letters

nature structural biology • volume 8 number 1 • january 2001 37

Mechanism-based designof a protein kinaseinhibitorKeykavous Parang1,2, Jeffrey H. Till2,3, Ararat J. Ablooglu2,4, Ronald A. Kohanski4, Stevan R. Hubbard3 and Philip A. Cole1

1The Johns Hopkins University School of Medicine, Department ofPharmacology and Molecular Sciences, Baltimore, Maryland 21205, USA.2These authors contributed equally. 3Skirball Institute of BiomolecularMedicine and Department of Pharmacology, New York University School ofMedicine, New York, New York 10016, USA. 4Mount Sinai School ofMedicine, Department of Biochemistry and Molecular Biology, New York,New York 10029, USA.

Protein kinase inhibitors have applications as anticancertherapeutic agents and biological tools in cell signaling.Based on a phosphoryl transfer mechanism involving a dis-sociative transition state, a potent and selective bisubstrateinhibitor for the insulin receptor tyrosine kinase was syn-thesized by linking ATPγS to a peptide substrate analog via atwo-carbon spacer. The compound was a high affinity com-petitive inhibitor against both nucleotide and peptide sub-strates and showed a slow off-rate. A crystal structure of thisinhibitor bound to the tyrosine kinase domain of theinsulin receptor confirmed the key design features inspiredby a dissociative transition state, and revealed that the link-er takes part in the octahedral coordination of an active siteMg2+. These studies suggest a general strategy for the devel-opment of selective protein kinase inhibitors.

Protein kinases play a critical role in cell signaling pathwaysby catalyzing the transfer of the γ-phosphoryl group from ATPto the hydroxyl groups of protein side chains1. The proteinkinase superfamily is one of the largest as ∼ 2% of eukaryoticgenes encode them. Because of their importance in contribut-ing to a variety of pathophysiologic states, including cancer,inflammatory conditions, autoimmune disorders, and cardiacdiseases, intense efforts have been made to develop specificprotein kinase inhibitors as biological tools and as therapeuticagents2. Considerable success has been achieved in developingpotent and selective nucleotide analog-based inhibitors thatinteract with individual protein kinases at their nucleotidebinding sites, and several compounds are in early phases ofhuman clinical trials2. In general, the protein substrate bindingsite has not been exploited for inhibitor design3. Moreover,mechanism-based approaches to generating protein kinaseinhibitors have been unsuccessful4–8. This stands in contrast tothe situation with many other important enzyme classes, suchas the proteases, where consideration of enzyme mechanismand structure has led to potent inhibitors, some of which areclinically useful drugs9.

Protein kinases follow ternary complex kinetic mechanismsin which direct transfer of the phosphoryl group from ATP toprotein substrate occurs10. For such mechanisms, designingcovalently linked bisubstrate analogs can be a powerfulapproach for making potent enzyme inhibitors9. However, pre-vious attempts to employ this strategy with protein kinaseshave met with mixed results4,8,11. A sophisticated effort reportedby Gibson and colleagues8 linked ATP directly to the Ser oxygen

of a protein kinase A (PKA) peptide substrate (kemptide) togenerate an inhibitor (Fig. 1a, compound 1). Compound 1 wasa weak inhibitor with an IC50 of 226 µM (compared to the Km

(ATP) of 10 µM and the Km (kemptide) of 15 µM) that wascompetitive versus ATP but noncompetitive versus peptide sub-strate. While not providing all the desired features of a bisub-strate analog, these results suggest that improvements ingeometry and electronic character around the atoms equivalentto the entering nucleophile and reacting phosphate would ben-efit inhibitor design.

Compound 1 has a Ser Oγ–P bond distance (1.7 Å) morecompatible with a fully associative reaction mechanism forphosphoryl transfer by a protein serine kinase (Fig. 1b). In sucha transition state, a bond is largely formed between the attack-ing oxygen and the reactive γ-phosphorus atom, while the bondto the departing ADP would not yet have been significantlybroken (Fig. 1b). In contrast, studies of two protein tyrosinekinases in which the nucleophilicity of the attacking hydroxylfrom the Tyr residue was varied have provided strong evidencefor the occurrence of a dissociative or metaphosphate-like tran-sition state6,12. A dissociative transition state for phosphoryltransfer catalyzed by a protein kinase, analogous to an SN1 reac-tion in organic chemistry, is one in which the importance of thenucleophilicity of the attacking hydroxyl is diminished anddeparture of the leaving group (ADP) is well advanced(Fig. 1b). Dissociative transition states are well established fornonenzymatic phosphate monoester phosphoryl transfer reac-tions but have been considered more controversial for the cor-responding enzyme catalyzed reactions13,14. A prediction forsuch a fully dissociative transition state is that the ‘reactioncoordinate distance’ between the entering nucleophilic oxygenand the attacked phosphorus should be ≥4.9 Å. This is based onthe assumptions that the γ-phosphoryl group moves toward theentering oxygen and that this nucleophile and the ADP arefixed, probably as they appear in the ground state ternary com-plex that precedes the transition state14. For a transition statewith greater associative character, the reaction coordinate dis-tance could be ≤3.3 Å, indicating compression14. High resolu-tion X-ray and NMR structures of various protein kinasecomplexes with nucleotide and peptide substrate analogs haveyielded conflicting values for such distances14–16, although thesestudies have, by necessity, employed unreactive substrateanalogs that could have affected the results.

In reconsidering the design of a bisubstrate analog based onthe above parameters of (largely) dissociative reaction mecha-nisms for protein kinases, we hypothesized that a peptide–ATPbisubstrate analog (Fig. 1a, compound 2) in which the distancebetween the tyrosine nucleophilic atom and the γ-phosphoruswas set to ∼ 5 Å (5.66 Å in an extended form; Fig. 1a) by a shortlinker might show more potent inhibition toward a proteintyrosine kinase than that of compound 1 toward PKA. A seconddesign feature was based on the fact that proton removal fromthe hydroxyl of the Tyr residue occurs late in the dissociativereaction mechanism6,12, and that the phenolic hydroxyl servesas a hydrogen bond donor to the conserved catalytic Aspresidue (Asp 1,132 of the insulin receptor). To exploit thisinteraction, the Tyr oxygen was replaced with a nitrogen atomthat could serve as a hydrogen bond donor while simultaneous-ly being incorporated into a tether. Our choice of target enzymefor the inhibitor was the insulin receptor protein tyrosinekinase (IRK) because: (i) efficient peptide substrates (includingIRS727) for this enzyme have been well characterized kinetical-ly6; (ii) solution studies have provided direct evidence for a dis-

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com

letters

38 nature structural biology • volume 8 number 1 • january 2001

sociative transition state for this enzyme6; (iii) a high resolutioncrystal structure of this enzyme in a ternary complex with pep-tide substrate and an ATP analog has been determined15; and(iv) no potent inhibitors have been reported for this importantsignaling enzyme or for the highly related insulin-like growthfactor receptor protein tyrosine kinase.

Synthesis and kinetic analysis of a bisubstrateinhibitor The synthesis of compound 2 was carried out as shown in Fig.1c. The protected peptide was assembled using solid-phasepeptide synthesis using Fmoc chemistry, in which the single Tyrof IRS727 was replaced with commercially availablenitrophenylalanine. The nitro group was reduced with SnCl2

and the aniline function was bromoacetylated under standardcoupling conditions. The peptide was cleaved from the resinand deblocked with trifluoroacetic acid, water, and thioanisoleand purified by reversed phase HPLC. The bromoacetylatedpeptide was reacted with ATPγS in aqueous solution at pH 7,which resulted in chemoselective displacement of the bromideby the phosphorothioate. The peptide–ATP conjugate 2 waspurified by reversed phase HPLC and characterized by electro-spray mass spectrometry and 1H NMR. While it was stable inaqueous solution at neutral pH and room temperature for>12 h, it was unstable in acidic solution (pH 2), decomposing

with a half life of ∼ 2 h. It could be stored in frozen solution at -80 °C for several weeks without detectable decomposition.

Kinase assays with compound 2 revealed it to be a potentinhibitor of IRK (Fig. 2). Using a steady state kinetic analysis,compound 2 was a linear competitive inhibitor versus bothnucleotide (Fig. 2a) and peptide (Fig. 2b) substrates with a Ki of370 nM (extrapolated to zero substrates), which is 190–760-fold lower than the Km values of the substrates (Km app (ATP) = 71 µM; Km app (IRS727) = 280 µM). In an alterna-tive analysis of compound 2, the IC50 is ∼ 10 µM at a concentrationof 1 mM ATP and 0.3 mM IRS727. By using a rapid dilutionanalysis, the koff was measured to be 0.013 s–1 (Fig. 2c), which isalso consistent with its tight binding behavior. The calculated second order rate constant kon for compound 2 binding to IRK is3.5 × 104 M–1 s–1, which is substantially slower than a predicteddiffusional encounter rate (108–109 M–1 s–1) and suggests that a‘slow’ conformational change may be responsible for reaching thetightly bound complex17. Compound 2 represents the mostpotent inhibitor yet reported for IRK, and its potency comparesfavorably to bisubstrate analog inhibitors designed for other(nonprotein kinase) enzymes9. The most potent bisubstrate ana-log inhibitors have binding free energies equal to the sum of thebinding energies of the two substrates. The binding energy ofcompound 2 (8.9 kcal mol–1) is approximately equal to the sum ofits two parts (4.1 kcal mol–1 for IRS727 peptide binding and

a

b

c

Fig. 1 Synthetic compounds and mechanisticschemes. a, Designed protein kinase bi-substrate analog inhibitors. Compound 1was synthesized and studied by Gibson andcolleagues8 as a potential PKA inhibitor.Compound 2 was designed based on a disso-ciative transition state for phosphoryl trans-fer. Distance between the anilino nitrogenand the γ-phosphorus was calculated usingChem3D assuming an extended conforma-tion of the acetyl linker. The peptidesequence was derived from IRS727 (ref. 27).Compound 3 was prepared to evaluate therelative contribution of the peptide residuestoward inhibition. b, Scheme illustratingassociative versus dissociative transitionstates for phosphoryl transfer. ROH is thenucleophile (tyrosine phenol in this work)attacking the γ-phosphoryl group of ATP,and ADP is the leaving group. The associa-tive transition state (path A) in this work isdefined as >50% bond formation betweenthe nucleophilic oxygen and the phospho-rus, which occurs with at least 50% leavinggroup residual bond formation present. Thedissociative transition state (path B) in thiswork is defined as <50% bond formationbetween the nucleophile and the phospho-rus, which occurs before the leaving groupphosphorus bond is at least 50% broken. c, Synthetic scheme for the preparation ofbisubstrate analog compound 2. R1, AcNH-Lys-Lys-Lys-Leu-Pro-Ala-Thr-Gly-Asp- ; R2, -Met-Asn-Met-Ser-Pro-Val-Gly-Asp-CO2H; R3,R1 with side chain protected residues; R4, R2

with side chain protected residues and Asplinked to Wang resin.

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com

letters

nature structural biology • volume 8 number 1 • january 2001 39

5.1 kcal mol–1 for ATPγS binding; A.J.A. and R.A.K., unpublishedresults), confirming that each substrate-like part of compound 2contributes to the binding affinity.

To further probe the basis of the potency of compound 2, thederivatized ATP analog compound 3 (Fig. 1a), which lacks thepolypeptide portion of compound 2, was prepared and testedas an IRK inhibitor. It was a linear competitive inhibitor of IRKversus ATP with a Ki of 114 µM (data not shown). This is onlytwo-fold tighter binding than ATPγS alone (Ki of 210 µM) and∼ 300-fold weaker as an inhibitor than compound 2, indicatingthat the peptide moiety is an essential contributor to theinhibitory potency of compound 2 and that the anilino tether-ing group present in compound 3 accomplishes little by itself.

The specificity of inhibition was also tested by assaying com-pound 2 as an inhibitor of the protein tyrosine kinase Csk18.The peptide binding selectivity for Csk is quite different fromIRK, and the IRS727 peptide moiety would be predicted tobind poorly to this enzyme19. As anticipated, compound 2 wasonly a modest inhibitor of Csk, with a Ki of ∼ 40 µM (Km (ATP)of 195 µM). Thus, the bisubstrate analog approach shouldallow for kinase selectivity based in part on peptide sequence.

Structure of compound 2 bound to insulin receptorkinase To understand the structural details of IRK inhibition by com-pound 2, we determined the crystal structure of compound 2bound to the core tyrosine kinase domain of the insulin recep-tor (cIRK) at 2.7 Å resolution (Table 1; Fig. 3). cIRK wasautophosphorylated in vitro to yield the activated form of cIRKin which Tyr 1,158, Tyr 1,162 and Tyr 1,163 in the activationloop are phosphorylated. The bisubstrate inhibitor was cocrys-tallized with cIRK, producing trigonal crystals with the sameunit cell dimensions as crystals of the ternary complex (cIRK–IRS727–MgAMP-PNP)15. The structure of the kinase in thebinary complex (cIRK–bisubstrate inhibitor) is very similar tothe structure in the ternary complex, with an overall root meansquare (r.m.s.) deviation of only 0.22 Å for 303 cIRK Cα atoms.

Several remarkable features were observed in the active siteof cIRK. First, two mechanism-based design components werevalidated: the distance between the γ-phosphorus and the anili-no nitrogen atom (corresponding to the phenolic oxygen ofTyr) was 5.0 Å, the expected reaction coordinate length for adissociative mechanism as discussed above; and a 3.2 Å hydro-gen bond between the anilino nitrogen and the key catalyticAsp (Asp 1,132) was evident. Serendipitously, the carbonyloxygen of the linker was coordinated to a Mg2+ in the active site(the O–Mg distance was 2.2 Å), replacing a water molecule atthat position in the ternary complex (Fig. 3c) that preserves theoctahedral coordination of this Mg2+. A second Mg2+ appears tohave a lower occupancy in the binary complex than wasobserved in the ternary complex. In the binary complex, theside chain of invariant Lys 1,030 was hydrogen bonded toinvariant Glu 1,047 as well as to the α-phosphate group ofATPγS (Fig. 3c). This is in contrast to the ternary complex in

which a hydrogen bond from Lys 1,030 to the α-phosphategroup was observed, but the Lys 1,030–Glu 1,047 distance was4.4 Å. A salt bridge between the invariant Lys and Glu may berequired for efficient phosphoryl transfer20.

The peptide moiety of the bisubstrate inhibitor binds tocIRK in a similar manner to that observed in the ternary com-plex15: an antiparallel β-strand interaction is present betweenresidues P+1 to P+3 of the peptide and Gly 1,169–Leu 1,171 ofthe activation loop, and the Met residues P+1 and P+3 are posi-tioned in shallow hydrophobic pockets on the surface of the C-terminal kinase lobe. Several IRS727 residues that had beendisordered in the ternary complex (P-3 to P-5 and P+4 to P+7)are visible in the complex with the bisubstrate inhibitor (sameunit cell), suggesting a more stable interaction (Fig. 3b). Thepeptide portion of the bisubstrate inhibitor outside of the corerecognition region (P-1 to P+3) follows a unique path (Fig. 3a).Interestingly, the residues N-terminal to the modified Tyr donot lie along the extended peptide binding groove definedlargely by residues in α-helix D, as observed for the proteinkinase inhibitor (PKI) bound to PKA21; the P-5 Pro of thebisubstrate inhibitor is adjacent to Gly 1,005 of the nucleotidebinding loop in the N-terminal lobe. This may be due to charge

Fig. 2 Kinetic analysis of the inhibition of IRK by bisubstrate analog com-pound 2. a, E/V versus 1/ATP in the presence of varying concentrations ofcompound 2. b, E/V versus 1/IRS727 in the presence of varying compound2. For (a), the apparent Ki (compound 2) was 550 ± 80 nM; for (b), the apparent Ki (compound 2) was 750 ± 90 nM. c, Product/Enzyme versus Time. This experiment monitors the phospho-peptide (product) formation after a rapid dilution of the enzyme–inhibitor complex to indirectly measure the koff for dissociation of com-pound 2 from IRK; koff = 0.013 ± 0.001 s-1.

a

b

c

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com

letters

40 nature structural biology • volume 8 number 1 • january 2001

repulsion between the three N-terminal Lys residues of thepeptide moiety and Arg 1,089 and Arg 1,092 of cIRK, which areexposed on α-helix D. In addition, residues at the C-terminusof the peptide moiety (P+4, etc.) lie in a shallow groove bound-ed by the activation loop and α-helices EF and G. This latterinteraction was unanticipated and could possibly be associatedwith slow release of the peptide.

ConclusionsIn summary, we have established a new approach to proteinkinase inhibition based in part on mechanistic and structuralconsiderations of a predicted dissociative transition state forthese enzymes. This method should be applicable to other pro-tein kinases, taking into account the distinctive peptide sub-strate sequence selectivities of the individual enzymes. Thiswould allow designer inhibitors to be generated as biologicaland structural tools once the peptide specificity for a proteinkinase has been established by one of several library-based

techniques22,23. A limitation of this approach in its currentform is the suboptimal pharmacokinetic properties of pep-tides and nucleotides as therapeutics. To overcome thisobstacle, phosphate surrogates and peptidomimetics couldbe substituted for the original constituents. Since differentprotein kinases can have overlapping peptide substrateselectivities, greater specificity for inhibition of individualkinases may also be achieved by synthetic refinements of thenucleotide and peptide moieties. By adding a new dimen-sion of molecular recognition to the already high potency ofnucleotide analog-based protein kinase inhibitors, thismechanism-based inhibitor approach could afford anentirely new class of therapeutically useful agents.

MethodsSynthesis of compounds 2 and 3. The nitrophenylalaninepeptide starting material (Fig. 1c) was assembled on Wang resinusing automated solid-phase peptide synthesis via the Fmoc

strategy (0.3 mmol scale). While still in the solid phase, the nitrogroup was reduced with SnCl2⋅2H2O (6 mmol) in dimethylform-amide for 24 h at room temperature with gentle stirring. It waswashed with dimethylformamide and CH2Cl2 and dried. The resinwas then treated with bromoacetic acid (2.5 mmol), diisopropylcar-bodiimide (2.5 mmol) in dimethylformamide for 5 h at room tem-perature with gentle mixing. The resin was washed and dried asabove prior to treatment with cleavage and deblocking conditions(5 ml trifluoroacetic acid, 1.5 ml CH2Cl2, 0.25 ml H2O, 0.1 mlthioanisole) for 30 min at room temperature. The particulate wasremoved by filtration through scintered glass and the filtrate wasconcentrated and treated with cold ether to precipitate the bromo-acetylated peptide. The crude bromoacetylated peptide waswashed once more with cold ether, dried under vacuum, and puri-fied by preparative reversed phase HPLC (gradient with H2O:CH3CN:0.05% (v/v) CF3CO2H, UV monitored at 214 nm) to yield 71 mg ofpure material. Electrospray mass spectrometry was used to confirmits structural identity. The purified bromoacetylated peptide(17 mg) was dissolved in 4:1 MeOH:H2O (3.6 ml) and treated withATPγS (adenosine 5′-O-3-thiotriphosphate, 0.02 mmol; Boehringer)for 24 h at room temperature. Compound 2 was subsequently puri-

Fig. 3 Crystal structure of the binary complex between cIRK and thebisubstrate inhibitor. a, Overall view of the binary complex in whichcIRK is shown in surface representation and the bisubstrate inhibitorin stick representation. The ATPγS moiety of the inhibitor is coloredgreen, the peptide moiety is colored red, and the linker connectingthe nucleotide and peptide is colored yellow. The cIRK surface issemi-transparent; the N-terminal lobe of cIRK partially masks thenucleotide. b, Stereo view of the Fo - Fc electron density (2.7 Å resolu-tion, 3σ contour) for the bisubstrate inhibitor computed after simu-lated annealing (1,000 K), omitting from the atomic model eitherATPγS + linker (blue map) or the peptide moiety (green map).Coloring of the bisubstrate inhibitor is the same as in (a). Selectedpeptide residues are labeled; Y′(P0) refers to the modified Tyr at theP0 position of the peptide. The purple sphere represents the Mg2+

and the red sphere represents the Mg2+-coordinating water mole-cule. c, Stereo view of the interactions between the inhibitor and keycatalytic residues. Superimposed are the cIRK–bisubstrate inhibitor(binary) complex and the cIRK–MgAMP-PNP–peptide (ternary) com-plex15. Oxygen atoms are colored red, nitrogen atoms blue, sulfuratoms green, and phosphorus atoms yellow. Bonds/carbon atoms arecolored orange for the binary complex and green for the ternarycomplex. Bonds and atoms of the ternary complex are semi-transpar-ent. Mg2+ ions and water molecules are represented as purple andred spheres, respectively. Hydrogen bonds and bonds to the Mg2+ arerepresented as dashed and solid black lines, respectively. Only themodified tyrosine from the peptide moiety of the inhibitor is shown.‘BSI’ indicates the bisubstrate inhibitor in the binary complex, and‘Y(P0)’ and ‘PNP’ indicate the substrate tyrosine and AMP-PNP,respectively, in the ternary complex. Selected secondary structuralelements (αC and β3) are shown. The figure was prepared withGRASP28, BOBSCRIPT29 and MOLSCRIPT30.

b

c

a

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com

letters

pound 2 for 10 min in reaction buffer and then diluted 1:50 intoreaction buffer containing 1 mM ATP and 250 µM peptide IRS939(another peptide substrate)6 and aliquots were taken to measurephosphopeptide formation at selected times. Data were fit to theequation17: P = vs(t-[1 - e-koff × t])/koff, where P is phosphopeptideconcentration, vs is steady state velocity, t is time and koff is thefirst-order rate constant for dissociation of compound 2 from theenzyme.

Crystal structure of cIRK with compound 2. Tris-phosphory-lated cIRK (residues 978–1283) was prepared as described15.Crystals were grown at 4 °C by vapor diffusion in hanging dropscontaining 1.0 µl of protein solution (280 µM cIRK (10 mg ml-1),400 µM of the bisubstrate inhibitor, and 1.2 mM MgCl2) and 1.0 µlof reservoir buffer (18% (w/v) polyethylene glycol 8000, 100 mMTris-HCl, pH 7.5). Crystals belonged to space group P3221 with unitcell dimensions a = b = 66.3 Å, c = 138.1 Å. Crystals were trans-ferred into a cryosolvent consisting of 30% (w/v) polyethyleneglycol 8000, 100 mM Tris-HCl, pH 7.5, and 15% (v/v) ethylene glycol. A data set was collected from a flash-cooled crystal on aRigaku RU-200 rotating anode equipped with a Rigaku R-AXIS IICimage plate detector. Data were processed using DENZO andSCALEPACK24. Rigid body, simulated annealing, positional and B-factor refinement were performed with CNS25 and model build-ing with O26. Bulk solvent and anisotropic B-factor correctionswere applied.

Coordinates. Coordinates have been deposited in the PDB (acces-sion code 1GAG).

AcknowledgmentsWe gratefully acknowledge support from the NIH (to R.A.K., S.R.H. and P.A.C.),and the Burroughs Wellcome Fund. We thank A. Mildvan for helpful discussionsand critical reading of the manuscript. X-ray equipment at The Skirball Institute ispartially supported by grants from The Kresge Foundation and The Hyde andWatson Foundation.

Correspondence should be addressed to S.R.H. email: hubbard@tallis.med.nyu.edu or P.A.C. email: pcole@jhmi.edu

Received 16 August, 2000; accepted 20 October, 2000.

1. Hunter, T. Cell 100, 113–127 (2000).2. Showalter, H.D. & Kraker, A.J. Pharmacol. Ther. 76, 55–71 (1997).3. Lawrence, D.S. & Niu, J. Pharmacol. Ther. 77, 81–114 (1998).4. G. Rosse, G. et al. Helvetica Chimica Acta 80, 653–670 (1997).5. Yuan, C.J., Jakes, S., Elliott, S. & Graves, D. J. Biol. Chem. 265, 16205–16209 (1990).6. Ablooglu, A.J. et al. J. Biol. Chem. 275, 30394–30398 (2000).7. Cushman, M. et al. Int. J. Pept. Protein Res. 36, 538–543 (1990).8. Medzihradszky, D., Chen., S.L., Kenyon, G.L. & Gibson, B.W. J. Am. Chem. Soc.

116, 9413–9419 (1994).9. Silverman, R.B. The organic chemistry of drug design and drug action (Academic

Press, New York; 1992).10. Ho, M., Bramson, H.N., Hansen, D.E., Knowles, J.R. & Kaiser, E.T. J. Am. Chem. Soc.

110, 2680–2681 (1988).11. Uri, A., Raidaru, G., Jarv, J. & Pia, E. Bioorg. Med. Chem. Lett. 9, 1447–1452 (1999).12. Kim, K. & Cole, P.A. J. Am. Chem. Soc. 120, 6851–6858 (1998).13. Admiraal, S.J. & Herschlag, D. J. Am. Chem. Soc. 122, 2145–2148 (2000).14. Mildvan, A.S. Proteins 29, 401–416 (1997).15. Hubbard, S.R. EMBO J. 16, 5572–5581 (1997).16. Brown, N.R., Noble, M.E.M., Endicott, J.A. & Johnson, L.N. Nature Cell Biol. 1,

438–443 (1999).17. Morrison, J.F. & Walsh, C.T. Adv. Enzymol. 61, 201–301 (1988).18. Grace, M.R., Walsh, C.T. & Cole, P.A. Biochemistry 36, 1874–1881 (1997).19. Sondhi, D., Xu, W., Songyang, Z., Eck, M.J. & Cole, P.A. Biochemistry 37, 165–172

(1998).20. Taylor, S.S. & Radzio-Andzelm, E. Structure 2, 345–355 (1994).21. Knighton, D.R. et al. Science 253 414–420 (1991).22. Songyang, Z. et al., Nature 373, 536–539 (1995).23. Till, J.H., Annan, R.S., Carr, S.A. & Miller, W.T. J. Biol. Chem. 269, 7423–7428

(1994).24. Otwinowski, Z. & Minor, W. Methods Enzymol. 276, 307–326 (1997).25. Brünger, A. et al. Acta Crystallogr. D 54, 905–921 (1998).26. Jones, T.A. Methods Enzymol. 115, 157–171 (1985).27. Shoelson, S.E., Chatterjee, S., Chandhuri, M. & White, M.F. Proc. Natl. Acad. Sci.

USA 89, 2027–2031 (1992).28. Nicholls, A., Sharp, K.A. & Honig, B. Proteins 11, 281–296 (1991).29. Esnouf, R.M. J. Mol. Graph. 15, 132–134 (1997).30. Kraulis, P.J. J. Appl. Crystallogr. 24, 946–950 (1991).

fied by reversed phase HPLC (gradient with H2O:CH3CN, neutral pH,UV monitored at 214 or 260 nm) and lyophilized to yield 12 mg ofproduct. Electrospray mass (negative ion) and 1H NMR spectra con-firmed the assigned structure. The concentration of compound 2 inaqueous solution was determined by amino acid analysis (HarvardUniversity Microchemistry Facility). The compound was stored atpH 7 in aqueous solution at -80 °C and periodically monitored fordecomposition by HPLC. Synthesis of compound 3 was carried outby reacting bromoacetanilide and ATPγS analogous to the produc-tion of compound 3. Compound 3 was purified by reversed phaseHPLC and the pure product characterized by 1H NMR, electrospraymass spectrometry and quantified by UV.

Kinetic analysis. Peptide phoshorylation assays were carried outusing activated recombinant IRK as described previously6. Thereaction buffer was 20 mM MgCl2, 0.5 mM DTT, 0.05% bovineserum albumin, 50 mM Tris-Acetate, pH 7. Inhibition by compound2 against variable concentrations of ATP was measured in 10 minreactions at 6 nM IRK and 120 µM IRS727 (peptide substrate). Tomeasure inhibition against variable concentrations of IRS727,5 min reactions were performed at 8 nM IRK and 80 µM ATP.Duplicate measurements agreed within ± 15%. Data for eachexperiment were globally fit to the equation for linear competi-tive inhibition: v = Vm(S) / [Km(1 + I / Ki) + S], where v is initial veloc-ity, Vm is maximal velocity, S is variable substrate concentration,and Km is the Michaelis constant. Inclusion of a Ki-intercept term(for a mixed competitive inhibition calculation) for either analysisdid not reduce standard errors in the global fit and led to veryhigh standard errors for the Ki-intercept terms, thereby addingconfidence to our selection of linear competitive inhibition as thebasis for data analysis. To determine the off-rate constant for com-pound 2, 1 µM IRK was pre-incubated with or without 5 µM com-

Table 1 X-ray data collection and refinement statistics

Data collectionResolution (Å) 30.0–2.7Observations 42,217Unique reflections 9,633Completeness1 (%) 96.5 (91.9)Rsym

1,2 (%) 11.4 (33.9)

RefinementNumber of atoms

Protein 2,370Inhibitor 125Water 23Mg2+ 2

Resolution (Å) 30.0–2.7Reflections 9,342Rcryst / Rfree

3 (%) 21.4 / 27.0R.m.s. deviations

Bond lengths (Å) 0.007Bond angles (°) 1.5B-factors4 (Å2) 1.3

Average B-factors (Å2)All atoms 20.6Protein 20.1Inhibitor (ATPγS + linker) 15.4Inhibitor (peptide) 34.2

1The overall (30.0–2.7 Å) value is given first, with the value in the highestresolution shell (2.8–2.7 Å) given in parentheses.2Rsym = 100 × Σ|I - <I>| / ΣI.3Rcryst = 100 × Σ||Fo| - |Fc|| / Σ|Fo|, where Fo and Fc are the observed and calcu-lated structure factors, respectively (Fo > 0σ). The Rfree was determinedfrom 5% of the data.4For bonded protein atoms.

nature structural biology • volume 8 number 1 • january 2001 41

©20

01 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://s

tru

ctb

io.n

atu

re.c

om

© 2001 Nature Publishing Group http://structbio.nature.com