Bioorganic & Medicinal Chemistry Letters 21 (2011) 4243–4247

Contents lists available at ScienceDirect

Bioorganic & Medicinal Chemistry Letters

journal homepage: www.elsevier .com/ locate/bmcl

Structural investigation into the inhibitory mechanisms of indomethacinand its analogues towards human glyoxalase I

Ming Liu a, Minggui Yuan a, Zhe Li a, Yuen-Kit Cheng b, Hai-Bin Luo a,⇑, Xiaopeng Hu a,⇑a School of Pharmaceutical Sciences, Sun Yat-Sen University, Guangzhou 510006, PR Chinab Department of Chemistry, Hong Kong Baptist University, Kowloon Tong, Hong Kong

a r t i c l e i n f o a b s t r a c t

Article history:Received 24 February 2011Revised 14 May 2011Accepted 20 May 2011Available online 1 June 2011

Keywords:Human glyoxalase IIndomethacinInhibitory mechanismKinetic analysisBinding free energyMolecular modeling

0960-894X/$ - see front matter � 2011 Elsevier Ltd. Adoi:10.1016/j.bmcl.2011.05.095

⇑ Corresponding authors. Tel.: +86 20 39943031 (H(X.H.).

E-mail addresses: luohb77@mail.sysu.edu.cn (H.-edu.cn (X. Hu).

In the present work, a combined study of kinetic analysis, molecular docking, and molecular dynamicssimulations on indomethacin and its analogues is performed to better understand their inhibitory mech-anisms towards human glyoxalase I (GLOI). A remarkable correlation (R2 = 0.974) was observed for sixinhibitors including indomethacin between their experimental inhibitory affinities and predicted bindingfree energy parameter (DGbind,pred). This suggests that DGbind,pred of a GLOI/inhibitor complex can be effi-ciently used to interpolate the experimental inhibitory affinity of a ligand of similar nature in the GLOIenzyme system. Energetic analyses revealed that electrostatic contribution plays an important role intheir inhibitory mechanisms, which reflects the significant contribution of the coordination bondbetween zinc and ligands. The present work highlights that indomethacin is a promising lead as GLOIinhibitors for further development since it may bind all subsites in the active site pocket of GLOI and sta-bilize the flexible loop (152–159).

� 2011 Elsevier Ltd. All rights reserved.

Numerous experimental and clinical evidences1–6 demonstratethat some of the non-steroidal anti-inflammatory drugs (NSAIDs),especially indomethacin, can enhance the cytotoxicity of a varietyof anticancer drugs when co-administered. One possible explana-tion of their anti-cancer effect is that they might come into effectvia a cyclooxygenase-independent pathway, in which the enzy-matic activity of human glyoxalase I (GLOI) is inhibited by indo-methacin with a Ki value of 18.1 lM (comparable to itsinhibitory affinity of 5 lM towards COX-2).7,8 It is proposed thatindomethacin is able to restore apoptosis in cancer cells by bindingto GLOI.8 As a ubiquitous detoxification pathway, GLOI (EC 4.4.1.5)along with glyoxalase II removes highly reactive a-oxoaldehydes,thus protects against cellular damage caused by glycation reac-tions.9,10 Abnormal expression or higher activity of GLOI has beendetermined in colon11, prostate12, lung13, and resistant human leu-kemia cells.14 Such over-expression or elevated activity of GLOI isinvolved in apoptosis resistance to anti-tumor agents in cancercells.12,14 It is shown by many in vitro and in vivo studies thatinhibitors of GLOI are promising agents in inhibiting carcinogene-sis and overcoming drug resistance by inducing elevated concen-trations of methylglyoxal in cancer cells.15–17

ll rights reserved.

.-B.L.); tel.: +86 20 39943032

B. Luo), huxpeng@mail.sysu.

The inhibition of NSAIDs towards GLOI provides a direction forthe development of novel effective GLOI inhibitors. Although theinhibitory activity of indomethacin was determined and the inter-actions between it and GLOI were probed by NMR titration exper-iments8, no detailed inhibitory mechanism is provided at astructural level. In the present work, a parallel study on indometh-acin and its analogues is performed both experimentally and theo-retically to better understand the inhibitory mechanisms of NSAIDstowards GLOI. The main objective herein is to study the quantita-tive relationship between the chemical structures of representativeindomethacin-type NSAIDs and two structurally related inhibitors(curcumin and bisdemethoxycurcumin we studied previously18)and their inhibitory affinities against GLOI. The quantitative rela-tionship is explained in terms of the physicochemical interactionsbetween the NSAID ligands and GLOI. The present study would aidin discovery of novel GLOI inhibitors based on the chemical skele-ton of indomethacin.

Kinetic measurement of GLOI activity was performed similarly asdescribed in our recent work18 with minor modification. The forma-tion of S-D-lactoylglutathione was measured by monitoring theincrement in absorbance at 240 nm at 30 �C with a thermostatedspectrophotometer. Reaction buffer containing 0.10 M sodiumphosphate (pH = 7.1) and hemithioacetal (MG-SG) was pre-incu-bated for 20 minutes at 30 �C. Concentrations of MG and GSH werecalculated and varied by using the equilibrium constant Kd

(3.0 mM) to obtain desired concentration of MG-SG. Excess free

Figure 1. Chemical structures and experimentally inhibitory affinities of inhibitorstowards human glyoxalase I. ⁄Ki values collected from the literature [8,18].

4244 M. Liu et al. / Bioorg. Med. Chem. Lett. 21 (2011) 4243–4247

GSH in the assay was kept at 0.10 mM. Reaction was initiated by theaddition of recombinant 6His-tagged human His-GLOI (3.0 nM) tothe reaction buffer. The NSAIDs inhibitors indomethacin, zomepirac,fenoprofen, and ketoprofen were purchased from Sigma. Ki values ofthese inhibitors were determined by Dixon plot.

All model calculations of this work were carried out based onthe X-ray crystal structure of human GLOI with bound s-(N-hydro-xy-N-P-iodophenylcarbamoyl) glutathione (HIPC-GSH, GSH refersto glutathione) (PDB code: 1QIN).19 Hydrogen atoms were addedaccording to the default protonation states of all residues at neutralpH. Residue His126 adjacent to the zinc ion is set to HID (Histidine,delta H). The partial atomic charge of the zinc ion was set to +2.Various docking methods (CDOCKER20a, ligandfit20a, and Surflex-dock20b) were compared to verify their credibility to the GLOI sys-tem by re-docking the substrate HIPC-GSH into the active sitepocket of GLOI. As a result, the optimal docking approach CDOCKERfor this system gave an RMSD value of 0.83 Å for HIPC-GSH to theX-ray crystal counterpart, which confirms the reliability of thisdocking protocol. Details of the docking procedure are describedin Appendix 1. From the top ten poses with the most negativeCDOCKER-Interaction-Energy, the optimal one based on both thedocking energy and cluster popularity for each inhibitor was cho-sen. Each system was prepared by using GLOI in complex withthe optimal docking pose of each inhibitor. The AMBER version10 program suite21 was used to perform all the MD (moleculardynamics) simulations. The AMBER ff03 force field was appliedas the parameters of the target protein GLOI, while the generalAMBER force field (GAFF)22 was used for the parameterization ofinhibitors. The partial atomic charges of the inhibitors were calcu-lated at the ab initio HF/6-31G⁄ level by Gaussian 0323 to generatethe electrostatic potential (ESP), which was then fitted to the re-stricted electrostatic potential (RESP) by the Antechamber modulein the AMBER package. The Zn2+ ion was treated by the ‘non-bonded model’ method.24–26 All initial models after docking werecompleted using the XleaP module included in AMBER. The com-plexes were neutralized by adding eight sodium counter ionswhich were outside the active site pocket, and solvated in a rectan-gular box of water molecules with solvent layers of 10 Å from thebox edges to solute surface.

Then the seven complexes (GLOI-indomethacin, GLOI-zome-pirac, GLOI-fenoprofen, and GLOI-ketoprofen, GLOI-tolmetin,GLOI-curcumin, and GLOI-bisdemethoxycurcumin) were energy-minimized to remove unfavorable steric strain relative to the forcefield adopted. Further MD simulations were carried out based onthe relaxed structures. Each complex was progressively heatedfrom 0 to 300 K in 50 ps and equilibrated for another 50 ps at300 K using the NVT (constant composition, volume, and tempera-ture) ensemble with a weak constraint of 10 kcal mol�1 Å�2. Final-ly, periodic boundary dynamics simulations of 6 ns were carriedout for the production step in an NPT (constant composition, pres-sure, and temperature) ensemble at 1 atm and 300 K. The temper-ature was controlled by the weak-coupling algorithm27, while thelong-range electrostatic interactions were treated by the Particle-Mesh-Ewald method28 with a non-bonded cut-off of 10 Å. All thesimulations proceeded with the SHAKE algorithm29 turned on.Output trajectory files were sampled every 2 ps for subsequentanalysis. The MM-PBSA approach30–36 encoded in AMBER was usedto estimate the binding free energy (DGbind, pred) in continuum sol-vent representation of the protein-ligand systems. A total of 100snapshots were taken from each last 1 ns trajectory with an inter-val of 10 ps. The entropy contribution was not considered becausethe structural similarity of the inhibitors made insignificant differ-ences at the studied temperature. The detailed information aboutthe MM-PBSA approach is provided in Appendix 2.

The reassessment of kinetic inhibition constants (Ki). The kineticinhibition constants (Ki) of five NASIDs towards GLOI have been

determined by others8 previously. The inhibitory affinities (seethe curves in Appendix 3) of four NASIDs except tolmetin werereassessed under identical conditions in the present work. Asshown in Figures 1 and 2, our experimental results lead to a consis-tent conclusion with those in the literature.

The MD trajectories of the complexes appear to be equilibrated. Tovalidate the dynamic stability of the model systems, the RMSD val-ues for the backbone atoms of the 6-ns MD trajectories were mon-itored using the X-ray crystal structure as a reference. Appendix 4shows that the RMSD values of the seven complexes reached con-vergence within 2 Å, which ensure the dynamic stability of MD tra-jectories of the complexes. Therefore, the subsequent analyseswere based on the MD trajectories between 5 and 6 ns.

Significantly linear correlations were observed between the pre-dicted binding free energies and the experimental ones. The predictedbinding free energies DGbind, pred (including the individual energyterm) were calculated by the MM-PBSA method (Table 1). Basedon DGbind,pred and the approximately estimated DGbind,exp

test fromour experimental Ki values (via DGtest

bind;exp � RT ln Ki), our statisticalanalysis achieved a significantly linear correlation (Fig. 2) with ahigh conventional regression coefficient (R2 = 0.974). This suggeststhat DGbind,pred of a GLOI/inhibitor complex can be used to prog-nosticate the experimental inhibitory affinity of a ligand towardsGLOI of similar structural features.

The performance of this model is better than expected. Brown et al.developed a series of plots to serve as a reality check for models of

Figure 2. Correlations of the experimental binding energy parameter DGbind,exp tothe theoretical counterpart DGbind,pred. R is the Pearson correlation coefficient and nis the number of ligands studied. DGbind,exp is approximately estimated viaDGbind,exp = RT ln Ki. The experimental Ki values are collected from the publishedexperimental data [8,18] (circle) and our kinetic measurements (triangle),respectively.

M. Liu et al. / Bioorg. Med. Chem. Lett. 21 (2011) 4243–4247 4245

a certain number of data points and potency span.37 For the datasetof this model, which spans a potency range of 2 log units and hassix points, it is estimated that the correlation coefficient is likelyto be between 0.8 and 0.9 in terms of experimental error only(r = 0.3) according to the research. Thus, our correlation coefficientof 0.974 is over expectation. However according to the statisticanalyses in the same work37, among 16 publications from 2006to 2007, 12 of the R2 values of them were better than expectedand two of them were worse than expected. Only two are in the ex-pected range. This result exemplified that it is a usual case in whichR2 values can be out of the reasonable range.

The predicted binding free energies can achieve better correlationswith the actual binding affinities of ligands than other molecular prop-erties. Several molecular properties such as molecular weight, log P,pKa, were used to study the dependence of actual binding affinitieson molecular properties. However, none of them produced bettercorrelations (Appendix 5) than that of the predicted binding energyby the MM-PBSA method since those molecular properties do nottake the receptor GLOI into consideration.

Electrostatic contribution (DEele,int) plays an important role in thebinding of ligands. Each of the energy terms involved in DGbind,pred

represents a specific contribution to the total binding free energy.Among the four terms (DEele,int, DEvdw, DGele,sol, and DGnonpol,sol),the internal electrostatic one DEele,int yields a reasonable correla-

Table 1Predicted binding free energies DGbind,pred (kcal/mol) obtained by the MM-PBSA method

indomethacin tolmetin zomepirac fenop

DGbind,pred �26.30 ± 3.61 �21.05 ± 4.26 �19.67 ± 3.11 �17.2

DGrefbind;exp

�6.48 �5.55 �5.43 �4.65

DGtestbind;exp

�6.31 — �4.75 �4.67

DEele,int �27.32 ± 5.77 �27.43 ± 6.97 �7.74 ± 3.14 �2.91DEvdw �38.88±3.00 �31.31 ± 3.30 �29.18 ± 2.12 �32.6DEMM �66.20 ± 4.95 �58.74 ± 5.81 �36.92 ± 3.82 �35.5DGele,sol 45.55 ± 3.93 42.36 ± 4.00 21.89 ± 2.72 23.00DGnonpol,sol �5.65 ± 0.12 �4.67 ± 0.13 �4.64 ± 0.19 �4.66DGsol 39.90 ± 3.89 37.69 ± 3.97 17.25 ± 3.71 18.34

DGbind,pred: The predicted binding free energy.DGref

bind;exp: The binding free energy estimated from the published experimental data [8].DGtest

bind;exp: The binding free energy estimated from our kinetic measurements.DEele,int: The Coulomb interactions, DEvdw: the van der Waals interactions, DGele,sol: theDGnonpol,sol: The nonpolar contribution to solvation, DGMM = DEele,int + DGvdw, and DGsol

tion (Appendices 6 and 7) with DGtestbind;exp , which emphasizes the rel-

ative importance of the internal electrostatic interactions in theinhibition of these ligands against GLOI. The studied GLOI inhibi-tors could be broadly divided into two groups in terms of DEele,int.Group-A is represented by indomethacin, tolmetin, curcumin, andbisdemethoxycurcumin, while Group-B by zomepirac, fenoprofen,and ketoprofen. The internal electrostatic energies of Group-A areremarkably more dominating than those of Group-B. These energydifferences of these two groups reflect diverse binding modes ofthese ligands.

Coordination with zinc is the underlying reason for the difference inthe internal electrostatic energies between Group A and Group B. Tofurther explain the energy differences between Group-A andGroup-B, the average distances between ligands and the zinc cat-ion were obtained from the last 1 ns trajectory. It turns out thatthe members of group A were within the coordination distanceof the zinc ion, whereas those of group B stayed out of the distancelimit of 2.4 Å. These average distances of the ligands to zinc versusthe internal electrostatic interaction energies and the predictedbinding free energies are shown in Appendix 6. As the distance be-tween ligand and zinc decreases, DEele,int and DGtest

bind;pred becomesmore negative. As a result, the members in Group-A exhibitedstronger inhibitory activities than those in Group-B. This peculiarrelationship between the ligand-zinc distance and the electrostaticinteraction energy strongly suggests that coordination with zinc isa crucial element in the ligand-binding process.

Similar structures of ligands do not necessarily correspond to sim-ilar binding poses due to the flexibility of the protein. The chemicalstructures of the representative NASIDs shown in Figure 1 are quitesimilar. They are comprised of two ring moieties (moiety A andmoiety B), which are connected by the carbonyl linker (except fen-oprofen). The interaction patterns were represented by averageconformations, which were generated by averaging 100 snapshotsfrom the last 1 ns MD trajectory. However, they exhibited diverseconformations and binding modes as simulation progresses. Thismay be attributed to the flexibility of the protein.

To better understand the intrinsic features of the ligand-bindingsite which account for both the energetic and binding difference ofthese molecules, the ligand-binding site of GLOI is subdivided intofour subsites (A, B, C, and D) as shown in Figure 3. Subsite A con-sists of Leu69B, Leu92B, Met179A, Met183A, and Phe71B and isrepresented by bulky hydrophobic groups. Subsite B contains thezinc ion and other residues which are responsible for coordination.Subsite C is located in the flexible loop area and displays fluctuat-ing conformation. It consists of Lys150A, Gly155A, Lys156A,Met157A, Leu160A, and Phe162A. Subsite D is formed by polar res-idues (Arg122A, Asn103B, and Arg37B) as well as nonpolar resi-dues (Phe67B and Trp170A).

rofen ketoprofen curcumin bisdemethoxycurcumin

1 ± 3.05 �16.72 ± 2.25 �30.68 ± 3.68 �27.40 ± 3.68�4.63 — —

�4.20 �6.82 �6.48

± 2.08 �2.09 ± 1.90 �55.75 ± 8.35 �48.87 ± 7.803 ± 2.04 �25.60 ± 2.44 �32.90 ± 4.05 �28.15 ± 4.065 ± 2.76 �27.69 ± 2.86 �88.65 ± 7.20 �77.02 ± 5.93± 3.26 15.12 ± 3.11 64.12 ± 5.49 54.80 ± 3.88± 0.14 �4.16 ± 0.25 �6.15 ± 0.16 �5.18 ± 0.12± 3.21 10.97 ± 2.94 57.97 ± 5.44 49.62 ± 3.86

polar contribution to solvation.= DGele,sol +DGnonpol,sol.

Figure 3. The ligand-binding site of human glyoxalase I. This site is subdivided intofour subsites (A, B, C, and D). The pocket which is displayed as a water-accessiblesurface (probe diameter 1.4 Å) is defined as residues found within 6 Å range of thecrystal ligand s-(N-hydroxy-N-P-iodophenylcarbamoyl) glutathione (HIPC-GSH).

Figure 5. Indomethain and curcumin (shown in sticks) in the ligand-binding site ofhuman glyoxalase I (shown in surface). Subsite A is fully occupied by curcumin(carbon: gray and oxygen: red), and partly occupied by indomethacin (carbon: cyan,oxygen: red, and nitrogen: blue). However indomethacin is bulky enough to take upboth subsite C and subsite D, whereas curcumin is not.

4246 M. Liu et al. / Bioorg. Med. Chem. Lett. 21 (2011) 4243–4247

Indomethacin is able to contact the four subsites. The bindingmode of indomethacin (at 10 ps intervals for the final 1 ns) is de-picted in Figure 4. On top of it, the carbonyl linker between moietyA and B forms a coordination bond of 2.24 Å with the zinc ion. Thishas been recognized as the key element of ligand binding. Thep-methylphenyl of indomethacin (moiety A) is buried inside sub-site A, surrounded by Leu160A, Met179A, Leu69B, and Phe62B.However, the pocket is not fully occupied comparing with curcu-min (Fig. 5), which has a larger group of moiety A to saturate thepocket. The B moiety of indomethacin is bulky enough to take upboth subsites C and D, delimited by Phe162A, Met35B, Met65B,Phe67B, and Thr101B. These interactions were also observed inNMR titration experiments.8 Besides, the carboxyl group of moietyB points toward the opening of the active site pocket andestablishes weak hydrogen bonds or electrostatic interactionswith Arg37B located in subsite D. However, the carboxylgroup of moiety B of indomethacin does not make strong contactwith those electropositive groups because it is not long and flexibleenough.

Figure 4. The binding pattern of indomethacin within the ligand-binding site ofhuman glyoxalase I. The red/blue/green numbers refer to the distances betweentwo atoms in Å. Blue dashes: coordination bond; red dashes: hydrogen bond; andgreen dashes: electrostatic interactions.

Indomethacin is the only compound that is able to occupy bothsubsites C and D simultaneously. In comparison, the other ligandsstudied here tend to contact with subsite C or subsite D exclu-sively. This explains why indomethacin possesses the most favor-able DEvdw (�38.88 kcal/mol) in binding. As shown in Figure 5,the moiety B of curcumin binds to subsite C, leaving subsite Dunoccupied. Tolmetin binds to GLOI in a similar manner as indo-methacin, except that the size of ring B is smaller than that of indo-methacin and subsite C is not filled, which results in weaker vander Waals interactions (DEvdw = �31.31 kcal/mol).

A flexible loop (152–159) was described to be responsible forthe binding of ligands in our previous study.18 The loop is locatedover the active site, and is proposed to be open in the absence of abound ligand. To illustrate this mechanism, both the trajectories ofthe protein with bound ligand and the apo one were analyzed. Theopen and close of the pocket can be monitored by the distance be-tween the CA atoms of Lys156A and Ile64B, which are located onthe two opposite sides of the opening of the pocket. The distancesare monitored in two different systems, as shown in Appendix 8. Itturns out that the indomethacin-GLOI complex exhibits a shorterdistance (about 2 Å) than the apo-protein, indicating that the looptends to be in a more closed conformation with indomethacin. TheB moiety of indomethacin is bulky and rigid, and occupies bothsubsite C and subsite D, which suggests that bulky and rigid com-ponents in moiety B are required to stabilize the flexible loop.

In the present study, both experimental and theoretical studieswere performed to explore the inhibitory mechanisms of indo-methacin and its analogues towards GLOI. A remarkable correla-tion (R2 = 0.974) was derived for the four structurally similarNSAIDs and two curcumins between the experimental bindingdata DGtest

bind;exp) and the theoretically estimated counterparts(DGbind,pred) by means of the MM-PBSA method, which suggeststhat DGbind,pred of a GLOI/inhibitor complex can be used to prog-nosticate the experimental inhibitory affinity towards GLOI of a li-gand of similar structural features. Energetic analysis revealed thatelectrostatic contribution (DEele,int) plays an important role in theirinhibitory mechanisms, which reflects the significance of the coor-dination bond between the zinc ion and ligand.

M. Liu et al. / Bioorg. Med. Chem. Lett. 21 (2011) 4243–4247 4247

The present study provides some relevant structural insights inthe design of GIOI inhibitors. The ligand-binding site of GLOI wasdivided into four subsites. Further structural analysis revealed thatindomethacin is the only compound binds significantly to all sub-sites simultaneously within the ligand-binding pocket of GLOI. Andthe flexible loop (152–159) tends to be in a more closed conforma-tion with bound indomethacin. Indomethacin, though possessingonly with moderate inhibitory affinity, may serve as a promisinglead skeleton for further modification because it binds all subsitessimultaneously, stabilizes the flexible loop and it has been investi-gated in clinical trials for decades. A case in point, modificationcould be attempted to enhance activity by increasing the size ofmoiety A, and lengthening the carboxyl group in moiety B to formfavorable contacts with Arg37B.

Acknowledgments

We thank the financial support from Natural Science Founda-tion of China (30800169), Natural Science Foundation of Depart-ment of Education in Guangdong Province (CXZD1006), NaturalScience Foundation of Guangzhou City (2010Y1-C531), and Funda-mental Research Funds for the Central Universities (10ykjc20).

Supplementary data

Supplementary data associated with this article can be found, inthe online version, at doi:10.1016/j.bmcl.2011.05.095.

References and notes

1. Coruzzi, G.; Venturi, N.; Spaggiari, S. Acta Biomed. 2007, 78, 96.2. Heath, C. W.; Thun, M. J.; Greenberg, E. R.; Levin, B.; Marnett, L. J. Cancer 1994,

74, 2885.3. Raveendran, R.; Heybroek, W.; Caulfield, M.; Lawson, M.; Abrams, S. M.;

Wrigley, P. F.; Slevin, M.; Turner, P. Hum. Exp. Toxicol. 1992, 11, 291.4. Maca, R. D. Anticancer Drug. Des. 1991, 6, 453.5. Ruegg, C.; Zaric, J.; Stupp, R. Ann. Med. 2003, 35, 476.6. Hull, M. A.; Gardner, S. H.; Hawcroft, G. Cancer Treat. Rev. 2003, 29, 309.7. Ulrich, C. M.; Bigler, J.; Potter, J. D. Nat. Rev. Cancer 2006, 6, 130.

8. Sato, S.; Kwon, Y.; Kamisuki, S.; Srivastava, N.; Mao, Q. A.; Kawazoe, Y.; Uesugi,M. J. Am. Chem. Soc. 2007, 129, 873.

9. Thornalley, P. J. Biochem. J. 1990, 269, 1.10. Thornalley, P. J.; Edwards, L. G.; Kang, Y.; Wyatt, C.; Davies, N.; Ladan, M. J.;

Double, J. Biochem. Pharmacol. 1996, 51, 1365.11. Clapper, M. L.; Hoffman, S. J.; Tew, K. D. Biochim. Biophys. Acta. 1991, 1096, 209.12. Davidson, S. D.; Cherry, J. P.; Choudhury, M. S.; Tazaki, H.; Mallouh, C.; Konno,

S. J. Urol. 1999, 161, 690.13. Sakamoto, H.; Mashima, T.; Sato, S.; Hashimoto, Y.; Yamori, T.; Tsuruo, T. Clin.

Cancer Res. 2001, 7, 2513.14. Ayoub, F. M.; Allen, R. E.; Thornalley, P. J. Biochem. Soc. Trans. 1993, 21, 168S.15. Creighton, D. J.; Zheng, Z. B.; Holewinski, R.; Hamilton, D. S.; Eiseman, J. L.

Biochem. Soc. Trans. 2003, 31, 1378.16. Sharkey, E. M.; O’Neill, H. B.; Kavarana, M. J.; Wang, H.; Creighton, D. J.; Sentz,

D. L.; Eiseman, J. L. Cancer Chemother. Pharmacol. 2000, 46, 156.17. Kalsi, A.; Kavarana, M. J.; Lu, T.; Whalen, D. L.; Hamilton, D. S.; Creighton, D. J. J.

Med. Chem. 2000, 43, 3981.18. Liu, M.; Yuan, M.; Luo, M.; Bu, X.; Luo, H.-B.; Hu, X. Biophys. Chem. 2010, 147,

28.19. Cameron, A. D.; Ridderstrom, M.; Olin, B.; Kavarana, M. J.; Creighton, D. J.;

Mannervik, B. Biochemistry 1999, 38, 13480.20. (a) Accelrys Discovery Studio 2.1, Accelrys Software Inc., San Diego, 2008.; (b)

Ajay, J. J. Comput. Aid. Mol. Des. 2007, 21, 281.21. D.A. Case, T.A. Darden, T.E. Cheatham, III, C.L. Simmerling, J. Wang, et al.,

AMBER 10, University of California, San Francisco, 2008.22. Wang, M.; Wang, W.; Kollman, P. A.; Case, D. A. J. Mol. Graph Model 2006, 25,

247; Hou, J.; Xu, J.; Liu, M.; Zhao, R.; Luo, H.-B.; Ke, H. PLoS ONE 2011, 6, e18092.23. M.J. Frisch, G.W. Trucks, H.B. Schlegel, G.E. Scuseria, M.A. Robb, et al.,Gaussian

03, Revision E.01, Gaussian, Inc., Pittsburgh PA, 2004.24. Aqvist, J.; Warshel, A. J. Mol. Biol. 1992, 224, 7.25. Wasserman, Z. R.; Hodge, C. N. Proteins 1996, 24, 227.26. Stote, R. H.; Karplus, M. Proteins 1995, 23, 12.27. Berendsen, H. J. C.; Postma, J. P. M.; Vangunsteren, W. F.; Dinola, A.; Haak, J. R. J.

Chem. Phys. 1984, 81, 3684.28. Darden, T.; York, D.; Pedersen, L. J. Chem. Phys. 1993, 98, 10089.29. Miyamoto, S.; Kollman, P. A. J. Comput. Chem. 1992, 13, 952.30. Kollman, P. A.; Massova, I.; Reyes, C.; Kuhn, B.; Huo, S. H.; Chong, L.; Lee, M.;

Lee, T.; Duan, Y.; Wang, W.; Donini, O.; Cieplak, P.; Srinivasan, J.; Case, D. A.;Cheatham, T. E. Accounts Chem. Res. 2000, 33, 889.

31. Kuhn, B.; Kollman, P. A. J. Med. Chem. 2000, 43, 3786.32. Wang, J. M.; Morin, P.; Wang, W.; Kollman, P. A. J. Am. Chem. Soc. 2001, 123,

5221.33. Chachra, R.; Rizzo, R. C. J. Chem. Theory Comput. 2008, 4, 1526.34. He, L.; He, F.; Bi, H.; Li, J.; Zeng, S.; Luo, H.-B.; Huang, M. Bioorg. Med. Chem. Lett.

2010, 20, 6008.35. Rizzo, R. C.; Toba, S.; Kuntz, I. D. J. Med. Chem. 2004, 47, 3065.36. Onufriev, A.; Bashford, D.; Case, D. A. Proteins 2004, 55, 383.37. Brown, S. P.; Muchmore, S. W.; Hajduk, P. J. Drug Discov. Today 2009, 14, 420.