Identification and Characterization of an Irreversible Inhibitor of CDK2

Summary Irreversible inhibitors that modify cysteine or lysine residues within a protein kinase ATP binding site offer, through their distinctive mode of action, an alternative to ATP-competitive agents. 4-((6-(Cyclohexylmethoxy)-9H-purin-2-yl)amino)benzenesulfonamide (NU6102) is a potent and selective ATP-competitive inhibitor of CDK2 in which the sulfonamide moiety is positioned close to a pair of lysine residues. Guided by the CDK2/NU6102 structure, we designed 6-(cyclohexylmethoxy)-N-(4-(vinylsulfonyl)phenyl)-9H-purin-2-amine (NU6300), which binds covalently to CDK2 as shown by a co-complex crystal structure. Acute incubation with NU6300 produced a durable inhibition of Rb phosphorylation in SKUT-1B cells, consistent with it acting as an irreversible CDK2 inhibitor. NU6300 is the first covalent CDK2 inhibitor to be described, and illustrates the potential of vinyl sulfones for the design of more potent and selective compounds.


In Brief
Irreversible inhibitors have a distinctive mode of action and offer an alternative route to competitive ATP inhibitors to target protein kinases. Anscombe et al. describe NU6300, a covalent CDK2 inhibitor that illustrates the potential of using vinyl sulfones to mediate irreversible inhibition.

INTRODUCTION
Cyclin-dependent kinases (CDKs) play significant roles in regulation of the eukaryotic cell cycle and in transcription (Lim and Kaldis, 2013;Malumbres and Barbacid, 2009). During G1 phase, CDK2 bound to cyclin E mediates phosphorylation of the retinoblastoma tumor suppressor protein (Rb), which results in activation of members of the E2F family of transcription factors, thereby assisting entry into S phase (Morgan, 2007). Following degradation of cyclin E during late G1, CDK2 pairs with cyclin A to phosphorylate and inactivate E2F, resulting in S-phase pro-gression. Deregulation of the cell cycle is a characteristic of most human tumors, occurring frequently through disruption of the Rb signaling circuit. Aberrant control of CDK activity has been directly linked to cancer development (Malumbres and Barbacid, 2009).
Chemical-genetic evidence has shown a difference in the cellular response to the absence of CDK2 and to small-molecule CDK2 inhibitors (Guha, 2012;Horiuchi et al., 2012;Malumbres and Barbacid, 2009). These observations suggest that CDK2 inhibitors may be appropriate for treating a subset of tumors with defined genetic characteristics. Strategies that exploit synthetic lethalities have also highlighted a potential role for CDK2 inhibitors. Combined administration of a phosphatidylinositol 3-kinase inhibitor and a CDK2 inhibitor demonstrated induction of apoptosis in malignant glioma xenografts (Cheng et al., 2012). CDK2 inhibitors may also have clinical utility in subsets of cancers such as high-grade serous ovarian carcinomas, which harbor amplifications in the CCNE1 gene that encodes its partner cyclin E (Etemadmoghadam et al., 2013).
The majority of protein kinase inhibitors in clinical trials are reversible competitive inhibitors that bind to the enzyme's ATP binding site. Protein kinase activation is accompanied by significant structural rearrangement of the kinase fold, and this has been exploited to identify compounds with increased selectivity and potency (Dar and Shokat, 2011;Zhang et al., 2009). In this context, irreversible inhibition may be extremely effective for the subset of kinases that encode residues within the active site that can be covalently modified and are not widely conserved (Barf and Kaptein, 2012;Leproult et al., 2011). Furthermore, covalent inhibitors can be useful tool compounds in target validation studies to investigate the cellular effects of selective protein kinase inhibition.
We present biochemical and structural studies that confirm 6-(cyclohexylmethoxy)-N-(4-(vinylsulfonyl)phenyl)-9H-purin-2-amine (NU6300) as the first example of an irreversible inhibitor of CDK2. We identify the site of covalent modification as Lys89, a residue that lies just outside the CDK2 ATP binding cleft and is not well conserved across the protein kinase family. Our studies define the inhibitor mode of action and show that selective irreversible CDK2 inhibition can be achieved in cells.

NU6300 Binds Covalently to CDK2
To determine whether NU6300 binds covalently to CDK2, recombinant CDK2/cyclin A was incubated overnight with NU6300 and then analyzed by electrospray ionization mass spectrometry (ESI-MS) ( Figure S1A). This analysis revealed an increase in the mass by 414 Da compared with the control CDK2 samples, supporting the formation of a covalent adduct.
A similar experiment was also carried out using a surface plasmon resonance (SPR) biosensor. By exposing immobilized CDK2/cyclin A to NU6300, the binding of NU6310 decreased. This competitive effect indicates that NU6300 blocks the inhibitor binding site. The effect was time dependent and relatively slow, with less than 50% reduction of the apparent binding capacity in 20 hr (Figures S1B-S1D).
To determine the kinetics of the interactions of NU6310 and NU6300 with immobilized CDK2/cyclin A, the sensor surface was exposed to five different concentrations of each inhibitor for different contact times (Figures 2A and 2B). On these relatively short time scales (compared with the experiment above), the interactions between CDK2 and either NU6300 or NU6310 appeared reversible, since the sensorgrams were well described by a simple 1:1 model and the dissociation was similar irrespective of the contact time. It enabled the estimation of the kinetic constants (k on , k off , K D ) for the formation of the non-covalent complex. The values were 0.545 ± 0.072 3 10 5 M À1 s À1 (k on ), and 0.0713 ± 0.0063 s À1 (k off ), respectively, yielding a K D of 1.31 ± 0.18 mM for NU6300, and 1.13 ± 0.10 3 10 5 M À1 s À1 (k on ) and 0.0809 ± 0.0070 s À1 (k off ), yielding a K D of 0.716 ± 0.012 mM for the interaction of NU6310 with CDK2. The formation of a covalent complex could not be detected on the time scales of these experiments, and the injection times could not be extended further for practical reasons. The inability to detect the formation of a covalent bond was not attributable to inhibitor instability, as they showed unchanged kinetic characteristics over several hours after preparation in aqueous buffer.
To confirm the results of the ESI-MS and SPR analysis, CDK2/cyclin A was incubated with either NU6300 or NU6310, then the samples and appropriate controls were dialyzed. The resulting CDK2/cyclin A activity was analyzed in an in vitro kinase assay against a C-terminal fragment of the retinoblastoma protein Rb. After an overnight incubation in the presence of NU6300, CDK2 activity was not recovered after dialysis (Figure 2C). However, the non-covalent inhibitor NU6310 was removed by this treatment, and the resulting CDK2 could phosphorylate Rb. The activity of NU6300 was also characterized in an alternative kinase assay format (ADP-Glo; , Promega) in which covalent inhibition of CDK2/cyclin A was allowed to proceed in a pre-incubation phase and was assessed in a subsequent activity assay, where enzyme and inhibitor were diluted such that the inhibitor was present at 20-fold below its IC 50 value. The samples and appropriate controls were incubated for 0, 10, 30, 60, and 120 min, prior to addition of ATP and peptide substrate (sequence HHASPRK, single-letter amino acid code), to initiate the kinase reaction. The results of the study indicate a time-dependent inhibition of CDK2/cyclin A by NU6300, with the extent of inhibition increasing linearly with time, consistent with irreversible inhibition occurring in the initial rate regime ( Figure 2D). The corresponding k inact for this process is 5.0 3 10 3 M À1 s À1 . Taken together, these results are consistent with a model in which the two inhibitors have equivalent micro-rate constants for their interaction with CDK2, but in which upon extended exposure NU6300 covalently modifies CDK2.

Identification of CDK2 Residues Covalently Modified by NU6300
Guided by the structure of the CDK2/cyclin A/NU6102 complex (Davies et al., 2002), the nucleophilic residues that are suitably positioned to react with the vinyl sulfone of NU6300 are Asp86, Lys88, and Lys89. These residues were individually mutated to an alanine, glutamate, or valine, respectively, and the resulting mutant CDK2/cyclin A complexes were analyzed by ESI-MS following overnight incubation with either NU6300 (exact mass 413.15) or DMSO ( Figure S2 and Table S1). After treatment with the inhibitor, the major CDK2 D86A and CDK2 K88E species were modified by addition of 414 and 412 Da, respectively, whereas the major species present in the CDK2 K89V sample acquired no additional mass. (D) Time-dependent inhibition of CDK2/cyclin A. Activity was measured using the ADP-Glo assay format against a peptide of sequence HHASPRK.
Error bars indicate SD of the measurements. See also Figure S1 and Table S1.
The bandshift experiment was repeated after incubating CDK2 K88E /cyclin A and CDK2 K89V /cyclin A with NU6300, NU6310, or DMSO followed by dialysis to remove unbound inhibitor ( Figure S2G). The two mutants and wild-type CDK2 recovered their activity following treatment with NU6310. CDK2 K89V /cyclin A recovered activity following incubation with NU6300 and subsequent dialysis. However, the CDK2 K88E mutant and wild-type CDK2 did not. These results prove that Lys89 is the preferred site of modification by NU6300.
To confirm this conclusion, NU6300 was co-crystallized with CDK2/cyclin A and the structure was resolved to 2.4 Å resolution (Table S2 and Figure 3). As has been previously observed within this inhibitor series, the purine ring makes a triplet of conserved hydrogen bonds with the backbone amide and carbonyl groups of Glu81 and the backbone carbonyl of Leu83 within the CDK2 hinge. The aniline moiety adopts a similar pose to that previously observed in the CDK2/NU6102 co-complex, packing against CDK2 through a p-p interaction with the peptide backbone between Asn85 and Asp86, thus positioning one of the sulfone oxygens to interact with the side chain of Asp86 (Davies et al., 2002). As a result, the vinyl moiety can react with the side-chain ε-amino group of Lys89 ( Figure 3B). In this region, the electron density map has continuous density between the side-chain amino group of Lys89 and the inhibitor's vinyl sulfone, indicating the formation of a covalent bond between them ( Figure 3C).

NU6300 Inhibits Rb Phosphorylation in Rb-Positive SKUT-1B Cells
Our results demonstrate that NU6300 is a covalent inhibitor of CDK2. Prior to determining its cellular activity, we carried out a screen to assess its selectivity against a panel of 131 protein kinases under conditions that would not distinguish between a reversible or irreversible mode of action. This screen identified 13 kinases that exhibited less than 25% activity in the presence of 1 mM NU6300 (Table S3A). These 13 kinases and 29 additional kinases, selected for their known sensitivity to other inhibitors within the purine series, were then tested in a modified screening format in which the assay was repeated following a 4-hr pre-incubation (Table S3B). A comparison of the IC 50 values revealed that in addition to CDK2, only Aurora A, Mst2, and GCK (MAP4K3) showed a >50% additional loss of activity after pre-incubation. This result suggests that they are also irreversibly inhibited by NU6300. An inspection of the Mst2 (PDB: 4LG4) structure revealed that it has a C-terminal helix that positions a lysine residue at a suitable distance and geometry for covalent modification by a molecule of NU6300 bound within the ATP binding site ( Figure S3). The structure of Aurora A (PDB: 2J4Z) does not suggest an immediate reactive group for the vinyl sulfone moiety (the residues equivalent to CDK2 Lys88 and Lys89 are Tyr219 and Arg220). However, conformational flexibility around the active site may promote such an interaction with a residue from another part of the structure (Martin et al., 2012). Taken together, these results suggest that NU6300 is not expected to have considerable off-target activity in cells.
The cellular potency of NU6300 was examined by measuring the inhibition of Rb phosphorylation following exposure of SKUT-1B cells to the inhibitor for 1 hr. Pre-incubation with . Crystal Structure of NU6300 Bound to CDK2/Cyclin A (A) CDK2 and cyclin A are rendered in ribbon representation and colored mint blue and lemon, respectively. NU6300 is shown in ball-and-stick representation, with carbon atoms in yellow. (B) Schematic representation of the hydrogen bonds made between backbone atoms of CDK2 residues Glu81, Leu83, and Asp86, located in the hinge region, and NU6300 to illustrate the binding mode. Hydrogen bonds are drawn as arrows. (C) Electron density map at the CDK2 active site. The 2F 0 -F c map is contoured at 0.2 e À A 3 . CDK2 and NU6300 carbon atoms are colored green and yellow, respectively. See also Figure S2 and Table S2. 50 mM NU6300 inhibited phosphorylation of Rb at Thr821, a known CDK2 phosphorylation site, by 43%. After a 1-hr drug washout in drug-free media, the inhibition of Rb phosphorylation was 33% of the untreated control, indicating that more than 75% of the inhibitory activity had been retained (Figure 4). This difference just reached statistical significance (p = 0.04). In contrast, NU6102 at 50 mM inhibited phosphorylation of Thr821 on Rb by 85%, but within 1 hr of washout only 40% inhibition was observed, which represented a highly statistically significant change (p = 0.0003) and indicated that just over half of the inhibitory activity had been lost. These data are consistent with NU6300 having irreversible activity against CDK2 in cells.

DISCUSSION
A number of CDK2-specific inhibitors with diverse pharmacophores have been structurally characterized . Our results suggest that these molecules could be modified by taking a similar approach to that described herein, thus generating more structurally diverse irreversible CDK2 inhibitors to explore the potential of CDK2 inhibition in combination chemotherapies. While we have demonstrated that NU6300, an irreversible inhibitor of CDK2, can reach and modulate its target within cells, it remains to be established whether this activity can enhance growth inhibition in a suitable cell line model.
A recent report has also demonstrated that CDK7 can be covalently modified following incubation with an inhibitor bound at the ATP binding site (Kwiatkowski et al., 2014), suggesting that this strategy may be applicable to the wider CDK family. For example, CDK1 and CDK5 both encode a lysine residue equivalent to CDK2 Lys89. Profiling NU6300 against a range of kinases revealed limited off-target activity. We hypothesize that unexpected cross-reactivity arises from the warhead forming covalent interactions with appropriately positioned amino acid side chains that originate from distinct parts of the kinase fold, and that this activity may be abrogated by judicious substitution on the vinyl sulfone moiety.

SIGNIFICANCE
Protein kinase inhibitors that bind covalently within the enzyme's active site offer an attractive alternative route for developing drugs against this clinically important protein family. CDKs have significant roles in regulating both the cell cycle and transcription in eukaryotic cells. They have been the subject of intensive studies resulting in a number of drugs entering clinical trials for cancer treatment. We describe the first example of an irreversible inhibitor that targets CDK2 by grafting a reactive vinyl sulfone moiety onto a potent reversible CDK2 inhibitor. The structure of the CDK2/ cyclin A/NU6300 complex reveals the inhibitor binding mode within the ATP binding site and confirms that the vinyl sulfone forms a covalent bond to the ε-amino group of Lys89. Furthermore, this structure suggests how the reactive moiety could be grafted onto other CDK2-selective pharmacophores to develop inhibitors that specifically target CDK2.   Kinase Assays CDK2/cyclin A kinase assays were carried out using a method modified from Brown et al. (1999) or by using the ADP-Glo assay (Promega) essentially as described by the manufacturers. A full description of the assay formats is provided in the Supplemental Experimental Procedures.

Interaction Analysis
The interaction experiments were performed using SPR biosensor technology, with Biacore S51 and T100 instruments, CM5 biosensor chips, and standard reagents (GE Healthcare). Full details can be found in the Supplemental Experimental Procedures.

Crystallography
The CDK2/cyclin A/NU6300 complex was crystallized as described by Davies et al. (2002). Data processing was carried out using programs of the CCP4 suite (CCP4, 1994), run through the CCP4i2 GUI. The structure was then solved by molecular replacement using Phaser (McCoy et al., 2007) and a high-resolution structure of a recruitment peptide bound to CDK2/cyclin A (PDB: 2CCH) as a search model. Structures were refined using REFMAC (Murshudov et al., 1997), interspersed with manual rebuilding in Coot (Emsley et al., 2010), including TLS (translation/libration/screw) refinement. Full details can be found in the Supplemental Experimental Procedures. The statistics for the datasets and crystallographic refinement are presented in Table S2.

Western Blotting
Western blot analysis was carried out as described previously (Thomas et al., 2011) using rabbit anti-T821 phospho-Rb antibody (Invitrogen) or mouse antihuman Rb antibody (BD Pharmingen) to detect phosphorylated and total retinoblastoma protein, respectively. Sample preparation is described in Supplemental Experimental Procedures.

ACCESSION NUMBERS
The coordinates and structure factors of CDK2/cyclin A/NU6300 have been deposited in the PDB with accession code PDB: 5CYI.

ACKNOWLEDGMENTS
We thank the beamline staff at The Diamond Light Source who provided excellent facilities for data collection, and E. Lowe and A. Basle for assistance with data collection and management. The authors would also like to thank A. Hole, A. Echalier, and R. Suckling for preparing CDK2 mutants, E. Homan for interpretation of SPR data, N. Brown for advice, and I. Taylor for technical support.

Supplemental Data
Supplemental Figure S1, associated with Figure 2 Supplemental Figure S2, associated with Figure 3 Supplemental Figure S3, associated with Figure 4 Supplemental Table S1, associated with Figure 2 Supplemental Table S2, associated with Figure 3 Supplemental  30094 in each panel derive from cyclin A, which is not modified by NU6300. The unmodified CDK2 has a mass of 34426 Da, which agrees well with the mass predicted from sequence (34421) and which includes in addition to the CDK2 sequence, the phosphate group on Thr160 and an N-terminal GPLGS sequence that is a cloning artefact.
Despite the high inhibitor concentration used in the incubation, only one molecule of NU6300 is incorporated into CDK2, and no modification of cyclin A is observed. (B) Incubation of CDK2 with 10 µM NU6300 for up to 20 h reduced the amount of bound NU6310 by almost 50%. (C, D) The SPR sensorgrams (reference subtracted) for the interaction between immobilised CDK2 and NU6310 after incubation with 10 µM NU6300 for the indicated times (C) and the corresponding sensorgrams for the interaction between CDK2 and NU6310 after incubation with buffer only (D) were qualitatively similar and only differed in signal levels.   Table S2. (G) Protein kinase activity of authentic CDK2 and single site CDK2 mutants following incubation with NU6300 or NU6310. Samples were taken at the indicated number of minutes after initiation of the reaction. Note that compared to authentic CDK2, the single site Lys88 and Lys89 mutants have a slower rate of reaction (Compare LHS panels). This drop in activity was expected because whereas the wild type protein was produced by co-expression with S. cerevisiae CAK1 (CDK-activating kinase), and hence stoichiometrically phosphorylated on Thr160, the mutant proteins were activated by S. cerevisiae CAK1 phosphorylation in vitro-a relatively inefficient process as judged from the results of ESI-MS (Table S1). (H) NU6300 bound at the CDK2 active site: "omit" electron density demonstrates a covalent link to  ice-blue ribbon representation) was superimposed on the structure of CDK2 bound to NU6300 (green ribbon representation). NU6300 attached to K89 of CDK2 is drawn in ball and stick/cyclinder mode with carbon atoms coloured green. K298, which forms part of a C-terminal helix of MST2, is drawn in cylinder mode with carbon atoms coloured ice blue.   602 and 34,860). The major CDK2 K88E species (34,234) undergoes partial modification to yield two proteins differing in mass by 414 Da suggesting that mutation of Lys88 to a glutamate affects the accessibility and/or activity of Lys89 as a substrate for NU6300. However, when Lys89 is mutated, there is very little modification of CDK2 ( Figure S2F). The CDK2 mutants were phosphorylated in vitro post-purification by S.
cerevisiae CAK rather than phosphorylation proceeding in vivo by co-expression of CDK2 and CAK in recombinant E. coli cells.
The MS results suggest that whereas phosphorylated CDK2 is the major species following in vivo phosphorylation of the wildtype protein, the in vitro phosphorylation reaction carried out on the CDK2 mutants is    was tested at a single concentration of 1µM against 131 protein kinases present in a Dundee protein kinase screen (Bain et al., 2007). Values given are % activity remaining and are the average of duplicate measurements: 0-25% (red), 26-50% (yellow), 51-80% (black) and above 81% (green). The kinase domain sequences (as defined by UniProt) of the 13 kinases that exhibit < 25% activity were aligned and the available structures were also superposed. An inspection of both comparisons showed that these kinases do not have a lysine residue at a position close to CDK2 Lys88 or Lys89. However, in the structure of tyrosine-protein kinase BTK (PDB ID: 3GEN) the residue equivalent to CDK2 Asp86 is Cys481 and therefore there is the potential for a covalent adduct to be formed through this side chain. Ephrin type-B receptor 3 (UniProt entry P54753) also encodes a cysteine residue (Cys717) that by sequence alignment would be predicted to be close to the CDK2 lysine pair. (B) Selected kinases were re-tested in the Dundee kinase screen against NU6300 at a single concentration of 0.5 µM both in the standard assay format and following a 4 hour pre-incubation in the presence of NU6300. Kinases which have very low activity after the 4 hour incubation so that results might not be wholly reliable are boxed in salmon. Kinases where the activity after a 4 hour pre-incubation is < 50% that in the standard assay format are highlighted in red.

Protein expression and purification
Site-directed mutagenesis of CDK2 was performed using the QuikChange II kit (Stratagene) following the manufacturer's instructions and verified by DNA sequencing.

ESI mass spectrometry
CDK2/cyclin A at 3 mg mL -1 was incubated overnight with 2.5 mM NU6300 or NU6310 (prepared as stocks at 50 mM in 100% DMSO) or DMSO only and then desalted using C4 ZipTips (Millipore) according to manufacturer's instructions. The samples in 1:1 (v/v) acetonitrile and water + 0.1% formic acid were introduced at a flow rate of 10 µL/min by electrospray ionisation (ESI) into a Micromass LCT orthogonal acceleration reflecting TOF mass spectrometer in positive ion mode. The mass spectrometer had been calibrated using myoglobin. The resultant m/z spectra were converted to mass spectra by using the maximum entropy analysis MaxEnt in the MassLynx suite of programmes.

Kinase assays
CDK2/cyclin A (200 µM) was incubated overnight with 1 mM inhibitor or DMSO, and then dialysed for 8 h with regular replacement of dialysis buffer into HEPES-buffered saline (50 mM HEPES, pH 7.5, 250 mM NaCl, 0.02% MTG). Kinase assays were performed at 8 µg mL -1 CDK2/cyclin A and 50 µg mL -1 GST-pRb (residues 792-928 of pRb fused at the Nterminus to glutathione-S-transferase) as substrate in 10 µl of buffer containing 50 mM Tris-HCl pH 7.5, 10 mM MgCl 2 , 1.0 mM ATP for 15 min at room temperature. Reactions were stopped by addition of SDS-PAGE loading buffer and analyzed by SDS-PAGE.
CDK2/cyclin A was also assayed using the ADP-Glo TM assay (Promega). 100 nM CDK2/A was incubated with 300 nM NU6300 for 0, 10, 30, 60, and 120 mins at 18 °C. The CDK2/cyclin A-NU6300 incubated solutions were then finally diluted 50-fold and the kinase reaction was initiated by the addition of ATP and CDK2/A peptide substrate HHASPRK (Enzo Scientific), resulting in a final assay concentration of 2nM CDK2/A, 6nM NU6300, 25 µM ATP and 50 µM peptide substrate. The kinase reaction was allowed to proceed for 30 mins, with the production of ADP detected via a luminescence signal produced using the ADP-Glo TM assay (Promega). Reactions were conducted in triplicate in 40 mM Tris pH7.5, 20 mM MgCl 2 , 0.1mg/ml BSA, and 1% DMSO in white low volume 384-well plates using a PheraStar plate reader (BMG). Data was plotted using SigmaPlot 12.0 to derive a reaction velocity of 5.98 x 10 -14 M.s -1 from which a k inact of 4.98 x 10 3 M -1 .s -1 was calculated.

Interaction analysis
CDK2 was immobilized to activated carboxylated dextran surfaces by amine coupling to give surface densities of 6000-25000 RU. All interaction experiments were performed at 25 °C in 10 mM phosphate pH 7.4, 137 mM NaCl, 3 mM KCl, with addition of 0.05 % Tween 20, 5 % (v/v) DMSO, and at a flow rate of 90 or 30 µL/min. For analysis of the time dependence of the irreversible interaction ( Figure S1 B-D), immobilized CDK2 was exposed to 10 µM NU6300 for 0 h, 4 h, and 20 h. A CDK2 surface incubated with buffer without NU6300 in parallel on the same chip was used as a reference. The binding capacity of the two surfaces was assayed by injections of 10 µM NU6310. For kinetic analysis (Figure 2A), the two test compounds were diluted in the running buffer and injected for 30-240 s over the immobilized CDK2 at increasing concentrations.
Sensorgrams or extracted report points from reference surfaces and blank injections were subtracted from the raw data prior to data analysis, using Biacore T100 evaluation software 2.0. A 1:1 interaction model was fitted globally to sets of sensorgrams recorded with different contact times and at different inhibitor concentrations in multi-cycle experiments. Kinetic parameters were determined from sensorgrams with 30 s and 60 s contact time. Standard deviations were based on at least 4 measurement series.

Crystallography
CDK2/cyclin A was mixed with a freshly prepared solution of NU6300 to achieve DMSO and inhibitor concentrations of 2% and 2 mM respectively, concentrated by ultrafiltration to a CDK2/cyclin A final concentration of circa 5 mg ml -1 , and then crystallized as described (Davies et al., 2002). Briefly, crystals were grown from a mother liquor containing 0.6-0.8 M KCl, 0.9-1.2 M (NH4) 2 SO 4 , and 100 mM HEPES (pH 7.0). Sitting drops were set up with a 1:1 ratio of protein to reservoir solution in a total initial volume of 0.5 or 1.0 µl.
Before data collection, crystals were briefly immersed in cryo-protectant (1 M sodium formate) before cryo-coolling. Data processing was carried out using programs of the CCP4 suite (CCP4, 1994). The structure of NU6300 bound to CDK2/cyclin A was solved by molecular replacement using Phaser (McCoy et al., 2007), using as the search model a high-resolution structure of a recruitment peptide bound to CDK2/cyclin A ( (Cheng et al., 2006) PDB code 2CCH). A single clear solution was found with an inhibitor bound to each of the two copies of the binary complex in the asymmetric unit. This solution was then subjected to rigid body refinement in REFMAC (Murshudov et al., 1997), to reveal unambiguous electron density in the CDK2 ATP-binding site, consistent with the expected shape of the inhibitor. A model of NU6300 was created using Coot (Emsley et al., 2010).
The inhibitor atoms were kept in all subsequent models during refinement carried out by additional rounds of manual rebuilding in Coot and restrained refinement in REFMAC5, including TLS refinement. Towards the end of refinement, waters were added using the Coot water picking utility and manually verified.

Western blotting
SKUT-1B cells (ATCC, Manassas, USA) were grown in MEM medium supplemented with non-essential amino acids, L-Glutamine, sodium pyruvate and 10% (v/v) foetal calf serum (Sigma, UK). SKUT-1B cells were incubated with NU6300 (50 µM), NU6102 (50 µM) or DMSO (as control) for 1 hour, then media containing the inhibitors was removed, washed once with PBS and fresh media was added. Cells were harvested at different time points after the washout and lysed by adding PhosphoSafe extraction reagent (Merck, UK) containing protease inhibitor cocktail (Roche, UK) at the manufacturer's recommended dilution. The harvested cell suspension was placed in an eppendorf tube on ice, centrifuged at 13,000x g for 5 min, and the supernatant (cell lysate) removed for analysis.
Subsequent western blot analysis was carried out as described in (Thomas et al., 2011) using rabbit anti-T821 phospho-Rb antibody (Invitrogen, Paisley, UK Cat No. 44-582G) or mouse anti-human Rb antibody (BD Pharmingen, Oxford, UK Cat No. 554136) to detect phosphorylated and total retinoblastoma protein, respectively.