Targeting Aurora2 Kinase in Oncogenesis: A Structural ...posed onto each other using the program SAP (19). The structural alignments from SAP were fine-tuned manually to better match

Targeting Aurora2 Kinase in Oncogenesis: A StructuralBioinformatics Approach to Target Validation andRational Drug Design1

Hariprasad Vankayalapati, David J. Bearss,Jose W. Saldanha, Ruben M. Munoz,Sangeeta Rojanala, Daniel D. Von Hoff, andDaruka Mahadevan2

Arizona Cancer Center, University of Arizona, Tucson, Arizona 85724[H. V., D. J. B., J. W. S., R. M. M., S. R., D. D. V. H., D. M.]; The Collegeof Pharmacy, University of Arizona, Tucson, Arizona 85721 [H. V.];Division of Mathematical Biology, NIMR, Mill Hill, London NW7 1AA,United Kingdom [J. W. S.]

AbstractThe aurora kinases are a novel oncogenic family ofmitotic serine/threonine kinases (S/T kinases) that areoverexpressed in a number of solid tumors, includingpancreas and colorectal cancer. A PSI-BLAST search[National Center for Biotechnology Information (NCBI)]with the sequence of the S/T kinase domain of humanaurora1 kinase [also known as AUR1, ARK2, AIk2, AIM-1, and STK12] and human aurora2 kinase (also knownas AUR2, ARK1, AIK, BTAK, and STK15) showed a highsequence similarity to the three-dimensional structuresof bovine cAMP-dependent kinase [Brookhaven ProteinData Bank code 1CDK], murine cAMP-dependentkinase (1APM), and Caenorhabditis elegans twitchinkinase (1KOA). When the aurora1 or aurora2 sequencewas input into the tertiary structure predictionprograms THREADER and 3D-PSSM (three-dimensionalposition-sensitive scoring matrix), the top structuralmatches were 1CDK, 1APM, and 1KOA, confirming thatthese domains are structurally conserved. Thestructural models of aurora1 and aurora2 were builtusing 1CDK as the template structure. Moleculardynamics and docking simulations, targeting the ATPbinding site of aurora2 with adenylyl imidodiphosphate(AMP-PNP), staurosporine, and six small molecular S/Tkinase inhibitors, identified active-site residues thatinteract with these inhibitors differentially. The dockedstructures of the aurora2–AMP-PNP andaurora2–staurosporine complexes indicated that theadenine ring of AMP-PNP and the indolocarbazolemoiety of staurosporine have similar positions and

orientations and provided the basis for the docking ofthe other S/T kinase inhibitors. Inhibitors withisoquinoline and quinazoline moieties were recognizedby aurora2 in which H-89 and 6,7-dimethoxyquinazolinecompounds exhibited high binding energies comparedwith that of staurosporine. The calculated bindingenergies for the docked small-molecule inhibitors werequalitatively consistent with the IC50 values generatedusing an in vitro kinase assay. The aurora2 structuralmodel provides a rational basis for site-directedmutagenesis of the active site; design of novel H-89,staurosporine, and quinazoline analogues; and thescreening of the available chemical database for theidentification of other novel, small-molecular entities.

IntroductionIn cell division, the M phase is composed of mitosis andcytokinesis. Progression through the M phase is dependenton several mitotic kinases, the activities of which are regu-lated by phosphorylation-dephosphorylation and proteoly-sis. Of the mitotic kinases, cyclin-dependent kinase 1 is themost prominent cell cycle regulator that orchestrates M-phase activities. However, a number of other mitotic proteinkinases that participate in M phase have been identified; theyinclude members of the polo, aurora, and NIMA (never inmitosis A) families and kinases implicated in mitotic check-points, mitotic exit, and cytokinesis (1). Aurora kinases are afamily of oncogenic S/T kinases3 that localize to the mitoticapparatus (centrosome, poles of the bipolar spindle, or mid-body) and regulate completion of centrosome separation,bipolar spindle assembly, and chromosome segregation.Three human homologues have been identified (aurora1,aurora2, and aurora3; Ref. 2), and they all share a highlyconserved catalytic domain located in the COOH terminus,but their NH2 terminal extensions are of variable lengths withno sequence similarity (3). The human aurora kinases areexpressed in proliferating cells and are also overexpressed innumerous tumor cell lines of the breast, ovary, prostate,pancreas, and colon (2, 4–6). It has been shown that aurora2acts as an oncogene and is able to transform both Rat1fibroblasts and mouse NIH3T3 cells in vitro, and aurora2transforms NIH3T3 cells grown as tumors in nude mice (3, 4).It has been proposed (2) that overexpression of aurora2 maydrive cells into aneuploidy, which in turn can accelerate theloss of tumor suppressor genes and/or amplification of on-

Received 8/23/02; revised 12/04/02; accepted 12/20/02.The costs of publication of this article were defrayed in part by thepayment of page charges. This article must therefore be hereby markedadvertisement in accordance with 18 U.S.C. Section 1734 solely to indi-cate this fact.1 Supported by NIH Grants CA95031 and CA88310 from the NationalCancer Institute.2 To whom requests for reprints should be addressed, at Arizona CancerCenter, University of Arizona, 1515 North Campbell Avenue, Tucson, AZ85724. Phone: (520) 626-6132 or 5622; Fax: (520) 626-2255 or 5623;E-mail: [email protected].

3 The abbreviations used are: S/T kinase, serine/threonine kinase; AMP-PNP, adenylyl imidodiphosphate; NCBI, National Center for Biotechnol-ogy Information; MD, molecular dynamic(s); PK, protein kinase; rms, rootmean square; SA, simulated annealing.

283Vol. 2, 283–294, March 2003 Molecular Cancer Therapeutics

Research. on January 20, 2020. © 2003 American Association for Cancermct.aacrjournals.org Downloaded from

http://mct.aacrjournals.org/

cogenes, leading to cellular transformation. This is likely tobe achieved when those cells that overexpress aurora2 es-cape mitotic check points, which allows inappropriate acti-vation of proto-oncogenes. Recently, we have shown up-regulation of aurora2 in a number of pancreatic cancer celllines and have validated this as a target by demonstratingthat antisense oligonucleotide treatment leads to a loss ofaurora2 activity and increased apoptosis (7). Therefore, au-rora2 is a potential target that requires further validation bythe rational design of novel small-molecular inhibitors (8) andtesting in pancreatic cancer cell lines and animal models forefficacy and potency.

To validate aurora2 as a drugable target in pancreaticcancer, we have undertaken a study to design specific in-hibitors of this mitotic S/T kinase. Here we describe a struc-ture-based design approach that uses three-dimensionalmodeling of aurora1 and aurora2 kinases. Homology mod-eling of PKs is well validated and documented (9–11). Wehave used a structural bioinformatic algorithm that uses PSI-BLAST (NCBI), THREADER, 3D-PSSM (three-dimensionalposition-sensitive scoring matrix), and SAP programs to de-termine the optimal template for homology modeling of au-rora1 and aurora2. The crystal structure of the activated formof bovine cAMP-dependent PK was used as the best tem-plate for homology modeling using MD simulations inINSIGHT II. The modeled aurora2 structure was docked withknown S/T kinase and aurora2 kinase inhibitors using thebinary complex of cAMP-dependent PK-Mn2�–AMP-PNP(12, 13). The calculated binding energies from the dockinganalysis are in agreement with experimental IC50 values ob-tained from an in vitro kinase assay, which uses histone H1phosphorylation to assess inhibitor activity.

Materials and MethodsSequence and Structure Analysis. The aurora2 (ARK1;residues 98–403) and aurora1 (ARK2; residues 35–344) ki-nase domain sequences were used as probes to search anonredundant database of sequences using PSI-BLAST(NCBI). Top-ranked sequences for which the three-dimen-sional structures of S/T kinase domains are available werethe porcine heart cAMP-dependent PK (1CDK; Ref. 12), re-combinant mouse cAMP-dependent PK (1APM; Ref. 14),and Caenorhabditis elegans twitchin kinase (1KOA; Ref. 15).The aurora1 and aurora2 domain sequences were analyzedusing the programs THREADER (16) and 3D-PSSM, (17),which compare primary sequences with all of the knownthree-dimensional structures in the Brookhaven Protein DataBank. The output is composed of the optimally aligned,lowest-energy, three-dimensional structures that are similarto the aurora kinases. The top-ranked structures were bovinecAMP-dependent kinase (1CDK), murine cAMP-dependentkinase (1APM) and twitchin kinase (1KOA) confirming thePSI-BLAST search. These three tertiary structures providethe three-dimensional templates for the homology modelingof aurora1 and aurora2. The crystal structure coordinates forthe above S/T kinase domains were obtained from the Pro-tein Data Bank (18). These domains were pairwise superim-posed onto each other using the program SAP (19). Thestructural alignments from SAP were fine-tuned manually to

better match residues within the regular secondary structuralelements. The three manually aligned S/T kinase domainsequences with their respective secondary structures wereviewed in Clustal X (20).

Homology Modeling and Refinement. The modelingsoftware used was INSIGHT II (version 2000, Accelrys Inc.)running on a Silicon Graphics Indigo2 workstation (SiliconGraphics Inc.) under the UNIX operating system. The crystalstructure coordinates of the ternary complex of cAMP-de-pendent kinase (1CDK; Ref. 12) were retrieved from theBrookhaven Protein Data Bank. The catalytic domain servedas the template for aurora1 and aurora2 modeling. The in-hibitory peptide PKI (5–24) and the solvent molecules weredeleted from the coordinate set before modeling. The twoMn2� ions in the active-site pocket were retained andreplaced by Mg2� ions. The model building procedurefor aurora1 and aurora2 entailed a deletion of residues282GNLKD286 from the COOH-terminal end of cAMP-de-pendent PK and a deletion of the last 30 residues from theCOOH terminus of cAMP-dependent PK (residues high-lighted in blue in Fig. 1). Ser284 and Leu228 were insertedusing the conformation of Lys204 of twitchin kinase (1KOA).All of the other side chains except for the identical residueswere mutated to those of aurora1 using the Homology mod-eler and Biopolymer modules in Insight II.4 This model servedas a template for the homology modeling (12, 13) of aurora2kinase. The main differences between aurora1 and aurora2are two insertions between Lys124-Lys125 and Ser388-Lys389.The peptide bonds between these residues were broken,and SGTPDIL and RVL residues were inserted in aurora1consecutively without altering the geometry of the rest of thesequence. The conformation of the inserted fragments wasobtained from the fragment database of low-energy confor-mations. Furthermore, all of the amino acid residues exceptfor the identical residues were mutated to those of aurora2.The mutated side chains were manually positioned to mini-mize steric hindrance with adjacent residues. In general,insertions and deletions were found in loop regions connect-ing regular secondary structures and, hence, did not alter thehydrophobic core of the structure.

After the model building processes were complete, a se-ries of minimizations were performed to relax the structure.Before model refinement, all of the close contacts caused bythe mutation of side chains were fixed by manually rotatingthe �1 and �2 torsion angles and keeping the side chaintorsion angles of the conserved residues in their originalconformation. Hydrogen atoms were supplied by Biopolymerwith CFF parameters (21), solvated by simulating with thedistance dielectric constant, � � 4rij (22, 23) and refined bythe steepest descent and conjugate gradient minimizationuntil convergence was reached, while restraining the positionof the heavy atoms. Finally, the entire system was subjectedto energy minimization using the 3000-step steepest descentfollowed by conjugate gradient minimization until an energy

4 INSIGHT II 2000. Molecular modeling software, Accelrys Inc., SanDiego, CA.

284 Targeting Aurora2 Kinase in Oncogenesis



gradient of less than 0.001 kcal/mol/Å was achieved andfrom which, constraints were gradually removed.

Model Evaluation. The final aurora2 model was exam-ined using profile-3D (24). The profile-3D and three-dimen-sional–one-dimensional score plots of the model were pos-itive over the entire length of protein in a moving-windowscan to the template structure. Additionally, PROCHECK (25)was used to verify the correct geometry of the dihedralangles and the handedness of the model-built structure. Forthe PROCHECK statistics, an overall G factor of �0.31,hydrogen bond energy of 1.6, and only 0.29 bad contacts per100 residues were observed, which is consistent with a goodquality structure comparable with the crystal structure. Theoverall fold of the homology model (Fig. 3A) was found to bevery similar to that of crystal structure template 1CDK, which

was used for additional energy refinement, MD simulations,and docking studies of other S/T kinase inhibitors.

MD Simulations. The energy-minimized aurora2 struc-ture was used as starting model for MD simulations (26, 27),which were performed in the canonical ensemble (NVT) at300°K using the CFF force field implemented in Discover_3(version 2.9.5).5 Dynamics were equilibrated for 10 ps withtime steps of 1 fs and continued for 10-ps simulations. Thenonbonded cutoff distance of 8 Å and a distance-dependentdielectric constant (� � 4rij) for water were used to simulatethe aqueous media. All of the bonds to hydrogen were con-

5 Discover_3 (Version 2.9.5). Molecular mechanics force fields, INSIGHTII, 2000. Molecular modeling software, Accelrys Inc., San Diego, CA.

Fig. 1. Structure-based sequence alignment inClustal X of the catalytic PK domains of aurora2(ARK1), aurora1 (ARK2), bovine cAMP-dependentPK (1CDK), murine cAMP-dependent PK (1APM),and C. elegans twitchin kinase (1KOA). The �-helices are �1 to �11 (black bars), �-strands are�1 to �11 (gray bars), identical residues (shadedand �), highly conserved residues (:), and similarresidues (.). Magenta, the active site residues;green, inserted residues; red, deleted residues;blue, N- extensions of aurora2 and -1 not includedin the modeling and the COOH-terminal residuesof 1CDK and 1APM not included in the modeling.Threonine197 (1CDK) is equivalent to Thr288 (auro-ra2 and 1, yellow) and is the autophosphorylationsite.

285Molecular Cancer Therapeutics



strained. Dynamic trajectories were recorded every 0.5 ps foranalysis. Individual simulations from the point of a stabletrajectory generated time-averaged structures. The resultinglow-energy structure was extracted, and energy was mini-mized to 0.001 kcal/mol/Å. To examine the conformationalchanges that occur during MD, the rms deviations werecalculated from trajectories at 0.5-ps intervals and comparedwith the C� backbone of cAMP-dependent PK (see Fig. 3B).The rms deviation for the two superimposed structures was0.42 Å. Furthermore, the rms deviations were calculated forthe protein backbone (0.37 Å) and the active-site pocket(0.41 Å) and were compared with crystal structure before thedocking experiments. The resulting aurora2 structure servedas the starting model for docking studies.

SA Docking. To find sterically reasonable binding geom-etries and to explore the interactions of the proposed ligandrecognition pocket, affinity docking with SA was performedin INSIGHT II (28, 29). The structures of ligands used fordocking were from five crystal structure complexes of cAMP-dependent PK bound to AMP-PNP, (12), staurosporine (30,31), H-89, H-7, H-8, (32, 33), and structures that were em-pirically built and energy minimized [KN-93 (34), ML-7 (35),and 6,7-dimethoxyquinazoline (8, 36); Fig. 2] in INSIGHT II.Partial charges were assigned to the ligands by the Gasteigermethod defined within INSIGHT II.5 Systematic conforma-tional searches were performed on each of the minimizedligands using 10-ps MD simulations at 300°K. For dockingwith AMP-PNP, the position of the ATP analogue was re-tained from its crystal structure with 1CDK, in which theadenine base served as a template for field-fit alignmentswith the indolocarbazole, fused biphenyl, and quinazolinemoieties. The ATP analogue was then removed from thefield-fit alignment, and each of the other ligands was dockedinto the active-site pocket with a position and orientation thatwere similar to those of AMP-PNP. The heavy atoms from theATP analogue were used as sphere centers in the input to thedocking procedure. To clarify the orientation of these inhib-itors in the active-site pocket, the electrostatic potential atthe van der Waals surface was determined using solventsurface calculations. Orientations with the lowest intermo-lecular potential energy were calculated. To explore the in-teraction of the ligands in the recognition pockets, SA dock-ing of complex formations was carried out withoutconstraints, to allow each of the protein-ligand complex sys-tems to evolve freely. To explore the effects of solvationimplicitly, a distance-dependent dielectric constant (� � 4rij)was used (22, 23). The nonbonded cell multipole method wasused for SA docking with input energy parameters. Dockingsimulations were performed at 500°K with 100 fs/stage (totalof 50 stages), quenching the system to a final temperature of300°K. The whole complex structure was energy minimizedusing 1000 steps. This provided 10 structures from SA dock-ing, and their generated conformers were clustered accord-ing to rms deviation. The lowest global structure obtainedwas used for computing intermolecular binding energies (37).Furthermore, we have validated the robustness of the affin-ity-docking methodology (27, 38) by comparative docking ofan ATP analogue, AMP-PNP, with 1CDK and aurora2 kinase.

The binding mode of these complexes are shown in Fig. 4, Aand B, respectively.

In Vitro Aurora Kinase Assay. Aurora2 immunoprecipi-tation kinase assays were performed as described previously(6). Briefly, cells were first homogenized in lysis buffer [150mM NaCl, 50 mM HEPES (pH 7.2), 1 mM EDTA, 1 mM EGTA,1 mM DTT, 0.1% Tween 20, 0.1 mM phenylmethylsulfo-nylfluoride, 2.5 �g/ml leupeptin, 0.1 mM sodium orthovana-date (Sigma Chemicals, St. Louis, MO)] at 4°C. Lysates werethen centrifuged at 10,000 � g for 10 min. Protein contentwas determined by the BCA protein assay (Pierce, Rockford,IL), and 200 �g was used for each sample. The supernatantswere immunoprecipitated for 12 h at 4°C with protein A-agarose beads precoated with saturating amounts of aurorakinase-2 polyclonal antibody. Immunoprecipitated proteinson beads were washed twice with 1 ml of lysis buffer andtwice with kinase buffer [50 mM HEPES (pH 7.0), 10 mM

MgCl2, 5 mM MnCl2, and 1 mM DTT]. The beads were thenresuspended in 40 �l of kinase buffer containing 10 �M ATP,5 �Ci of [�-32P]ATP (6000 Ci/mmol; 1 Ci � 37 GBq; Amer-sham Corp., Arlington Heights, IL), and 4 �g of the histoneH1 protein substrate. The samples were then incubated for30 min at 30°C with occasional mixing. For dose-response

Fig. 2. The structures of the ATP analogue and S/T kinase inhibitorsevaluated for inhibition of aurora2 kinase activity. These are AMP-PNP,staurosporine, H-89, H-8, H-7, KN-93, ML-7, and 6,7-dimethoxyquinazo-line.




studies to determine IC50, serial dilutions of the inhibitorswere used (10–200 �M). The samples were then boiled inpolyacrylamide gel sample buffer containing SDS and sep-arated by electrophoresis. Phosphorylated proteins werequantified after exposure to autoradiographic film (Labscien-tific, Inc., Livingston, NJ) by densitometry using ImageQuantversion 5.1 (Molecular Dynamics Computing Densitometer,Sunnyvale, CA). The IC50 values (Table 1) were calculatedfrom the phosphorylation profile of the substrate histone H1protein.

ResultsAurora Kinases Possess a Conserved yet DistinctS/T Kinase Catalytic DomainThe sequence identity and similarity between aurora1 andaurora2 kinases and cAMP-dependent PK are 30 and 60%,respectively, and that between aurora1 and aurora2 kinasesand twitchin kinase is 17 and 54%, respectively. The aurora1and aurora2 domain sequences were aligned with respect tothe structural alignments obtained in Clustal X. Fig. 1 showsthe secondary structural elements of the aligned kinase do-mains. The critical catalytic residues involved in the transferof the �-phosphoryl group of ATP to the substrates are highlyor absolutely conserved between the aurora kinases, cAMP-dependent PK, and twitchin kinase, and these residues arehighlighted in Fig. 1. In the prototype cAMP-dependent PKcatalytic domain, they are 50GXGXXGX57, 72K, 91E, 121E,166D, 168K, 170E, 171N, 184DFG, 197T (autophosphorylationsite), 206APE, 220D, and 280R (12). The main differences be-tween cAMP-dependent PK and the aurora kinases are: (a)the glycine-rich nucleotide binding motif 50GTGSFGRV57 ischanged to 140GKGKFGNV147; (b) 206APE is changed to297PPE299; (c) 284S and 228L in aurora kinases are inserted; (d)residues GNLKD are deleted from the COOH-terminal end ofcAMP-dependent PK; and (e) the last 30 residues are deleted

from the COOH terminus of cAMP-dependent PK. Theseresults indicate that the aurora kinases possess a conservedS/T kinase catalytic domain similar to that of the cAMP-dependent PKs. However, the changed glycine-rich nucle-otide-binding sequence motif, V123A213, E127T217, insertionof Ser284 in aurora2 and Leu228 in aurora1 (between Ser283

and Arg285) and A206P297 in the active site provide importantstructural differences between these two kinases that mayexplain their distinct substrate specificities.

Each S/T kinase structure is composed of a small NH2-terminal �-sheet domain and a large COOH-terminal �-hel-ical domain, with the catalytic cleft between the two domainsat which the substrate and ATP bind in the presence of Mg2�

ions (12). The crystal structure determined for bovine cAMP-dependent PK is phosphorylated in the activation loop and isin complex with AMP-PNP (Fig. 2) and the pseudo-substratepeptide PKI (5–24). This catalytically competent enzyme is ina state immediately before phosphoryl transfer, with the cleftbetween the NH2- and COOH-terminal domains in a closedconformation. The tertiary structure modeling of aurora1 andaurora2 on the catalytically competent cAMP-dependent PKstructure has defined pertinent structural features in the au-rora kinases and has guided our structure-based rationaldesign process. Fig. 3A shows the model-built structure ofaurora2. Secondary structural elements are indicated as redcylinders (�-helix), yellow arrows (�-sheet), green ribbons(coil), and blue ribbons (turns). The C� backbone superim-position of the final minimized structure of aurora2 (blue) oncAMP-dependent PK (1CDK; red) is shown with a rms devi-ation of 0.42 Å (Fig. 3B). Because of the two COOH-terminaldeletions in the aurora kinases compared with that of cAMP-dependent PK, the aurora kinases have evolved a morecompact catalytic kinase core domain that may affect theirsubstrate specificity and activity.

Fig. 3. A, the homology model ofaurora2 kinase. Secondary struc-tural elements are labeled as�-helix (red cylinder), �-sheet (yel-low arrow), coil (green ribbon), andturns (blue ribbons). B, the C�backbone superimposition of thefinal minimized structure of auro-ra2 (blue) on cAMP-dependent PK(1CDK; red) with an rms deviationof 0.23 Å.




Docking of S/T Kinase Inhibitors Coupled to an inVitro Kinase Assay Defines Distinct ChemicalMoieties as Building Blocks for Potent Inhibitorsof Aurora2

Aurora2–AMP-PNP Complex. Affinity docking experi-ments indicate that AMP-PNP in complex with aurora2 ex-hibits similar conformational orientations to that of the crystalstructure of cAMP-dependent PK bound to AMP-PNP (12).This comparative binding mode provides validation of affinitydocking and a rationale for docking simulations of other S/Tkinase inhibitors. Because of the change in sequence motifof the glycine-rich loop and residues interacting with the ATPanalogue, AMP-PNP binds with an affinity in the ATP-bindingpocket of aurora2 that is lower than that with 1CDK (Table 1).The crystal structure of the 1CDK–AMP-PNP complex (12)revealed that the high binding affinity of the ATP analogue inthe nanomolar range is caused by a number of interactions,including polar, �,�-phosphoryl coordination to Mn2� ions,and hydrogen bonds with specific enzyme residues. Fromour docking and MD simulations, we observed a network ofhydrogen bonding interactions with aurora2, which, in turn,

stabilized the docked structure. The adenine base, its 6-a-mino group, and ring N1 atoms interact with Glu211 andAla213, which have strong hydrogen bonding interactions:Glu211-CAO. . . H2N of adenine (1.97 Å, energy �2.34 kcal/mol) and Ala213-NH. . . N of ring N1 (2.26 Å, energy �2.65kcal/mol), respectively. Weaker hydrogen bonding interac-tions for AMP-PNP with Glu121 (2.90 Å) and Val123 (3.20 Å)were observed in the 1CDK crystal structure complex. Incomparative docking experiments, the docking of AMP-PNPwith 1CDK revealed that the conformation and binding modeof AMP-PNP shows high similarity to that of the crystalstructure (Fig. 4B). Another important finding from dockingsimulations suggests that the �-phosphoryl group seems toplay a significant role in hydrogen bonding interactions withthe �NH atoms of Phe144 and Gly145 in aurora2. The 1CDKcrystal structure complex shows that the 2� hydroxyl groupof the ribose moiety has potential hydrogen bonding inter-actions with Glu127 and Glu170. However, in the aurora2–AMP-PNP complex, the 2�OH group of the ribose moiety hasa H-bonding interaction with Thr217 (Thr217-O. . . H of 2�OH-

Fig. 4. A, binding mode of AMP-PNP inthe ATP binding site of aurora2 kinase. Thepurine base occupies the adenine bindingsite and the �- and �-phosphoryl groupsare buried in the Gly/Lys pocket. Blackdashed lines, the hydrogen bonds; dis-tances are in Å. B, binding mode of AMP-PNP in the ATP binding site of 1CDK. Thepotential hydrogen bonding interactionsand the active-site residues involved inbinding are shown.




group, 2.40 Å). The conformation of AMP-PNP within theactive site of aurora2 is maintained by seven hydrogen bondsformed arising from the ATP-binding site and the Gly/Lyspocket (Fig. 4). These hydrogen bonding interactions werecomputed from the final MD trajectories of the docked com-plex structure. The absence of the Glu260 hydrogen bondinginteraction with the ribose moiety of aurora2 is attributable tothe orientation of the �,�-phosphoryl groups in the Gly/Lyspocket that prevent a strong electrostatic interaction withLys162. The �- and �-phosphoryl groups are also involved inclose contacts with Lys143, Asn261, and Asp274 of the activesite. The absence of critical H-bonding interactions withLys162 and Glu260 presumably leads to an affinity binding ofAMP-PNP (�69.3 kcal/mol) with the active-site residues ofaurora2 that is lower than that with 1CDK (Table 1). Thesesequence and structural differences between cAMP-de-pendent PK and aurora2 kinase provide guidelines to betterdefine the active-site pocket and to design novel and morespecific aurora2 kinase inhibitors.

Aurora2–Staurosporine Complex. In the case of affinitySA docking simulations on staurosporine (Fig. 2), the position

and orientation within the active-site pocket of aurora2 wereidentical to those of the cAMP-dependent PK–staurosporinecomplex (1STC; Ref. 30). In particular, the hydrogen bondinginteractions computed from affinity were identical to those inthe crystal structure of the cAMP-dependent PK–staurospo-rine complex (Fig. 5). The amino acid residues that recognizeand interact with staurosporine are Gly140, Val147, Glu181,Glu211, Ala213, Thr217, Asn261, and Asp274. A strong intermo-lecular hydrogen bond to Ala213 �NH with the lactam car-bonyl group (2.03 Å) of staurosporine was observed in all ofthe structures obtained from MD simulations. However,Glu211-CAO. . . HN of the lactam amide hydrogen bondinginteraction, present in the cAMP-dependent PK–staurospo-rine complex, was not observed in the initial docked aurora2–staurosporine complex. This is most likely attributable to theindolocarbazole and pyranosyl moieties of staurosporine,which induced conformational changes in the neighboringactive site residues of aurora2. The staurosporine-inducedconformational changes were investigated by calculating therms between the initial and final complex structures obtainedfrom MD simulations. The 0.23-Å difference observed in the

Fig. 5. The binding mode of au-rora2–staurosporine (color-by-atom type) complex. The activesite residues of the protein in-volved in interactions are shown;black dashed lines, the hydrogenbonds.

Table 1 The measured (IC50) and calculated energies (kcal/mol) of the ATP analogue and inhibitors tested for aurora2 kinase binding

Ligand Total energyBindingenergy

(kcal/mol)IC50

a (�M) No. ofH-bonds

AMP-PNP �183.7 �69.3 NAb 7Staurosporine �236.2 �91.6 0.3 � 0.2 2

H-89 �219.2 �103.5 103.5 � 3.5 2H-8 �183.2 �69.4 ND 1H-7 �138.6 �41.3 ND 1

KN-93 �151.8 �66.2 �1000 0ML-7 �139.4 �54.1 428.5 � 96.5 0

6,7-dimethoxyquinazoline �216.8 �104.9 0.00785 3

a IC50 values measured in triplicate.b NA, not available; ND, not determined.




active-site pocket of aurora2 on staurosporine binding isshown in Table 2. An analysis of the low-energy trajectoryfrom the final run of simulations reveals that the amide protonforms an H-bond interaction with Glu211 (2.72 Å). It is inter-esting to note that the indolocarbazole moiety exhibits amore stable orientation in all of the structures generatedduring MD simulations. The phenyl ring syn to the methoxygroup is positioned in the ATP-binding pocket and is in closecontact with Glu181 and Asp274 residues and are within 2.50–3.00 Å, whereas the phenyl ring anti to methoxy group isoriented into the Gly–Lys pocket. The pyranosyl methylaminogroup is positioned within H-bonding distance to Thr217 (1.97Å) and Glu260 (3.32 Å). On the basis of these observationsand experimental data (Table 1), staurosporine (a nanomolarPK inhibitor) binds to aurora2 with a high binding energy of�91.6 kcal/mol (see Table 1).

Aurora2–H-series Complexes. We have analyzed themode of binding of three isoquinoline H-series PK inhibitors,H-89, H-8, and H-7 (Fig. 2), with aurora2 kinase (32, 33). Theconformations of these isoquinolines were obtained from thecrystal structure complexes of cAMP-dependent PKs andwere subjected to affinity docking experiments without al-tering their crystallographic geometries. AMP-PNP served asa template for SA docking, and an analysis of dynamic tra-jectories revealed that the isoquinoline moiety occupied theATP binding site and is positioned similarly to that of theadenine base of AMP-PNP. The N-alkyl chain of H-8 andH-89 extended into the �,�-phosphoryl site. In all of theH-series isoquinolines, we observed that the nitrogen ringinteracts with Ala213, which is located close to Glu211 at thehinge region and is in the purine binding site of AMP-PNP.Although the H-89 isoquinoline structure is different from thatof the AMP-PNP, it has similar structural elements, includingan isoquinoline nitrogen atom at position 13, two �NHgroups at positions 4 and 17, and a sulfonyl group at position1. These play an important role in the recognition of theactive-site pocket of aurora2 and represent another class ofS/T kinase inhibitors with an IC50 of 107 �M (Table 1). Anumber of SA docking simulations with different initial con-figurations identified a binding mode with a binding energy(�103.5 kcal/mol) substantially higher than that observed forAMP-PNP, staurosporine, and other isoquinoline structures.In this model, the N13 atom is positioned such that it H-bonds with the Ala213 �NH group (�N. . . HN-Ala213, 2.80 Å),which mimics the 1CDK Val123 amide H-bonding interaction(Fig. 6). Unlike the �,�-phosphoryl group, the bromophenylmoiety is located deep in the Gly/Lys pocket and exhibitsclose hydrophobic interactions (3–3.50 Å) with Phe144 andGly145. The carbonyl group of Asn261 in aurora2 forms astrong hydrogen bonding interaction with the N17 amideproton, and Glu260 is also in close contact with the sameamide proton. The isoquinoline ring and sulfonamide moietymake a number of close contacts with Gly140, Gly145, andVal147. Moreover, the styrene moiety is also in close contactwith Lys141 and Gly142. These interactions determine a stablebinding mode observed in several dynamic trajectories an-alyzed within �10 kcal/mol of that of the low-energy globalstructure generated during MD simulations.

For the H-8 and H-7 compounds (data not shown), thedocking is based on the aurora2–H-89 complex. The N13isoquinoline ring atom involved in H-bonding interactionswith the Ala213 amide group is within a distance of 1.95–2.45 Å. The replacement of the bromophenyl styrene moi-ety of H-89 with the N-alkyl side chain and piprazine groupin the case of H-8 and H-7 leads to less potent inhibitorsof aurora2. Although the isoquinoline and sulfonyl moietiesexhibit structural similarity, the H-series of compoundslacks a number of van der Waals and hydrophobic con-tacts with the enzyme. Such interactions play an importantrole for high binding affinity, as observed in the crystalstructures of cAMP-dependent PK in complex with inhib-itors. As shown in Fig. 6, the simulations demonstratethose structural changes that alter the shape of the sulfo-nyl substituents, leading to significant changes in bindingenergies (Table 1). Thus, the �NH of the N-methyl groupof H-8, oriented to the Asp274 residue and its methylfunction, are positioned within 3.5– 4 Å. The sulfonyl sub-stituent had only one contact with Vall47. Whereas thepiprazine moiety is positioned within the ribose bindingsite of AMP-PNP, its �NH group is 3.5 Å away from Lys141.These results suggest that although H-8 and H-7 com-pounds bind to aurora2, the higher affinity of H-89 can beexplained by its strong hydrogen bonding and hydropho-bic interactions with both the ATP-binding site and theresidues within the Gly/Lys pocket.

Aurora2–6,7-Dimethoxyquinazoline Complex. Althoughquinazoline-containing compounds have been reported tobe very potent and selective EGFR receptor tyrosine ki-nase inhibitors (39, 40), recent data suggest that thesecompounds also bind to S/T kinases (8). The 6,7-dime-thoxyquinazoline which binds to aurora2 with an IC50 valueof 0.00785 �M (8), represents another class of high-affinityaurora2 inhibitors. The aurora2–H-89 complex providedthe basis for docking the quinazoline ring and, subse-quently, the H-89 molecule was removed. The entire au-rora2– 6,7-dimethoxyquinazoline complex was optimizedby several cycles of SA docking simulations and minimi-zation using the CFF force field. From SA docking simu-lations, it was observed that the quinazoline moiety waspositioned in an orientation that was similar to that of theadenine base of AMP-PNP and was anchored deep in theATP binding site by forming three hydrogen bonds (Fig. 7).The N1 atom of quinazoline and N3 atom of the pyrimidinerings are involved in potential hydrogen bonding interac-tions with Ala213 (Ala213-NH. . . N1, 2.78 Å, energy �1.61kcal/mol) and Asp274 (Asp274-NH. . . N3, 2.05 Å, energy

Table 2 The calculated rms deviation (Å) for the superimposed aurora2and aurora2-staurosporine complex

The rms deviations were calculated for the active-site residues and theC� backbone.

Aurora2 Aurora2-staurosporinecomplex

Active-site amino acids 0.4126 0.6432Backbone 0.3692 0.4256C�-trace 0.3913 0.4693




�1.29 kcal/mol). Additionally, the �CAO of Asp274

is involved in hydrogen bonding interactions with 2-(N-benzoylamino) �NH group (Fig. 7). The oxygen atom of the7-methoxy substituent is also positioned within hydrogenbonding distance with the backbone amide of Ala213.Glu211 has strong hydrogen bonding interactions withAMP-PNP and staurosporine, but no such an interactionwas observed with the quinazoline compound. This isattributable to the lack of a donating functional group andstrong steric interaction of methoxy group with Glu211. The2-(N-benzoylamino) group had a similar position and ori-entation to that of the �-phosphoryl group of AMP-PNP orthe bromophenyl group of H-89 and interacts with Phe144

and Gly145, which in turn interacts with Lys143 located 3.50Å from the carbonyl group. In the final model, we observedthat the 5-aminopyrimidine moiety and the correspondingH-89 sulfonyl moiety positioned similarly with in the activesite. As shown by the AMP-PNP model, the similar mannerin which H-89 and quinazoline (Fig. 8) bind indicates thatthe polar, nonpolar, and hydrogen bonding interactions

are essential for the high activity observed for these com-pounds. Such a large number of polar, nonpolar, andhydrogen bonding contacts with the active-site residuesare observed for 6,7-dimethoxyquinazoline, which exhibitsa high binding energy (�104.9 kcal/mol) compared withthat of H-89 (Table 1). These results suggest that the rigidsubstituents of the �- and �-phosphoryl positions are welltolerated.

Aurora2 in Complex with KN-93 and ML-7. The bindingmodes of KN-93 and ML-7 (data not shown) were alsoexplored using the above models. ML-7 was positionedand oriented similarly to H-7. The 4-methoxyphenyl sulfo-nyl amide moiety of KN-93 was found to occupy the ATPbinding site of aurora2. The interaction of the chlorophe-nylstyrene group with Asp274 and Asn261 and its bindingmode is dissimilar to that observed for the inhibitor classof H-89 and H-8 bound to aurora2. Interestingly, KN-93did not exhibit any hydrogen bonding interactions. Spe-cifically, the presence of the methyl (amino) methyl andhydroxyethyl groups had significant steric clashes with

Fig. 6. The stereo-view of the binding modeof the aurora2–H89 complex. The isoquinolinering is positioned deep in the ATP binding siteand interacts with A213 and N261. The bro-mophenyl moiety extends into the Gly/Lyspocket. Dotted lines, the two hydrogen bonds;labels, the active site residues.

Fig. 7. The binding mode of theaurora2–6,7-dimethoxyquinazo-line complex. The active sitebackbone of the protein is shownas a blue ribbon, the quinazolinering (color-by-atom) anchoreddeep in the purine binding siteand is involved in hydrogenbonding interactions with E211and D274 (black dashed lines;see “Results” for explanation).The active site residues involvedin the interactions are labeledand clipped for a better view ofthe binding site.




Gly140, Val147, and Asp274, presumably because of its rigidconformation.

DiscussionThe development of potent and selective nucleotide ana-logue-based inhibitors that interact with individual PKs attheir ATP binding site are being evaluated in various stagesof clinical trials (41). Given the proof of concept that theinhibition of dysregulated PKs in human cancers is poten-tially achievable and could provide clinical efficacy, we havetargeted human aurora2 kinase in pancreatic cancer. Aniterative structure-based small-molecule drug design ap-proach in combination with an in vitro kinase assay has beendeveloped to test the potency and efficacy of novel aurora2kinase inhibitors.

The tertiary structure models of aurora1 and aurora2 werebuilt using the catalytic domain of PKA complexed with a20-amino acid substrate analogue inhibitor and Mn2�-ATP(12). The cAMP-dependent PK structure revealed a two-domain PK with a deep cleft between the domains. Mn2�-ATP and a portion of the inhibitor peptide occupy the activesite cleft. The NH2-terminal smaller domain is associatedwith nucleotide binding and its largely antiparallel �-sheetstructure constitutes a nucleotide binding motif. The COOH-terminal larger domain is predominantly �-helical with a sin-gle �-sheet at the domain interface. This domain is primarilyinvolved in peptide binding and catalysis. Most of the invari-ant amino acids in this conserved catalytic core are clusteredat the sites of nucleotide binding and catalysis. The tertiarystructure of the aurora2 model built on 1CDK has significantidentity and similarity indicating a conserved catalytic core.The main differences in the active site of aurora2 and cAMP-dependent PK are the changed glycine-rich nucleotide-bind-ing motif 50GKGKFGNV57, which should be sufficient to pro-vide distinct substrate specificities and deletion of the last 30residues from the COOH terminus of cAMP-dependent PK(Fig. 1) and to provide for a more compact aurora2 catalytic

kinase core domain that may be relevant to its functionallocalization to mitotic spindles.

A number of crystal structures of cAMP-dependent PK incomplex with ATP analogues and pseudo-substrates (42, 43)have provided details as to the mechanisms of inactivationby modulation of one or more of the four conserved struc-tural elements; these are: (a) inhibition of the ATP bindingpocket; (b) distortion of the glycine-rich loop; (c) alteration ofthe position of �-helix C; and (d) alteration of the conforma-tion of the phosphorylation-dependent activation segment.The crystal structure determined for bovine cAMP-depend-ent PK is phosphorylated in the activation loop and is incomplex with AMP-PNP and the pseudo-substrate peptidePKI (5–24). This catalytically competent enzyme is in a stateimmediately before phosphoryl transfer, with the cleft be-tween the NH2- and COOH-terminal domains in a closedconformation. The tertiary structure modeling of aurora1 andaurora2 on the catalytically competent cAMP-dependent PKstructure has defined pertinent structural features in the au-rora kinases that has guided the rational design of novelchemical entities.

To gain insight into the binding mode of nucleoside ana-logue-based inhibitors, we have docked known S/T kinaseinhibitors. The potency of these docked inhibitors has beenevaluated in an in vitro kinase assay. The complex structureof AMP-PNP with cAMP-dependent PK and aurora2 from SAdocking simulations reveals a network of hydrogen bondinginteractions. The sequence differences in the active-sitepocket between the two kinases explains the lower bindingenergy of the AMP-PNP–aurora2 complex. This is attributa-ble to the weak hydrogen bonding interactions of the adeninebase with Glu211, Ala213, and Glu127Thr217 mutation in auro-ra2, the absence of a hydrogen bonding interaction of the2�OH group of ribose with Lys162 and Glu260 and the orien-tation of the �,�-phosphoryl groups in the Gly/Lys pocket(Fig. 4). However, in the case of staurosporine, the positionand orientation in the active site was identical to that of the

Fig. 8. The superposed struc-tures of staurosporine (ash), 6,7-dimethoxyquinazoline (blue),H-89 (red), and AMP-PNP (ma-genta) in the ATP-binding pocketof aurora2. The enzyme activesite is clipped and residuesshown in color-by-atom type.




cAMP-dependent PK–staurosporine complex (1STC; Ref.30). In particular, the hydrogen bonding interactions com-puted from affinity were identical with the mostly conservedactive-site residues (Fig. 5) and accommodated staurospo-rine by minor conformational changes during MD simulations(Table 2).

In the case of the H-series of compounds (H-89, H-8, andH-7) the isoquinoline moiety occupied the ATP binding site andis positioned similarly to that of the adenine base of AMP-PNP.The N-alkyl chain of H-89 extended into the �,�-phosphoryl siteand the isoquinoline nitrogen ring interacts with Ala213. Al-though the H-89 isoquinoline structure is different from that ofthe AMP-PNP, it has similar structural elements including anisoquinoline nitrogen atom at position 13 which is involved in ahydrogen bonding interaction with Ala213 (Val123 in 1CDK). Thebromophenyl moiety is located deep in the Gly/Lys pocket andexhibits a hydrophobic interaction with Phe144 (Fig. 6). Further-more, the strong hydrogen bonding and hydrophobic interac-tions with both the ATP and Gly/Lys pocket supports the cal-culated binding energy (Table 1). The aurora2–H-89 complexwas used to model 6,7-dimethoxyquinazoline, and the quinaz-oline moiety is positioned similar to that of the adenine base ofAMP-PNP and anchored deep in the ATP binding site (Fig. 7).The 2-(N-benzoylamino) group had similar positions and orien-tations to those of the �-phosphoryl group of AMP-PNP andinteracts with Gly145, which in turn is in close proximity toPhe144. The N1 atom of quinazoline and N3 atom of pyrimidinering exhibits a strong hydrogen bonding interaction with Ala123

and Asp274. In quinazoline, the 2-substituted amide hydrogen isin close contact with Gly145 and its 5-substituted amide hydro-gen has strong interactions with Gly140 and Val147. Further-more, the aminopyrimidine and benzoly carbonyl groups arepositioned close to Asp274. These interactions explain its highbinding energy (–104.9 kcal/mol) when compared with that ofH-89. Compounds KN-93 and ML-7 (data not shown) essen-tially had poor interactions with aurora2, mainly because of theirrigid conformations and lack of hydrogen bonding interactions.The 4-chlorophenyl-2-propenyl-(methylamino)methylphenylsubstituent of KN-93 at the 2-position is close to the hydrophilicpocket and appears to be sterically crowded. The 4-positionprojects toward the Gly/Lys pocket in the region occupied bythe bromostyrene moiety of H-89. Taken together, these resultssuggest that SA docking complexes of aurora2 with H-89 and6,7-dimethoxyquinazolines furnish a number of guidelines foranalogue design.

For the docked structures, the calculated binding energiesfrom the highest to the lowest are 6,7-dimethoxyquinazoline �H-89 � staurosporine � H-8 � AMP-PNP � KN-93 � ML-7 �H-7. For the in vitro kinase assay the IC50 values from thehighest to the lowest are 6,7-dimethoxyquinazoline � stauro-sporine � H-89 � ML-7 � KN-93 (Table 1). These resultsqualitatively validate the modeling and docking studies de-scribed for this series of S/T kinase inhibitors. Molecular mod-eling was used to evaluate and select potential substituents forthe isoquinoline and quinazoline templates. A de novo smallfragment and available chemical database library using LUDI4

(44) and FlexX (45) virtual docking computational procedureswere performed using SYBYL (Tripos, St. Louis, MO) to gener-ate novel compounds specific to aurora2. The high-score com-

pounds from FlexX having a quinazoline moiety were selected,and are currently being synthesized and screened, for aurora2kinase inhibitory activity.

Recently, the crystal structure of the human aurora 2 kinase(46) was published (Brookhaven Protein Data Bank entry 1MUOon hold). The structure determination was based on molecularreplacement using an aurora2 kinase homology model, whichwas based on 1CDK. This essentially validates the model fordocking studies undertaken in this report. The main conclu-sions from the crystal structure were that the homology modelis consistent with an overall kinase fold and active site exceptfor a lack of electron density of the activation loop, whichincludes a conserved tryptophan. Although the coordinates ofthe crystal structure of aurora2 are not available, the bindingmode of the docked AMP-PNP in the ATP binding site aresimilar in 1CDK and the modeled aurora2.

In conclusion, we have developed a structural bioinfor-matic approach that has facilitated the tertiary structuremodeling of aurora2, studied the binding mode of ATP ana-logue and inhibitors, and evaluated these compounds usingan in vitro kinase assay. A structural analysis of the dockedATP analogue and S/T kinase inhibitors has provided guide-lines for the rational design of novel aurora2 kinase inhibitors.Furthermore, using the available chemical database andFlexX, we have identified a series of novel compounds thatare currently been synthesized for in vitro and in vivo testing.

AcknowledgmentsWe thank Drs. Laurence H. Hurley and Emmanuelle J. Meuillet-May forhelpful discussions and Dr. David M. Bishop for preparing, proofreading,and editing the manuscript and figures.

References1. Nigg, E. A. Mitotic kinases as regulators of cell division and its check-points. Nat. Rev. Mol. Cell. Biol., 2: 21–32, 2001.

2. Bischoff, J. R., and Plowman, G. D. The Aurora/Ipl1p kinase family:regulators of chromosome segregation and cytokinesis. Trends Cell Biol.,9: 454–459, 1999.

3. Bischoff, J. R., Anderson, L., Zhu, Y., Mossie, K., Ng, L., Souza, B.,Schryver, B., Flanagan, P., Clairvoyant, F., Ginther, C., Chan, C. S. M.,Novotny, M., Slamon, D. J., and Plowman, G. D. A homologue of Dro-sophila aurora kinase is oncogenic and amplified in human colorectalcancers. EMBO J., 17: 3052–3065, 1998.

4. Giet, R., and Prignet, C. Aurora/Ipl1p-related kinases, a new oncogenicfamily of mitotic serine–threonine kinases. J. Cell Sci., 112: 3591–3601, 1999.

5. Tatsuka, M., Katayama, H., Ota, T., Tanaka, T., Odashima, S., Suzuki,F., and Terada, Y. Multinuclearity and increased ploidy caused by over-expression of the aurora- and Ipl1-like midbody-associated protein mi-totic kinase in human cancer cells. Cancer Res., 58: 4811–4816, 1998.

6. Zhou, H., Kuang, J., Zhong, L., Kuo, W. L., Gray, J. W., Sahin, A.,Brinkley, B. R., and Sen, S. Tumour amplified kinase STK15/BTAK in-duces centrosome amplification, aneuploidy and transformation. Nat.Genet., 20: 189–193, 1998.

7. Han, H., Bearss, D. J., Browne, L. W., Calaluce, R., Nagle, R. B., and VonHoff, D. D. Identification of differentially expressed genes in pancreatic can-cer cells using a cDNA microarray. Cancer Res., 62: 2890–2896, 2002.

8. Austen, M., John, K., Henri, F., and George, B. Preparation of quina-zolines as aurora 2 kinase inhibitors. PCT Int. Appl, WO0121596 UnitedKingdom: 2001.

9. Sun, L., Tran, N., Liang, C., Tang, F., Rice, A., Schreck, R., Waltz, K.,Shawver, L. K., McMahon, G., and Tang, C. Design, synthesis, and eval-uations of substituted 3-[(3- or 4-carboxyethylpyrrol-2-yl)methylidenyl]in-




dolin-2-ones as inhibitors of VEGF, FGF, and PDGF receptor tyrosinekinases. J. Med. Chem., 42: 5120–5130, 1999.

10. Krystal, G. W., Honsawek, S., Kiewlich, D., Liang, C., SVasile, S., Sun,L., McMahon, G., and Lipson, K. E. Indolinone tyrosine kinase inhibitorsblock kit activation and growth of small cell lung cancer cells. CancerRes., 61: 3660–3668, 2001.

11. Knighton, D. R., Cadena, D. L., Zheng, J., Eyck, L. F. T., Taylor, S. S.,Sowadski, J. M., and Gill, G. N. Structural features that specify tyrosinekinase activity deduced from homology modeling of the epidermal growthfactor receptor. Proc. Natl. Sci. USA, USA., 90: 5001–5005, 1993.

12. Bossemeyer, D., Engh, R. A., Kinzel, V., Ponstingl, H., and Huber, R.Phosphotransferase and substrate binding mechanism of the cAMP-de-pendent protein kinase catalytic subunit from porcine heart as deducedfrom the 2.0 Å structure of the complex with Mn2� adenylyl imidodiphos-phate and inhibitor peptide PKI(5–24). EMBO J., 12: 849–859, 1993.

13. Zheng, J., Trafny, E., Knighton, D., Xuong, N-H., Taylor, S., Ten Eyck,L., and Sowadski, J. 2.2 Å refined crystal structure of the catalytic subunitof cAMP-dependent protein kinase complexed with MnATP and a peptideinhibitor. Acta Crystallogr. D Biol. Crystallogr., 49: 362–365, 1993.

14. Knighton, D. R., Zheng, J. H., Ten Eyck, L. F., Ashford, V. A., Xuong,N. H., Taylor, S. S., and Sowadski, J. M. Crystal structure of the catalyticsubunit of cyclic adenosine monophosphate-dependent protein kinase.Science (Wash. DC), 253: 407–414, 1991.

15. Kobe, B., Heierhorst, J., Feil, S. C., Parker, M. W., Benian, G. M.,Weiss, K. R., and Kemp, B. E. Giant protein kinases: domain interactionsand structural basis of autoregulation. EMBO J., 15: 6810–6821, 1996.

16. Jones, D. T. GenTHREADER: an efficient and reliable protein fold rec-ognition method for genomic sequences. J. Mol. Biol., 287: 797–815, 1999.

17. Kelley, L. A., MacCallum, R. M., and Sternberg, M. J. E. Enhancedgenome annotation using structural profiles in the program 3D-PSSM. J.Mol. Biol., 299: 499–520, 2000.

18. Sussman, J. L., Lin, D., Jiang, J., N. Manning, N. O., Prilusky, J.,Ritter, O., and Abola, E. E. Protein data bank (PDB): database of three-dimensional structural information of biological macromolecules. ActaCrystallogr. D Biol. Crystallogr., 54: 1078–1084, 1998.

19. Taylor, W. R., and Orengo, C. A. Protein structure alignment. J. Mol.Biol., 208: 1–22, 1989.

20. Thompson, J. D., Gibson, T. J., Plewniak, F., Jeanmougin, F., andHiggins, D. G. The CLUSTAL X windows interface: flexible strategies formultiple sequence alignment aided by quality analysis tools. Nucleic AcidRes., 24: 4876–4882, 1997.

21. Maple, J. R., Hwang, M-J., Jalkanen, K. J., Stockfisch, T. P., andHagler, A. T. Derivation of class II force fields: V. Quantum force field foramides, peptides, and related compounds. J. Comp. Chem., 19: 430–458, 1998.

22. Garemyr, R., and Elofsson, A. Study of the electrostatics treatment inmolecular dynamics simulations. Proteins Struct. Funct. Genet., 37: 417–428, 1999.

23. Orozco, M., Laughton, C. A., Herzyk, P., and Neidle, S. Molecular-mechanics modelling of drug-DNA structure: the effects of differing di-electric treatment on helix parameters and comparison with a fully sol-vated model. J. Biomol. Struct. Dyn., 8: 359–373, 1990.

24. Luthy, R., Bowie, J. U., and Eisenberg, D. Assessment of proteinmodels with three-dimensional profiles. Nature (Lond.), 356: 83–85, 1992.

25. Laskowski, R. A., MacArthur, M. W., Moss, D. S., and Thornton, J. M.PROCHECK: a program to check the stereochemical quality of proteinstructures. J. Appl. Cryst., 26: 283–291, 1993.

26. Marhefka, C. A., Moore, I. I., B. M., Bishop, T. C., Kirkovsky, L.,Mukherjee, A., Dalton, J. T., and Miller, D. D. Homology modeling usingmultiple molecular dynamics simulations and docking studies of the hu-man androgen receptor ligand binding domain bound to testosterone andnonsteroidal ligands. J. Med. Chem., 44: 1729–1740, 2001.

27. Kiyama, R., Tamura, Y., Watanabe, F., Tsuzuki, H., Ohtani, M., andYodo, M. Homology modeling of gelatinase catalytic domains and dock-ing simulations of novel sulfonamide inhibitors. J. Med. Chem., 42: 1723–1738, 1999.

28. Luty, B. A., Wasserman, Z. R., Stouten, P. F. W., Hodge, C. N.,Zacharias, M., and McCammon, J. A. A molecular mechanics/grid method

for evaluation of ligand-receptor interactions. J. Comp. Chem., 16: 454–464, 1995.

29. Stouten, P. F. W., Frommel, C., Nakamura, H., aand Sander, C. Aneffective solvation term based on atomic occupancies for use in proteinsimulations. Mol. Simulation, 10: 97–120, 1993.

30. Prade, L., Engh, R. A., Girod, A., Kinzel, V., Huber, R., and Bossem-eyer, D. Staurosporine-induced conformational changes of cAMP-de-pendent protein kinase catalytic subunit explain inhibitory potential.Structure (Lond.), 5: 1627–1637, 1997.

31. Link, J. T., Raghavan, S., Gallant, M., Danishefski, S. J., Chou, T. C.,and Ballas, L. M. Staurosporine and ent-staurosporine: the first totalsynthesis, prospects for a regioselective approach, and activity profile.J. Am. Chem. Soc., 118: 2825–2842, 1996.

32. Engh, R. A., Gorod, A., Kinzel, V., Huber, R., and Bossemeyer, D.Crystal structures of catalytic subunit of cAMP-dependent protein kinasein complex with isoquinolinesulfonyl protein kinase inhibitors H7, H8, andH89. J. Biol. Chem., 271: 26157–26164, 1996.

33. Xu, R-M., Carmel, G., Kuret, J., and Cheng, X. Structural basis forselectivity of the isoquinoline sulfonamide family of protein kinase inhib-itors. Proc. Natl. Acad. Sci. USA, 93: 6308–6313, 1996.

34. Anderson, M. E., Braun, A. P., Wu, Y., Lu, T., Wu, Y., Schulman, H.,and Sung, R. J. KN-93, an inhibitor of multifunctional Ca��/calmodulin-dependent protein kinase, decreases early after-depolarizations in rabbitheart. J. Pharmacol. Exp. Ther., 287: 996–1006, 1998.

35. Matsuoka, I., Nakahata, N., and Nakanishi, H. Selective inhibition ofcollagen-induced arachidonic acid liberation by 1-(5-iodonaphthalene-1-sulfonyl)-1H-hexahydro-1,4-diazepine hydrochloride (ML-7), a myosinlight chain kinase inhibitor, in washed rabbit platelets. Biochem. Pharma-col., 54: 1019–1026, 1997.

36. Austen, M., and John, K. Preparation of quinazolines as aurora 2kinase inhibitors. PCT Int. Appl. WO 0121597 (UK)., 208pp, 2001.

37. Åqvist, J., Medina, C., and Samuelsson, J-E. A new method forpredicting binding affinity in computer-aided drug design. Protein Eng., 7:385–391, 1994.

38. Riccio, A., Tamburrini, M., Giardina, B., and Prisco, G. Moleculardynamics analysis of a second phosphate site in the hemoglobins of theseabird, south polar skua. Is there a site-site migratory mechanism alongthe central cavity. Biophys J., 81: 1938–1946, 2001.

39. Bridges, A. J., Zhou, H., Cody, D. R., Rewcastle, G. W., McMicheal,A., Hollis Showalter, H. D., Fry, D. W., Kraker, A. J., and Denny, W. A.Tyrosine kinase inhibitors. 8. An unusually steep structure–activity rela-tionship for analogues of 4-(3-bromoanilino)-6, 7-dimethoxyquinazoline(PD 153035), a potent inhibitor of the epidermal growth factor receptor.J. Med. Chem., 39: 267–276, 1996.

40. Bridges, A. J. Chemical inhibitors of protein kinases. Chem. Rev., 101:2541–2572, 2001.

41. Scapin, G. Structural biology in drug design: selective protein kinaseinhibitors. Drug Discovery Today, 7: 601–611, 2002.

42. Narayana, N., Cox, S., Xuong, N-H., Ten Eyck, L. F., and Taylor, S. S.A binary complex of the catalytic subunit of cAMP-dependent proteinkinase and adenosine further defines conformational flexibility. Structure(Lond.), 5: 921–935, 1997.

43. Narayana, N., Diller, T. C., Koide, K., Bunnage, M. E., Nicalaou, K. C.,Brunton, L. L., Xuong, N-H., Ten Eyck, L. F., and Taylor, S. S. Crystalstructure of the potent natural product inhibitor Balanol in complex withthe catalytic subunit of cAMP-dependent protein kinase. Biochemistry,38: 2367–2376, 1999.

44. Bohm, H. J. The computer program LUDI: a new method for the denovo design of enzyme inhibitors. J. Comput-Aided Mol. Des., 6: 61–78,1992.

45. Rarey, M., Kramer, B., Lenguer, T., and Klebe, G. A. A fast flexibledocking method using an incremental construction algorithm. J. Mol.Biol., 261: 470–489, 1996.

46. Cheetham, G. M. T., Knegtel, R. M. A., Coll, J. T., Renwick, S. B.,Swenson, L., Weber, P., Lippke, J. A., and Austen, D. A. Crystal structureof aurora-2, an oncogenic serine/threonine kinase. J. Biol. Chem., 277:42419–42422, 2002.




2003;2:283-294. Mol Cancer Ther Hariprasad Vankayalapati, David J. Bearss, José W. Saldanha, et al.

1Drug DesignBioinformatics Approach to Target Validation and Rational Targeting Aurora2 Kinase in Oncogenesis: A Structural

Updated version

http://mct.aacrjournals.org/content/2/3/283

Access the most recent version of this article at:

Cited articles

http://mct.aacrjournals.org/content/2/3/283.full#ref-list-1

This article cites 39 articles, 11 of which you can access for free at:

Citing articles

http://mct.aacrjournals.org/content/2/3/283.full#related-urls

This article has been cited by 5 HighWire-hosted articles. Access the articles at:

E-mail alerts related to this article or journal.Sign up to receive free email-alerts

SubscriptionsReprints and

[email protected] at

To order reprints of this article or to subscribe to the journal, contact the AACR Publications

Permissions

Rightslink site. (CCC)Click on "Request Permissions" which will take you to the Copyright Clearance Center's

.http://mct.aacrjournals.org/content/2/3/283To request permission to re-use all or part of this article, use this link



http://mct.aacrjournals.org/content/2/3/283.full#ref-list-1

http://mct.aacrjournals.org/content/2/3/283.full#related-urls

http://mct.aacrjournals.org/cgi/alerts

mailto:[email protected]



Documents

Targeting Aurora2 Kinase in Oncogenesis: A Structural ...posed onto each other using the program SAP (19). The structural alignments from SAP were fine-tuned manually to better match