cyclin D versus Cdk2/cyclin A involved the D1 subdomain and the MRAIL pocket of the two cyclins. Specifically, p27-D1LH bound significantly more weakly to Cdk4/ cyclin D than to Cdk2/cyclin A (⌬⌬G ϭ ϩ2.0 kcal mol Ϫ1 ). The MRAIL pocket of cyclin A recruits substrates, including Rb and p107, through interactions with multiple copies of the RXL motif within their sequences (28). In addition, p27 contains a single RXL motif within subdomain D1, which promotes rapid binding to and inhibition of Cdk2/cyclin A (15). It is well appreciated that the five RXL motifs within Rb play a limited role in its recruitment and phosphorylation by Cdk4/cyclin D; recruitment is mediated by interactions between the LXCXE motif of cyclin D and pocket domain B of Rb (40,41) and, under certain circumstances, by an incompletely understood interaction between cyclin D and a motif, including Leu-901 within the C-terminal domain of Rb (40). Nonetheless, because we observed a weak interaction between p27-D1LH, which con- Interactions between p27 and Cdk4/Cyclin D AUGUST 26, 2011 • VOLUME 286 • NUMBER 34 tains the RXL motif, and Cdk4/cyclin D, we investigated the contribution of the five RXL motifs in Rb C to its phosphorylation by Cdk4/cyclin D. We performed kinase assays using Rb C and a mutant version of Rb C in which the Arg or Lys residue in each of the five RXL motifs was mutated to Ala (Rb C -5A). As expected, phosphorylation of Rb C by Cdk2/cyclin A was significantly reduced (by 80%) through mutation of the five RXL motifs (supplemental Fig. 7A). In contrast, although Cdk4/cyclin D phosphorylated Rb C less efficiently than Cdk2/cyclin A, mutation of the RXL motifs did not affect the extent of Rb C phosphorylation. ITC results showed directly that Rb C bound tightly to Cdk2/cyclin A (K d value of 11.2 M) and that Rb C -5A did not (supplemental Fig. 7B). In contrast and consistent with the biochemical results, ITC results showed that neither Rb C nor Rb C -5A directly bound to Cdk4/cyclin D (supplemental Fig. 7C). These results support past findings that the C-terminal RXL motifs of Rb do not significantly contribute to recruitment of Rb for phosphorylation by Cdk4/cyclin D.

DISCUSSION
p27 is an intrinsically disordered protein that can bind to the full repertoire of Cdk/cyclin complexes that regulate progression of the cell division cycle (14). In normal cells, p27 translation and protein stability are maximal in G 0 and early G 1 phase and decrease gradually during the G 1 phase due to mitogenic signaling (42). During the G 1 -to-S phase transition, several Cdk/cyclin complexes are sequentially activated (first Cdk4 (and Cdk6)/cyclin D, then Cdk2/cyclin E, and finally Cdk2/cyclin A). These kinase complexes, in turn, sequentially phosphorylate distinct sites on Rb (43)(44)(45) as well as other substrates. p27 interacts with these three complexes and differentially regulates their kinase activity. For example, unmodified p27 is a potent inhibitor of Cdk2/cyclin A. However, phosphorylation (of p27) on Tyr-88 and sometimes also on Tyr-74 relieves kinase inhibition and initiates a phosphorylation/ubiquitination cascade that ultimately eliminates p27 and drives cells into S phase (18,46). Although it is known that p27 can be associated with active Cdk4/cyclin D complexes (19) and that this kinase activity requires tyrosine phosphorylation of p27 (20), the molecular details of how p27 interacts with Cdk4/cyclin D and how tyrosine phosphorylation modulates these interactions are poorly understood. The results reported herein provide insights into the structural and thermodynamic nature of these interactions and also clarify the nature of interactions between Cdk4/cyclin D and RXL motif-containing polypeptides.
Our results confirm that Rb is not recruited to Cdk4/cyclin D through high affinity interactions between the RXL motifs (of Rb) and the MRAIL pocket of cyclin D. ITC results showed that p27-D1LH, containing a single RXL motif, bound to Cdk4/cyclin D with a K d value of 592 nM, which reflects a relatively weak interaction in comparison with that with Cdk2/cyclin A (with a K d value of about 20 nM). Rb C , with five RXL motifs, bound weakly to Cdk2/cyclin A (with a K d value of about 11.2 M), whereas an interaction with Cdk4/cyclin D was not detected. These results suggest that the RXL motifs within Rb C , a Cdk substrate, have been evolutionarily "tuned" to have relatively low affinity for the MRAIL pocket on cyclin A and to not bind this pocket on cyclin D. In contrast, we propose that the RXL motif within the D1 subdomain of p27 has evolved to competitively bind with high affinity to this site so as to inhibit substrate recruitment (to Cdk2/ cyclin A). Conservation of many features of the MRAIL pocket between cyclin A and cyclin D confers the ability of this optimal motif to weakly bind cyclin D and to enhance the binding of the entire KID to the Cdk4/cyclin D complex. These observations illustrate the different roles of linear motifs in molecular recognition and their ability to be tuned for different purposes as follows: substrate recruitment in the case of the RXL motifs in Rb and inhibition of substrate recruitment in the case of p27 (and p21). This situation is reminiscent of functional differences between cellular linear motifs and motifs that have arisen through convergent evolution in viral proteins and serve to competitively overtake cellular signaling systems (47).
Although the details of interactions between p27 and Cdk4/ cyclin D remain incompletely understood, our observations suggest that tyrosine residues within p27 are more readily phosphorylated by NRTKs such as Abl when bound to Cdk4/cyclin D than to Cdk2/cyclin A. For example, the binding of p27-LHD2 to Cdk4/cyclin D was enthalpically less favorable than to Cdk2/cyclin A and was associated with a smaller entropic penalty due to folding upon binding ( Table 2) (15). These findings suggest that the D2 subdomain of p27 is more dynamically associated with Cdk4 than with Cdk2. This is consistent with observations that p27 crystallized in association with Cdk2/cyclin A and that high resolution NMR spectra were obtained for p27 when bound to Cdk2/cyclin A (18). In contrast, we were unable to obtain crystals for p27 bound to Cdk4/cyclin D and observed that p27 resonances disappeared in the presence of Cdk4/cyclin D, consistent with resonance broadening due to conformational exchange. 3 The primary structures of Cdk4 and Cdk2 are very similar (supplemental Fig. 8A); however, there are four amino acid differences within the N-terminal domains of the two kinases that may give rise to the differences observed by ITC in their interactions with the D2 subdomain of p27. In particular, Ile-70 and Tyr-77 of Cdk2 interact with hydrophobic residues within the ␤-hairpin of p27 subdomain D2 (supplemental Fig. 8, B and C), and these residues are replaced by Cys-78 and Thr-90 in Cdk4 (supplemental Fig. 8, D and E), which expose more limited hydrophobic surface areas for interactions with p27. Although Cdk4 is catalytically less efficient than Cdk2 toward Rb (Cdk4/cyclin D was used at a 12-fold higher concentration than Cdk2/cyclin A in the Rb C phosphorylation assay illustrated in supplemental Fig. 7A), the enhanced accessibility of p27 to tyrosine phosphorylation within Cdk4/ cyclin D complexes may provide a mechanism for efficient transmission of mitogenic signals via activation of cyclin D expression and NRTKs such as Abl, Src, and Lyn (18,46). These signals lead to activation of Cdk4/cyclin D and hypophosphorylation of Rb. This then triggers activation of Cdk2/cyclin E and Cdk2/cyclin A complexes, which are the catalytic engines that drive hyperphosphorylation of Rb and entry into S phase of the cell division cycle. The heightened accessibility of Tyr-88 and possibly Tyr-74 within p27 to phosphorylation by NRTKs may have evolved to enable initiation of this signaling cascade.