Introduction

Necroptosis is a programmed necrotic cell death mechanism observed in various human pathologies, such as neonatal brain hypoxia [1], acute pancreatitis [2, 3], cerebral ischemia [4–6], Alzheimer’s disease (AD), and Parkinson’s disease (PD) [7]. It is also involved in myocardial infarction, retinal injuries, lethal systemic inflammatory response syndrome, skin inflammation, acute pancreatitis, and viral infection [8, 9]. Receptor-interacting serine/threonine protein kinase 1 (RIPK1) and RIPK3 are the two crucial regulatory proteins involved in this pathway [10–13]. In vitro and in vivo studies have established RIPK1 as a specific protein target of the necroptotic pathway [14–18]. Inhibition of RIPK1 leads to reduced necroptotic cell death and increased cell survival rate [14, 19–21].

Genetic studies have revealed that the targeted deletion of RIPK3 would rescue caspase-8 dependent necroptosis and embryonic lethality in mice [13, 22, 23]. Moreover, RIPK3 was suggested as a necroptotic target in RIP1-deficient cells [2, 24]. RIP1-independent and RIP3-dependent necroptosis can occur in RIP1-deficient mouse embryonic fibroblast cells [2] and also in RIP1/caspase-8 double-knockdown L929 cells [24]. Therefore, there is a need to identify dual inhibitors that may target both RIPK1 and RIPK3 proteins. Pharmacological agents that target both the RIPK1 and RIPK3 proteins might offer more effective cellular protection in diseases involving necroptosis. Information regarding the active-site features of these proteins might assist in discovering novel structurally diverse inhibitors. To the best of our knowledge, computational methods have not been used to explore the conformational flexibility of the RIPK3 binding site or to identify RIPK3 inhibitors.

Structure-based pharmacophore modeling is an important event in most in silico drug-design strategies. The use of multiple protein structures for pharmacophore extraction is considered an improvement over the standard technique of using a single structure. It was proposed that a multicomplex-based pharmacophore model derived from many holoprotein (protein–ligand complex) structures would determine all of the key protein–ligand interactions [25]. In addition, an ensemble pharmacophore derived from multiple apoprotein (protein only) and holoprotein structures was proposed that could extract all of the features available at the binding site of the protein [26]. The screening of multitarget ligands using pharmacophores derived from multiple structures of each protein is the preferred choice, particularly when the conformations of these structures vary considerably and the ligands bind to them with different pharmacophoric features [27]. This could result in the effective screening of compounds with different scaffolds.

In general, the pharmacophores derived from multiple structures of a protein are combined to develop a merged pharmacophore or an ensemble pharmacophore in order to decrease the number of pharmacophore-based screenings and to reduce the computational time [26, 27]. These pharmacophores contain all of the features present at the binding sites of individual structures. Compound screening using these pharmacophores could highlight ligands that contain the features of individual structures. However, ligands that contain a combination of features present in different structures might also be obtained, which increases the number of false positives enormously (Fig. 1). The presence of these unwanted compounds with “cross-features” of multiple structures might in turn lead to increased docking time and a greater computational load. Therefore, there is a need to filter out these compounds before preparing the final ligand dataset for docking.

Fig. 1
figure 1

Limitation of a merged or ensemble pharmacophore. C(P1) and C(P2) are groups of compound hits that contain the features of the pharmacophores P1 and P2, respectively. C(P3) is a group of false-positive compounds that contain features of both pharmacophores (“cross-features”)

Full size image

For instance, assume that a hypothetical protein contains two structures with different conformations; P1 and P2 are the pharmacophores derived from them, respectively (Fig. 1). P1 contains features 1, 2, and 3. P2 contains features 1, 4, and 5. A merged pharmacophore or an ensemble pharmacophore (EP) derived by combining these two pharmacophores contains features 1, 2, 3, 4, and 5 (Fig. 1). Theoretically, screening a compound dataset using this EP should result in two groups of compound hits, C(P1) and C(P2), with the features of any of the two pharmacophores P1 and P2. However, in practice, in addition to these two groups of compound hits, a group of unwanted compounds denoted C(P3) can be obtained that contain cross-features of P1 and P2, e.g., 1, 3, and 4 (Fig. 1). Employing these compounds in the subsequent docking step could increase both the computational time and load.

In the study reported in the present paper, we derived two ensemble pharmacophores, one from the RIPK1 protein structures in the PDB and the other from the modeled protein structures of RIPK3 (Fig. 2). At first, ensemble pharmacophore-based screening (EPS) of ZINC database compounds was carried out using the RIPK1 ensemble pharmacophore. The compound hits thus obtained could contain cross-features. Therefore, in this work we propose a novel method that we call post ensemble pharmacophore-based screening filtration (PEPSF), and we used this method to separate the compound hits into groups that each contain features of an individual pharmacophore (Fig. 2). The compounds excluded by PEPSF contained cross-features and were not considered for further analysis. Next, these groups of compounds were subjected to EPS using RIPK3 EP. Then, PEPSF was carried out on the compound hits, leading to different groups of compounds containing pharmacophore features of both the RIPK1 and RIPK3 proteins. These groups of compounds were subjected to multiple dockings with their corresponding RIPK1 and RIPK3 protein structures (Fig. 2). Finally, MD simulation was carried out to study the behavior of the top-ranking RIPK1 and RIPK3 dual inhibitors. This strategy allows effective multitarget inhibitors to be rapidly identified, since it removes the false-positive compounds and greatly reduces the number of compounds that are virtually screened, thereby minimizing the computational time and load.

Fig. 2
figure 2

Workflow used in this study to identify RIPK1 and RIPK3 dual inhibitors. An ensemble pharmacophore was derived from RIPK1 structures and validated, and this was employed to screen the database compounds. The compound hits were subjected to PEPSF, leading to three groups of ligands, with each group corresponding to the pharmacophore features of each RIPK1 structure. These ligands were screened using the RIPK3 ensemble pharmacophore. The ligand hits thus obtained were subjected to PEPSF, yielding 12 groups. These ligands then underwent multiple dockings, and MD simulations were carried out on the top-ranked protein–ligand complexes

Full size image

Methods

RIPK1 protein structures

The protein–ligand complexes of RIPK1 were retrieved from the PDB (PDB IDs: 4ITH, 4ITI, and 4ITJ) and were prepared using the protein preparation wizard of Schrodinger (Schrodinger, LLC, New York, NY, USA). Missing hydrogens were added, proper bond orders were assigned, and water molecules further than 5 Å from the heterogeneous groups were deleted. The H-bonds were optimized and the protein structures were minimized to the default root mean square deviation (RMSD) value of 0.30 Å. The minimized protein structures were further considered for pharmacophore analysis and docking.

RIPK3 protein structure modeling

The protein sequence of the kinase domain of human RIPK3 was retrieved from the UniProt database (ID: Q9Y572). Homology modeling of the 3D structure of RIPK3 was carried out using the Prime v2.2 module of the Schrodinger suite (Schrodinger, LLC). Side chains and hydrogen atoms were added and the stability of the homology model was validated through PROCHECK and the Ramachandran plot hosted by the Structural Analysis and Verification Server (SAVES) (http://nihserver.mbi.ucla.edu/SAVS/).

MD simulation and clustering

To obtain diverse conformations, the modeled RIPK3 protein was subjected to MD simulation using the Gromacs v.4.5.5 software package with the Gromos 53a6 force field, as described previously [26].

Prior to executing the MD simulation, 20,000 steps of steepest descent minimization in vacuum were carried out on the system, followed by explicit solvation in a cubic box containing simple point charge (SPC/E) water molecules. During the preparation procedure for the MD simulation, the ionizable residues of the protein were protonated without artifacts and the system was neutralized by the addition of Na+ and Cl− counterions and a salt concentration of 0.1 M NaCl. All of the hydrogen-related bond lengths were fixed with the Linear Constraint Solver (LINCS) algorithm, and the long-range electrostatic interactions were treated with the Particle Mesh Ewald (PME) method [28]. Further, all of the bond lengths involving H-atoms were constrained using the SHAKE algorithm, and the solvated system was subjected to 20,000 steps of steepest descent energy minimization. Then the system was equilibrated for 100 ps with an NVT and NPT ensemble equilibration protocol for about 50,000 steps. Finally, extensive MD simulation was performed for 10 ns with a 2.0-fs time step under constant temperature and pressure (310 K and 1 bar, respectively).

Clusters were generated by comparing the root mean square deviation (RMSD) cutoff values obtained for different snapshots during the MD simulation trajectory. The energy of each conformation was compared with the lowest energy within the trajectory to estimate the relative possibility of the conformation. Finally, representative structures that were closest to the average conformation were selected from each of the clusters. These protein structures were prepared as described in the previous section, and were considered for RIPK3 pharmacophore analysis and docking.

Receptor grid generation and docking

Glide energy grids were generated for all of the prepared protein structures. In the case of RIPK1, the co-crystallized ligand was differentiated from the active site of the receptor. The grid was defined by a rectangular box surrounding the co-crystallized ligand. For RIPK3 structures, the grid was generated such that it covered the active site of the protein. These grids were employed in all of the docking studies.

Docking was performed with the Glide module using the OPLS 2005 force field [29]. Default settings were selected for all of the docking calculations, and the Glide XP descriptor information was used to deduce energy terms such as H-bond interactions, electrostatic interactions, hydrophobic enclosure, and π–π stacking interactions (Glide v.5.7, Schrodinger, LLC). Finally, post-docking minimization was carried out to optimize the ligand geometries.

E-pharmacophore extraction

Structure-based pharmacophores were extracted from RIPK1 and RIPK3 protein structures using the energy-optimized pharmacophore (e-pharmacophore) script of Schrodinger [30]. Fragment-based e-pharmacophores were extracted from RIPK1 and RIPK3 protein structures.

For fragment-based e-pharmacophore extraction, the Glide fragment library was docked to each of the proteins using Glide XP. The resultant pose viewer files were used in both cases to generate pharmacophores via the e-Pharmacophore script in the Schrodinger software. A standard set of six pharmacophore features, i.e., H-bond acceptors (A), H-bond donors (D), hydrophobic groups (H), negatively ionizable regions (N), positively ionizable regions (P), and aromatic rings (R), were generated with the Phase module. The generated pharmacophore sites were ranked based on their energies, and the most favorable sites were selected for the pharmacophore hypothesis.

Ensemble pharmacophore construction

The e-pharmacophores extracted from the RIPK1 proteins were superposed and they were combined to generate an ensemble pharmacophore, as described previously [26]. Similarly, fragment-based e-pharmacophores extracted from RIPK3 proteins were combined to build an ensemble pharmacophore. The ensemble pharmacophore contained all of the features that the individual e-pharmacophores had in common, as well as the features that varied among the e-pharmacophores. If common features relating to the same amino acid at the binding site were present, they were considered a single feature, so only one of them was represented in the ensemble pharmacophore.

RIPK1 ensemble pharmacophore-based screening

Ensemble pharmacophore-based screening (EPS) of RIPK1 was carried out by employing its ensemble pharmacophore to screen the lead-like compounds (∼2 million unique structure records) in the ZINC database [31] using the Phase module. The database molecules were filtered explicitly with a distance-matching tolerance of 2.0 Å, and matching of a minimum of four sites was required. The database hits were ranked based on their fitness scores, which measured how well the aligned ligand conformers matched the hypothesis.

The RIPK1 ensemble pharmacophore was validated using the enrichment factor (EF). The enrichment at a given percentage of the database screened is the most popular and the simplest metric method used to evaluate the effectiveness of a virtual screening method [32, 33]. The EF is defined as

$$ {\displaystyle {\mathrm{EF}}^{x\%}}=\frac{n\left(\mathrm{active}\ \mathrm{at}\ x\%\right)/n\left(x\%\right)}{n\left(\mathrm{all}\ \mathrm{active}\right)/n\left(\mathrm{all}\right)}, $$

where n(active at x%) is the number of active ligands obtained in the top x% of the database screened, n(x%) is the number of compounds screened in the top x% of the database, n(all active) is the number of active ligands in the entire database, and n(all) is the number of compounds in the entire database [34].

All of the known RIPK1 inhibitors were randomly seeded into the ZINC database and EPS was carried out on these compounds, as described previously [26].

RIPK1 post-EPS filtration

The compound hits obtained through EPS contained the features of individual pharmacophores. However, false-positive compounds that contained cross-features were also present. In order to filter out these compounds, PEPSF was carried out, which separated the compounds into groups with features from individual pharmacophores.

The output of RIPK1 EPS obtained using Phase, when viewed in the form of a table, contained a column named “matched ligand sites.” This gave information on the pharmacophore features present in each compound hit that matched with the RIPK1 ensemble pharmacophore features. This table was exported in the form of an Excel worksheet, and the compound hits were sorted based on their matching sites. These sites were compared with each of the RIPK1 fragment-based e-pharmacophores, and the compounds with sites that matched with any of these pharmacophores were considered. Compounds with sites that matched with more than one pharmacophore were considered false positives since they contained cross-features, and were removed from the study. Thus, each of the three RIPK1 fragment-based e-pharmacophores had a corresponding group of compounds containing features that were complementary to that e-pharmacophore (i.e., there were three groups in total).

RIPK3 ensemble pharmacophore-based screening

The ensemble pharmacophore of RIPK3 was used to screen the three groups of compounds obtained through RIPK1 PEPSF. This resulted in compound hits that contained the features of both the RIPK1 and RIPK3 proteins. The RIPK3 ensemble pharmacophore could not be evaluated using EF since there were no reported RIPK3 inhibitors when this study was carried out.

RIPK3 post-EPS filtration

In order to remove the compounds that contained cross-features and did not match with any of the four pharmacophores derived from the RIPK3 structures, the compound hits that were obtained through RIPK3 EPS were subjected to PEPSF as described in the previous section.

Docking-based virtual screening

Using RIPK3 PEPSF, the database hits obtained from RIPK3 EPS were separated into different groups that contained pharmacophore features of any of the three RIPK1 proteins and also any of the four RIPK3 proteins. Multiple dockings were carried out to study the binding of the compounds in each group with their respective RIPK1 and RIPK3 protein structures (Fig. 2).

Prior to docking, the compounds obtained through RIPK3 PEPSF were subjected to ligand preparation by the LigPrep module of the Schrodinger suite (LigPrep v.2.3, Schrodinger, LLC). The ligands were processed to assign suitable protonation states at a physiological pH of 7.2 ± 0.2, and conformer generation was carried out with ConfGen torsional sampling using the OPLS 2005 force field. Docking was performed using Glide in the Virtual Screening Wizard of the Schrodinger suite. All of the docking calculations were performed as mentioned in an earlier section. The docking results (comprising the compounds listed in the order in which they were docked) as well as the Glide scores were obtained.

MD simulation of lead compounds

Based on the docking results and Glide scores of the compounds that were found to bind to both the RIPK1 and RIPK3 proteins, the top two ranked protein–ligand complexes (RIPK1–lead 1, RIPK3–lead 1, RIPK1–lead 2, and RIPK3–lead 2) were selected to carry out 20-ns MD simulations as described in the earlier section. In addition, the RIPK1–necrostatin-1 (Nec-1) and RIPK3–Nec-1 complexes were subjected to MD simulation to study the binding of Nec-1 with both proteins. The topologies and parameters for the ligands were generated using PRODRG [35]. The Gromacs and Visual Molecular Dynamics (VMD) packages were used to analyze the MD simulation results [36].

Results and discussion

RIPK3 protein

The 3D structure of human RIPK3 protein was modeled using the crystal structure of mouse RIPK3 kinase domain (PDB ID: 4 M66) as the template. The kinase domain of RIPK3 contains approximately 211 amino acids and its sequence is 70 % similar to its mouse counterpart. The Ramachandran plot indicated that 99.5 % of the amino acid residues in the modeled structure are present in the allowed region (Fig. 3).

Fig. 3
figure 3

Ramachandran plot of the modeled RIPK3 protein structure

Full size image

Clustering of the conformations obtained through MD simulation of the RIPK3 protein produced four clusters (Fig. 4). A representative structure from each cluster was selected, so four RIPK3 protein structures were used in this study.

Fig. 4
figure 4

Clusters of the RIPK3 protein conformations produced during MD simulation

Full size image

E-pharmacophores

Fragment-based e-pharmacophores were derived from the three RIPK1 protein structures, as described previously [26]. The fragment-based e-pharmacophore derived from 4ITH consisted of eight features (2A, 2D, 1P, 1H, and 2R) (Fig. 5a); the amino acids Met 67, Leu 70, Val 76, Leu 78, and Asp 156 were H-bond acceptors and donors. The e-pharmacophore derived from 4ITI consisted of nine features (3A, 3D, 1H, and 2R) (Fig. 5b); Leu 70, Val 76, Leu 78, Asp 156, and Ser 161, were the H-bond acceptors and donors. The e-pharmacophore derived from 4ITJ consisted of ten features (3A, 3D, 1P, 1H, and 2R) (Fig. 5c), where Met 67, Leu 70, Val 76, Ile 154, Asp 156, and Ser 161 were the H-bond acceptors and donors. Thus, the fragment-based e-pharmacophores may have identified all of the possible features present at the RIPK1 binding site.

Fig. 5a–c
figure 5

Fragment-based e-pharmacophores derived from the three RIPK1 protein structures: a from 4ITH; b from 4ITI; c from 4ITJ. The letters A, D, P, H, and R represent acceptor, donor, positive, hydrophobic, and aromatic features, respectively

Full size image

The fragment-based e-pharmacophore derived from RIPK3 structure 1 consisted of six features (2A, 2D, and 2R) (Fig. 6a), where the H-bond acceptors and donors were the amino acids Lys 37, Thr 81, Ala 146, and Glu 47. The e-pharmacophore derived from RIPK3 structure 2 consisted of eight features (3A, 1D, and 4R) (Fig. 6b); Gly 17, Asp 147, Met 84, and Arg 46 were the H-bond acceptors and donors. The e-pharmacophore derived from RIPK3 structure 3 consisted of eight features (2A, 2D, 1 N, 1P, and 2R) (Fig. 6c), where the H-bond acceptors and donors were Gly 17, Lys 131, Thr 81, Ala 50, Lys 37, Asp 129, and Asp 147, while the e-pharmacophore derived from RIPK3 structure 4 consisted of four features (2A, 1D, and 1H) with Ser 133, Lys 131, and Asp 147 as H-bond acceptors and donors (Fig. 6d).

Fig. 6a–d
figure 6

Fragment-based e-pharmacophores derived from the four RIPK3 protein structures: a from structure 1; b from structure 2; c from structure 3; d from structure 4. The letters A, D, N, P, H, and R represent acceptor, donor, negative, positive, hydrophobic, and aromatic features, respectively

Full size image

Ensemble pharmacophores

The ensemble pharmacophore of RIPK1 was created by combining all of the fragment-based e-pharmacophore features derived from the three RIPK1 proteins, as described previously [26]. It consisted of eleven features (3A, 6D, and 2R) (Table 1) corresponding to the amino acids Met 67, Leu 70, Val 76, Leu 78, Ile 154, Asp 156, and Ser 161 (Fig. 7). Thus, this ensemble pharmacophore covered all of the important amino acids at the binding site of RIPK1 that may be involved in protein–ligand interactions.

Table 1 Ensemble pharmacophore features of RIPK1 and their respective three-dimensional coordinates
Full size table
Fig. 7
figure 7

RIPK1 ensemble pharmacophore derived by combining the three RIPK1 e-pharmacophores. The letters A, D, and R represent acceptor, donor, and aromatic features, respectively

Full size image

The ensemble pharmacophore of RIPK1 was validated using the early enrichment factors at 1 and 5 % of the database screened. EF1% and EF5% were 30 and 8, respectively, indicating that the ensemble pharmacophore contained the features present in the active ligands.

The RIPK3 ensemble pharmacophore was constructed by combining all of the fragment-based e-pharmacophore features extracted from the four RIPK3 protein structures. It consisted of seventeen features (8A, 5D, 1P, 1H, and 2R) (Fig. 8) (Table 2) corresponding to the amino acids Gly 17, Lys 37, Arg 46, Glu 47, Ala 50, Thr 81, Met 84, Asp 129, Lys 131, Ser 133, Ala 146, and Asp 147. Thus, the pharmacophore features corresponding to all of the important amino acids in the RIPK3 protein structures that might interact with the ligands were retrieved through the ensemble pharmacophore.

Fig. 8
figure 8

RIPK3 ensemble pharmacophore derived by combining the four RIPK3 e-pharmacophores. The letters A, D, P, H, and R represent acceptor, donor, positive, hydrophobic, and aromatic features, respectively

Full size image
Table 2 Ensemble pharmacophore features of RIPK3 and their respective three-dimensional coordinates
Full size table

RIPK1 EPS and PEPSF

The RIPK1 ensemble pharmacophore was employed to screen the ZINC database of compounds. Approximately 750,000 compound hits were retrieved, which were subjected to PESF. This resulted in 450,000 compounds. PEPSF eliminated nearly 300,000 of those compounds, greatly reducing the computational time required for the docking step.

RIPK3 EPS and PEPSF

The compounds retrieved through RIPK1 PEPSF were subjected to RIPK3 EPS and PEPSF. Approximately 300,000 compounds were retrieved through EPS and 100,000 compounds were obtained after PEPSF. These compounds could contain pharmacophore features of both RIPK1 and RIPK3 and did not harbor any cross-features. Interestingly, Nec-1 exhibited pharmacophore features of the RIPK1 protein only. Finally, applying PEPSF yielded 12 groups of compounds (Fig. 2): group 1 contained pharmacophore features of RIPK1 protein 1 and RIPK3 protein 1, group 2 contained features of RIPK1 protein 1 and RIPK3 protein 2, group 3 contained features of RIPK1 protein 1 and RIPK3 protein 3, group 4 contained features of RIPK1 protein 1 and RIPK3 protein 4, group 5 contained features of RIPK1 protein 2 and RIPK3 protein 1, group 6 contained features of RIPK1 protein 2 and RIPK3 protein 2, group 7 contained features of RIPK1 protein 2 and RIPK3 protein 3, group 8 contained features of RIPK1 protein 2 and RIPK3 protein 4, group 9 contained features of RIPK1 protein 3 and RIPK3 protein 1, group 10 contained features of RIPK1 protein 3 and RIPK3 protein 2, group 11 contained features of RIPK1 protein 3 and RIPK3 protein 3, and group 12 contained features of RIPK1 protein 3 and RIPK3 protein 4.

Multiple dockings

Different groups of compounds were subjected to multiple dockings with their respective RIPK1 and RIPK3 protein structures (Fig. 2). The resultant top-ranked compounds showed good fitness and Glide scores and also exhibited good interactions with the desired amino acids of both the RIPK1 and RIPK3 proteins (Table 3). Lead 1 showed H-bonds with the Met 67, Val 76, and Asp 156 amino acids of RIPK1 (Fig. 9a), and with Lys 37 and Ser 133 of RIPK3 (Fig. 10a). Lead 2 showed H-bonding with Val 76, Asp 156, and Ser 161 of RIPK1 (Fig. 9b), and with Lys 131 and Asp 147 of RIPK3 (Fig. 10b). Similarly, lead 3 exhibited H-bonds with Leu 70, Ile 154, Asp 156, and Ser 161 of RIPK1 (Fig. 9c), and with Ser 133 and Asp 147 of RIPK3 (Fig. 10c). All of the amino acids of RIPK1 and RIPK3 that formed H-bonds with the lead compounds complemented the features present in their respective ensemble pharmacophores.

Table 3 Output from the docking of the top three ranking lead molecules with the RIPK1 and RIPK3 proteins
Full size table
Fig. 9a–c
figure 9

Output from the docking of the top-ranking ligands with RIPK1: a ligand ZINC01659029; b ligand ZINC71828321; c ligand ZINC72165687

Full size image
Fig. 10a–c
figure 10

Output from the docking of the top-ranking ligands with RIPK3: a ligand ZINC01659029; b ligand ZINC71828321; c ligand ZINC72165687

Full size image

Validation of leads

The binding and interaction of the top two ranked leads with the RIPK1 and RIPK3 proteins were validated through MD simulation studies. These leads remained in the binding sites of each protein throughout the simulations. The potential energy and the backbone RMSD for each protein and for each ligand complexed with each protein were steady and they stabilized during the course of the simulation (Figs. 11 and 12). However, Nec-1 was only found to bind to RIPK1; this complex remained stable throughout the simulation. It did not bind to RIPK3, so the corresponding complex was not stable during the simulation. This indicates that Nec-1 may not inhibit RIPK3 (Fig. 13), in line with a previous study which reported that Nec-1 only inhibits RIPK1, not RIPK3 [19]. This result represents a validation that the compounds identified using our strategy may inhibit both RIPK1 and RIPK3.

Fig. 11a–d
figure 11

MD simulation results for the binding of lead 1 to the RIPK1 and RIPK3 proteins: a RMSD of the RIPK1 protein; b RMSD of the lead 1 with RIPK1; c RMSD of the RIPK3 protein; d RMSD of lead 1 with RIPK3

Full size image
Fig. 12a–d
figure 12

MD simulation results for the binding of lead 2 to the RIPK1 and RIPK3 proteins: a RMSD of the RIPK1 protein; b RMSD of lead 2 with RIPK1; c RMSD of the RIPK3 protein; d RMSD of lead 2 with RIPK3

Full size image
Fig. 13a–d
figure 13

MD simulation results for the binding of Nec-1 to the RIPK1 and RIPK3 proteins: a RMSD of the RIPK1 protein; b RMSD of Nec-1 with RIPK1; c RMSD of the RIPK3 protein; d RMSD of Nec-1 with RIPK3

Full size image

Conclusions

Emerging evidence has confirmed that necroptosis does not necessarily require the interaction of RIPK1 and RIPK3 and the formation of a RIPK1–RIPK3 complex. Both of these proteins can instigate cell death independently, and are thus important targets in neurological disorders. In this study, a novel strategy involving multiple ensemble pharmacophores, PEPSF, and multiple dockings was developed to identify inhibitors of both RIPK1 and RIPK3. Recent studies have reported that employing multiple structures, with different conformations, of a single protein may permit the identification of structurally diverse potential inhibitors. Merged pharmacophores and ensemble pharmacophores are derived from these structures in order to minimize the screening time, since the compound database can be screened in a single step (rather than needing to screen the compounds multiple times, which is necessary when using multiple pharmacophores). However, the compound hits may include numerous false positives with cross-features; docking these false positives may result in a great deal of wasted computational time. Consequently, the computational time that was saved by using merged pharmacophores and ensemble pharmacophores could be lost at the docking phase. PEPSF is able to circumvent this problem by filtering out all of the false positives and grouping the compounds according to individual pharmacophores.

The top-ranked compounds identified in the present study had pharmacophore features that were complementary to the binding sites of both RIPK1 and RIPK3. These compounds exhibited stable interactions and formed stable complexes with both proteins throughout the course of each MD simulation. Further, Nec-1 could only bind to RIPK1; it did not form a stable complex with RIPK3. Therefore, the present strategy might facilitate the rapid in silico identification of effective multitarget inhibitors, and the compounds identified in this study may prove to be potential RIPK1 and RIPK3 dual inhibitors.