You are on page 1of 14

DOI: 10.1002/minf.

201200124

Computational Investigation of SENP:SUMO ProteinProtein Interaction for Structure Based Drug Design
Ashutosh Kumar[a] and Kam Y. J. Zhang*[a]

Abstract : SUMO specific proteases (SENPs) are cysteine proteases that carry out the proteolytic processing of SUMO from its pro form as well as its deconjugation from substrate proteins. SENPs have been implicated in various cancers including prostate cancer, thyroid cancer and colon cancer. Therefore, the inhibition of SENPs is an attractive strategy for the treatment of cancer. However, the current SENP inhibitor development strategies target catalytic site and involve the usage of reactive functionalities to facilitate the covalent binding with a catalytic cysteine, which makes them less desirable for therapeutic purposes. Based on the available structural knowledge about the interaction of

SENPs with various SUMO paralogues, an alternative approach for inhibiting SENPs could be via targeting SENP:SUMO protein-protein interaction. Here we have investigated the protein-protein interaction between SENP and SUMO as a target for structure based drug design using pocket prediction, ligand binding hotspot mapping, molecular dynamics simulation and in silico alanine mutagenesis. Finally, we have provided recommendations for the structure based design of SENP:SUMO protein-protein interaction inhibitors. Our study indicates that the SENP inhibitors targeting SENP:SUMO protein-protein interaction is a viable alternative strategy to existing inhibitors targeting the enzymatic site.

Keywords: Sumoylation SUMO specific protease Protein-protein interactions Computational fragment mapping Molecular dynamics simulation

1 Introduction
Sumoylation is a posttranslational modification that plays an important role in a wide range of cellular processes including DNA replication and repair, chromosome packing and dynamics, genome integrity, nuclear transport, signal transduction and cell proliferation.[1,24] Sumoylation involves the covalent attachment of a small ubiquitin like modifier (SUMO) protein to e-amino group of lysine residues in specific target proteins via an enzymatic cascade that requires three steps of enzymatic reactions to attach the SUMO protein to its substrate. This cascade requires a sequential action of a set of enzymes, which includes an activating enzyme E1, a conjugating enzyme E2 and a ligase E3. Initially, SUMO protein is covalently attached to the active site cysteine of SUMO-activating enzyme subunit 2 of the SUMO E1 via a thioester bond. In the next step, SUMO protein is transferred from SUMO E1 to SUMO E2, also known as Ubc9, again with the formation of a thioester linkage between the C-terminal glycine in SUMO protein and the active site cysteine in Ubc9. Finally, Ubc9 catalyzes the attachment of SUMO protein to the lysine residue of substrate proteins. The efficiency of this step is increased by SUMO E3 protein that associates both with Ubc9 and substrate protein to stimulate protein sumoylation.[2,5] The conjugation of SUMO to e-amino group of lysine residues in substrate proteins is a reversible process, which is achieved by the action of SUMO specific proteases (SENPs).[610] Six SENPs (SENP1, SENP2, SENP3, SENP5, SENP6 and SENP7) have been identified in mammals that differ
Mol. Inf. 2013, 32, 267 280

with each other on the basis of their sequence homology, substrate specificity and subcellular localization.[610] All SENPs possess isopeptidase activity and cleave the isopeptide bond between the substrate protein lysine and C-terminal glycine of SUMO to release conjugated SUMO proteins from substrates. Also, SENPs play an important role in the proteolytic processing of SUMOs, which needs to be proteolytically processed from its pro form to mature form.[610] SENPs cleave inactive or pro form of SUMO at the C-terminus by its hydrolase activity to expose two glycine residues and thereby generate active or mature SUMO. Over the past few years, several studies have implicated that SENPs were involved in various cancers including prostate cancer,[1114] thyroid cancer[15] and colon cancer.[16] Cheng et al.[14] reported that SENP1 was over-expressed in more than 50 % of their studied samples of prostatic intraepithelial neoplasia and prostate cancer. They also demonstrated the role of SENP1 in prostate cancer development in transgenic mice. Bawa-Khalfe et al.[11] also reported the high expression of SENP1 in human prostate cancer speci[a] A. Kumar, K. Y. J. Zhang Zhang Initiative Research Unit, Advanced Science Institute, RIKEN 2-1 Hirosawa, Wako, Saitama 351-0198, Japan phone: + 81-48-467-8792, fax: + 81-48-467-8790 *e-mail: kamzhang@riken.jp Supporting information for this article is available on the WWW under http://dx.doi.org/10.1002/minf.201200124.

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

267

Full Paper
mens. They also demonstrated the correlation between SENP1 expression and hypoxia-inducing factor 1a (HIF1a) expression. Xu et al.[16] showed significantly increased SENP1 expression in colon cancer tissues. Their research also showed that SENP1 regulates colon cancer cell growth by upregulating the expression of CDK inhibitors. The above mentioned studies suggest that SENPs play an important role in the development of cancer especially prostate cancer and can be attractive targets for drug development against prostate cancer. However, upto now only a few inhibitors or activity based probes have been identified for SENPs.[1721] Initial SENP inhibitor development strategies were based on the use of a full or truncated form of SUMO carrying a reactive functional group at the C-terminal glycine. Hemelaar et al.[17] reported the synthesis of C-terminally modified SUMO derivatives by using an intein-based method to attach vinyl sulfones, which function as electrophilic traps. Borodovsky et al.[18] used a similar strategy to synthesize several peptide vinyl sulfones harboring various portions of the C-terminus of ubiquitin like modifiers Nedd8, SUMO1, FAT10, Fau, and APG12. In recent years, several small molecular SENP inhibitors and activity based probes have been developed. These include compounds identified by Albrow et al.[19] that effectively inhibited SENP 1, 2, 5, and 7. The identified SENP inhibitors contain azaepoxide and acyloxymethyl ketone reactive groups, which bind covalently to the active site cysteine. Qiao et al.[20] also designed and synthesized a series of benzodiazepine based SENP1 inhibitors, which showed IC50 value upto 9.2 mM for the most potent compound. Recently, Do et al.[21] reported the synthesis and biochemical propbrota erties of activity based peptidyl probes for SENP1 and SENP2 that harbor a glycine derived flouromethylketone moiety at its C-terminus. This reactive chemical moiety was expected to target the active site cysteine. The common characteristics of all SENP inhibitors developed so far is that they target the catalytic site and possess reactive chemical functionalities to bind covalently to the active site cysteine.[1721] Although, some of these inhibitors can be effectively used as probe molecules to study the biological mechanism of desumoylation but the presence of a reactive functional group makes them undesirable for therapeutic uses. The discovery of small molecules with therapeutic potential against SENPs faces several challenges. The mechanistic feature of SENPs active site presents a major challenge to drug development. As SENPs belong to a family of cysteine proteases, the active site of all SENPs prefers an electrophilic carbonyl group in the binding ligand. Also, the chemistry behind peptide cleavage within a protease class overlaps, which makes it difficult to identify selective inhibitors. Another major challenge is arrangement of residues within the active site. A careful analysis of SENPs crystal structures revealed that the active site of SENPs possesses a small channel formed by the positioning of three (Trp, Val and active site cysteine) residues.[2226] This channel is formed upon the closing of a cleft near the 268
www.molinf.com

A. Kumar, K. Y. J. Zhang

active site cysteine after substrate binding. This tunnel like cavity only allows very small molecules like C-terminal GlyGly sequence to enter into it. Small molecules that target sites other than the enzymatic site and are still able to inhibit the catalytic activity of SENPs could be useful for therapeutic purposes. But the lack of structural information about non-enzymatic sites that can be targeted for the inhibition of SENP biological activity hampers the development of small molecule inhibitors and impedes rational drug design. Under these circumstances, it is desirable to use computational methods for the identification and characterization of non-enzymatic sites. The availability of sufficient structural information about different SENPs may also facilitate the computational analysis.[2226] There are 17 crystal structures (nine, six and one for SENP1, SENP2 and SENP7 respectively) of SENPs in apo form and in complex with different SUMO paralogues and substrate protein RanGAP1 available upto now. Here we report the identification of non-enzymatic functional sites on the surface of SENP1 and SENP2 that can be modulated by small molecules and can be targeted by structure based drug design. Specifically, we have identified possible protein ligand interaction hotspots at SENP:SUMO protein-protein interface. Furthermore, we have explored the druggability of these sites using computational fragment mapping calculations. This study also identified structural features that could be exploited to design potent SENP inhibitors.

2 Materials and Methods


2.1 Multiple Sequence Alignment and Phylogenetic Analysis

The protein sequences for the catalytic domain of six SENPs were downloaded from the National Center for Biotechnology Information (NCBI) protein database (http:// www.ncbi.nlm.nih.gov). The sequences were aligned using ClustalW[27] using Gonnet scoring matrix. The alignment figure was generated using ESPript2.2 webserver.[28] The phylogenetic analysis was performed using Phylogeny.fr web server.[29]
2.2 Identification of Putative Binding Pockets

In order to determine all possible cavities on the surface of SENP1 and SENP2, we have used Metapocket2.0 pocket prediction algorithm.[30,31] Metapocket2.0 is a consensus program, which combines the results of LIGSITEcs (LCS), PASS (PAS), SURFNET (SFN), Q-Site Finder (QSF), GHECOM (GHE), POCASA (PCS), FPOCKET (FPK) and Concavity (CON) binding site prediction algorithms. The SENP1 and SENP2 apo crystal structures (PDB code 2CKG and 1TH0) were used for pocket prediction using Metapocket2.0. As bound structures of SENP1 and SENP2 were also available, so the apo structures were also compared with bound structures for possible conformational variability. The root mean squared deviation (RMSD) calculated after superimposing
Mol. Inf. 2013, 32, 267 280

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

SENP:SUMO Protein-Protein Interaction for Structure Based Drug Design

apostructure with bound structure is given as Supporting Information Table T1 and the comparison revealed no significant difference in apo and bound structure of SENP1 and SENP2. Water molecules and other heteroatoms were removed prior to Metapocket2.0 analysis.

Table 1. A summary of SENP:SUMO structures used for molecular dynamics simulation and binding free energy calculation. Protein complex SENP1-SUMO1RanGAP1 SENP1-SUMO2 SENP1-SUMO3 SENP2-SUMO1 SENP2-SUMO2RanGAP1 SENP2-preSUMO3 PDBid Resolution Remark 2IY0 2CKH 1TGZ 2IO3 2IO1 2.77 3.20 2.80 3.20 2.60 Structure is C602S mutant Crystal structure Modeled structure[a] Crystal structure Crystal structure Crystal structure

2.3 Computational Fragment Mapping

FTMap server (http://ftmap.bu.edu) was used for computational fragment mapping to identify druggable hotspot regions in SENP1 and SENP2. Computational fragment mapping calculations were performed on PDB code 2CKG and 1TH0 for SENP1 and SENP2 respectively. In FTMap algorithm,[32] sixteen fragment probes with varying hydrophobicity and hydrogen bonding capabilities are used to map protein surface for identifying energetically favorable binding regions. The organic probe molecules used are ethanol, isopropanol, isobutanol, acetone, acetaldehyde, dimethyl ether, cyclohexane, ethane, acetonitrile, urea, methylamine, phenol, benzaldehyde, benzene, acetamide and N,N-dimethylformamide. Initially, a fast Fourier transform approach is used to dock fragments onto the protein surface and their interaction energy is calculated using an energy function that is dominated by the terms for attractive and repulsive van der Waals, electrostatic term obtained after solving PoissonBoltzmann equation, nonpolar term and a term that represents structure based pairwise interaction potential. Low energy fragment protein complexes are then clustered followed by the determination of regions of the protein surface where low-energy fragment clusters of multiple fragment types colocalize.

[a] The SENP1-SUMO3 structure was modeled by superimposing SUMO3 (from PDBid: 2IO1) on SENP1-SUMO2 crystal structure (PDBid: 2CKH). SUMO2 was then deleted from the crystal structure and SENP1-SUMO3 structure was refined by energy minimization and saved as a single PDB file.

Amber10[35] with Amber force field (ff03). Analysis was performed by Ptraj module of Amber tool.[35]

2.5 Binding Free Energy Calculation and MM-GBSA Binding Free Energy Decomposition

Molecular mechanics Poisson Boltzmann surface area (MMPBSA)[36,37] and molecular mechanics generalized Born surface area (MM-GBSA) methods[3840] as implemented in AMBER10 were used to calculate binding free energy between SENP and SUMO proteins. Binding free energy is given by the following equation DGbinding GSENP:SUMO GSENP GSUMO The above equation can be conceptually summarized as

2.4 Molecular Dynamics Simulation

Twenty nanosecond molecular dynamics (MD) simulations were carried out for each SENP1 and SENP2 structure under study. A total of six SENP : SUMO structures (SENP1 and SENP2 in complex with SUMO1/SUMO2/SUMO3) were used for molecular dynamics simulations and an overview of structures is presented in Table 1. In every case, the proteins were immersed in the rectangular truncated octahedron filled with 8 TIP3P water molecules and neutralized by adding Na + or Cl ions. Initially, the protein system was minimized by 500 steps of steepest descent followed by 2000 steps of conjugate gradient. After the minimization, the system was gradually heated in a canonical ensemble from 0 to 300 K over 50 ps and then equilibrated for 500 ps. Finally, a 20 ns MD simulation was performed under a constant temperature of 300 K. The long-range electrostatic interactions were dealt by employing partical mesh Ewald (PME) method.[33] All hydrogen atoms were constrained using SHAKE algorithm[34] and the time step was set to 2 fs. Trajectory snapshots were taken at each 1 ps, which were used for final analysis. All simulations were performed by the SANDER module of
Mol. Inf. 2013, 32, 267 280

DGbinding DGMM DGpolar solvation DGnonpolar solvation T DS and DGMM DE vdw DE elec DE int where DGMM is the molecular mechanics binding free energy between SENP and SUMO. DEvdw, DEelec, DEint accounts for differences in van der Waals energy, electrostatic energy and internal energy. DGpolar solvation is the polar contribution to solvation free energy and could be computed by either solving PoissonBoltzmann equation or by using generalized Born model. DGnonpolar solvation is the nonpolar contribution to solvation free energy and could be estimated using the following equation: DGnonpolar solvation g SASA b where SASA is the solvent accessible surface area calculated
www.molinf.com

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

269

Full Paper
using LCPO method.[41] The g and b are empirical constants with default values of 0.0072 and 0 respectively. The calculations were performed on 200 snapshots generated from a 20 ns molecular dynamics trajectory. The contribution of each residue to binding free energy was calculated by decomposing the MM-GBSA binding free energy per residue. We have not considered the entropy (TDS) in this calculation as we were mainly interested in identifying the relative energy contributions of amino acids in SENP:SUMO complex formations.
2.6 In Silico Alanine Mutagenesis

A. Kumar, K. Y. J. Zhang

binding and thereby in complex formation. In silico mutagenesis was performed using MM-PB(GB)SA alanine scanning utility as implemented in AMBER10. In MM-PB(GB)SA alanine scanning approach, mutant trajectory is initially generated from the wild type molecular dynamics trajectory by truncating the side chains and then used to recalculate various energetic contributions. The interface residues of SENPs that are within 5 from SUMO in SENP:SUMO complex were mutated to alanine for all six molecular dynamics calculations.

In silico alanine mutagenesis was also used to unveil the role of amino acid residues at SENP : SUMO protein-protein interface that contribute significantly to free energy of

3 Results and Discussion


In sumoylation pathway, SENPs catalyze the deconjugation of SUMO from substrate proteins. They also proteolytically

Figure 1. Phylogenetic analysis and multiple sequence alignment of the catalytic domain of SENPs. The sequence numbering corresponds to SENP1.

270

www.molinf.com

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Mol. Inf. 2013, 32, 267 280

SENP:SUMO Protein-Protein Interaction for Structure Based Drug Design

process SUMO from its inactive form to active form. There are six members of SENPs, which can be divided into three subfamilies based on their sequence, substrate specificity and cellular location. The sequence alignment of the catalytic domain of all the SENPs is shown in Figure 1 and it reveals high sequence homology among SENPs of the same subfamily. Phylogenetic analysis of all six members of SENP was also carried out using Phylogeny.fr web server[42] that resulted in three clusters as shown in Figure 1, which account for each of the subfamily of SENPs. The first subfamily consists of SENP1 and SENP2, which have broad specificity for all mammalian SUMOs, i.e SUMO1, SUMO2 and SUMO3 (There is no in vivo evidence for SUMO4 conjugation to protein substrates.[43,44]) SENP1 and SENP2 are localized in nucleoplasm and are able to carry out deconjugation of a large number of protein substrates. SENP3 and SENP5 constitute the second subfamily and are specific to SUMO2 or SUMO3. SENP6 and SENP7 make the third subfamily and as we can see in Figure 1 they have an insertion in the catalytic domain. SENPs play an important role in the development of cancer especially prostate cancer[1114] and have therapeutic potential. The difficulty in targeting the catalytic site of SENPs prompted us to investigate alternative sites in SENPs that can be exploited for structure based drug design. In the current study, we have first identified small molecule binding hotspots other than the catalytic site using pocket prediction and ligand binding hotspot profiling. These hotspots were then prioritized using molecular dynamics simulation and computational alanine mutagenesis. Finally, we have provided recommendations for

the structure based design of inhibitors targeting identified hotspots residues. Here, we have restricted our study to SENP1 and SENP2 because of the availability of crystal structures in apo form and in complex with SUMO and substrate protein (nine and six crystal structures for SENP1 and SENP2 respectively).
3.1 Identification of Putative Binding Sites Using Pocket Prediction Algorithms

The first step in our investigation was to identify all putative small molecule binding pockets in SENP1 and SENP2 that can be further prioritized for structure based drug design. The identification and analysis of small molecule binding pockets is very important in structure based drug design efforts. There are many pocket prediction programs available that are mainly based on geometry alone or algorithms including physicochemical and energy based considerations.[45,46] As both geometry and energy based pocket prediction algorithms have merits and demerits,[45,46] we have used Metapocket2.0[30,31] pocket prediction algorithm to locate the binding pocket on the surface of SENP1 and SENP2 in this study. Metapocket2.0 consensus results were generated after clustering top three pockets from each of the pocket prediction program. The pocket prediction resulted in the identification of 6 and 8 Metapocket2.0 clusters on the surface of SENP1 and SENP2 respectively. The Metapocket2.0 results are presented in Table 2 and top 5 Metapocket2.0 sites each for SENP1 and SENP2 are presented in Figure 2. Metapocket2.0 predicted the largest and top

Table 2. Predicted pockets by Metapocket2.0 for SENP1 and SENP2. Site Z-Score Ranking by different pro(Pocket sites) grams[a] GHE-1, FPK-3, LCS-2, SFN-2, PCS-1, PAS-2, CON-1 GHE-2, SFN-1, LCS-1, FPK-1, PCS-3, PAS-1, CON-2, QSF-2 PAS-3, LCS-3, QSF-1 SFN-3 QSF-3 SFN-1, LCS-1, PAS-1, GHE-1, CON-1 Amino acid residues

SENP1 A1 13.10 (7) A2 A3 A4 A5 SENP2 B1 12.85 (8) 3.33 (3) 2.95 (1) 0.61 (4) 16.31(5)

Phe448, Arg449, Leu450, Asn472, Met475, Asn476, Met479, His491, Ala492, Phe493, Asn494, Phe496, Phe497, Arg511, Trp512, Thr513, Lys514, Lys515, Val516, Pro527 Glu419, Phe420, Pro421, Tyr474, Arg481, Lys608, Asp611, Cys612, Lys615, Arg617, Pro618, Ile619, Asn620, Phe621, His625, Phe629, Arg632, Glu636, Leu642, Leu643 Met477, Glu480, Arg481, Glu484, Arg640, Lys641, Leu642 Phe435, Gln458, Leu460, Asn461, His462, Asn464 Phe420, Pro421, Glu422, Ile423, Glu428, Gln624, His625, Tyr628, Lys631

B2

7.88(3)

B3 B4 B5

4.39 (4) 3.52 (5) 0.51 (3)

Phe393, Lys394, Leu395, Asp413, Glu414, Asn417, Met420, Asn421, Val424, Glu425, Asn427, Lys428, Ala434, Leu435, His436, Val437, Phe438, Ser439, Phe441, Arg456, Trp457, Thr458, Lys459, Gly460, Val461, Gln466 FPK-1, SFN-2, PCS-2 Trp410, Tyr443, Pro444, Lys447, Ser448, Ile473, His474, Arg475, Lys476, Val477, His478, Trp479, Ser480, Tyr493, Asp495, Met497, Gln499, Lys500, Gly501, His502, Arg503, Ile504, Gln542, Asn544, Gly545, Ser546, Cys548, Gly549, FPK-2, LCS-2, GHE-2, QSF-1 Pro433, Ile468, Val483, Asp485, Arg487, Lys488, Cys490, Lys492, Met534, Ile559, Asp562, Lys563, Pro564, Ile565, Glu538, Ile539, Arg561 QSF-2, CON-2, PCS-1, SFN-3, Asp364, Leu365, Leu366, Tyr419, Arg426, Lys429, Tyr432, Gln430, Lys554, Asp557, GHE-3 Tyr558, Arg561, Phe575, Arg576, Lys578, Met579, Glu582, Gln587, Leu588, Leu589 QSF-3, LCS-3, PCS-3 Ala392, Phe393, Leu422, Glu425, Arg426, Lys429, Gln430, Gln586, Gln587, Leu588, His585

[a] Programs are abbreviated as follows : LIGSITEcs (LCS), PASS (PAS), SURFNET (SFN), Q-Site Finder (QSF), GHECOM (GHE), POCASA (PCS), FPOCKET (FPK) and Concavity (CON)

Mol. Inf. 2013, 32, 267 280

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

www.molinf.com

271

Full Paper

A. Kumar, K. Y. J. Zhang

Figure 2. Predicted pockets by Metapocket2.0 for (A) SENP1 (B) SENP2 (C) SENP1 : SUMO1 complex and (D) SENP2 : SUMO1.

ranked pocket (A1) with a z-score of 13.10 comprised of Phe448, Arg449, Leu450, Asn472, Met475, Asn476, Met479, His491, Ala492, Phe493, Asn494, Phe496, Phe497, Arg511, Trp512, Thr513, Lys514, Lys515, Val516 and Pro527 amino acid residues for SENP1. Metapocket2.0 predictions for SENP2 also identified this pocket (B1) as top ranked pocket with a z-score of 16.31 and consists of Phe393, Lys394, Leu395, Asp413, Glu414, Asn417, Met420, Asn421, Val424, Glu425, Asn427, Lys428, Ala434, Leu435, His436, Phe438, Ser439, Val437, Phe441, Arg456, Trp457, Thr458, Lys459, Gly460, Val461 and Gln466 amino acid residues. As shown in Figure 2 this pocket is different from the catalytic site and is near the major contact region for protein-protein interaction between SENPs with SUMO proteins.[2226] The loop from Phe66 to Gln69 in SUMO1 binds in this region and it is conserved in all SUMO1, SUMO2 and SUMO3. The sequence alignment of SUMO paralogues is given as a supplementary figure S1. The top ranked pocket is enriched in residues with larger side chains such as Phe, Trp, His, Arg, 272
www.molinf.com

Lys and Asn. The Phe and Trp are generally involved in the formation of hydrophobic and stacking interactions whereas His, Arg, Lys and Asn form hydrogen bond and salt bridges in protein-protein and protein-ligand complexes. These residues are found to occur with higher preference in ligand binding sites[47] and protein-protein interfaces.[48] Some of the important residues from this pocket are shown as sticks in Figures 2A and 2B for SENP1 and SENP2 respectively.
3.2 Computational Fragment Mapping

To further access the druggability and to elucidate ligand binding hotspots, computational fragment mapping was performed using FTMap server.[32] The FTMap approach of identifying small molecule binding hotspots is analogous to fragment based screening of a protein using experimental methods like NMR and X-ray crystallography. The results are shown in Figure 3 and it shows that FTMap was able to
Mol. Inf. 2013, 32, 267 280

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

SENP:SUMO Protein-Protein Interaction for Structure Based Drug Design

Figure 3. Computational fragment mapping results showing the distribution of ligand binding hotspots. The top 10 consensus sites are shown for (A) SENP1 and (B) SENP2. The most populated consensus sites are shown at the proteinprotein interface of (C) SENP1 : SUMO1 and (D) SENP2 : SUMO1.

find several consensus sites (CSs) and the top 10 CSs are presented in Figure 3. CSs are the overlapping regions of different probe clusters in proteins.[32] FTMap results are in excellent agreement with Metapocket2.0 pocket prediction with 4 out of 10 CSs were found at Metapocket2.0 predicted largest site in both SENP1 and SENP2 at the SENP:SUMO protein-protein interface including the most populated consensus site. Sequence alignment (Figure 1) also shows that this region is highly conserved among all SENPs highlighting its significant role in SENP:SUMO complex formation.
3.3 Molecular Dynamics Simulation and Binding Free Energy Calculations

Pocket prediction and ligand binding hotspot identification highlighted a region in SENP1 and SENP2 that can be exploited for structure based drug design. This potential hotspot falls at SENP:SUMO protein-protein interface, which is
Mol. Inf. 2013, 32, 267 280

quite large and spans about 30 . The protein-protein interfaces are generally large, however, only specific side chains in the interface may substantially contribute to binding free energy.[49,50] To understand the interactions of SUMO proteins with SENP1 and SENP2, binding free energy calculations were carried out for SENP:SUMO protein-protein complexes. It will also unveil hotspot residues that contribute significantly to the formation of SENP:SUMO complexes. Identification of key residues that contribute significantly to the formation of SENP:SUMO proteinprotein complex was carried out following two approaches: (a) free energy decomposition analysis, (b) in silico alanine mutagenesis. These two approaches make use of MM-PB(GB)SA methods implemented in AMBER for calculating the binding free energy between SENP and SUMO proteins. MMPB(GB)SA is one of the commonly used method to identify energetic hotspots at the protein-protein interface and has been used previously by various groups.[5160]

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

www.molinf.com

273

Full Paper

A. Kumar, K. Y. J. Zhang

Figure 4. Time evolution of root mean squared deviation (RMSD) of backbone atoms of SENP : SUMO complexes (A) SENP1 : SUMOs and (B) SENP2 : SUMOs.

3.3.1 Stability of Molecular Dynamics Simulation

To access the quality of our molecular dynamics simulation, the stability of the SENP:SUMO complex structures was checked by monitoring the RMSD of backbone atoms throughout the production run trajectory from their initial conformation. It can be seen from Figure 4 that the relative fluctuation from starting structure in RMSD is very small (less than 1 ) after the initial rise and the system has been stabilized in all the six SENP:SUMO protein complexes. To further test the convergence of simulation root mean square fluctuations (RMSF) were calculated to investigate the mobility of the protein and are presented in Figure 5. The RMSF were calculated using all backbone atoms for all

six structures from SENP1 and SENP2 and Figure 5 shows that both these proteins share similar RMSF distributions. The majority of fluctuations in both proteins corresponds to N-termini and some loop regions. Apart from these regions RMSF is comparatively low (less than 1 ) specifically at the ligand binding hotspot region identified using pocket prediction and FTMap calculations.

3.3.2 SENP : SUMO Binding Free Energy Calculations

The binding free energies for all six SENP:SUMO complexes were calculated using MM-PB(GB)SA as implemented in AMBER. The total binding free energy calculated from MMPB(GB)SA method is presented in Table 3 and the detailed

Figure 5. Per residue root mean square fluctuation (RMSF) of backbone atoms of SENP : SUMO complexes (A) SENP1 : SUMOs and (B) SENP2 : SUMOs.

274

www.molinf.com

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Mol. Inf. 2013, 32, 267 280

SENP:SUMO Protein-Protein Interaction for Structure Based Drug Design Table 3. Summary of binding free energy calculated from MM-PBSA and MM-GBSA methods for all six SENP:SUMO complex structures. Structure SENP1-SUMO1 SENP1-SUMO2 SENP1-SUMO3 SENP2-SUMO1 SENP2-SUMO2 SENP2-SUMO3 DGMM-PBSA binding Complex 11252.3 8699.7 8621.6 7669.8 10853.0 8156.3 SENP 9070.8 6006.9 5988.7 5418.3 8359.6 5487.9 SUMO 2033.3 2589.5 2503.7 2117.8 2339.4 2519.3 Delta 148.3 103.3 129.3 133.7 154.0 149.2 DGMM-GBSA binding Complex 11263.2 8732.5 8650.9 7715.8 10890.8 8197.4 SENP 9085.8 6026.8 5996.7 5462.4 8396.8 5513.5 SUMO 2050.9 2609.7 2533.9 2145.9 2364.5 2550.0 Delta 126.5 96.0 120.3 107.5 129.5 133.9

contribution of vdW, electrostatics and solvation energy are listed in Supporting Information Table T2. The MM-PBSA binding free energies of SENP1 : SUMO1, SENP1 : SUMO2 and SENP1 : SUMO3 are estimated to be 148.3 kcal/mol, 103.3 kcal/mol and 129.3 kcal/mol respectively while the MM-GBSA binding free energies are estimated as 126.5 kcal/mol, 96.0 kcal/mol and 120.3 kcal/mol for SENP1 : SUMO1, SENP1 : SUMO2 and SENP1 : SUMO3 respectively. Additionally, the MM-PBSA binding free energies of SENP2 : SUMO1, SENP2:SUMO2 and SENP2:SUMO3 are estimated to be 133.7 kcal/mol, 154.0 kcal/mol and 149.2 kcal/mol respectively while the MM-GBSA binding free energies are estimated as 107.5 kcal/mol, 129.5 kcal/mol and 133.9 kcal/mol for SENP2 : SUMO1, SENP2 : SUMO2 and SENP2 : SUMO3 respectively. The entropic term was ignored for the calculation of binding free energy as we were only interested in the relative ranking and contribution of energetic components. It is important to note here that the estimated binding free energies here do not represent absolute binding free energy values. MM-PB(GB)SA approach generally allows rapid estimation of relative variations in binding free energy and usually represent good correlation with experimental affinities.[59,6164] Here, the estimated binding free energies are larger than experimental affinities but the overall ranking of the binding affinities of SUMO paralogues against SENP2 matches well with the experimental activity.[22,65,66] According to MM-PB(GB)SA calculations, the relative affinity of SUMO paralogues against SENP1 ranks as follows: SUMO1 > SUMO2 < SUMO3 while against SENP2 ranks as SUMO1 < SUMO2  SUMO3. However, the affinity of SUMO2 against SENP1 is underestimated by our MMPB(GB)SA calculation.[22,65,66] It was expected that small molecule binding hotspots identified by pocket prediction and computational fragment mapping are the regions that contribute significantly to binding free energy. Therefore, to identify small molecule binding hotspots at SENP:SUMO protein-protein interface MM-GBSA residue-wise decomposition was carried out. We considered the amino acid residues that make significant contributions to the predicted binding free energy by forming crucial intermolecular interactions as SENP : SUMO interaction hotspots. The residuewise decomposition of SENP:SUMO binding free energy for all six protein-protein complexes is shown in Figure 6.

As can be seen from Figure 6 significant contributions of > 4.0 kcal/mol to binding free energy come from Asp468, Phe496, Arg511 and Trp512 in SENP1 and corresponding Asp413, Phe441, Arg456 and Trp457 in SENP2. All of these residues are located at the predicted hotspot region at SENP : SUMO protein-protein interface and are involved in the formation of strong intermolecular interactions with SUMO paralogues. The most significant contribution to predicted binding free energy originates from Arg511 of SENP1 and corresponding Arg456 of SENP2 with a contribution of more than 10 kcal/mol for nearly all cases (except SENP2:SUMO2 simulation where it is around 8 kcal/mol). This Arg511 (SENP1) or Arg456 (SENP2) forms a stable hydrogen bond with Glu67 of SUMO1 and similar Asp62 of SUMO2 and SUMO3 with occupancy of more than 40 % in all molecular dynamics trajectories (Supporting Information Table T3). Another amino acid residue Asp468 in SENP1 and corresponding Asp413 in SENP2 interact with Arg63 of SUMO1 via the formation of a hydrogen bond. This hydrogen bond was also conserved in molecular dynamics simulation and had an occupancy of more than 50 % in all the trajectories (Supporting Information Table T3). The binding of SUMO1, SUMO2 and SUMO3 to both SENP1 and SENP2 is also driven by hydrophobic interactions and electrostatic interactions. As can be seen from Figure 7, Arg63 of SUMO1 and the corresponding Arg58 in SUMO2 and SUMO3 contribute more than 6 kcal/mol to the binding free energy. One more noticeable contribution is from Tyr91 of SUMO1 and Phe86 of SUMO2 and SUMO3. The Trp512 of SENP1 and the corresponding Trp457 of SENP2 make strong hydrophobic contacts with Tyr91 of SUMO1 and similar Phe86 of SUMO2 and SUMO3. Other significant contributions belong to C-terminal of SUMOs and correspond to the catalytic site.
3.4 In Silico Alanine Mutagenesis

In silico alanine scanning mutagenesis estimates the contribution of each residue to binding free energy by mutating each residue at the protein-protein interface to alanine and recalculates the binding free energy over the mutant trajectory. In silico alanine scanning is similar to experimental alanine scanning which is generally used to identify to protein-protein interaction hotspot.[67,68] In this study, MMPB(GB)SA based alanine scanning approach was used to
www.molinf.com

Mol. Inf. 2013, 32, 267 280

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

275

Full Paper

A. Kumar, K. Y. J. Zhang

Figure 6. The contribution of each residue in SENP1/SENP2 to binding free energy calculated by MM-GBSA binding free energy decomposition analysis for (A) SENP1-SUMOs and (B) SENP2-SUMOs.

Figure 7. The contribution of each residue in SUMO1/SUMO2/SUMO3 to binding free energy calculated by MM-GBSA binding free energy decomposition analysis for (A) SENP1-SUMOs and (B) SENP2-SUMOs.

systematically screen SENP-SUMO interface and thereby providing structural and energetic consequences of mutations. The in silico alanine approach generally does not account for absolute binding free energy changes due to single or multiple mutations in a protein. However, this ap276
www.molinf.com

proach is very good in estimating the relative changes and depending on the system investigated, good qualitative agreement has been found between experiments and computations.[6972]

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Mol. Inf. 2013, 32, 267 280

SENP:SUMO Protein-Protein Interaction for Structure Based Drug Design

Figure 8. In silico alanine mutagenesis for SENP : SUMO complexes using single trajectory method of MM-PB(GB)SA alanine scanning. (A) MM-PBSA changes for SENP1 : SUMOs (B) MM-PBSA changes for SENP2 : SUMOs (C) MM-GBSA changes for SENP1 : SUMOs and (D) MM-GBSA changes for SENP2 : SUMOs.

The 30 interface residues of SENP1 and 26 interface residues of SENP2 were subjected to alanine scanning. The results of in silico alanine scanning mutagenesis for all six SENP:SUMO protein systems are presented in Figure 8 and the positive and negative values of DDG indicate the unfavorable and favorable contributions respectively. We can see by comparing Figure 8 and Figure 6 that binding free energy decomposition calculation and in silico alanine scanning mutagenesis give comparable results. Overall, in silico alanine scanning profile is very similar for all the six SENP:SUMO complexes. According to both the MM-GBSA and MM-PBSA calculations, the mutation of Asp468, Phe496, Arg511 and Trp512 in SENP1 and corresponding residues Asp413, Phe441, Arg456 and Trp457 in SENP2 to alanine significantly affected the SENP:SUMO binding affinity. In silico alanine scanning calculation revealed Asp468 and Arg511 in SENP1 and Asp413 and Arg456 in SENP2 are the most important residues for SENP:SUMO complex formation. Mutation of these residues to alanine leads to significant decrease in the binding free energy by a value of more than 10 kcal/mol in all the cases.
Mol. Inf. 2013, 32, 267 280

3.5 Overlap of Ligand Binding Hotspots Identified by Different Methods

The pocket prediction, fragment mapping, binding free energy decomposition and computational alanine scanning outlined ligand binding hotspots on the SUMO binding interface of SENP. To determine the overlap between different methods, amino acid residue overlap frequencies (ROf) were calculated by counting the number of occurrences of SENP:SUMO interface residues at ligand binding hotspot according to each method. The residues of SENPs that are within 5 from SUMO in SENP:SUMO complex are considered as interface residues. Metapocket predicted top ranked pocket residues and the residue within 4 of FTMap CSs site were used to calculate residue propensities. A cutoff of > 4 kcal/mol was used as criteria for a residue to provide significant contribution to binding free energy. For computational alanine scanning, mutated residue that resulted in 4.0 kcal/mol change in MM-PB(GB)SA binding free energy was considered as hotspot residue. Figure 9 presents the histogram obtained by plotting residue overlap frequencies for SENP : SUMO interface residues. As can
www.molinf.com

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

277

Full Paper

A. Kumar, K. Y. J. Zhang

Figure 9. Residue overlap frequencies (ROf) for (A) SENP1 and (B) SENP2 calculated as a means to determine the overlap of different hotspot prediction methods. The prediction methods include pocket prediction, fragment mapping, MM-GBSA decomposition for SENPSUMO1/2/3, MM-GBSA alanine scanning for SENP-SUMO1/2/3 and MM-PBSA alanine scanning for SENP-SUMO1/2/3.

be seen in Figures 9A and 9B, the highest ROf were found for Asp468, Phe496, Arg511 and Trp512 of SENP1 and the corresponding Asp413, Phe441, Arg456 and Trp457 of SENP2. These residues were consistently identified as hotspot residues by different computational methods and are crucial for the formation of SENP : SUMO complex.
3.6 Implication for Drug Design Targeting SENP:SUMO Protein-Protein Interface

A number of computational approaches can be used to identify inhibitors of SENP:SUMO interaction. One approach is pharmacophore modeling that has been successfully used to discover small molecule protein-protein interaction inhibitors (SMPPIIs).[73] Molecular dynamics simulation and binding free energy decomposition studies resulted in the identification of residues of SUMO1, SUMO2 and SUMO3 that contribute significantly to the formation of SENP:SUMO protein-protein complex. As can be seen from Figure 7 significant contributions to binding free energy originate from Arg63 of SUMO1 and Arg58 of SUMO2 and SUMO3. Also, Glu67 and Tyr91 of SUMO1 and Asp62 and Phe86 of SUMO2 and SUMO3 contribute to binding free energy via strong electrostatic and hydrophobic interactions with Asp468, Phe496, Arg511 and Trp512 of SENP1 and the corresponding Asp413, Phe441, Arg456 and Trp457 of SENP2. These pharmacophoric features in these residues can be used to derive pharmacophore models which can be further used to screen small molecule databases such as 278
www.molinf.com

ZINC.[74, 75] We have attempted to make one such pharmacophore that can be utilized to screen small molecular databases for future drug discovery efforts. A pharmacophore model generated using MOE (Molecular Operating Environment (MOE), version 2011.10; Chemical Computing Group Inc.: Montreal, Quebec, Canada, 2010) is shown in Figure 10A and consists of one hydrogen bond acceptor feature, two hydrogen bond donor features and 4 hydrophobic/aromatic features. These pharmacophoric features correspond to key molecular interactions formed by Arg63, Glu67 and Tyr91 of SUMO1 and Arg58, Asp62 and Phe86 of SUMO2 and SUMO3. The hydrophobic or aromatic feature resembles the sidechain of Tyr91 of SUMO1 and Phe86 of SUMO2 and SUMO3. To check whether the developed pharmacophore can retrieve compounds with matching features, we have used this pharmacophore to screen ~ 4 million compounds from Namiki Shoji collection (www.namikis.co.jp). About 3325 hit molecules were obtained that match at least five features of the pharmacophore. The pharmacophore with one of the hit molecules mapped on it is shown in Figure 10B.

3 Conclusions
Inhibition of SENPs offers excellent therapeutic value due to their critical role in cancer development. Up till now the inhibitor development efforts were focused towards the catalytic site which is tricky because of the cysteine proMol. Inf. 2013, 32, 267 280

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

SENP:SUMO Protein-Protein Interaction for Structure Based Drug Design

Figure 10. Pharmacophore modeling using the key molecular interactions formed by SUMO paralogues (ribbon representation) with SENP (surface representation) identified using binding free energy decomposition and in silico alanine mutagenesis. (A) Pharmacophoric features. Magenta sphere corresponds to hydrogen bond donor, cyan corresponds to hydrogen bond acceptor and green/brown sphere corresponds to Aromatic/hydrophobic feature (B) One hit compound matching 5 out of 7 pharmacophoric features.

tease active site architecture and its preference for electrophilic carbonyl group in the binding ligand. In the present work, we have identified an alternative site as a target for structure based drug design using various computational methods including pocket prediction, computational fragment mapping, molecular dynamics simulation, and in silico alanine scanning. Our study highlighted the ligand binding and energetic hotspot regions on SENP:SUMO protein-protein interface that can be targeted. Different SUMO paralogues interact with SENP1 and SENP2 via both hydrogen bonding and hydrophobic interaction. The amino acid residues Asp468, Phe496, Arg511 and Trp512 in SENP1 and the corresponding Asp413, Phe441, Arg456 and Trp457 in SENP2 make crucial interactions with Arg63, Glu67 and Tyr91 of SUMO1 and Arg58, Asp62 and Phe86 of SUMO2 and SUMO3 respectively. The results of binding free energy decomposition studies also showed that these residues contribute significantly to the binding free energy and are crucial for SENP:SUMO interaction. The significance of these residues for SENP:SUMO interaction is also displayed by in silico alanine scanning mutagenesis. The MM-PB(GB)SA binding free energy significantly decreases upon the mutation of these amino acid residues to alanine. Using small molecule inhibitors to disrupt SENP : SUMO interaction can be a good therapeutic strategy. A number of approaches can be used to identify SMPPIIs of SENP : SUMO interaction. One starting point is the virtual screening of available chemicals which can be assayed for the inhibition of SENP : SUMO protein-protein interaction. Taking advantage of these key interacting residues identified in the SENP : SUMO interface, a pharmacophore modeling approach can be used to screen for compounds that mimic these critical interactions.

Acknowledgements
We thank RIKEN Integrated Cluster of Clusters (RICC) at RIKEN for the supercomputing resources used for this study. We thank Dr. Minoru Yoshida and Dr. Akihiro Ito for stimulating discussions. We thank members of our lab for help and discussions. We acknowledge the Initiative Research Unit Program from RIKEN, Japan for funding.

References
[1] D. Baba, N. Maita, J. G. Jee, Y. Uchimura, H. Saitoh, K. Sugasawa, F. Hanaoka, H. Tochio, H. Hiroaki, M. Shirakawa, Nature 2005, 435, 979. [2] R. Geiss-Friedlander, F. Melchior, Nat. Rev. Mol. Cell Biol. 2007, 8, 947. [3] E. S. Johnson, Ann. Rev. Biochem. 2004, 73, 355. [4] B. Palancade, V. Doye, Tr. Cell Biol. 2008, 18, 174. [5] A. D. Capili, C. D. Lima, Curr. Opin. Struct. Biol. 2007, 17, 726. [6] D. Mukhopadhyay, M. Dasso, Tr. Biochem. Sci. 2007, 32, 286. [7] R. T. Hay, Tr. Cell Biol. 2007, 17, 370. [8] M. Drag, G. S. Salvesen, IUBMB Life 2008, 60, 734. [9] Y. Zuo, J.-K. Cheng, Asian J. Androl. 2009, 11, 36. [10] E. T. H. Yeh, J. Biol. Chem. 2009, 284, 8223. [11] T. Bawa-Khalfe, J. Cheng, S. H. Lin, M. M. Ittmann, E. T. Yeh, J. Biol. Chem. 2010, 285, 25859. [12] J. Cheng, X. Kang, S. Zhang, E. T. Yeh, Cell 2007, 131, 584. [13] T. Yamaguchi, P. Sharma, M. Athanasiou, A. Kumar, S. Yamada, M. R. Kuehn, Mol. Cell Biol. 2005, 25, 5171. [14] J. Cheng, T. Bawa, P. Lee, L. Gong, E. T. Yeh, Neoplasia 2006, 8, 667. [15] C. Jacques, O. Baris, D. Prunier-Mirebeau, F. Savagner, P. Rodien, V. Rohmer, B. Franc, S. Guyetant, Y. Malthiery, P. Reynier, J. Clin. Endocrinol. Metab. 2005, 90, 2314. [16] Y. Xu, J. Li, Y. Zuo, J. Deng, L. S. Wang, G. Q. Chen, Cancer Lett. 2011, 309, 78.

Mol. Inf. 2013, 32, 267 280

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

www.molinf.com

279

Full Paper
[17] J. Hemelaar, A. Borodovsky, B. M. Kessler, D. Reverter, J. Cook, N. Kolli, T. Gan-Erdene, K. D. Wilkinson, G. Gill, C. D. Lima, H. L. Ploegh, H. Ovaa, Mol. Cell Biol. 2004, 24, 84. [18] A. Borodovsky, H. Ovaa, W. J. Meester, E. S. Venanzi, M. S. Bogyo, B. G. Hekking, H. L. Ploegh, B. M. Kessler, H. S. Overkleeft, ChemBioChem 2005, 6, 287. [19] V. E. Albrow, E. L. Ponder, D. Fasci, M. Bekes, E. Deu, G. S. Salvesen, M. Bogyo, Chem. Biol. 2011, 18, 722. [20] Z. Qiao, W. Wang, L. Wang, D. Wen, Y. Zhao, Q. Wang, Q. Meng, G. Chen, Y. Wu, H. Zhou, Bioorg. Med. Chem. Lett. 2011, 21, 6389. [21] C. Dobrota, D. Fasci, N. D. Hadade, G. D. Roiban, C. Pop, V. M. Meier, I. Dumitru, M. Matache, G. S. Salvesen, D. P. Funeriu, ChemBioChem 2012, 13, 80. [22] L. Shen, M. H. Tatham, C. Dong, A. Zagorska, J. H. Naismith, R. T. Hay, Nat. Struct. Mol. Biol. 2006, 13, 1069. [23] L. N. Shen, C. Dong, H. Liu, J. H. Naismith, R. T. Hay, Biochem. J. 2006, 397, 279. [24] Z. Xu, S. F. Chau, K. H. Lam, H. Y. Chan, T. B. Ng, S. W. Au, Biochem. J. 2006, 398, 345. [25] D. Reverter, C. D. Lima, Structure 2004, 12, 1519. [26] D. Reverter, C. D. Lima, Nat. Struct. Mol. Biol. 2006, 13, 1060. [27] M. A. Larkin, G. Blackshields, N. P. Brown, R. Chenna, P. A. McGettigan, H. McWilliam, F. Valentin, I. M. Wallace, A. Wilm, R. Lopez, J. D. Thompson, T. J. Gibson, D. G. Higgins, Bioinformatics 2007, 23, 2947. [28] P. Gouet, E. Courcelle, D. I. Stuart, F. Metoz, Bioinformatics 1999, 15, 305. [29] A. Dereeper, V. Guignon, G. Blanc, S. Audic, S. Buffet, F. Chevenet, J. F. Dufayard, S. Guindon, V. Lefort, M. Lescot, J. M. Claverie, O. Gascuel, Nucleic Acids Res. 2008, 36, W465. [30] B. Huang, OMICS 2009, 13, 325. [31] Z. Zhang, Y. Li, B. Lin, M. Schroeder, B. Huang, Bioinformatics 2011, 27, 2083. [32] R. Brenke, D. Kozakov, G. Y. Chuang, D. Beglov, D. Hall, M. R. Landon, C. Mattos, S. Vajda, Bioinformatics 2009, 25, 621. [33] T. Darden, D. York, L. Pedersen, J. Chem. Phys. 1993, 98, 10089. [34] J.-P. Ryckaert, G. Ciccotti, H. J. C. Berendsen, J. Comput. Phys. 1977, 23, 327. [35] D. A. Case, T. Darden, T. E. Cheatham, C. Simmerling, J. Wang, R. E. Duke, R. Luo, M. Crowley, R. Walker, W. Zhang, K. M. Merz, B. Wang, S. Hayik, A. Roitberg, G. Seabra, I. Kolossvry, K. F. Wong, F. Paesani, J. Vanicek, X. Wu, S. R. Brozell, T. Steinbrecher, H. Gohlke, L. Yang, C. Tan, J. Mongan, V. Hornak, G. Cui, D. H. Mathews, M. G. Seetin, C. Sagui, V. Babin, P. A. Kollman, Amber 10 Users Manual, University of California, San Francisco, CA 2008. [36] J. Srinivasan, T. E. Cheatham, P. Cieplak, P. A. Kollman, D. A. Case, J. Am. Chem. Soc. 1998, 120, 9401. [37] P. A. Kollman, I. Massova, C. Reyes, B. Kuhn, S. Huo, L. Chong, M. Lee, T. Lee, Y. Duan, W. Wang, O. Donini, P. Cieplak, J. Srinivasan, D. A. Case, T. E. Cheatham 3rd, Acc. Chem. Res. 2000, 33, 889. [38] D. Bashford, D. A. Case, Ann. Rev. Phys.Chem. 2000, 51, 129. [39] V. Tsui, D. A. Case, Biopolymers 2000, 56, 275. [40] T. Simonson, Curr. Opin. Struct. Biol. 2001, 11, 243. [41] J. Weiser, P. S. Shenkin, W. C. Still, J. Comput. Chem. 1999, 20, 217. [42] A. Dereeper, V. Guignon, G. Blanc, S. Audic, S. Buffet, F. Chevenet, J.-F. Dufayard, S. Guindon, V. Lefort, M. Lescot, J.-M. Claverie, O. Gascuel, Nucleic Acids Res. 2008, 36, W465. [43] K. M. Bohren, V. Nadkarni, J. H. Song, K. H. Gabbay, D. Owerbach, J. Biol. Chem. 2004, 279, 27233.

A. Kumar, K. Y. J. Zhang [44] D. Owerbach, E. M. McKay, E. T. Yeh, K. H. Gabbay, K. M. Bohren, Biochem. Biophys. Res. Commun. 2005, 337, 517. [45] S. Perot, O. Sperandio, M. A. Miteva, A. C. Camproux, B. O. Villoutreix, Drug Discov. Today 2010, 15, 656. [46] S. Henrich, O. M. Salo-Ahen, B. Huang, F. F. Rippmann, G. Cruciani, R. C. Wade, J. Mol. Recogn. 2010, 23, 209. [47] S. Soga, H. Shirai, M. Kobori, N. Hirayama, J. Chem. Inf. Model. 2007, 47, 400. [48] C. Yan, F. Wu, R. L. Jernigan, D. Dobbs, V. Honavar, Protein J. 2008, 27, 59. [49] T. Clackson, J. Wells, Science 1995, 267, 383. [50] A. A. Bogan, K. S. Thorn, J. Mol. Biol. 1998, 280, 1. [51] C. Tintori, N. Veljkovic, V. Veljkovic, M. Botta, Proteins 2010, 78, 3396. [52] T. Venken, D. Daelemans, M. De Maeyer, A. Voet, Proteins 2012, 80, 1633. [53] O. A. Adekoya, N. P. Willassen, I. Sylte, J. Biomol. Struct. Dyn. 2005, 22, 521. [54] O. A. Adekoya, N. P. Willassen, I. Sylte, J. Struct. Biol. 2006, 153, 129. [55] H. Gohlke, D. A. Case, J. Comput. Chem. 2004, 25, 238. [56] A. Metz, C. Pfleger, H. Kopitz, S. Pfeiffer-Marek, K. H. Baringhaus, H. Gohlke, J. Chem. Inf. Model. 2012, 52, 120. [57] Q. Cui, T. Sulea, J. D. Schrag, C. Munger, M. N. Hung, M. Naim, M. Cygler, E. O. Purisima, J. Mol. Biol. 2008, 379, 787. [58] S. Wong, R. E. Amaro, J. A. McCammon, J. Chem. Theory Comput. 2009, 5, 422. [59] N. Bharatham, S. W. Chi, H. S. Yoon, PLoS One 2011, 6, e26014. [60] P. Nimmanpipug, C. Khampa, V. S. Lee, S. Nangola, C. Tayapiwatana, J. Mol. Graph. Model. 2011, 31, 65. [61] D. Spiliotopoulos, A. Spitaleri, G. Musco, PLoS One 2012, 7, e46902. [62] N. Basdevant, H. Weinstein, M. Ceruso, J. Am. Chem. Soc. 2006, 128, 12766. [63] D. J. Cole, E. Rajendra, M. Roberts-Thomson, B. Hardwick, G. J. McKenzie, M. C. Payne, A. R. Venkitaraman, C.-K. Skylaris, PLoS Comput. Biol. 2011, 7, e1002096. [64] N. G. Azoia, M. M. Fernandes, N. M. Micalo, C. M. Soares, A. Cavaco-Paulo, Proteins: Struct. Funct. Bioinf. 2012, 80, 1409. [65] J. Mikolajczyk, M. Drag, M. Bks, J. T. Cao, Z. e. Ronai, G. S. Salvesen, J. Biol. Chem. 2007, 282, 26217. [66] N. Kolli, J. Mikolajczyk, M. Drag, D. Mukhopadhyay, N. Moffatt, M. Dasso, G. Salvesen, K. D. Wilkinson, Biochem. J. 2010, 430, 335. [67] O. Keskin, B. Ma, R. Nussinov, J. Mol. Biol. 2005, 345, 1281. [68] A. Whitty, G. Kumaravel, Nat. Chem. Biol. 2006, 2, 112. [69] I. S. Moreira, P. A. Fernandes, M. J. Ramos, J. Comput. Chem. 2007, 28, 644. [70] H. Zou, C. Luo, S. Zheng, X. Luo, W. Zhu, K. Chen, J. Shen, H. Jiang, J. Phys. Chem. B 2007, 111, 9104. [71] V. Zoete, O. Michielin, Proteins: Struct. Funct. Bioinf. 2007, 67, 1026. [72] R. T. Bradshaw, B. H. Patel, E. W. Tate, R. J. Leatherbarrow, I. R. Gould, Protein Eng. Des. Select. 2011, 24, 197. [73] A. Voet, K. Zhang, Curr. Pharm. Des. 2012, 18, 4586. [74] J. J. Irwin, B. K. Shoichet, J. Chem. Inf. Model. 2005, 45, 177. [75] J. J. Irwin, T. Sterling, M. M. Mysinger, E. S. Bolstad, R. G. Coleman, J. Chem. Inf. Model. 2012, 52, 1757. Received: October 19, 2012 Accepted: February 5, 2013 Published online: March 11, 2013

280

www.molinf.com

 2013 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim

Mol. Inf. 2013, 32, 267 280

You might also like