Disruption of protein–protein interactions: hot spot detection, structure-based virtual screening and in vitro testing for the anti-cancer drug target – survivin

Abstract

Survivin is a member of the inhibitor of the apoptosis (IAP) family of proteins, and plays a crucial role in both cell division and apoptosis. As it is overexpressed in many human solid tumors, it has become an attractive drug target for cancer therapy. Survivin is involved in protein–protein interactions (PPI) with several of its substrate proteins, and disruption of these interactions could be a possible means to target the function of survivin in cancer. To this end, we sought and were able to detect hot spot residues in the survivin dimer and Chromosomal Passenger Complex (CPC; survivin/borealin/INCENP) using simple knowledge based physical models implemented in the Robetta server (http://robetta.bakerlab.org/), KFC server (http://kfc.mitchell-lab.org/) and the HotRegion database (http://prism.ccbb.ku.edu.tr/hotregion/). Then, extensive molecular dynamics simulations were applied to generate an ensemble of conformations and were used to quantitatively estimate the binding free energy of the identified hot spot residues using MM-PBSA alanine scanning mutagenesis and per-residue energy decomposition. Based on the frequency of occurrence of the hot spot residues and the estimated binding free energy, the survivin dimer and CPC interface residues were designated as “hot spots” and “warm spots”. Finally, based on the identified hot spots of survivin (Leu6A, Trp10A, Leu98A, Phe101A, Asp105A, and Arg106A), a pharmacophore model was derived and used to virtually screen database compounds to identify indinavir as potential inhibitor that could target survivin PPI. A preliminary biochemical investigation shows that the treatment of MDA-MB-231 breast cancer cells with indinavir resulted in Aurora B and XIAP downregulation and caspase-3 activation, hallmarks of survivin PPI inhibition.

---

Introduction

Protein–protein interactions (PPIs) play a major role in many biological functions, including gene expression, cell growth and morphology, nutrient uptake, signalling networks, and apoptosis.¹ One report estimated the total PPI in a single cell to number around 6 500 000, and these interactions are very specific for a particular biological function.² Various advances in molecular and cell biology have illuminated the function and roles of these PPI in both normal and diseased states. Growing evidence clearly implicates deregulation of PPIs in various disease conditions such as cancer and neurodegenerative disease.³ As a result, deregulated PPIs have become an alternate drug target to that of the traditional targets (enzymes, receptors, ion channels and transporters) for various diseases, including cancer.^4–6 Developing small molecule inhibitors for these PPIs is pursued both in pharmaceutical companies and also in academia. However, rational design of small molecule PPI inhibitors is difficult due to incomplete structural and functional knowledge of the PPIs. Furthermore, designing small molecule inhibitors for PPIs is more challenging because of flat and featureless large hydrophobic interface regions.^4,7,8 Nevertheless, growing evidence indicate that PPIs are suitable targets for therapeutic intervention, and several small molecule inhibitors of PPIs have been recently reported.^9–12 Hence, an understanding of the structural and functional importance of PPIs in the diseased state could help in the design of small molecule inhibitors of PPIs as effective therapeutic intervention.¹³

Understanding the structural aspects of PPI requires identification of the interface residues that contribute specificity and selectivity to their partner proteins. Most of the bimolecular interaction (including protein–protein, protein–receptor and protein–ligand interactions) depends on only the few clustered residues which actually contribute to the majority of the binding free energy for the complex formation. A residue or a subset of residues in PPI interface region that plays a crucial role in accounting for the high binding free energy are called “hot spots”.^13–15 Detection of such hot spot residues in deregulated PPIs has become an important task for drug designers, which could eventually help in the structure-based design of therapeutics for various diseases.¹⁶ Moreover, it is ascertained that hot spots are evolutionarily conserved and have a high propensity for ligand binding.¹⁷ Hence, targeting these residues could disable the crucial interactions of PPIs in many diseases.

A number of experimental and computational methods have been reported for the detection of the hot spot residues in PPIs. Alanine scanning mutagenesis (ASM) has been shown to accurately predict hot spot residues in PPIs through systematic mutation of selected residues to alanine.^18,19 Upon mutation, the binding free energy of the mutated residue is recalculated using thermodynamic properties. If the binding free energy of the mutant protein drops by two or more kcal mol⁻¹ compared to the native protein, then the mutated residue is considered as hot spot, and assumed to be vital for the protein structure and function. However, high throughput screening of PPIs by experimental ASM is very expensive and time consuming; moreover, it is very difficult to mutate all the interface residues and determine the binding free energy. For these reasons, various computational methods have been developed to predict the hot spots at PPI interface regions.^20–24 These tools use different approaches and parameter optimization methods to predict the hot spots, and can be classified as simple free energy based methods,^20,23 molecular dynamics based methods²¹ and machine learning approaches.²² These in silico tools are fast, reliable, and successfully used by various scientific groups.^25–29

Survivin is an IAP family protein and plays a multifunctional role in cell division and apoptosis by interacting with its substrate proteins, including Aurora B, INCENP, borealin, CDK4, heat shock proteins (HSP), X-linked inhibitor of apoptosis protein (XIAP), caspase enzymes, and other proteins.^30–35 Over expression of survivin has been observed in many of human solid tumors, but not in normal adult cells.^36–38 Its expression level and activity correlates with cancer recurrence, transformation, resistance to chemo- and radiation therapy; and is associated with shorter patient survival time. Hence, survivin has become a universal drug target for small molecule inhibitors, which could disrupt the survivin interaction with other substrate proteins, and cause mitotic arrest and cell death.^39,40

Numerous survivin inhibitors have been identified using various in vitro and in silico methods, including antisense oligonucleotides, ribozymes, small interfering RNAs, dominant-negative mutant, cancer vaccine, and small-molecule inhibitors.^40–43 Most of the reported small molecule inhibitors were developed to inhibit survivin expression. For example, the reported small molecule inhibitor, YM155 (ref. 44 and 45) inhibits survivin expression level by binding to its transcription factor site (SP1). Unfortunately, YM155 was found to have only modest efficacy on human cancers in phase I and phase II clinical trials. Another reported inhibitor, FL118,⁴⁶ also selectively inhibits survivin promoter activity and diminishes its expression level. Antisense oligonucleotide LY2181308 (ref. 47 and 48) inhibits survivin expression level by binding to its mRNA but therapeutic evaluation of this compound alone and in combination with docetaxel for prostate cancer revealed no beneficial effect. However, only a few small molecule inhibitors^49–51 (S12, LLP3, LLP9 and LQZ-7F, Fig. 1) were reported to interrupt survivin functions by directly binding to its PPI interface (survivin dimer & CPC complex interface) regions. Thus, new survivin inhibitors which could disrupt survivin protein functions by interfering with the PPIs are urgently required.

Fig. 1 Reported survivin inhibitors.

Survivin is involved in the regulation of cell division (chromosome alignment, histone modification, and cytokinesis) through the formation of chromosomal passenger complex (CPC: a complex of survivin, borealin, and INCENP protein); moreover, survivin dimer is reported to be involved in an anti-apoptotic mechanism.^30,52 In order to disrupt the CPC protein and survivin dimer PPIs, the hot spot residues at their interface regions were identified using simple free energy based physical methods like Robetta server (http://robetta.bakerlab.org/), KFC server (knowledge-based FADE and contacts; http://kfc.mitchell-lab.org/) and HotRegion database (http://prism.ccbb.ku.edu.tr/hotregion/). To confirm the identified residues as hot spots, molecular dynamics (MD) simulation was performed to quantitatively measure the relative binding free energy of mutant as well as the native protein using in silico alanine scanning mutagenesis. The results helped us to describe the energetically important hot spot residues that contribute to the specificity and affinity in the PPI complex formation. Furthermore, a structure-based pharmacophore model was derived based on the identified hot spots and used for virtual screening to identify indinavir as potential hit for disruption of survivin PPIs. Thus, the hot spots identified in the computational study could serve as a starting point for structure-based drug design. The methodology followed in this work is represented in Fig. 2.

Fig. 2 Work flow depicting the methodology adopted for hot spot prediction and PPI inhibitor identification for the anti-cancer drug target – survivin.

Results and discussion

Detection of hot spots using online servers

Identification of the hot spot residues at the PPI interface using computational methods identifies the residues which directly impart affinity and specificity to their partner proteins, and could assist in understanding the structural and functional role played by these hot spot residues. This information could be exploited in the rational design and development of drugs. As mentioned above, survivin is overexpressed in various human solid tumors but undetected in normal adult tissues, and thus targeting survivin function by developing PPI inhibitors could be an effective treatment modality for cancer treatment. With this goal in mind, here we have used computational methods to detect the hot spot residues in survivin homo dimer (PDB ID: 3UIH), and Chromosomal Passenger Complex (PDB ID: 2QFA), which includes survivin, borealin and INCENP proteins.

To detect the hot spot residue in survivin dimer and CPC, three freely available web-based services, including Robetta server, KFC server and HotRegion database were used. As predicting the hot spots using a single method might give inaccurate results, multiple methods were used to improve the result accuracy. All the three methods use their own set of characteristic functions and automatically identify the hot spot residues at PPI interface region. The Robetta and KFC servers implement in silico alanine scanning mutagenesis and machine learning approaches, respectively; whereas the HotRegion database implements experimentally derived hot spot information and protein structural properties (residues pair potential, solvent accessible surface area of monomer and complex) to identify the hot spot in a given protein complex structure. All hot spot predictions were subjected to further investigation using molecular dynamics simulations, to evaluate its relative binding free energy contribution for complex formation. The predicted hot spots of survivin dimer and CPC using the above mentioned three methods are given in ESI Tables S1 and S2,† respectively.

For survivin dimer, seven residues (Leu6, Trp10, Phe93, Leu98, Phe101, Leu102 and Asp105) were predicted as hot spots; whereas in CPC complex, 21 survivin residues (Leu6, Trp10, Phe13, Phe93, Leu96, Leu98, Phe101, Leu102, Asp105, Arg106, Arg108, Lys110, Ile113, Thr117, Lys120, Lys121, Phe124, Val131, Arg132, Ile135 and Leu138), were predicted as hot spots. Among them, seven residues – Leu6, Trp10, Phe93, Leu98, Phe101, Leu102 and Asp105 – are the common hot spot residues in survivin dimer and CPC complex.

Per-residue energy decomposition analysis and in silico alanine scanning mutagenesis (ASM) to detect hot spots

Normally, proteins are dynamic in nature and adopt different conformations, depending on their function. Thus, using different conformations of the protein would help to identify the hot spots more accurately instead of using a single crystal structure conformation. Thus, to obtain different conformations of the protein complex, 50 ns molecular dynamics simulations were carried out using SANDER module of the Amber software suite. The obtained conformers were used to determine the relative binding free energy contribution of the previously identified hot spot residues using per-residue energy decomposition and in silico alanine scanning mutagenesis analysis. These methods use a well-defined approximate MM-PBSA (Molecular Mechanics – Posisson Boltzmann Solvent Accessible) method to derive the relative binding free energy changes of native protein as well as mutant protein. The results obtained using these methods usually correlate well with experimental results and are widely employed by many scientific groups.^53–56

After a 50 ns molecular dynamics simulation, the obtained trajectory snapshots (ensemble of conformations) were used for analysis using the CPPTRAJ and PTRAJ modules. Then, the stability of survivin dimer and CPC proteins were analyzed by measuring the root mean square deviation (RMSD) of the backbone atoms from their initial coordinates. The RMSD of survivin dimer initially increased from 2 to 6 Å for a period of 16 ns, and then maintained stability up to 28 ns. Subsequently, the protein RMSD increased to 9 Å due to conformational changes in C-α6 helix region of survivin. Afterwards, it maintained stability throughout the simulation with minor fluctuations; whereas the CPC protein RMSD increased from 1 to 2.3 Å for a period of 4.5 ns and then stabilized for the remaining period of simulation. The RMSD plots of both complexes are shown in Fig. 3. From the RMSD plot, it is clear that the CPC complex is much more stable than the survivin homo dimer. This could be due to larger binding interface for CPC protein, as compared to survivin homo dimer.^30,57 This is further confirmed from binding free energy decomposition analysis.

Fig. 3 Protein backbone root mean square deviation (RMSD) graph for survivin dimer (black) and CPC protein (red) during 50 ns molecular dynamics simulation.

After the molecular dynamics simulation, the last 4 ns trajectory snapshots (4000 frames) of survivin dimer and CPC proteins were subjected to binding free energy decomposition analysis using MM-PBSA method. The estimated total binding free energy and the component energies for survivin dimer and CPC protein are shown in Tables 1 and 2, respectively. The estimated total binding free energy values of survivin dimer and CPC protein are −56.2280 and −186.0680 kcal mol⁻¹, respectively. It should be noted that the binding free energy values of survivin dimer and CPC protein do not represent the absolute binding free energy, as entropic terms are excluded from the calculation. Therefore, from these binding free energy values we can summarize that the CPC protein is much more stable compared to survivin homo dimer, which may be due to the extended interface region with borealin and INCENP protein.

Table 1 Binding free energy and its components derived from MD simulation for survivin dimer

Energy components^a	Complex kcal mol^−1	Receptor (A chain) kcal mol^−1	Ligand (B chain) kcal mol^−1	ΔGBIND kcal mol^−1 (complex – receptor-ligand)
a E VDWAALS = van der Waals energy contribution calculated from MM force field; EELE = electrostatic energy contribution calculated from MM force field; EPB = electrostatic contribution to the solvation free energy calculated by PB; ECAVITY = nonpolar energy contribution to the solvation free energy calculated by an empirical model; GGAS = average interaction energy of complex, receptor and ligand in gas phase; GSOLVENT = average interaction energy of complex, receptor and ligand in solvent; ΔGBIND = binding free energy difference.
E VDWAALS	−1818.9236	−892.2472	−861.1856	−65.4907
E ELE	−21 695.8046	−10 716.7287	−10 931.2786	−47.7973
E PB	−5088.1170	−2674.9043	−2476.6614	63.4488
E CAVITY	97.6908	52.2123	51.8672	−6.3887
G GAS	−23 514.7282	−11 608.9760	−11 792.4642	−113.2880
G SOLVENT	−4990.4262	−2622.6920	−2424.7942	57.0600
Total	−28 505.1543	−14 231.6680	−14 217.2584	−56.2280

---

Table 2 Binding free energy and its components derived from MD simulation for CPC protein

Energy components^a	Complex kcal mol^−1	Receptor (A chain) kcal mol^−1	Ligand (BC chain) kcal mol^−1	ΔGBIND kcal mol^−1 (complex – receptor–ligand)
a E VDWAALS = van der Waals energy contribution calculated from MM force field; EELE = electrostatic energy contribution calculated from MM force field; EPB = electrostatic contribution to the solvation free energy calculated by PB; ECAVITY = nonpolar energy contribution to the solvation free energy calculated by an empirical model; GGAS = average interaction energy of complex, receptor and ligand in gas phase; GSOLVENT = average interaction energy of complex, receptor and ligand in solvent; ΔGBIND = binding free energy difference.
E VDWAALS	−1892.1252	−916.3650	−766.0449	−209.7153
E EEL	−17 775.4636	−10 487.6946	−6832.7839	−454.9851
E EPB	−5181.5682	−2985.4815	−2698.0310	501.9443
E ECAVITY	82.1482	54.5359	50.9241	−23.3118
G GAS	−19 667.5888	−11 404.0596	−7598.8288	−664.7004
G SOLVENT	−5099.4200	−2930.9456	−2647.1069	478.6325
Total	−24 767.0088	−14 335.0052	−10 245.9356	−186.0680

---

Once the total binding free energy of the complex was calculated, per-residue energy decomposition using the MM-PBSA method was performed. This reveals how the individual residues of the native protein contribute to the total binding free energy of the complex. The identified hot spot per-residue energy values of survivin dimer and CPC proteins are depicted in Fig. 4. It should be noted that the positive and negative per-residue energy values represent the unfavorable and favorable energy contribution, respectively towards complex formation. It is clear that the hot spot residues of survivin dimer, Trp10A, Leu98A and Leu102 (from chain A), Trp10B and Leu98B (from chain B) are contributing more than −2 kcal mol⁻¹ binding free energy. While, the remaining interface residues: Leu6, Phe93A, (from chain A), and Phe93B, Phe101B, Leu102B (from chain B) are consistently contributing ∼−1 kcal mol⁻¹ binding free energy. Whereas in the CPC protein (Fig. 4B), the survivin hot spot residues Leu6A, Trp10A, Phe101A, Leu102A, Arg106A, Arg108A, Ile113A, Thr117A, Lys120A, Val131A, Arg132A, Ile135A and Leu138A are contributing more than −2 kcal mol⁻¹ binding free energy for the complex formation. These results suggest that the identified hot spot residues inherently contribute to the binding free energies for protein structural stability.

Fig. 4 Hot spot residue-wise energy contribution for the complex. (A) Survivin dimer hot spot residues energy contribution and (B) CPC protein hot spot residues energy contribution (contribution from borealin and INCENP hot spot residues are shown in ESI Fig. S1†).

Following the total binding free energy and per-residue energy decomposition analysis, in silico alanine scanning mutagenesis (ASM) was carried out to determine if the identified hot spot residues contribute significant binding free energy to the protein structural stability. This was carried out by mutating the individual hot spot residues of survivin dimer and CPC protein to alanine, and then calculating the change in binding free energy (ΔΔG). The binding free energy difference of native and mutant proteins were estimated over the same set of trajectory snapshots used for free energy decomposition analysis, using the MM-PBSA ASM script. The estimated ΔΔG values of survivin dimer and CPC proteins are shown in ESI Tables S3 and S4,† respectively.

Next, ASM results were analyzed based on the following criteria:⁵⁸ (i) if the binding free energy (ΔΔG) of the identified residues exceeds 4 kcal mol⁻¹, then these residues are considered as “hot spots” because these are very significant for protein structural stability, and moreover these residues are anchored deeply and protected from solvent exposure; (ii) if the ΔΔG is between 2 and 4 kcal mol⁻¹, then these residues are considered as “warm spots”, as these residues are moderately significant for protein structural stability; (iii) finally, if the ΔΔG is less than 2 kcal mol⁻¹, then these residues are considered as “null spots,” and are assumed to be less important for protein structural stability. Generally, both the warm and null spots exist in the surrounding regions of the hot spots and protect them from solvent exposure.

Based on the above analysis criteria, in the survivin dimer (Table S3, ESI†), residues Trp10A, Leu98A, Trp10B, Leu98B and Asp105B significantly (ΔΔG > 4 kcal mol⁻¹) contribute to the stability of the complex and are categorized as “hot spots,” and the residues Leu6A, Phe93A, Leu102A Phe93B and Leu102B (ΔΔG = 2 to 4 kcal mol⁻¹) are classified as “warm spots”. The remaining interface residues are considered to be “null spots”. In the case of CPC protein (Table S4, ESI†), the survivin residues Trp10A, Leu98A, Asp105A, Arg106A, Lys110A, Ile113A, Thr117A, Lys121A, Phe124A, and Ile135A are considered as “hot spots” and residues Leu6A, Phe101A, Leu102A, Lys120A, Val131A Arg132A and Leu138A are considered as “warm spots”. The remaining interface residues are considered as “null spots”. Hence, the in silico ASM results revealed the most important residue for proteins structural stability in survivin dimer and CPC protein.

Elucidation of survivin hot spots based on consensus of all the methods

Every computational method has its own pros and cons; choosing one over the other depends on a multitude of factors including speed and accuracy. Here, five different computational methods (Robetta server, KFC server, HotRegion database, per-residue energy decomposition and in silico ASM) were used to predict the hot spot residues in survivin dimer and the CPC complex interface region. Analysis of results from all the five methods will help to improve the accuracy of the predicted hot spots, and the hot spot information derived could be used further for the design of PPI inhibitors. The results obtained from the five methods are summarized in Tables 3 and 4, for survivin dimer and CPC protein, respectively. Comparison of the results suggests that in the survivin dimer interface, residues Leu98A, Trp10B, Leu98B are designated as “hot spot” by all five methods and residues Trp10A and ASP105B are designated as “hot spots” in at least two of the five methods. In contrast, residues Leu6A, Phe93A, Leu102A, Phe93B and Leu102B are designated as “warm spot” in at least two of the five methods.

Table 3 Comparison of results from five different methods used for hot spot detection in survivin dimer

Residue name	Robetta server^a	KFC server^a	HotRegion database^a	Per-residue energy (ΔG kcal mol^−1)	Alanine scanning mutagenesis (ΔΔG kcal mol^−1)	Residue status
a NH denotes non hot spot and H denotes hot spot.
Leu6A	NH	NH	H	−1.88	2.87	Warm spot
Trp10A	H	NH	H	−3.34	4.83	Hot spot
Phe93A	NH	NN	H	−1.54	2.62	Warm spot
Leu98A	H	H	H	−6.82	8.89	Hot spot
Phe101A	NH	H	H	−0.80	1.66	Null spot
Leu102A	NH	H	NH	−2.16	2.87	Warm spot
Asp105A	NH	H	NH	−0.36	0.91	Null spot
Leu6B	NH	NH	H	0.14	0.58	Null spot
Trp10B	H	H	H	−2.22	7.89	Hot spot
Phe93B	NH	NH	H	1.33	3.14	Warm spot
Leu98B	H	H	H	−7.14	5.07	Hot spot
Phe101B	NH	H	H	−1.44	1.61	Null spot
Leu102B	NH	H	NH	−1.77	2.47	Warm spot
Asp105B	H	NH	NH	−0.34	7.11	Hot spot

---

Table 4 Comparison of results from five different methods used for hot spot detection in CPC protein

Residue name	Robetta server^a	KFC server^a	HotRegion database^a	Per-residue energy ΔG kcal mol^−1	Alanine scanning mutagenesis (ΔΔG kcal mol^−1)	Residue status
a NH denotes non hot spot and H denotes hot spot.
Leu6A	H	H	H	−2.71	3.74	Warm spot
Trp10A	H	H	H	−2.97	4.39	Hot spot
Phe13A	NH	H	H	−0.61	−0.54	Null spot
Phe93A	NH	NH	H	−1.83	0.84	Null spot
Leu96A	NH	H	H	−2.35	1.19	Null spot
Leu98A	H	H	H	−7.93	5.14	Hot spot
Phe101A	H	H	H	−1.85	2.08	Warm spot
Leu102A	H	H	H	−2.66	3.57	Warm spot
Asp105A	H	H	H	1.36	7.96	Hot spot
Arg106A	H	H	NH	−9.36	16.15	Hot spot
Agr108A	H	NH	NH	−1.45	1.87	Null spot
Lys110A	H	H	NH	−7.06	11.37	Hot spot
Ile113A	H	H	NH	−4.49	4.60	Hot spot
Thr117A	H	H	NH	−2.40	6.40	Hot spot
Lys120A	H	H	NH	−0.16	2.60	Warm spot
Lys121A	H	H	NH	−1.77	5.43	Hot spot
Phe124A	H	H	NH	−2.68	5.97	Hot spot
Val131A	H	H	NH	−3.14	3.39	Warm spot
Arg132A	H	NH	NH	−2.33	2.01	Warm spot
Ile135 A	H	H	H	−3.51	5.86	Hot spot
Leu138A	H	H	NH	−3.05	3.45	Warm spot

---

In CPC protein, the survivin residues Trp10A, Leu98A, Asp105A, Arg106A, Lys110A, Ile113A, Thr117A, Lys121A and Phe124A are designated as hot spots by the in silico ASM method and at least three out of the remaining four methods. Residues Leu6A, Phe101A, Leu102A, Lys120A, Val131A, Arg132A, Leu138A, are designated as warm spots and also detected as hot spots in at least two out of the remaining four methods. The predicted hot spots, warm and null spots of survivin dimer and CPC proteins are depicted in Fig. 5 and 6, respectively. Consensus from different in silico methods has helped us to identify the most important hot spot residues involved in survivin PPI. Disruption of these residue interactions with their partner proteins is expected to have therapeutic value for treating cancer and related diseases.

Fig. 5 Mapping of the energetically important residues in survivin dimer interface region. Hot spots, warm spots and null spots are represented in red, gold and light green color, respectively.

Fig. 6 Mapping of the energetically important residues in CPC interface region. Hot spots, warm hot spots and null spots are represented in red, gold and light green color, respectively. The CPC complex partners survivin, borealin and INCENP are represented in blue, purple and gray color, respectively (for clarity purpose only survivin hot spot residues are shown; refer to ESI Fig. S2,† for mapping of borealin and INCENP hot spot residues).

Structure based pharmacophore generation and virtual screening

In order to show that the information derived from the hot spot detection at the survivin–partner protein interface could be utilized for the development of PPI inhibitors, we undertook a model generation and virtual screening exercise. Different computational methods could be employed to design small molecule inhibitors of protein–protein interaction. One such approach is the development of structure-based pharmacophore (SBP), which has been successfully used to discover the inhibitors of protein–protein interactions by other researchers.^9,59 SBP can be derived using a known active site (or) hot spot residue of a given protein complex, and can be considered as a complementary computational technique to docking methods to identify potential hits. Here, using the Catalyst module of Discovery studio pharmacophore features were derived based on the survivin dimer and CPC protein hot/warm spot residues (Leu6A, Trp10A, Leu98A, Phe101A, Asp105A and Arg106A). These residues are clustered at the survivin dimer interface as well as in the CPC interface region where the borealin C-terminal contacts the survivin protein. These hot spot residue regions would be an ideal place to disrupt the survivin PPIs. Based on these residues, an eight feature pharmacophore model was derived using the Catalyst module of Discovery studio, which contains two hydrogen bond acceptors (A1 and A2), two hydrogen bond donors (D1 and D2) and four hydrophobic features (HP1, HP2, HP3 and HP4). The hot spot residues and their corresponding pharmacophore features are shown in Fig. 7A. From this figure, it is clear that the hydrophobic features were derived from Leu6 (HP1), Trp10 (HP2) and leu98 (HP3 and HP4) side chains and the acceptor features were derived from Trp10 (A1) and Arg106 (A2) side chains. While, the two donor features of the pharmacophore model were derived from the main chains of Leu98 (D1) and Asp105 (D2). Next, using the pharmacophore model, virtual screening (VS) of approved drugs from the Drug Bank database⁶⁰ was performed to identify hits, by matching at least five features of the model. The VS process retrieved 75 hits from the Drug Bank database and these were subjected to further in situ ligand minimization. Ligand geometry optimization in the presence of protein complex is an essential step in structure-based drug discovery methods.^61–64 This process employs specific molecular mechanics based force field parameters to correct the geometries of the protein and ligand atoms, and also removes the steric clashes, if any, between them. Here, the selected 75 hits from VS were subjected to in situ ligand minimization using Discovery studio. This was followed by ligand binding free energy calculation to determine the strength of the binding between the hit molecule and the survivin protein at the defined hot spot site. Based on the binding free energy values, the hits were ranked and the top scoring hits were further analyzed by matching them to the pharmacophore model.

Fig. 7 (A) Structure-based pharmacophore model developed from the hot spot information derived for the anti-cancer drug target – survivin. The model mapped with the survivin hot spot residues. HP1, HP2 and HP3 – hydrophobic (cyan), A1 and A2 – acceptor (green), D1 and D2 – donor (magenta) features. (B) Pharmacophore model overlaid with the best hit (indinavir) identified from the Drug Bank (C) 2D representation of indinavir with matching pharmacophore features (D) 3D binding mode of indinavir with survivin (E) 2D interaction map of indinavir with survivin.

Pharmacophore mapping analysis revealed that the top scoring hits matched well to the pharmacophore model without much deviation. The top scoring ten hits were mapped to the pharmacophore model and are shown in ESI Table S5.† The top hit (indinavir, −217.119 kcal mol⁻¹) mapped onto the pharmacophore model is shown in Fig. 7B and C. The indinavir pyridine N occupies the A2 site, the piperazine carboxyl amide NH occupies the D1 site; the indene OH group occupies the D2 site; and the indene carboxyl amide NH group occupies the A1 site. The benzyl and N-tert-butyl groups occupy HP2 and HP3 sites, respectively. Pharmacophore mapping analysis suggests that the structural features of the hit indinavir match well with six out of the eight pharmacophore features of the model and snugly fit into the survivin interface region (ESI, Fig. S3†).

Next, the binding mode of indinavir and nine other hits with survivin was investigated. Analysis revealed that the predicted hot and warm spots were actively involved in making hydrogen, hydrophobic and electrostatic interactions with indinavir (Fig. 7D and E). As shown in figure, indinavir forms two hydrogen bond interactions with survivin hot spots: one hydrogen bond between the amide carbonyl group (linked to indane ring) of indinavir and the indole NH group of Trp10; and a second hydrogen bond between the central hydroxyl group of indinavir and side chain carboxylic group of Asp105. Additionally, it was also observed that the residues: Ala9, Trp10, Phe13, Leu98, Phe101, Leu102 and Ala109 showed extensive hydrophobic contacts with indinavir. Moreover, the negative and positively charged hot spot residues, Asp105, Arg106, Arg108 and Lys112 were actively involved in electrostatic interactions with indinavir. The in silico analysis suggest that, indinavir, a protease inhibitor approved for the treatment of HIV, could inhibit the survivin PPIs by binding to the interface region.

Further analysis of the binding mode of the remaining hits with survivin revealed that the detected hot and warm spots were also actively involved in various types of interactions. The Trp10 indole NH group makes a hydrogen bond interaction with the ketone carbonyl group of Iloprost, and the catechol hydroxyl group of arbutamine. In the same way, the Asp105 side chain carboxyl group makes a hydrogen bond interaction with the hydroxyl group on the tetrahydro-pyran ring of mupirocin, and the benzylic hydroxyl group of salmeterol. Furthermore, the indole ring of Trpy10 is involved in a π–stacking interaction with the aromatic rings of iloprost and indacaterol. In addition, the predicted hot and warm spot residues Leu6, Trop10, Leu98, Phe101 and Leu102 were extensively involved in making hydrophobic contacts with all the identified hits. The charged residues, Asp105 and Arg106 were involved in making wide range of electrostatic interactions with all the hits. The 3D and 2D binding mode, and the hot and warm spots involved in various interactions of the hits are mentioned in Table S6, ESI.†

Recently, Berezov et al. used a ‘cavity induced allosteric modification’ (CIAM) algorithm to identify a small molecule inhibitor S12 (ref. 49) which binds to a cavity near the survivin dimerization zone; S12 makes extensive hydrophobic contacts with Leu98 from one monomer and Leu6, Trp10, Phe93, Phe101, and Leu102 from the neighbouring monomer of the survivin dimer. Moreover, the reported inhibitor LQZ-7F⁵¹ interaction with survivin also showed that the residues Trp10, Phe93, Leu98 and Phe101 are essential for making hydrophobic and π–stacking interactions. The interaction of LLP3, LLP6 and LLP9 (ref. 50) with survivin also showed similar hydrophobic and π–stacking interactions with Leu6, Trp10, Phe93, Leu98, Phe101 and Leu102. Thus, the predicted hot and warm spot residues, Leu6, Tpr10, Leu98, Phe101, Leu102, Asp105 and Arg106 are very important for maintaining stability of the survivin dimer and the CPC complex. Designed small molecule inhibitors which can make extensive hydrophobic and π–stacking interaction with the predicted aromatic hot and warm spots, and also forms hydrogen and electrostatic interactions with positive and negative charged hot and warm spots could disrupts survivin function.

In vitro testing of indinavir

Indinavir was tested against MB-231 breast cancer cell lines (Fig. 8) in a cell viability assay, and the results revealed that indinavir was cytotoxic to MB-231 cell lines with a half maximal inhibitory activity (IC₅₀) at 510 μM concentrations (Fig. 8A). Next, mechanistic investigations using western blot analysis was carried out to understand the cytotoxicity of indinavir. As mentioned before, survivin forms CPC – complex with two other proteins – INCENP and borealin – and regulates the phosphorylation and functions of Aurora B. Disruption of this complex leads to cell endo-replication and cell death.^30,65,66 Here, western blot analysis revealed that indinavir treatment decreased the expression of phosphor-Aurora B in MDA-MB-231 cells (Fig. 8B). In addition, fluorescence microscopy revealed that indinavir induced nucleus enlargement, a typical morphological change observed in cells with Aurora B hypo-phosphorylation/inhibition induced endo-replication, in MDA-MB-231 cells (Fig. 8C). It is reported that survivin also forms PPI complex with XIAP, another member of IAP. The survivin–XIAP complex shows enhanced anti-apoptotic activity by inhibiting caspase 9 enzyme activity, leading to the inhibition of the downstream effector caspase 3. Interestingly, survivin–XIAP complex formation is reported to enhance the stability of XIAP by decreasing the ubiquitination/proteasomal destruction of XIAP; conversely, disruption of survivin–XIAP complex results in degradation of XIAP.^35,39 Western blot analysis revealed that indinavir treatment did not alter the expression of survivin in MDA-MB-231 cells. However, it decreased the expression of XIAP and increased the expression of the cleaved caspase 3 in MDA-MB-231 cells in a time-dependent manner (Fig. 8B). Taken together, these results indicate that indinavir, identified from in silico studies, imparts anti-proliferative and apoptotic activity in MDA-MB-231 cells, and thus possibly through inhibitions of survivin PPI such as survivin–Aurora B and survivin–XIAP.

Fig. 8 Effect of indinavir on MDA-MB-231 breast cancer cell lines. (A) Cell viability in MDA-MB-231 cells after treatment with various concentration of indinavir for 72 h. MTT assay was used to determine the cell viability of cells treated with or without indinavir. (B) Western blot analysis of MDA-MB-231 breast cancer cell lines after treatment with indinavir for 24 h and 48 h. Equal protein loading was verified by actin. The numbers under the blot are intensity of the blot relative to that of the control. (C) Fluorescence microscopy showing the nucleus of MDA-MB-231 cells treated with or without indinavir for 48 h. Nucleus was stained blue with Hoechst 33342.

Conclusion

Survivin is over-expressed in many human solid tumors but not in normal adult tissues, and thus disruption of survivin interaction with their substrate proteins constitutes an interesting therapeutic option for the treatment of cancer. Currently, only a few small molecule inhibitors that target survivin have been identified, none of which gave rise to clinically useful drugs, and so novel small molecule inhibitors are urgently required. To this end, we used a simple, knowledge based physical model to detect energetically important hot spot residues in survivin dimer and Chromosomal Passenger Complex (CPC) proteins. Then, using a molecular dynamics simulation, the detected hot spots residues were extensively investigated by per-residue energy decomposition and in silico alanine scanning mutagenesis. Analysis of the results revealed that residues Trp10A, Leu98A, Trp10B, Leu98B and Asp105B are designated as hot spots in the survivin–dimer interface; and survivin residues Trp10A, Leu98A, Asp105A, Arg106A, Lys110A, Ile113A, Thr117A, Lys121A, Phe124A and Ile135A are designated as hot spots in the CPC protein.

The hot spot residues are crucial for intermolecular interactions and contribute a significant amount of binding free energy towards survivin complex formation. Accordingly, these results were used to direct structure-based drug design for PPI inhibitor identification by generation of a pharmacophore model, and screening of the Drug Bank database. One of the hits identified from the virtual screening, indinavir, was tested in MDA-MB-231 breast cancer cell line and found to show anti-proliferative and apoptotic activity. Preliminary biochemical results indicate that the mode of action is by disruption of survivin PPIs. Further detailed investigation is underway to understand the mode of action of indinavir's anti-proliferative activity. Moreover, comparison of the binding mode of indinavir with other reported survivin inhibitors revealed that the detected hot and warm spot residues, Leu6, Trp10, Leu98, Phe101, Leu102 and Asp105 are involved in various types of interactions including hydrogen, hydrophobic, π-stacking and electrostatic interactions.

In summary, our study proves that it is possible to use the information derived from hot spot detection for the identification of PPI inhibitors. Large scale VS (e.g., using ZINC database) using pharmacophore models derived from different combinations of hot spots, followed by testing of the hits could help to identify novel inhibitors of survivin PPI for the treatment of cancer in the future.

Experimental section

Detection of hot spot residues in survivin dimer and CPC

Hot spots from PPI interface can be detected using a variety of in silico approaches. In this work, the online web services – Robetta server (http://robetta.bakerlab.org/),²⁰ KFC server (http://kfc.mitchell-lab.org/)²² and HotRegion database (http://prism.ccbb.ku.edu.tr/hotregion/)²⁴ were used to detect the hot spots in survivin dimer as well as in CPC protein. The downloaded PDB files of survivin dimer and CPC protein were prepared and submitted to these servers. These methods are fast, reliable, and use a simple free energy function to predict the hot spots based on their knowledge based physical models.

Molecular dynamics simulation

To access the stability of survivin dimer and CPC protein, molecular dynamics simulations were carried out using the SANDER module of the Amber 11 software suite⁶⁷ over a period of 50 ns by applying the ff99SB force field. Prior to simulation, the protein structures were downloaded from protein data bank and were processed using PyMol molecular visualizer. The connectivity data and other hetero atoms were removed except for the zinc ion, which is essential for protein structural stability.⁶⁸ It was parameterized using a non-bonded dummy atom approach.⁶⁹ Then proteins were subjected to leap program for addition of hydrogens and missing heavy atoms, and solvated using a truncated octahedron TIP3P water box. Further, the two systems were neutralized by addition of 10 Na⁺ ions. To remove steric clashes, minimization was carried out in two steps: in the first step, the water molecules were minimized and proteins were kept at fixed position whereas in the second step, the solvent as well as proteins were minimized. After minimization, the two systems were gradually heated from 0 K to 300 K over a period 50 ps with position restrain at a constant volume (NVT) using Langevin thermostat. Then, the systems were equilibrated over a period of 500 ps at constant pressure (NPT) of 1 bar using Barendsen. Finally, 50 ns production run was carried out for both the systems at a temperature of 300 K.

During the simulation period, periodic boundary condition was applied and time step was set to 2 fs. All the hydrogen bond interactions were constrained using the SHAKE algorithm and long range electrostatic interaction were monitored by employing particle Mesh Ewald (PME) approach. Trajectory frames were saved for each 1 ps and were analyzed by the CPPTRAJ and PTRAJ modules of the Amber software suite.

Binding free energy decomposition and alanine scanning mutagenesis

Binding free energy decomposition of survivin dimer and CPC protein was carried out using well-known molecular dynamics based MM-PBSA method. This method is widely employed to calculate the binding free energies of biomolecule interactions, including protein–ligand and protein–protein interactions. The binding free energy of a given protein complex could be calculated using the following eqn (1)

ΔG_binding = G_complex − (G_receptor + G_ligand) (1)

Further, the eqn (1) could be summarized as eqn (2)

ΔG_binding = ΔG_MM + ΔG_{polar solvation} + ΔG_{non polar solvation} − TΔS (2)

The term ΔG_MM, is the molecular mechanics binding free energy of protein–protein or protein–ligand complex and could be calculated as sum of energy differences in van der Waals, electrostatic and internal energies terms of bond, angle and dihedrals as shown in eqn (3).

ΔG_MM = ΔE_vdw + ΔE_ele + ΔE_int (3)

The second terms in eqn (2), ΔG_{polar solvation}, is the polar contribution to solvation free energy and could be calculated by solving Posisson Boltzmann equation. Whereas, the third term, ΔG_{non polar solvation}, is the non-polar contribution to solvation free energy and could be computed from molecular solvent-accessible surface area (SASA) using LCPO method. The final term TΔS is the entropic contribution to solvation free energy, and is not estimated here as we are only interested in computing relative binding energy contribution of each amino acid to protein complex formation.

The ensemble conformations (trajectory snapshots) of survivin dimer and CPC protein were generated by a molecular dynamics simulation, from which their respective protein unbound receptors and ligand structures were extracted to calculate the binding free energy and per-residue energy contribution for their complex formation. Following per-residue energy decomposition the same ensemble conformations were used to generate the mutant trajectory by replacing respective protein complex interface residues with alanine. Then, the MM-PBSA alanine scanning script was used to estimate the relative binding free energy change (ΔΔG) between the wild type and alanine mutant protein by the following eqn (4).

ΔΔG_binding = ΔG_{alanine mutant protein} − ΔG_{wild type protein} (4)

Structure-based pharmacophore generation and virtual screening

A structure (hot spot residue) based pharmacophore was generated using the Catalyst module of Discovery studio. For this purpose, survivin hot spot residues (Leu6A, Trp10A, Leu98A, Phe101A, Asp105A, and Arg106A) were used as the active site, and defined the sphere around these residues. Using the find features option, 375 pharmacophore features were identified in this region. Next, using hierarchical clustering, merging and excluded volume, a pharmacophore model with eight features (four hydrophobic, two acceptors and two donors) was generated. The pharmacophore model was used to screen approved drugs (http://www.drugbank.ca/) by matching at least 5 features to the pharmacophore hypothesis using screen library option of Discovery studio module. In total, 75 hits from the Drug Bank were selected for further analysis.

In situ ligand minimization and binding free energy calculation

The 75 hits identified by the structure-based pharmacophore screening were subjected to in situ ligand minimization to remove the steric clashes between the protein and ligand atoms. Then, the ligand binding free energies were calculated so their affinities towards the protein complex could be ascertained. While performing minimization and ligand binding free energy calculation, all the parameters were kept as default and the CHARMm force field was applied. Then, all the ligands were ranked based on their binding free energy values, and the top 10 compounds were selected for further analysis.

In vitro testing of indinavir

Survivin expressing human breast adenocarcinoma MDA-MB-231 cells were obtained from ATCC and treated with indinavir (Sigma). The MTT assay was firstly used to determine the anti-proliferative effect of indinavir in vitro. Cells were seeded on 96-well plates for 24 h before being treated with either water (control, re-constitutive medium used for indinavir) or indinavir for 72 h. Cell viability was quantified by measuring the absorbance of the solution at 570 nm wave length by a spectrophotometer. The percentage of viable cells for each treatment group was calculated by adjusting the untreated control group to 100%. Duplicate wells were assayed for each condition and repeated at least three times. The IC₅₀ value resulting from 50% inhibition of cell viability was calculated graphically as a comparison with the growth of the control group.

Western blot analysis was then used to investigate any molecular changes induced by indinavir in MDA-MB-231 cancer cells. Cells (untreated and indinavir treated) were lysed using Cellytic™ M cell lysis reagent (Sigma) that contained 1 mM PMSF, 1 mM NaF and protease inhibitor cocktail (Roche). Equal amounts of protein were subjected to SDS-PAGE on either a 10 or 12% polyacrylamide gel. The resolved proteins were transferred onto a PVDF membrane (Millipore), which was then exposed to 5% non-fat dried milk in TBS-Tween buffer for an hour at room temperature before incubation overnight at 4 °C with primary antibodies. Antibodies against survivin and XIAP were from R&D system (cat# AF886 and AF8221); antibodies against β-actin were obtained from Millipore (cat# MAB1501); antibodies against phosphor-Aurora B were obtained from Abcam (cat# ab61074); antibodies against cleaved caspase 3 were obtained from Cell Signaling (cat# 9664P). The PVDF membrane was then washed three times with TBS containing 0.05% Tween-20 before incubation for an hour at room temperature with HRP-conjugated secondary antibodies. Immuno-complexes were finally detected with chemiluminescence reagents (Millipore). Experiments were repeated at least twice.

Fluorescence microscopy was performed to detect any morphological difference of the nucleus in between MDA-MB-231 cells treated with and without indinavir. Cells were treated with or without 1 X IC₅₀ indinavir for 48 h and subsequently incubated with Hoechst 33342. Nuclei (stained blue) were viewed under a UV-enabled fluorescence microscope. Same magnification was used to take every single image of nuclei in cells treated with or without indinavir.

Acknowledgements

Financial support from the Department of Biotechnology, India for MSC to purchase the server (DBT's Twining Program for the North East, BT/246/NE/TBP/2011/77) and for SS (DBT-JRF/2012-13/80), are gratefully acknowledged. We greatly appreciate the help extended by Prof. Devadasan Velmurugan and Mr Manish Kesherwani, for running molecular dynamics simulation using the GPU enbled DELL Server faclility, CAS in Crystallography & Biophysics, University of Madras, Chennai.

Notes and references

1. G. Sowmya and S. Ranganathan, Protein Pept. Lett., 2014, 21, 779–789.
2. M. A. Klein, ACS Med. Chem. Lett., 2014, 5, 838–839.
3. D. P. Ryan and J. M. Matthews, Curr. Opin. Struct. Biol., 2005, 15, 441–446.
4. J. A. Wells and C. L. McClendon, Nature, 2007, 450, 1001–1009.
5. D. C. Fry and L. T. Vassilev, J. Mol. Med., 2005, 83, 955–963.
6. P. Chene, Mol. Cancer Res., 2004, 2, 20–28.
7. B. O. Villoutreix, M. A. Kuenemann, J. L. Poyet, H. Bruzzoni-Giovanelli, C. Labbe, D. Lagorce, O. Sperandio and M. A. Miteva, Mol. Inf., 2014, 33, 414–437.
8. A. A. Ivanov, F. R. Khuri and H. Fu, Trends Pharmacol. Sci., 2013, 34, 393–400.
9. W. Guo, J. A. Wisniewski and H. Ji, Bioorg. Med. Chem. Lett., 2014, 24, 2546–2554.
10. S. Patel and M. R. Player, Expert Opin. Invest. Drugs, 2008, 17, 1865–1882.
11. F. Gonzalez-Lopez de Turiso, D. Sun, Y. Rew, M. D. Bartberger, H. P. Beck, J. Canon, A. Chen, D. Chow, T. L. Correll, X. Huang, L. D. Julian, F. Kayser, M. C. Lo, A. M. Long, D. McMinn, J. D. Oliner, T. Osgood, J. P. Powers, A. Y. Saiki, S. Schneider, P. Shaffer, S. H. Xiao, P. Yakowec, X. Yan, Q. Ye, D. Yu, X. Zhao, J. Zhou, J. C. Medina and S. H. Olson, J. Med. Chem., 2013, 56, 4053–4070.
12. M. R. Arkin, Y. Tang and J. A. Wells, Chem. Biol., 2014, 21, 1102–1114.
13. I. S. Moreira, P. A. Fernandes and M. J. Ramos, Proteins, 2007, 68, 803–8012.
14. T. Clackson and J. A. Wells, Science, 1995, 267, 383–386.
15. A. A. Bogan and K. S. Thorn, J. Mol. Biol., 1998, 280, 1–9.
16. E. Cukuroglu, H. B. Engin, A. Gursoy and O. Keskin, Prog. Biophys. Mol. Biol., 2014, 116, 165–173.
17. D. Kozakov, D. R. Hall, G. Y. Chuang, R. Cencic, R. Brenke, L. E. Grove, D. Beglov, J. Pelletier, A. Whitty and S. Vajda, Proc. Natl. Acad. Sci. U. S. A., 2011, 108, 13528–13533.
18. G. A. Weiss, C. K. Watanabe, A. Zhong, A. Goddard and S. S. Sidhu, Proc. Natl. Acad. Sci. U. S. A., 2000, 97, 8950–8954.
19. K. L. Morrison and G. A. Weiss, Curr. Opin. Chem. Biol., 2001, 5, 302–307.
20. D. E. Kim, D. Chivian and D. Baker, Nucleic Acids Res., 2004, 32, W526–W531.
21. I. S. Moreira, P. A. Fernandes and M. J. Ramos, J. Comput. Chem., 2007, 28, 644–654.
22. S. J. Darnell, L. LeGault and J. C. Mitchell, Nucleic Acids Res., 2008, 36, W265–W269.
23. D. M. Kruger and H. Gohlke, Nucleic Acids Res., 2010, 38, W480–W486.
24. E. Cukuroglu, A. Gursoy and O. Keskin, Nucleic Acids Res., 2012, 40, D829–D833.
25. Y. T. Wang and W. J. Lee, Mol. BioSyst., 2012, 8, 3274–3280.
26. V. Lafont, M. Schaefer, R. H. Stote, D. Altschuh and A. Dejaegere, Proteins, 2007, 67, 418–434.
27. E. Ciglia, J. Vergin, S. Reimann, S. H. Smits, L. Schmitt, G. Groth and H. Gohlke, PLoS One, 2014, 9, e96031.
28. K. D. Dubey and R. P. Ojha, J. Mol. Model., 2012, 18, 1679–1689.
29. A. Kumar and K. Y. Zhang, Mol. Inf., 2013, 32, 267–280.
30. A. A. Jeyaprakash, U. R. Klein, D. Lindner, J. Ebert, E. A. Nigg and E. Conti, Cell, 2007, 131, 271–285.
31. S. Shin, B. J. Sung, Y. S. Cho, H. J. Kim, N. C. Ha, J. I. Hwang, C. W. Chung, Y. K. Jung and B. H. Oh, Biochemistry, 2001, 40, 1117–1123.
32. H. S. de Souza, G. A. West, N. Rebert, C. de la Motte, J. Drazba and C. Fiocchi, Gastroenterology, 2012, 143, 1017–1026 e1019.
33. A. Mirza, M. McGuirk, T. N. Hockenberry, Q. Wu, H. Ashar, S. Black, S. F. Wen, L. Wang, P. Kirschmeier, W. R. Bishop, L. L. Nielsen, C. B. Pickett and S. Liu, Oncogene, 2002, 21, 2613–2622.
34. A. Suzuki, M. Hayashida, T. Ito, H. Kawano, T. Nakano, M. Miura, K. Akahane and K. Shiraki, Oncogene, 2000, 19, 3225–3234.
35. T. Dohi, K. Okada, F. Xia, C. E. Wilford, T. Samuel, K. Welsh, H. Marusawa, H. Zou, R. Armstrong, S. Matsuzawa, G. S. Salvesen, J. C. Reed and D. C. Altieri, J. Biol. Chem., 2004, 279, 34087–34090.
36. R. Kannangai, J. Wang, Q. Z. Liu, F. Sahin and M. Torbenson, Int. J. Gastrointest. Cancer, 2005, 35, 53–60.
37. N. K. Sah, Z. Khan, G. J. Khan and P. S. Bisen, Cancer Lett., 2006, 244, 164–171.
38. S. Small, G. Keerthivasan, Z. Huang, S. Gurbuxani and J. D. Crispino, Leukemia, 2010, 24, 1920–1926.
39. C. H. Cheung, C. C. Huang, F. Y. Tsai, J. Y. Lee, S. M. Cheng, Y. C. Chang, Y. C. Huang, S. H. Chen and J. Y. Chang, OncoTargets Ther., 2013, 6, 1453–1462.
40. M. S. Coumar, F. Y. Tsai, J. R. Kanwar, S. Sarvagalla and C. H. Cheung, Cancer Treat. Rev., 2013, 39, 802–811.
41. N. Singh, S. Krishnakumar, R. K. Kanwar, C. H. Cheung and J. R. Kanwar, Drug Discovery Today, 2015, 20, 578–587.
42. F. Aranda, E. Vacchelli, A. Eggermont, J. Galon, C. Sautes-Fridman, E. Tartour, L. Zitvogel, G. Kroemer and L. Galluzzi, Oncoimmunology, 2013, 2, e26621.
43. H. Dong, L. Yao, W. Bi, F. Wang, W. Song and Y. Lv, J. Cancer Res. Ther., 2015, 11, 717–722.
44. A. Rauch, D. Hennig, C. Schafer, M. Wirth, C. Marx, T. Heinzel, G. Schneider and O. H. Kramer, Biochim. Biophys. Acta, 2014, 1845, 202–220.
45. K. Yamanaka, T. Nakahara, T. Yamauchi, A. Kita, M. Takeuchi, F. Kiyonaga, N. Kaneko and M. Sasamata, Clin. Cancer Res., 2011, 17, 5423–5431.
46. F. Li, Am. J. Cancer Res., 2014, 4, 304–311.
47. P. Wiechno, B. G. Somer, B. Mellado, P. L. Chlosta, J. M. Cervera Grau, D. Castellano, C. Reuter, M. Stockle, J. Kamradt, J. Pikiel, I. Duran, S. Wedel, S. Callies, V. Andre, K. Hurt, J. Brown, M. Lahn and B. Heinrich, Eur. Urol., 2014, 65, 516–520.
48. R. Natale, F. Blackhall, D. Kowalski, R. Ramlau, G. Bepler, F. Grossi, C. Lerchenmuller, M. Pinder-Schenck, J. Mezger, S. Danson, S. M. Gadgeel, Y. Summers, S. Callies, V. Andre, M. Das, M. Lahn and D. Talbot, J. Thorac. Oncol., 2014, 9, 1704–1708.
49. A. Berezov, Z. Cai, J. A. Freudenberg, H. Zhang, X. Cheng, T. Thompson, R. Murali, M. I. Greene and Q. Wang, Oncogene, 2012, 31, 1938–1948.
50. S. N. Chettiar, J. V. Cooley, I. H. Park, D. Bhasin, A. Chakravarti, P. K. Li, C. Li and N. K. Jacob, Bioorg. Med. Chem. Lett., 2013, 23, 5429–5433.
51. J. Qi, Z. Dong, J. Liu, R. C. Peery, S. Zhang, J. Y. Liu and J. T. Zhang, Cancer Res., 2016, 76, 453–462.
52. J. Du, A. E. Kelly, H. Funabiki and D. J. Patel, Structure, 2012, 20, 185–195.
53. G. Rastelli, A. Del Rio, G. Degliesposti and M. Sgobba, J. Comput. Chem., 2010, 31, 797–810.
54. V. Zoete and O. Michielin, Proteins, 2007, 67, 1026–1047.
55. J. Guo, X. Wang, H. Sun, H. Liu and X. Yao, J. Mol. Model., 2012, 18, 1421–1430.
56. H. Gohlke, C. Kiel and D. A. Case, J. Mol. Biol., 2003, 330, 891–913.
57. E. Bourhis, S. G. Hymowitz and A. G. Cochran, J. Biol. Chem., 2007, 282, 35018–35023.
58. N. F. Bras, P. A. Fernandes and M. J. Ramos, Chem. Biol. Drug Des., 2014, 83, 427–439.
59. C. Barillari, G. Marcou and D. Rognan, J. Chem. Inf. Model., 2008, 48, 1396–1410.
60. D. S. Wishart, C. Knox, A. C. Guo, S. Shrivastava, M. Hassanali, P. Stothard, Z. Chang and J. Woolsey, Nucleic Acids Res., 2006, 34, D668–D672.
61. B. R. Brooks, C. L. Brooks 3rd, A. D. Mackerell Jr, L. Nilsson, R. J. Petrella, B. Roux, Y. Won, G. Archontis, C. Bartels, S. Boresch, A. Caflisch, L. Caves, Q. Cui, A. R. Dinner, M. Feig, S. Fischer, J. Gao, M. Hodoscek, W. Im, K. Kuczera, T. Lazaridis, J. Ma, V. Ovchinnikov, E. Paci, R. W. Pastor, C. B. Post, J. Z. Pu, M. Schaefer, B. Tidor, R. M. Venable, H. L. Woodcock, X. Wu, W. Yang, D. M. York and M. Karplus, Journal of computational chemistry, 2009, 30, 1545–1614.
62. T. Fu, Z. Jin, Z. Xiu and G. Li, Curr. Pharm. Des., 2013, 19, 2293–2307.
63. G. Sliwoski, S. Kothiwale, J. Meiler and E. W. Lowe Jr, Pharmacol. Rev., 2014, 66, 334–395.
64. M. R. Reddy, C. R. Reddy, R. S. Rathore, M. D. Erion, P. Aparoy, R. N. Reddy and P. Reddanna, Curr. Pharm. Des., 2014, 20, 3323–3337.
65. F. H. Gohard, D. J. St-Cyr, M. Tyers and W. C. Earnshaw, Open Biol., 2014, 4, 140163.
66. G. Vader, R. H. Medema and S. M. Lens, J. Cell Biol., 2006, 173, 833–837.
67. D. A. Case, T. Darden, T. Cheatham, C. L. Simmerling, J. Wang, R. E. Duke, R. Luo, R. Walker, W. Zhang and K. Merz, Amber 11, University of California, 2010.
68. M. A. Verdecia, H. Huang, E. Dutil, D. A. Kaiser, T. Hunter and J. P. Noel, Nat. Struct. Biol., 2000, 7, 602–608.
69. Y. P. Pang, Proteins, 2001, 45, 183–189.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: 10.1039/c5ra22927h

---