Chemical IN04 Inhibits the Kinase Domain not the ROC Domain of LRRK1: Results from Homology Modeling and Molecular Docking

Abstract

Background:

Bone loss is the most common reason for broken bones among the elderly. An ideal agent for treatment of bone loss should have both osteoclast inhibitory and osteoblast stimulatory functions. Leucine rich repeat kinase 1 (LRRK1) is a novel target for alternative anti-resorptive drugs to treat osteoporosis and osteoporotic fractures. Recently a chemical IN04, Methyl 3-[({([5-(3,5-dimethoxyphenyl)-1,3,4-oxadiazol-2-yl]-thio}-acetyl)-amino]-benzoate, has been identified as a potential LRRK1 inhibitor.

Objective:

The aim of this work is to investigate how the chemical IN04 interacts with LRRK1 and inhibits its activity.

Methods:

A structural model of LRRK1 kinase domain was constructed with SWISS-MODEL. The human protein kinase ROCO4 (PDB ID: 4YZN) was chosen as the template based on sequence homology, structural and phylogenetic analysis. In addition, a homology model of the LRRK1 ROC domain was also prepared based on the LRRK2 ROC domain structure (PDB ID: 2ZEJ). The interactions of IN04 with the active sites in the LRRK1 kinase domain and ROC domain were investigated by SwissDock.

Results:

IN04 was docked into the active site of the LRRK1 kinase domain with similar interactions as ATP comparable to the ligand bound to homologous kinases. Many rational binding modes of IN04 to LRRK1 kinase domain were investigated and the most likely binding pose containing multiple hydrogen bonds and a salt bridge was discovered. However, IN04 cannot fit into the GDP-binding site of the ROC domain.

Conclusion:

Chemical IN04 inhibits LRRK1 by binding to the active site of the kinase domain but not the ROC domain.

Graphical Abstract:

Chemical IN04 is an inhibitor of the kinase domain not the ROC domain of LRRK1.

1. INTRODUCTION

Bone loss occurs with age in part because of imbalance between the processes of osteoclast-mediated bone resorption and osteoblast-medicated bone formation. The ideal agents for management and treatment of bone loss should have both osteoclast inhibitory and osteoblast stimulatory functions. However, such drugs are currently not available. At present, most available drugs function to limit bone resorption, either directly or indirectly targeting osteoclasts. These antiresorptive agents include bisphosphonates, a selective estrogen receptor modulator raloxifene, and a humanized monoclonal antibody specific to RANKL (receptor activator of nuclear factor-κB ligand).

However, treatment with bisphosphonates results in suppression of both bone resorption and bone formation. Long term treatment with bisphosphonate drugs may impair bone healing after tooth extraction, cause jaw osteonecrosis, and increase the risk for atypical fractures of the femur [1–3]. Treatment of postmenopausal women with an anti-RANKL monoclonal antibody was effective, however the bone turnover rebounds after 2 years’ discontinuation of injection [4]. Therefore, novel anti-resorptive molecules that avoid anti-anabolic actions are needed as therapeutics to increase bone mass and reduce osteoporotic fractures.

In our previous studies, we have demonstrated that mice with disruption of the leucine rich repeat kinase 1 (LRRK1) gene displayed severe vertebral and long bone osteopetrosis resulting from the dysfunction of mature osteoclasts. While osteoclast-mediated bone resorption in LRRK1 knockout mice is reduced dramatically, bone formation is not and respond to anabolic parathyroid hormone treatment, but they are resistant to ovariectomy-induced bone loss [5]. Consistent with the phenotypic changes in LRRK1 knockout mice, a patient with an autosomal recessive mutation of LRRK1 suffered from severe osteosclerosis confined to the metaphysis of the long and short tubular bones due to osteoclast dysfunction [5–6]. These mouse and human genetic studies strongly suggest that LRRK1 plays a critical role in regulating osteoclast function. Therefore, LRRK1 is a novel target for alternative anti-resorptive drugs to treat osteoporosis and osteoporotic fractures.

Direct high throughput screening of small molecular libraries for LRRK1 inhibitors by testing inhibitory phosphorylation or autophosphorylation via a LRRK1 kinase assay is not feasible because the key direct biological substances have not been identified yet. Although the 3D structures of the ROCO4 superfamily including LRRK2 have been partially resolved and used as receptors for structure-based drug screening [7–9], the crystal structure of LRRK1 kinase domain has not been resolved yet. Recently, we performed structure homology modeling and virtual screening, and identified a potential LRRK1 inhibitor IN04 [10], Methyl 3-[({([5-(3,5-dimethoxyphenyl)-1,3,4-oxadiazol-2-yl]-thio}-acetyl)-amino]-benzoate. IN04 treatment of osteoclasts significantly impaired their ability to resorb bone with no effect on osteoclast formation in in vitro osteoclast differentiation and pit assays [10]. However, a detailed IN04 interaction with LRRK1 kinase has not been elucidated yet. In this study, we analyzed how IN04 interacts with homology models of the kinase and ROC domains of the LRRK1 kinase. Our study supports previous in vitro function studies of IN04 on osteoclasts and warrants future tests as a clinical drug candidate.

2. MATERIALS AND METHOD

2.1. Homology modeling and molecular docking

Because crystal structure of the LRRK1 kinase domain (KD) and ROC domain has not been reported, we performed protein structure homology modeling with SWISS-MODEL. Based on a BLASTP search and/or the rank in SWISS-MODEL templates search, protein kinase ROCO4 complexed with a ligand named compound 19 (PDB ID: 4YZN) was chosen as the modeling template for LRRK1-KD modeling and LRRK2 ROC domain dimer complexed with GDP (PDB ID: 2ZEJ) was chosen as the template for LRRK1-ROC modeling.

For the molecular docking, the PDB file of the proteins such as LRRK1-KD and the .mol2 file of small molecule such as IN04 were loaded into the online server of SwissDock, and the potential ligand binding sites were revealed.

3. RESULTS AND DISCUSSION

3.1. Homology Modeling of LRRK1 kinase domain (KD)

LRRK1 is a 2015 amino acid residues protein which belongs to the ROCO family and contains a serine/threonine kinase domain. Five most homologous kinases with known 3D protein structures deposited in the Protein Data Bank (PDB) were identified by a BLASTP search using LRRK1 KD covering residues 1222–1526 as the query. A sequence alignment was made using UniproKB alignment (Fig. 1A). These five homologous kinases have 27.50% – 32.18% sequence identity with LRRK1 KD. E-values (expectation values) are in the range of 2e⁻³¹ to 2e⁻²¹.

Fig. 1. Sequence comparison of LRRK1 with its homologues.

(A) Sequence alignment of LRRK1-KD (residue 1222–1526) with 5 homologous kinases. Human protein kinase ROCO4, mixed-lineage kinase MLK1, Mitogen-activated protein kinase kinase kinase MLK4, Serine/threonine-protein Kinase ROCO4, and transforming growth factor-beta (TGF-beta)-activated kinase 1 TAK1) are identified from on the Protein Data Bank (PDB) based on a BLASTP search. PDB IDs for the 5 kinases (in the above order) are 4YZN, 3DTC, 4UYA, 4F0F and 2EVA, respectively. Identical residues are colored in the darkest blue and highly similar residues are colored in lighter blues. Most conserved motifs/regions (GXGXXG motif, hinge region, HRD motif and DFG motif) that are reported to participate in ATP binding are indicated. Sequence identity percentage between LRRK1-KD and the 5 homologous kinases (in the above order) are 30.64, 31.25, 32.18, 29.79, and 27.5; and E-values (expectation values) are 2e-31, 3e-29, 5e-29, 2e-25, and 2e-21, respectively. (B) Sequence alignment of the LRRK1-ROC (residue 634 – 835) domain with the LRRK2-ROC (residue 1335 – 1515) domain. Sequence identity and similarity between these two sequences are 27% and 48%, respectively. Identical residues are highlighted by blue color.

A typical protein kinase fold comprises a smaller N-terminal lobe and a larger C-terminal lobe connected by a flexible hinge region. This hinge region is located in the cleft between the 2 lobes and forms part of the catalytic active site that adenosine binds. In addition to the hinge region, three motifs named GxGxxG motif (or P loop), DFG motif (or base of the activation loop), and catalytic HRD motif have also been reported in positioning and/or catalysis of the substrate [11].

Since there is no LRRK1 KD crystal structure reported yet, a LRRK1 KD structure model was constructed using the SWISS-MODEL web server [12–16]. Human protein kinase ROCO4 complexed with a ligand named compound 19 (PDB ID: 4YZN) [17] was chosen as the modeling template based on four parameters: 1. Its structural resolution (1.5 Å) is the highest and its GMQE (global model quality estimation) value (0.60) [16] is the greatest among the 50 top-ranking templates; 2. It has been reported that mutations of two Phe residues substituted with Leu in the human protein kinase ROCO4 do not change the overall structure or activity of the protein [17]; 3. A phylogenetic tree based on the sequences of the 50 top-ranking templates indicates that the human protein kinase ROCO4 is very close to the LRRK1 kinase domain (Fig. 2). 4. The ROCO4 kinase shares 30.6% sequence identify with LRRK1 KD.

Fig. 2. A phylogenetic tree of homologous kinases of LRRK1-KD.

The tree was obtained with the web server (ngphylogeny.fr) using sequences of LRRK1 KD and the 50 top-ranking protein templates searched by the web server SWISS-MODEL. The templates are labeled using their PDB IDs. Target LRRK1 KD and the template human protein kinase ROCO4 are framed in box. The tree scale is shown below the phylogenetic tree.

The LRRK1-KD model (Fig. 3A) resembles the typical protein kinase fold. Its smaller N-terminal lobe is a 5-stranded anti-parallel β-sheet comprising residues 1241–1334 in which two α-helices connect the strands β3 and β4. The C-terminal lobe covers residues 1342–1522 and consists of mainly α-helices.

Fig. 3. Homology model of LRRK1 kinase domain (KD) and its comparison with 5 homologous kinases.

(A) The structure of LRRK1-KD modeled by SWISS-MODEL shown in cartoon, N terminus and C terminus are indicated; GxGxxG motif, Hinge region, HRD motif and DFG motif are shown in red, blue, pink and gray, respectively; (B) Structural superimposition of the LRRK1-KD model shown in cyan and 5 homologous kinases with ligand complexed shown in different colors with PDB ID 4YZN (green), 3DTC (light blue), 4F0F (purple), 4UYA (orange) and 2EVA (yellow). (C) Structural superimposition for the four conserved ATP binding motifs from the LRRK1-KD model and the 5 homologous kinases shown in ribbons by different colors as in (B).

According to the structure assessment module in the modeling results, the QMEAN Z-score of our model (−3.38) indicates a good agreement between the modeled structure and experimental structures of proteins with a similar size [15]. This model has a score of 1.96 for the MolProbity, which is a log-weighted combination of the clash score, percentage Ramachandran not favored and percentage bad side-chain rotamers, giving one number that reflects the crystallographic resolution at which those values would be expected [18]. The backbone dihedral angle distributions of all amino acid residues (Ramachandran Plot) were 90.71% in preferred regions, 5.36% in allowed regions and 3.93% in Ramachandran outliers.

3.2. Molecular Docking Results

Next, we applied the online molecular docking program SwissDock, which is a web implementation of the EADock DSS algorithm [19–20], to attest the interactions of the potential small molecule inhibitor named IN04 with LRRK1. First, we checked if this web server works for this purpose. There is a small molecule called 70 (Mol70) reported to inhibit the activity of LRRK2 by binding to its ROC domain [21]. Using SwissDock, autodocking was done for Mol70 into the LRRK2 ROC domain, which is taken out from the reported LRRK2 ROC-GDP complex structure (PDB ID: 2ZEJ). The result showed that Mol70 is docked to a very similar position as GDP in the LRRK2-ROC domain (Fig. 4A). We also found that Mol70 can be manually docked by overlaying it with GDP in the LRRK2-ROC domain without any structural clashing. Therefore, we concluded that the web server SwissDock can be used as a tool to identify molecular inhibitors to LRRK1.

Fig. 4. Molecular docking by SwissDock reveals that the binding of IN04 to LRRK1-is similar with ATP binding to homologous kinases.

(A) Mol70 (shown in stick and cyan) is docked to a very similar position as GDP in the LRRK2-ROC domain shown in purple. LRRK2-ROC complexed with GDP structure is shown in cartoon. (B) The LRRK1-ROC model and the clash of IN04 with the GDP binding site from the auto docking result. IN04 (shown in stick and cyan) is positioned different from GDP (shown in yellow) and clashes with the LRRK1-ROC structure. The LRRK1-ROC model dimer is built by SWISS-MODEL using LRRK2 ROC domain dimer structure (PDB ID: 2ZEJ) as the template. (C) Location of IN04 in LRRK1-KD model from the group No.6 solution of LRRK1-KD-IN04 autodocking results by SwissDock and a schematic diagram of the small molecule IN04. GxGxxG motif, Hinge region, HRD motif and DFG motif are shown in different colors as in Fig. 3.

To test if IN04 inhibits LRRK1 activity by competing with GTP/GDP to bind at the active site of the ROC domain, we first built a homology model of the LRRK1 ROC domain since its structure is not reported yet, using the LRRK2 ROC domain structure as the template with the SWISS-MODEL web server (Fig. 4B) [12–16]. The LRRK1 ROC domain shares 27% sequence identity and 48% similarity to the ROC domain of LRRK2 (Fig. 1B). Therefore, the structural model of LRRK1 ROC domain has the high quality of a good homology model. The MolProbity score of the model is 2.23. The backbone dihedral angle distributions of all amino acid residues (Ramachandran Plot) were 94.12% in preferred regions, 4.81% in allowed regions and 1.07% in Ramachandran outliers. Next, we did autodocking for IN04 into the LRRK1 ROC domain model by running SwissDock to test whether it has the similar binding mode as GDP. The docking results showed that most docking solutions (not shown) put IN04 at the positions away from the GDP-binding site in the LRRK1 ROC domain, and the only solution with the IN04 molecule docked at the position close to the GDP-binding site would cause IN04 to clash with the ROC domain (Fig. 4B).

Then we checked if IN04 can be docked into the ATP binding site of LRRK1 kinase domain. We first identified the amino acid residues participating in the kinase-ligand interaction for the 5 homologous kinases (Fig. 5A–5E) to locate the ATP or analogy binding site, which contains GxGxxG motif, Hinge region, HRD motif, and DFG motif [17, 22–25]. Each motif/region from the 5 kinase structures was aligned spatially and it was found that they can be superimposed well especially for the hinge region and HRD motif (Fig. 3B, 3C), suggesting that these parts constitute the conservative ATP binding core in these kinases.

Fig. 5. The interactions between the corresponding ligand and the ATP binding pocket in the 5 homologous kinases of LRRK1-KD.

(A) human protein kinase ROCO4 with compound 19 (PDB ID: 4YZN) [17]; (B) mixed-lineage kinase MLK1 with compound 16 (PDB ID: 3DTC) [26]; (C) mitogen-activated protein kinase MLK4 with ATPgammaS (PDB ID: 4UYA) [23]; (D) serine/threonine-protein kinase ROCO4 (PDB ID: 4F0F) [25]; and (E) transforming growth factor-beta (TGF-beta)-activated kinase 1 TAK1 with adenosine (PDB ID: 2EVA) [22]. The webserver Protein-Ligand Interaction Profiler (PLIP) is used for analyses [27]. On the left panel, the amino acid residues interacting with the ligand are labeled. Hydrogen and halogen bonds are shown in dotted lines. Information for each interaction including residue number, atom of the ligand, and distance between the 2 interacting atoms are listed on the right panel. For water bridges, the distances are shown in a W-A / W-D format where W-A means water-acceptor and W-D means water-donor. The positively charged residue lysine / arginine at the corresponding site with the Lys167 of LRRK1-KD is highlighted with a bold font.

To assess the KD model’s quality at the ATP binding site, we checked the structural similarity for the 4 conservative motifs among the KD model and the 5 kinases from the superimposition (Fig. 3A). The 4 motifs match each other well in all 6 structures and the KD model almost coincides in Hinge region, HRD motif, and DFG motif with 4YZN, which was used as the modeling template (Fig. 3C). Then we searched the Protein Data Bank for ligand-complexed crystal structures of serine/threonine kinases where LRRK1 kinase domain belongs. The crystal structure of aurora-α kinase complexed with AMPPNP (PDB ID: 2DWB) was spatially aligned with our model of LRRK1-KD. It showed that the 4 conservative motifs of both structures were superimposed very well (data not shown). Therefore, we concluded that the LRRK1-KD model is qualified to be used for ligand docking at the ATP binding site.

Based on the molecular docking by the SwissDock web server, 32 clusters with a total 256 binding possibilities for IN04 to KD were sought out. First, we checked the position of IN04 docked in the KD model for each cluster and selected clusters 2, 3, 4, 5, 6, 7, 9, 13, 15, 29 and 31 from the 32 clusters because these 6 clusters locate at the conserved ATP binding site of KD. The binding modes of IN04 to the LRRK1-KD model within each cluster are very similar, if not the same. Then we used a score named FullFitness from the SwissDock web server to assess the binding modes. The smaller the FullFitness is, the more favorable energy the predicted binding mode has [19]. Top 6 ligand binding modes were ranked and positions of the ligand were compared with the 5 ligands bound to the homologous kinase structures (Fig. 6A). Interactions between IN04 and LRRK1-KD for each binding mode were analyzed in detail (Fig. 6B–6G).

Fig. 6. Analysis of the top 6 solutions from molecular docking for LRRK1-KD and ligand IN04.

(A) On the left panel, overlays of IN04 from the top 6 docking results with the ligands from 5 homologous kinases shown in the same colors as in Fig. 3. Each solution is presented as the cluster number followed by the rank number within the cluster. On the right panel, the ranking list based on the FullFitness score. Analysis of interactions between LRRK1-KD and the docked ligand IN04 from the top 6 binding resolutions: (B) 2–1; (C) 3–1; (D) 4–0; (E) 3–6; (F) 5– 1; and (G) 6–0. On the left panel, the ligand and binding amino acid residues are shown in orange and green, respectively. The number of each residue is labeled. Hydrogen bonds and salt bridge are shown in yellow and purple dotted lines, respectively. Information for each interaction including residue number, atom of the ligand, and distance between the 2 interacting atoms are listed on the right panel. For hydrogen bonds, the distances are shown in a H-A / D-A format in which H-A means water-acceptor and D-A means donor-acceptor. The webserver Protein-Ligand Interaction Profiler (PLIP) is used [27].

Although mode No. 2–1 is ranked first with the smallest Fullfitness value, which is 0.942 and smaller than the 2^nd ranked mode No. 3–1 (Fig. 6A), there are only 1 hydrogen bond and 3 hydrophobic interactions between the ligand and the KD model (Fig. 6B) at this mode. From rank 2 to rank 6, Fullfitness values are close to each other, so we focused on the binding mode and ligand-protein interaction. Finally, No. 3–6 and No. 6–0 were selected for further analysis because they have the binding poses most similar to the ligands in the 5 homologous kinases and contain the most interactions between ligand IN04 and LRRK1-KD (Fig. 6E, 6G): 3 hydrophobic interactions and 4 hydrogen bonds for No. 3–6; 4 hydrophobic interactions, 2 hydrogen bonds and an unanticipated salt bridge for No. 6–0.

We analyzed the IN04-KD interaction for each binding mode in cluster 3 and cluster 6 in detail. The KD residues ranging from 1222 to 1526 were re-numbered as 1–305 hereafter for simplicity. For the cluster 3, we found that the KD residues that participate in the ligand binding are Gly31, Ser32, Ile36, Ala116, Leu172 and Asp188 (Fig. 7). These residues all belong to the 4 highly conserved ATP binding motifs in KD (Fig. 1). At the most, there can be 3 hydrophobic interactions and 4 hydrogen bonds between ligand IN04 and KD. Strikingly, for cluster 6, we found a strong salt bridge between the side chain amine group of KD residue Lys167 in the HRD motif and the carboxylate group at one tail of the ligand IN04 is always predicted in all of the 8 binding modes, giving the ligand IN04 a converse orientation compared with that in cluster 3 (Fig. 7–8). Since there are no salt bridges in any of the other 5 homologous kinases (Fig. 5, Table 1) and 2DWB (Table 1) at the corresponding residue site, we think that this interaction may be essential in properly positioning IN04 into the ATP-binding site of LRRK1 KD. Moreover, there are several more residues predicted to be involved in the ligand binding in cluster 6 solutions (Leu27, Gly31, Ile36, Ala47, Lys49, Ala116, Lys167, Asn170, Leu172, Asp188, Glu204 and Thr206) than those in cluster 3 solutions (Fig. 7–8). These residues cover the 4 conserved motifs (Ala47 and Lys49 in a β-sheet after the GxGxxG motif in the KD model, (Fig. 1) and another highly conserved region APE motif (Glu204 and Thr206) predicted for ligand-binding. This suggests that the interaction between IN04 and KD may contain up to 7 hydrogen bonds. Based on this analysis, we conclude that IN04 likely has a higher affinity/selectivity than ATP to KD, in agreement with previous results showing that IN04 at 16 nM completely blocked ATP binding to human LRRK1 kinase domain in an in vitro pulldown assay [10].

Fig. 7. Analysis of the interaction between ligand IN04 and LRRK1-KD for the top 8 binding modes from the cluster 3 of the docking results.

The webserver Protein-Ligand Interaction Profiler (PLIP) is used [27]. The amino acid residues interacting with the ligand are labeled. Hydrogen bonds are shown in dotted lines.

Fig. 8. Analysis of the interaction between ligand IN04 and LRRK1-KD for the top 8 binding modes from the cluster 6 of the docking results.

The webserver Protein-Ligand Interaction Profiler (PLIP) is used [27]. The amino acid residues interacting with the ligand are labeled. Hydrogen bonds and salt bridges are shown in dotted lines.

Table 1.

Interactions of the corresponding ligand with the residue Lys167 of LRRK1-KD and its equivalent residues in the 5 homologous kinases.

Protein	Residue	Interaction	Distance to ligand (Å)
KD model	Lys 167	Salt bridge	3.10
4YZN	Arg 1156	No salt bridge	9.90
3DTC	Lys 270	No salt bridge	7.86
4UYA	Lys 265	H bond	3.23
4F0F	Arg 1156	H bonds	3.30 / 3.01
2EVA	Lys 158	No salt bridge	8.63
2DWB	Lys 258	No salt bridge	4.60

There are 3 methyl (-CH3) groups in the structure of IN04 and in our experiments we found that the solubility of IN04 is very low even with some DMSO. So next we checked if deleting one, two or three -CH3 groups from the IN04 structure will enhance its docking into the LRRK1-KD modeled structure. Molecular docking by SwissDock was done for 5 IN04 variants which has one or more -CH3 group(s) replaced by H atom(s), and the top 10 docking solutions based on the Fullfitness value are listed (Fig. 9). The results showed that if all the 3 -CH3 groups are replaced, there will be the smallest Fullfitness value (Fig. 9E), suggesting that without -CH3 groups, the small molecule IN04 may bind LRRK1-KD better.

Fig. 9. Fullfitness and delta G values for the top 10 docking solutions of IN04 or variants into LRRK1 kinase domain.

Structural formulas of IN04 or variants with one -CH₃ (Me(a), Me(b)), two -CH₃ (2Me(a), 2Me(b)) or three -CH₃ (3Me) replaced with H atom(s) are shown on the top of the table.

CONCLUSION

We conclude that the ligand IN04 identified recently can inhibit the LRRK1 activity by binding to the ATP binding site in the LRRK1 kinase domain, but IN04 is likely not an inhibitor of the LRRK1 ROC domain. Our findings support previous in vitro studies showing that IN04 at 16 nM completely blocked ATP binding to human LRRK1 kinase domain in an in vitro pulldown assay [10]. These studies support chemical IN04 as a potential drug for the treatment of bone loss and warrant future in vivo functional tests.