The assay employed here is based on phosphorylation of a peptide termed LRRKtide derived from the cytoskeleton-associated moesin protein (Jaleel et al., 2007) and is adapted to a phosphor imaging readout (Asensio and Garcia, 2003; Taymans et al., 2011), as described in Materials and Methods and shown in Figure 1. The quality of the chosen assay is given by the average Z' factor for this assay which we determined to be 0.82 (Figure 2), well within the range of 0.5–1 which is considered an excellent value for screening assays (Zhang et al., 1999). A panel of 160 kinase inhibitors was tested in the LRRK2 in vitro kinase activity assay using GST-LRRK2r_970−2527 at one concentration (10 μM) (Figure 1, quantifications given in Table 1). Of these 160 compounds, 35 compounds from three compound classes (A) 9-methyl-N-phenylpurine-2,8-diamine, (B) N-phenylquinazolin-4-amine, and (C) 1,3-dihydroindol-2-one analogs were selected for further testing in the in vitro assay as well as for in silico analysis (see further). These three compound classes were chosen because they contain compounds with a common core or scaffold possessing a wide variety of activities. Additionally, known LRRK2 reference compounds were tested including LRRK2-IN1, CZC-25146 and the pan-kinase inhibitor staurosporine as well as compound CDK1/2 inhibitor III which displayed highest potency in inhibiting LRRK2. The selected compounds were retested with full length recombinant enzyme at 100 μM, 10 μM, 1 μM and 100 nM. For those compounds that inhibited LRRK2 kinase activity more than 50% at 1 μM, a pIC₅₀ [=−log(IC50)] value was determined as described in Materials and Methods (Figure 2 and Table 2). It can be noted that for known LRRK2 inhibitory compounds that we included in our testing, our adapted LRRKtide phosphorylation assay yielded similar potencies to those previously published (comparative examples of obtained IC50 values are: LRRK2-IN1 3.54 nM vs. 13 nM (Deng et al., 2011), CZC-25146 1.78 nM vs. 4.76 nM (Ramsden et al., 2011), with values given from this study and from published studies, respectively; for the LRRK2-IN1 comparison, the slightly lower value obtained here may be due to the ATP concentrations used which are 10 μM in the present study and 100 μM in the original study).

Using the same panel of 160 kinase inhibitors, cellular activity was assayed by monitoring dephosphorylation of LRRK2 at Ser935 induced by kinase inhibitor treatment (5 μM for 2 h) of the SH-SY5Y cell line with stable expression of LRRK2 as described in Materials and Methods and shown in Figure 3. The mean Z' factor for the dual detection immune-dotblot assay used here is 0.65 (Figure 4). A total of 20 compounds were found to reduce Ser935 phosphorylation levels to less than 50% of control levels (Tables 1, 2), all are ATP-binding site competitive compounds. None of the 20 non ATP-competitive compounds of the panel (see Materials and Methods) induce more than 50% dephosphorylation of LRRK2 at 5 μM although AG490 shows 49.32% dephosphorylation of LRRK2 at 5 μM. Representative dot blot images and bar diagrams are depicted in Figure 3, exact quantifications are given in Table 1. For the selected compounds, the IC50 was determined (Figure 4 and Table 2).

Correlations between the in vitro and cellular activities for each compound were investigated by drawing up correlation plots for these two parameters and performing linear regression analysis as described in Materials and Methods (Figure 5). This analysis showed a significant correlation between in vitro and cellular activity (with Pearson's r coefficient of −0.7953).

We constructed, optimized and quality improved a 3D homology model of the LRRK2 kinase domain as described in detail in Materials and Methods. Based on their sequence identity with the LRKK2 kinase domain, the tyrosine-kinase like (TKL) kinases B-Raf (PDB 3OG7; Bollag et al., 2010), MLK1 (PDB 3DTC; Hudkins et al., 2008), and IRAK-4 (PDB 2NRU; Wang et al., 2006) were selected as templates to model LRRK2 kinase (see Table 3 for an overview of TKL kinases with available 3D structures and their sequence identity with the LRRK2 kinase domain). The alignment between the LRRK2 kinase domain and these three kinases is given in Figure 6A. The final homology model colored by conserved kinase motifs is shown in Figure 6B; the final model colored by quality of each predicted amino-acid position is given in Figure 6C and was determined as described in the Materials and Methods section. The ATP-binding groove lies at the interface of the N-and C-terminal lobes (Huse and Kuriyan, 2002; Nolen et al., 2004).

A preliminary docking step, where staurosporine was docked in the LRRK2 ATP-binding site, was applied to optimize the local environment to get the most optimal binding pose during the subsequent docking step. 35 compounds from three compound classes (A) 9-methyl-N-phenylpurine-2,8-diamine, (B) N-phenylquinazolin-4-amine, and (C) 1,3-dihydroindol-2-one analogs were docked into this active site using three class-specific pharmacophore models (Figures 7A–C) as described in detail in Materials and Methods. A summary of the in silico docking scores, given as the Generalized-Born Volume Integral/Weighted Surface area dG (GBVI/WSA dG) is given in Table 2. To evaluate the correlation between docking and in vitro and cellular compound activities, receiver operating characteristic plots describing the trade-off between sensitivity and specificity were constructed using the GVBI/WSA dG computed values and the measured in vitro activities in the LRRKtide assay (IC50 of 1 μM or better is scored active) or the measured pS935 cellular dephosphorylation (a measure of greater than 50% dephosphorylation at 5 μM is scored as active). Receiver operating characteristic plots which trace above the diagonal signify docking enrichment, with best docking for those receiver operating characteristic plots furthest above the diagonal. An indicative parameter of the receiver operating characteristic plot is the area under the curve (AUC) with values above 0.5 indicating a valid correlation between in silico and measured activity values. The receiver operating characteristic plots were determined for the LRRK2 kinase model as well as the three separate kinase structures which were used as templates for constructing the LRRK2 kinase model. These receiver operating characteristic plots, displayed in Figures 7D–K clearly show that of these four models, only docking results obtained with the LRRK2 kinase domain model itself have predictive value for in vitro kinase activity and cellular pS935 dephosphorylation activity for at least three different structural classes of LRRK2-active compounds. Illustrative LRRK2 kinase—ligand binding poses for representative active compounds representing the three different classes are given in Figure 8.

LRRK2 kinase activity was assessed using an isotopic peptide substrate assay essentially as described in reference (Taymans et al., 2011). In short, recombinant LRRK2 was incubated with 6 μCi γ−³²P-ATP [3000 Ci/mmol; Perkin Elmer (USA)], 200 μM LRRKtide (RLGRDKYKTLRQIRQ) (Jaleel et al., 2007) [Enzo Life Sciences (USA)], 10 μM ATP and kinase inhibitor (see below) or dimethylsulfoxide (DMSO) solvent per 40 μl reaction in 1× kinase buffer for 30 min at 30°C. The composition of 1x kinase buffer is: Tris 25 mM pH 7.5, MgCl₂ 10 mM, dithiothreitol (DTT) 2 mM, Triton 0.02%, beta-glycerophosphate 5 mM, Na₃VO₄ 0.1 mM. DMSO content of each in vitro kinase reaction was 1 %. For the single dose testing (at 10 μM of kinase inhibitor), the LRRK2 enzyme used was GST-tagged truncated LRRK2 containing residues 970 to 2527 (Life Technologies). Reactions were stopped by the addition of 500 mM EDTA containing bromophenol blue. Reactions were spotted to P81 Whatman phosphocellulose paper (GE Healthcare) and washed four times 10 min in 75 mM phosphoric acid. LRRKtide phosphorylation levels were measured via autoradiography (Asensio and Garcia, 2003).

A commercially available panel of 160 kinase inhibitors (EMD4Biosciences, Inhibitor select panel) was initially screened at one concentration (10 μM). The inhibitor panel contains cell-permeable and previously characterized inhibitors which together target all branches of the kinome. All inhibitors are confirmed cell permeable, with the exception of PKCβ Inhibitor, PKR Inhibitor—Negative Control, Alsterpaullone—2-Cyanoethyl, Cdk1/5 Inhibitor and JNK Inhibitor IX which are of unknown permeability. Compounds in the inhibitor panel are mostly ATP-binding site competitive inhibitors, although 20 compounds are labeled as non ATP-competitive compounds (including 1 allosteric compound). These compounds are: Bcr-abl Inhibitor, AG 490, AG 112, Akt Inhibitor X, AG 1024, Akt Inhibitor V—Triciribine, Akt Inhibitor VIII, Isozyme-Selectiv—Akti-1/2, Chelerythrine Chloride, MEK1/2 Inhibitor, MNK1 Inhibitor, KN-62, Cdk4 Inhibitor II—NSC 625987, ERK Inhibitor III, MK2a Inhibitor, MEK Inhibitor I, Sphingosine Kinase Inhibitor, PD 98059, GSK-3b Inhibitor I, KN-93 and the allosteric inhibitor IGF-1R Inhibitor II. Information on the branch of the kinome targeted is provided together with experimental results in Table 1; further details on each inhibitor are available from the supplier. Then, selected molecules were further tested at lower doses. For those selected compounds that were found to inhibit >50% at 1 μM, an IC₅₀ (inhibitor concentration yielding 50% inhibition) was determined. To this end, compounds were tested in triplicate with full length LRRK2 protein (Taymans et al., 2011; Civiero et al., 2012) at concentrations of 10 μM, 1 μM, 300 nM, 100 nM, 30 nM, 10 nM, 3 nM, 1 nM, 100 pM, 10 pM, and solvent. An inhibition curve was fitted and IC50s calculated using GraphPad Prism version 5.01 for Windows (GraphPad Software, San Diego, CA, USA). IC50s are expressed as pIC50 [=−log(IC50)].

To assess LRRK2 cellular activity, an SH-SY5Y stable cell line was first generated. SH-SY5Y cells were cultured in Dulbecco's modified Eagle's medium supplemented with 15% fetal bovine serum and 1× non-essential amino acids (Gibco) at 37°C and 5% CO₂. Lentiviral vectors (LVs) encoding human 3xFlag-LRRK2 under control of the cytomegalovirus (CMV) promoter and co-expressing a hygromycin selection marker via an internal ribosomal entry site element were prepared and used for cellular transduction as previously described (Civiero et al., 2012). Following selection in medium containing 200 μg/ml hygromycin, cells were expanded for use in experiments.

Cells were plated out into 96-well plates. When wells were >80% confluent, cells were treated with kinase inhibitors by dilution of the compounds into the cell culture medium to the desired final concentration. Following a 2 h incubation of the cells with kinase inhibitors, cells were immediately rinsed in PBS and lysed in lysis buffer [20 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 0.5% Tween 20 or 1% Triton X-100, 2.5 mM sodium pyrophosphate, 1mM beta-glycerophosphate, 1 mM NaVO₄, protease inhibitor cocktail and phosphatase inhibitor cocktail (Roche)]. Lysates were centrifuged for 30 min at 14000 × g. Supernatant was spotted to hydrated pvdf membranes and LRRK2 phospho-Ser935 levels as well as total LRRK2 levels were sequentially determined by immunoblot detection using the rabbit monoclonal anti-phospho-S935-LRRK2 antibody [Epitomics, clone UDD2 10(12)] and the mouse monoclonal anti-LRRK2 antibody N138/6 (Neuromab) or flag-M2 antibody followed by incubation with appropriate secondary horseradish peroxidase coupled antibodies and chemiluminescent detection using ECL plus HRP substrate (Thermo Scientific, Rockford, IL, USA). Densitometric analysis of the immunoreactive spots was performed using Aida analyzer v1.0 (Raytest, Straubenhardt, Germany). Phosphorylation levels were determined by the ratio of phospho-LRRK2 to total LRRK2, normalized to solvent controls.

To determine Z' of the in vitro LRRKtide assay or the cellular Ser935 dephosphorylation assay, the following formula was used: Z′=1−(3σ+c+3σ−c| μ−c−μ+c |) where σ_+c, σ_−c, μ_+c, and μ_−c are the standard deviation (σ) and mean (μ) of the positive control samples (+c, LRRK2-IN1 10 μM for the LRRKtide assay, CZC 10 μM treated samples in the pS935 dephosphorylation assay) or negative control samples (−c, DMSO treated samples). Results are based on values from 3–10 replicates from the same assay run.

All the following steps were conducted using MODELLER 9v9 (Sali and Blundell, 1993), unless otherwise stated. First, human TKL kinases with a DFG-in activation loop conformation (i.e., active conformation) available at that time in the “Kinase Database” implemented in Molecular Operating Environment (MOE, Chemical Computing Group Inc., 1010 Sherbooke St. West, Suite #910, Montreal, QC, Canada, H3A 2R7, 2013) were structurally aligned using the alignment.salign command. Afterwards, this structure domain was superimposed on the LRRK2 kinase domain. The TKL kinases B-Raf (Rapidly accelerated fibrosarcoma) (PDB 3OG7; Bollag et al., 2010) and MLK1 (Mixed Lineage Kinase 1) (PDB 3DTC; Hudkins et al., 2008) and—for one short stretch of 11 amino-acids in the N-terminal region of LRRK2 kinase (corresponding to LRRK2 amino acids 1872–1882 found just N-terminal of the P-loop, see Figure 6A)—IRAK-4 (Interleukin-1 Receptor–Associated Kinase 4) (PDB 2NRU; Wang et al., 2006) displayed the highest identity with the LRRK2 kinase domain and were chosen as templates. The final alignment between B-Raf, MLK1, IRAK-4, and the LRRK2 kinase domain is shown in Figure 6A. A 3D model of LRRK2 kinase domain was calculated by satisfaction of spatial restraints and screened for unfavorable regions by computing the Discrete Optimized Protein Energy (DOPE) score per residue. The alignment was modified using iterative alignment-modeling-evaluation steps until no improvement could be found. For the top-scoring alignment, multiple models were computed and subjected to a rough refining procedure: each model is optimized with the variable target function method with conjugate gradients and further refined using molecular dynamics with simulated annealing. The LRRK2 kinase domain model with the best DOPE score, GA341 score and Modeler objective function was selected.

Explicit hydrogen atoms were added and the model was subjected to a more thorough refining procedure with MOE using the AMBER99 force field with Born solvation model. First, all inconsistencies and outliers were selected (as observed from values in the Ramachandran plot, backbone bond angles and lengths, the rotamer strain energy, and atom clashes), the other residues were potentially fixed and an energy minimization (EM) was performed with backbone atoms restrained to 100. The EM was terminated when the RMS gradient fell below 0.1. A final EM was performed on all atoms with all backbone atoms restrained to 10. This minimization was terminated when the RMS gradient fell below 0.1.

Model quality has been checked by computational methods, giving us a good validation of the reliability of the model. The Ramachandran Plot assured very good confidence: only 0.7% residues in the disallowed region and 2.1% residues outside generously allowed regions. 2.9% of the residues had unfavorable bond angles and 1.1% had unfavorable dihedrals. However, most of these residues were oriented away from the active site (Figure 6C). Assessment of model quality using Meta-MQAPII (Pawlowski et al., 2008) gave absolute global deviations, expressed as RMS deviation (3.47 Å) and Global Distance Test Total Score (65.98), for the model vs. the unknown true structure, indicating a medium quality model. The Meta-MQAPII score per residue is shown in Figure 6C. The only unfavorable regions were the loop regions, especially the activation loop, explained by the lack of a good template for these regions (Figure 6A). Overall, the activation loop is also rather flexible for kinases (un-crystallized region). Since these unfavorable modeled loops are not part of the ligand binding site, we proceeded with the LRRK2 kinase domain homology model. PyMOL and MOE were used as a visualization tool. The model is freely available upon request to the corresponding authors.

The kinase inhibitor database was supplied as a two-dimensional structure data file by EMD4Biosciences (USA). Using the MOE Structure Database Tools (sdwash, sdcharges, and sdstereo commands), the database was curated. Based on the fact that different stereoisomers may have different activities, molecular docking simulations were carried out for both stereoisomers. Generation of 3D structures was done via the energy minimize command using default settings. We used the MMFF94x force field. All data were stored in a MOE molecular database file.

Common substructures, based on the analysis of PDB kinase ligands by Ghose and co-workers (Ghose et al., 2008), were found using Instant JChem (ChemAxon, Hungary).

After homology modeling with MODELLER (Sali and Blundell, 1993), the generated alternative conformations of the ligands were docked into the active site using MOE. A preliminary docking step, where staurosporine was docked in the LRRK2 active site, was applied to optimize the local environment to acquire the most optimal binding pose in the subsequent docking step. The crystal structures B-Raf (PDB 3OG7), MLK1 (PDB 3DTC), and IRAK-4 (PDB 2NRU) were protonated using Protonate3D module of MOE.

For docking simulations, initial binding conformations were generated for the purine, quinazoline, and oxindol derivatives (termed compound classes A–C) of the kinase inhibitor database. These initial binding conformations were refined using pharmacophore models for these three compound classes (Figures 7A–C). The pharmacophore models were based on structural elements, shared by all derivatives in one class that are essential for interaction with LRRK2. During the refinement step, ligands that fulfilled the pharmacophore hypothesis were allowed to advance and an optimized binding conformation was saved in a MOE molecular database file. Ligands that didn't satisfy the pharmacophore requirements were excluded from subsequent steps. The DOCK module was used and default settings were applied. A force field based scoring function was used: GBVI/WSA dG. After docking, the results were collected by receptor (i.e., docking values obtained with each of the four kinase structures tested). For each ligand the best scoring (e.g., lowest energy) docking pose was kept.

Receiver operating characteristic plots are useful as a graphical illustration of the performance of the in silico docking strategy as they can evaluate the computed docking values together with measured activities (either in vitro or cellular). Here, we plotted the receiver operating characteristic plots as false positive rate (equivalent to the 1—specificity) vs. the true positive rate (equivalent to the sensitivity), therefore docking strategies which plotted on average above the diagonal can be considered to have predictive value (this is also reflected by the AUC which is >0.5 for predictive docking strategies). In more detail, for each ligand, the activity/inactivity was indicated by adding 1 and 0 respectively based on the in vitro LRRKtide assay or cellular pS935 dephosphorylation assay. Ligands with at least 50% inhibitory activity at 1 μM in the LRRKtide assay or at least 50% inhibitory activity at 5 μM in the cellular pS935 dephosphorylation assay were designated as active. A receiver operating characteristic plot was generated for each receptor (i.e., for each kinase model) using CROC v1.0 (Swamidass et al., 2010). Plots were made with Deltagraph V7 (2014 RedRock Software, Salt Lake City, USA).

To evaluate the correlation between the different activities obtained for tested kinase inhibitors, pairs of activity values were plotted against each other in GraphPad Prism 5.01 (GraphPad Software Inc.). Linear regression analysis was performed and a trendline was drawn as well as the 95% confidence band. Finally, Pearson's r-coefficient was calculated and a two tailed correlation significance test performed (GraphPad). The level of statistical significance was set at P < 0.05.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.