Pharmacophoric Site Identification and Inhibitor Design for Autotaxin

Autotaxin (ATX) is a potential drug target that is associated with inflammatory diseases and various cancers. In our previous studies, we have designed several inhibitors targeting ATX using computational and experimental approaches. Here, we have analyzed topological water networks (TWNs) in the binding pocket of ATX. TWN analysis revealed a pharmacophoric site inside the pocket. We designed and synthesized compounds considering the identified pharmacophoric site. Furthermore, we performed biological experiments to determine their ATX inhibitory activities. High potency of the designed compounds supports the predictions of the TWN analysis.


Introduction
Autotaxin (ATX) is an enzyme that hydrolyzes lysophatidylcholine (LPC) to lysophosphatidic acid (LPA) in the extracellular matrix [1,2]. The LPA produced in the process binds to the LPA receptors. This leads to a number of signal transduction processes such as cell proliferation, cell survival, cell migration and secretion of cytokines and chemokines [3,4]. These signal transduction processes are associated with several pathological conditions such as angiogenesis, chronic inflammation, fibrosis, metastasis and various cancers. The LPA-related signal transduction processes have been actively studied in the inflammatory diseases including asthma, arthritis, fibrosis of the lungs, liver and skin and various cancers [5][6][7][8][9]. Several ATX inhibitors have been reported up till now [10][11][12] but no ATX inhibitor has been approved for treatment yet. GLPG-1690 reported by Galapagos NV [13] is merely one ATX inhibitor that is under clinical trials for the treatment of idiopathic pulmonary fibrosis. This compound has completed Phase I (ClinicalTrials.gov Identifiers: NCT02738801, NCT03143712, NCT02179502, and NCT03515382) and Phase II (ClinicalTrials.gov Identifier: NCT02738801) clinical trials. It is currently in Phase III clinical trials (ClinicalTrials.gov Identifiers: NCT03733444 and NCT03711162).
Proteins in the biological environment are surrounded by water molecules. They play an important role in the structure, stability, folding, dynamics and function of proteins [14,15]. Water molecules present in different sites of biomolecules establish hydrogen bonds with the proteins and contribute to the stability of their three-dimensional structures [14]. Thermodynamic changes in the aqueous environment affect the stability of biomolecules. Water molecules contribute to the catalytic function of proteins and their dynamics [15,16]. They control protein folding through hydrophobic collapse [15]. Binding sites of the proteins in the apo condition are filled with several water molecules that play an important role in the recognition of the ligands [17,18]. They effect the binding process between proteins and small molecule ligands [19][20][21][22]. Water molecules can be displaced by the ligands or they may act as bridges to stabilize the formation of a complex [23][24][25]. Displacement of water molecules from the active site to the bulk region occurs when a ligand binds to a protein. The thermodynamics of MD simulations and TWN analysis were performed as described in the methodology section. The results showed 35.1% and 64.9% TWNs for the head and linker sites, respectively. Interestingly, most of the water molecules in the linker site (75%) were located near the CO group (L') at the point where the linker and the head are connected. This suggested that water is always present in that vicinity in any form. In contrast with the head and linker sites, we could not identify the TWNs in the tail site. This site is narrow and the space is limited. Thus, water molecules could not form a ring structure. Moreover, this site lies in the hydrophobic pocket and possesses hydrophobic characteristics. TWNs around the L' position are represented by red spheres in Figure 2a. As can be seen in Figure 2b, there is a loop consisting of the residues His242, Leu243, Arg244, and Gly245 nearby the L' position. Arg244 is located at a position where it is capable of forming a hydrogen bond with the ligand. Additionally, we calculated the shape similarity between the ligand and the water molecules. Similarity in shape is used to compare the geometric characteristics of a ligand and the water molecule networks. Further details and in-house codes for the shape similarity are provided in our previous research paper [41]. Water networks whose centers of mass were located within 2 Å from PF-8380 were extracted for the shape similarity calculation. A similarity of 23.4% was obtained between the ligand and TWNs when the entire ligand structure was considered. However, 73.7%  MD simulations and TWN analysis were performed as described in the methodology section. The results showed 35.1% and 64.9% TWNs for the head and linker sites, respectively. Interestingly, most of the water molecules in the linker site (75%) were located near the CO group (L') at the point where the linker and the head are connected. This suggested that water is always present in that vicinity in any form. In contrast with the head and linker sites, we could not identify the TWNs in the tail site. This site is narrow and the space is limited. Thus, water molecules could not form a ring structure. Moreover, this site lies in the hydrophobic pocket and possesses hydrophobic characteristics. TWNs around the L' position are represented by red spheres in Figure 2a. As can be seen in Figure 2b, there is a loop consisting of the residues His242, Leu243, Arg244, and Gly245 nearby the L' position. Arg244 is located at a position where it is capable of forming a hydrogen bond with the ligand. Additionally, we calculated the shape similarity between the ligand and the water molecules. Similarity in shape is used to compare the geometric characteristics of a ligand and the water molecule networks. Further details and in-house codes for the shape similarity are provided in our previous research paper [41]. Water networks whose centers of mass were located within 2 Å from PF-8380 were extracted for the shape similarity calculation. A similarity of 23.4% was obtained between the ligand and TWNs when the entire ligand structure was considered. However, 73.7% similarity was observed when the analysis was focused on the L' region. This result also indicated Additionally, we calculated the shape similarity between the ligand and the water molecules. Similarity in shape is used to compare the geometric characteristics of a ligand and the water molecule networks. Further details and in-house codes for the shape similarity are provided in our previous research paper [41]. Water networks whose centers of mass were located within 2 Å from PF-8380 were extracted for the shape similarity calculation. A similarity of 23.4% was obtained between the ligand and TWNs when the entire ligand structure was considered. However, 73.7% similarity was observed when the analysis was focused on the L' region. This result also indicated that hydrophilic interactions involving hydrogen bonds at the L' position could be important for the ATX inhibitory activity.

Design, Synthesis and Activity Evaluation
Based on the TWN analysis, hydrophilic interactions at the L' position were supposed to be important. To confirm this hypothesis, we designed and synthesized compounds which could form hydrogen bonds with Arg244. Finally, we verified their ATX inhibitory activities through in vitro bis-pNPP assay.
In our previous study, compound 1 showed an IC 50 value of 250.0 nM in the bis-pNPP assay [18]. Probable binding mode for this compound is shown in Figure 3a. It can be seen that head and tail regions of the compound occupy the catalytic site and the hydrophobic pocket, respectively. Both head and tail regions showed hydrogen-bonding interactions, however, no hydrogen bond was observed for the linker region ( Figure 3b). Residues Thr209 and Trp275 were involved in hydrogen-bonding interactions with compound 1. We used compound 1 as a reference for designing compounds 2-4. Structures, IC 50 values, and predicted physicochemical properties of the designed compounds are summarized in Table 1. The synthesis of the compounds is shown in Schemes 1-3.
that hydrophilic interactions involving hydrogen bonds at the L' position could be important for the ATX inhibitory activity.

Design, Synthesis and Activity Evaluation
Based on the TWN analysis, hydrophilic interactions at the L' position were supposed to be important. To confirm this hypothesis, we designed and synthesized compounds which could form hydrogen bonds with Arg244. Finally, we verified their ATX inhibitory activities through in vitro bis-pNPP assay.
In our previous study, compound 1 showed an IC50 value of 250.0 nM in the bis-pNPP assay [18]. Probable binding mode for this compound is shown in Figure 3a. It can be seen that head and tail regions of the compound occupy the catalytic site and the hydrophobic pocket, respectively. Both head and tail regions showed hydrogen-bonding interactions, however, no hydrogen bond was observed for the linker region ( Figure 3b). Residues Thr209 and Trp275 were involved in hydrogenbonding interactions with compound 1. We used compound 1 as a reference for designing compounds 2-4. Structures, IC50 values, and predicted physicochemical properties of the designed compounds are summarized in Table 1. The synthesis of the compounds is shown in Schemes 1-3.   As shown in Figure 3b, the predicted binding mode of compound 1 did not show any hydrogen bond with Arg244. TWN analysis suggested that this interaction could be significant. Accordingly, we interchanged NH and CO positions in compound 2. In the docking results, the CO group of compound 2 was located closer to the Arg244 as compared to the CO group of compound 1. It was able to form a hydrogen bond with Arg244. Figure 4 shows a comparison of the binding modes of compounds 1 and 2. As shown in the figure, CO group of compound 2 was positioned nearer to the TWNs observed around the L' region. Based on these results, compound 2 was anticipated to be more potent than compound 1. As expected, compound 2 (IC 50 = 44.50 nM) was found to be five-fold more potent than compound 1 (IC 50 = 250.00 nM) in the bis-pNPP assay.
Compound 3 was designed by replacing the piperidine scaffold of compound 1 with a 2,3-dihydro-1H-isoindole moiety. This change produced a difference in the overall length of the linker and thereby affected the position of the linker CO group. The binding mode of compound 3, predicted by docking, indicated that a new fragment would enable the CO group of the linker to form hydrogen bonds with Arg244. As shown in Figure 5a, the linker CO group of compound 3 overlaps well with the TWNs of the L' region, and has good orientation towards TWNs. Thus, we believed that compound 3 would be able to form a strong hydrogen bond with Arg244. It can be seen in Figure 5b that compound 3 formed a hydrogen bond with Arg244. Moreover, as expected, compound 3 showed an IC 50 value of 3.86 nM. In addition, it demonstrated an IC 50 value of 827.00 nM in the human plasma assay.    Figure 3. (a) Binding mode of compound 1 (orange stick model) inside the ATX binding pocket. It occupies the catalytic site as well as the hydrophobic pocket. (b) 2D interaction diagram of compound 1 (orange line model). Hydrogen bonds and hydrophobic interactions with the active site residues are displayed as green and pink dashed lines, respectively. Electrostatic interaction with catalytic zinc ion is represented by gray dashed line. We interchanged NH and CO positions of compound 3 to design compound 4. As shown in Figure 6, compound 4 was able to form a hydrogen bond with Arg244 in the docking results. However, it may not be as strong as compound 3 due to a slight change in CO position and orientation. Also, docking results suggested a slight change in the interactions of the head region. The head group of compound 3 showed a pi-anion interaction with Asp311, while the head group of compound 4 showed hydrophobic interaction with Leu243. Thus, activity of compound 4 was expected to be slightly lower than compound 3. As anticipated, activity of compound 4 (IC 50 = 12.70 nM) was marginally reduced in comparison to the compound 3 (IC 50 = 3.86 nM).  As shown in Figure 3b, the predicted binding mode of compound 1 did not show any hydrogen bond with Arg244. TWN analysis suggested that this interaction could be significant. Accordingly, As shown in Figure 3b, the predicted binding mode of compound 1 did not show any hydrogen bond with Arg244. TWN analysis suggested that this interaction could be significant. Accordingly, we interchanged NH and CO positions in compound 2. In the docking results, the CO group of compound 2 was located closer to the Arg244 as compared to the CO group of compound 1. It was able to form a hydrogen bond with Arg244. Figure 4 shows a comparison of the binding modes of compounds 1 and 2. As shown in the figure, CO group of compound 2 was positioned nearer to the TWNs observed around the L' region. Based on these results, compound 2 was anticipated to be more potent than compound 1. As expected, compound 2 (IC50 = 44.50 nM) was found to be five-fold more potent than compound 1 (IC50 = 250.00 nM) in the bis-pNPP assay. Compound 3 was designed by replacing the piperidine scaffold of compound 1 with a 2,3dihydro-1H-isoindole moiety. This change produced a difference in the overall length of the linker and thereby affected the position of the linker CO group. The binding mode of compound 3, predicted by docking, indicated that a new fragment would enable the CO group of the linker to form hydrogen bonds with Arg244. As shown in Figure 5a, the linker CO group of compound 3 overlaps well with the TWNs of the L' region, and has good orientation towards TWNs. Thus, we believed that compound 3 would be able to form a strong hydrogen bond with Arg244. It can be seen in Figure 5b that compound 3 formed a hydrogen bond with Arg244. Moreover, as expected, compound 3 showed an IC50 value of 3.86 nM. In addition, it demonstrated an IC50 value of 827.00 nM in the human plasma assay. we interchanged NH and CO positions in compound 2. In the docking results, the CO group of compound 2 was located closer to the Arg244 as compared to the CO group of compound 1. It was able to form a hydrogen bond with Arg244. Figure 4 shows a comparison of the binding modes of compounds 1 and 2. As shown in the figure, CO group of compound 2 was positioned nearer to the TWNs observed around the L' region. Based on these results, compound 2 was anticipated to be more potent than compound 1. As expected, compound 2 (IC50 = 44.50 nM) was found to be five-fold more potent than compound 1 (IC50 = 250.00 nM) in the bis-pNPP assay. Compound 3 was designed by replacing the piperidine scaffold of compound 1 with a 2,3dihydro-1H-isoindole moiety. This change produced a difference in the overall length of the linker and thereby affected the position of the linker CO group. The binding mode of compound 3, predicted by docking, indicated that a new fragment would enable the CO group of the linker to form hydrogen bonds with Arg244. As shown in Figure 5a, the linker CO group of compound 3 overlaps well with the TWNs of the L' region, and has good orientation towards TWNs. Thus, we believed that compound 3 would be able to form a strong hydrogen bond with Arg244. It can be seen in Figure 5b that compound 3 formed a hydrogen bond with Arg244. Moreover, as expected, compound 3 showed an IC50 value of 3.86 nM. In addition, it demonstrated an IC50 value of 827.00 nM in the human plasma assay. We interchanged NH and CO positions of compound 3 to design compound 4. As shown in Figure 6, compound 4 was able to form a hydrogen bond with Arg244 in the docking results. However, it may not be as strong as compound 3 due to a slight change in CO position and The head group of compound 3 showed a pi-anion interaction with Asp311, while the head group of compound 4 showed hydrophobic interaction with Leu243. Thus, activity of compound 4 was expected to be slightly lower than compound 3. As anticipated, activity of compound 4 (IC50 = 12.70 nM) was marginally reduced in comparison to the compound 3 (IC50 = 3.86 nM). In addition, we calculated the shape similarity between the designed compounds and the TWNs in the same way as for PF-8380. The shape similarities for compounds 1, 2, 3 and 4 considering the whole ligand structure were found to be 22.9%, 21.3%, 17.1%, and 21.6%, respectively. The shape similarities concentrated in the L' region were 64.1%, 64.1%, 65.5%, and 65.5% for the compounds 1, 2, 3 and 4, respectively. The shape similarity results for the L' region appear to be very similar. However, the results can be interpreted differently when compared with the similarity obtained for the whole structure. Although compound 3 showed a low similarity of 17.1% for the overall structure, a high similarity of 65.5% was obtained for the analysis focused on the L' position. Given the lower similarity for the overall structure, the similarity at the L' position can be considered relatively high, compared to the other compounds. Through this analysis, it can be interpreted that the CO group of compound 3 is in a position where it is properly overlapped with the TWNs.

Protein Preparation
Protein structures were obtained from the RCSB Protein Data Bank (PDB) [51]. Except the two zinc ions, co-crystallized ligand and water molecules were removed from the crystal structure. Missing residues, bond order, and partial charges were checked using the protein preparation protocol of Discovery Studio 2018 (BIOVIA, San Diego, CA, USA). Amino acids were assumed to be in an ionized state at pH 7.4.

MD Simulation
MD simulation was performed using GROMACS 4.5.3 [52]. Topological files were generated using the GROMOS96 43a1 force field [53]. Protein was immersed in a cubic box of SPC/E water model [28] with a margin distance of 10 Å. The system was neutralized by adding 3 Clcounter ions. Energy minimization was carried out for 500 steps using the steepest descent method. Equilibration was performed to stabilize temperature (300 K) and pressure (1 bar) of the system. Initially, equilibration was conducted at 300 K temperature for 100 ps under NVT ensemble (canonical ensemble). Afterwards, equilibration was conducted at 1 bar pressure for 200 ps under NPT ensemble (isothermal-isobaric ensemble). Subsequently, production run was carried out for 10 ns at 300 K temperature and 1 bar pressure. Coordinate trajectories were saved every 10 ps. Temperature and pressure were regulated by Berendsen thermostat [54] and Parrinello-Rahman barostat [55], respectively. The cutoff for short-range non-bonded interactions was set to 1.2 nm. Long-range In addition, we calculated the shape similarity between the designed compounds and the TWNs in the same way as for PF-8380. The shape similarities for compounds 1, 2, 3 and 4 considering the whole ligand structure were found to be 22.9%, 21.3%, 17.1%, and 21.6%, respectively. The shape similarities concentrated in the L' region were 64.1%, 64.1%, 65.5%, and 65.5% for the compounds 1, 2, 3 and 4, respectively. The shape similarity results for the L' region appear to be very similar. However, the results can be interpreted differently when compared with the similarity obtained for the whole structure. Although compound 3 showed a low similarity of 17.1% for the overall structure, a high similarity of 65.5% was obtained for the analysis focused on the L' position. Given the lower similarity for the overall structure, the similarity at the L' position can be considered relatively high, compared to the other compounds. Through this analysis, it can be interpreted that the CO group of compound 3 is in a position where it is properly overlapped with the TWNs.

Protein Preparation
Protein structures were obtained from the RCSB Protein Data Bank (PDB) [51]. Except the two zinc ions, co-crystallized ligand and water molecules were removed from the crystal structure. Missing residues, bond order, and partial charges were checked using the protein preparation protocol of Discovery Studio 2018 (BIOVIA, San Diego, CA, USA). Amino acids were assumed to be in an ionized state at pH 7.4.

MD Simulation
MD simulation was performed using GROMACS 4.5.3 [52]. Topological files were generated using the GROMOS96 43a1 force field [53]. Protein was immersed in a cubic box of SPC/E water model [28] with a margin distance of 10 Å. The system was neutralized by adding 3 Cl − counter ions. Energy minimization was carried out for 500 steps using the steepest descent method. Equilibration was performed to stabilize temperature (300 K) and pressure (1 bar) of the system. Initially, equilibration was conducted at 300 K temperature for 100 ps under NVT ensemble (canonical ensemble). Afterwards, equilibration was conducted at 1 bar pressure for 200 ps under NPT ensemble (isothermal-isobaric ensemble). Subsequently, production run was carried out for 10 ns at 300 K temperature and 1 bar pressure. Coordinate trajectories were saved every 10 ps. Temperature and pressure were regulated by Berendsen thermostat [54] and Parrinello-Rahman barostat [55], respectively. The cutoff for short-range non-bonded interactions was set to 1.2 nm. Long-range electrostatics were calculated by the particle-mesh Ewald method [56]. A linear constraint solver (LINCS) algorithm [57] was used for constraining bond lengths. Periodic boundary conditions were applied in all the dimensions.

TWN Analysis
Water molecules form a number of cyclic water-ring networks through hydrogen-bonding between themselves. These networks are known as TWNs. Here, we focused on a small unit of water networks including three-to six-membered rings. The interactions between water molecules are commonly modeled by Lennard-Jones and Coulomb potentials [58]. The interaction potential energy between water molecules (a and b) is calculated by the equation: The electrostatic attraction is expressed as the Coulombic force between two charges, q i and q j , located at a distance of r ij . The van der Waals interaction is expressed using parameters, A and C. The distance between oxygen atoms is indicated by r oo . Parameters A and C were selected to produce reasonable structural and energetic results for liquid water. Parameter values are as follows. A = 582,000 kcal Å 12 mol −1 , C = 595 kcal Å 6 mol -1 , q i = 0.834e, and q j = 0.417e. The hydrogen bond between water molecules was determined using the energy criteria of −2.25 kcal mol −1 , which is close to the minimum value of the potential pair energy distribution [58]. More details about TWNs are provided in our earlier studies [33,[39][40][41][42].
In the TWN analysis, it was presumed that the distribution of water molecules could characterize the binding site. We retrieved the ATX crystal structure with a resolution of 1.899 Å (PDB ID 3WAX) [50] from the PDB. Although this structure was obtained from Mus musculus, it shares very high sequence identity and similarity (91.9 and 94.7%, respectively) with human ATX structures (PDB ID 4ZGA) [59]. As shown in Figure S1 of the Supplementary Materials, the majority of amino acid residues within their binding sites are identical. Accordingly, MD simulation was performed on PDB ID 3WAX in the apo state for TWN analysis. After the simulation, 200 trajectories were obtained from the stable region of the root mean square deviation (RMSD) plot ( Figure S2 of Supplementary Materials). Reasonably stabilized RMSD curve during 3-5 ns suggested that 200 extracted trajectories were suitable for further analysis. Water molecules present within 25 Å from the binding site were extracted for the TWN analysis. PF-8380 is a potent ATX inhibitor whose binding mode is known [45]. This compound was placed inside the binding site of ATX, and TWNs were analyzed around it.

Molecular Docking
Molecular docking studies were performed on the same ATX structure (PDB ID 3WAX) [50] that was used for the TWN analysis. Prior to docking, protein preparation was carried out through the previously discussed process. Compounds were built and optimized using the prepared ligand protocol. Momany-Rone partial charges [60] were applied to the protein and ligand structures. Energy minimization was performed using the CHARMM force field [61]. The CDOCKER protocol [62] of Discovery Studio 2018 (BIOVIA, San Diego, CA, USA) was used for docking. The original ligand of the crystal structure was considered for defining the binding site. Volume of the binding site was found to be 387.25 Å 3 . A simulated annealing process was performed with 2000 heating steps for the target temperature, 700 K and 5000 cooling steps for the cooling target temperature, 300 K. Then, CDOCKER energy was obtained. Finally, the binding mode of the ligands was carefully selected based on the protein-ligand interaction.

General Information
Solvents, reagents and starting materials were purchased from the commercial supplier. All reaction procedures were monitored by thin layer chromatography (TLC). The products were confirmed by 1 H-NMR and 13 C-NMR spectra. They were detected using the AVANVE III 600 spectrometer ( 1 H = 600 MHz, 13 C = 150 MHz). Compounds were dissolved in deuterated dimethylsulfoxide (DMSO-d 6 ) and chemical shifts were presented in ppm (d) relative to TMS as internal standard. 1 H-NMR spectra for compounds 1-4 are provided in Figures S3-S6 of the Supplementary Materials, respectively. 13 C-NMR spectra for compounds 1-4 are provided in Figures S7-S10 of the Supplementary Materials, respectively. High resolution mass spectra (HRMS) were obtained by AB SCIEX TripleTOF ® 5600 plus system (Seoul, Korea). HRMS-TOF scan results for the compounds are given in Table S1  20 mmol) were added. The reaction mixture was stirred at room temperature for 16 h. The reaction was quenched by adding ice/water and the resulting solid was filtered, then triturated with ether to afford the (3,5-dichlorophenyl)methyl 4-(4-{[(tert-butoxy)carbonyl]amino}benzamido)piperidine-1-carboxylate (2-c) as light yellow solid (883 mg, 92%). Next, 4N HCl in dioxane solution (4 mL) was added to an ice-cooled solution of 2-c (882 mg, 1.69 mmol) in dichloromethane (4 mL). The reaction mixture was stirred at room temperature for 4 h. The solvent was removed under reduced pressure. The resulting residue was dissolved in saturated sodium bicarbonate aqueous solution (30 mL) and extracted with ethyl acetate (30 mL × 3). The organic layer was collected, dried over anhydrous sodium sulfate, filtered and concentrated under reduced pressure to afford the (3,5-dichlorophenyl)methyl 4-(4-aminobenzamido)piperidine-1-carboxylate (2-d) as light yellow solid (710 mg, 99%). Formic acid (0.13 mL) was added to chlorosulfonyl isocyante (500 mg, 3.53 mmol) slowly at 0 • C and the mixture was stirred at room temperature for 2 h. The reaction mixture was concentrated under reduced pressure and the resulting solid (400 mg) was added to an ice-cooled solution of 2-d (30 mg, 0.071 mmol) and N,N-diisopropylethylamine (25 µL, 0.14 mmol) in DMF (1 mL). The reaction mixture was stirred at room temperature for 18 h, then it was diluted with ethyl acetate (20 mL) and washed with water (20 mL × 5). The organic layer was dried over anhydrous sodium sulfate, filtered and concentrated under reduced pressure. The resulting residue was purified by column chromatography (eluent = methanol:dichloromethane = 2:98 → 5:95) to afford the title compound as white solid ( To an ice-cooled solution of N-(4-aminophenyl)sulfamide (1-a) (14 mg, 0.066 mmol) in DMF (1 mL) and 2-{[(3,5-dichlorophenyl)methoxy]carbonyl}-2,3-dihydro-1H-isoindole-5-carboxylic acid (1-c) (30 mg, 0.082 mmol), N,N-diisoprolyethylamine (0.07 mL, 0.41 mmol) and benzotriazol-1-yl-oxytripyrrolidinophosphonium hexafluorophosphate (PyBOP, 64 mg, 0.12 mmol) were added. The reaction mixture was stirred at room temperature for 15 h under nitrogen atmosphere. After completion, water (15 mL) was added to the reaction mixture and it was extracted with ethyl acetate (30 mL × 3). The organic layer was washed with brine, dried over anhydrous sodium sulfate, filtered, and concentrated under reduced pressure. The resulting residue was triturated with diethyl ether and methanol to afford the title compound as white solid (20 mg, 57%). 1

Ex Vivo Human Plasma Assay
First, 5 µL of each test compound solution (10 mM, 100% dimethyl sulfoxide) was diluted 100 times with 495 µL of methanol and then 3 µL of a diluted methanol solution (1% dimethyl sulfoxide) was added to a human or rat serum solution. Next, 12 µL of the test compound solution was mixed into 48 µL of 100% serum in order to make 5 concentrations, and each tube was placed in a constant temperature water bath at 37 • C for 15 min. Subsequently, 10 mg/mL 18: 1 LPC solution (50% ethanol) was diluted to 375 µg/mL with serum and 2 µL was dispensed into each tube to make 50 µL. Each tube was reacted in a constant temperature water bath at 37 • C for 3 h. Then, 100 µL of a 0.5 µM 17:0 LPA solution (Chloroform/Methanol/Water = 65/35/8) was added to the reaction tube. Reaction mixture was centrifuged at 14,000 rpm, at 4 • C for 10 min in a centrifuge. Thereafter, 100 µL of 50% methanol solution was first dissolved in a 96-well polypropylene plate (Agilent Technologies 5042-1385, Santa Clara, CA, USA), and then 50 µL of the supernatant was carefully transferred from the centrifuged tube to the plate. Then, it was covered with a well cap (Thermo 276011, Thermo Fisher Scientific Solutions LLC, Seoul, Korea) and analyzed by LC-MS/MS (Agilent 1260). IC 50 values were calculated using the Grafit5 software after calculating the % inhibition values by the formula: 100 − (3 h serum + test solution/3 h serum + control group) × 100.

Conclusions
In continuation with our previous efforts to develop potent ATX inhibitors, we have reported novel ATX inhibitors in the present study. We used an in-house computational method to analyze the water networks in the binding site of ATX. This analysis led to the identification of a pharmacophoric site in the binding pocket. We designed new compounds considering the pharmacophoric site. They showed high inhibitory activities against ATX.