1321N1 astrocytoma cells and HEK-293 cells were cultured (at 37°C, 5% CO₂) in Dulbecco’s modified Eagle’s medium (DMEM/F-12 with Glutamax) supplemented with 10% foetal bovine serum (FBS) and 200 unit/ml of penicillin and streptomycin antibiotics (Fisher Scientific). For stable cell line generation, cells were transfected using FuGene HD Transfection Reagent (Promega) according to the manufacturer’s protocol with plasmids encoding either human P2X4 with a C-terminal (His)₁₀ tag (Young et al., 2008), human P2X7 with a C-terminal EYMPME tag (kindly provided by R.A. North) or rat P2X7 C-terminally tagged with a GFP-(His)₈ tag, generated by PCR amplification of the coding sequence for rat P2X7 from a pcDNA-based expression vector (Young et al., 2007), adding XhoI and EcoRI sites at the 5′- and 3′- ends respectively, followed by restriction digestion and ligation into a zebrafish P2X4-GFP expression vector based on pEGFP-C1 (kindly provided by Eric Gouaux; Kawate and Gouaux, 2006) to replace the zebrafish P2X4 coding region with rP2X7. Transfected clones were grown in DMEM/F-12 with Glutamax medium supplemented with G-418 (Geneticin®, Fisher Scientific), 600 μg/ml and 150 μg/ml during selection and maintenance respectively. Each clone was tested for P2X4 or P2X7 receptor expression by Western blot and clones, which displayed receptor expression were assessed for functionality using the Ca²⁺ influx assay (P2X4) or the Yopro-1 uptake assay (P2X7). HEK-293 cells stably expressing human P2X7 (Adinolfi et al., 2010) were a kind gift from Elena Adinolfi.

In silico simulations were performed on a MAC pro 2.80 GHz Quad-Core Intel Xeon running Ubuntu 12.04 LTS. Graphical representations were generated with MOE (Molecular Operating Environment, 2014.09; Chemical Computing Group ULC, Canada). Structure files retrieved from the RCSB Protein Data Bank (http://www.rcsb.org) as PDB file format were used as template to build P2X4 [PDB ID 4DW1 (Hattori and Gouaux, 2012)] and P2X7 [PDB ID 6U9W (McCarthy et al., 2019)] receptor homology models using MOE 2014.09, single template mode and default settings with AMBER12:EHT force field, including the crystallized ATP molecules for the induced-fit mode. Each model was checked visually and via the Rampage Server, Molprobity (Lovell et al., 2003) and PROCHECK (Laskowski et al., 1993) to exclude gross errors in protein geometry. For validation, ATP coordinates in the crystal structure were used to set the centroid coordinated for a box of 10 Å³ grid used in the docking. FlexX 2.1.3 with default parameters (Rarey et al., 1996), PLANTS (Korb et al., 2006) and Glide Standard Precision (https://www.schrodinger.com/products/glide) simulations were used for the docking. 20 aco ants (for the explorative docking algorithm) and standard parameters were set for simulations with PLANTS. 10 or more poses generated in each simulation were visually inspected.

Preparation of compound structure was performed with the LigPrep module (https://www.schrodinger.com/products/ligprep), using the Specs library of commercially available compounds (∼300,000 entries; Specs.net). Briefly, hydrogens were added and possible ionization states were generated for pH 7.0 ± 2.0 using the OPLS_2005 forcefield; possible tautomers were generated for each molecule as well as all possible enantiomers for structures with non-specified chiralities. The center of the binding region was defined by either manual selection of the pocket residues or by selecting the superposed ligand. Ligands were initially docked using the Glide High throughput Screening (HTVS) protocol, the top 15% best scored poses were re-docked with the Glide Standard Precision protocol, and then rescored (without performing ulterior docking) using the Glide Extra Precision (XP) protocol, FlexX and PLANTS (in parallel). A consensus score from all 3 re-scorings was calculated (Bassetto et al., 2013), and poses that ranked best according to Glide XP, FlexX and PLANTS—were selected for visual inspection. Docking of GP-25 and analogues were performed on the human P2X7 and human P2X4 models using Glide XP protocol using a grid appropriate for 14 Å³ ligand length, with crystallized ATP coordinates as grid centroid. Poses were visually inspected before final selection of compounds for assay. Criteria in the choice of ligands included i) ability to entertain multiple H-bond or ionic interactions with key residues in the ATP binding pocket and ii) at least a partial overlap between the docked ligand conformation and the superposed adenine of the ATP. Additional interactions were also a favorable criteria.

Cells were plated the day before assay into poly-lysine-coated 96-well plates at 40–65,000 cells/ml. To load cells with calcium-sensitive fluorescent dye, culture medium was replaced with modified Ringer’s buffer (140 mM NaCl, 10 mM HEPES, 10 mM Glucose, 1 mM MgCl₂, 1 mM CaCl₂, 2.5 mM KCl, .5% BSA, pH = 7.4) containing 2.6 μM FLUO4-AM (Fisher Scientific) and 250 μM Probenecid (Sigma), and cells were incubated for 20–30 min. Immediately prior to assay, buffer was replaced with fresh modified Ringer’s buffer containing 500 μM Probenecid. Compounds were dissolved in DMSO at 1,000× the required assay concentration prior to diluting in assay buffer and the final DMSO concentration did not exceed .1%. A Fluoroskan Ascent FL plate reader (Fisher Scientific) equipped with a solution dispenser and an appropriate filter pair (excitation: 485 nm, emission: 538 nm) was used to record 5 s baseline followed by P2X4-mediated calcium influx measurements for 20–25 s after ATP stimulation (20 μL injection of solutions dissolved in modified Ringer’s buffer). The amplitude of ATP responses was calculated as ΔF/F0, where ΔF = F1 − F0, subtracting the fluorescent background as suggested by Bootman et al. (2013).

Cells were plated as for the calcium influx assay (Section 2.4). Culture medium was replaced with low divalent cation containing extracellular solution (ECS-LD; 147 mM NaCl, 10 mM HEPES, 13 mM Glucose, .2 mM CaCl₂, 2 mM KCl. pH = 7.4 in MilliQ water) and 5 μM Yopro-1. Compounds were dissolved in DMSO at 1,000× the required assay concentration prior to diluting in assay buffer and the final DMSO concentration did not exceeded .1%. The baseline was measured with a BMG Clariostar instrument for 5–10 cycles, followed by addition of ATP (or BzATP) and recording for at least 30 further cycles (approx. 1 min per data point). The collected responses were normalized to the first data-point recorded after ATP addition. The initial gradient of the dye uptake curve was used to calculate the response to ATP.

Whole-cell patch-clamp traces were recorded at room temperature using patch pipettes pulled from borosilicate glass (World Precision Instruments, Sarasota, FL, United States). Pipettes were filled with internal solution (IS) containing (in mM): 145 NaCl, 10 EGTA and 10 HEPES, with the pH adjusted to 7.3 with NaOH and had resistances of 3–5 MΩ. Cells were constantly perfused in a bath with ECS-LD. Currents were recorded at a holding membrane potential of −60 mV using an Axon Instruments Axopatch Multiclamp 700A amplifier and Digidata 1322A A/D interface (Molecular Devices, Sunnyvale, CA, United States). ATP working solutions were prepared fresh and solutions with compounds had a final concentration of DMSO equal or lower than .1% (.1% DMSO was included as vehicle control). ATP applications were made at 90-s intervals. Solutions were applied to patch-clamped cells with the help of a rapid perfusion system (RSC-160, Biologic, Claix, France), allowing solution exchange times in the range 20–100 m.

To assess hit compound toxicity, CellTiter-Blue Cell Viability Assay (Promega, Southampton, United Kingdom) was used as recommended by the manufacturer on not transfected HEK-293. Cells were seeded in 96-well plates (100 μL at a concentration of 30–35,000 cells/ml) in culture media and a day later media was replaced with 100 µL fresh media containing 30 µM of tested compounds (10 µM of A740003). After 24 h incubation at 37 °C, 5% CO₂, 20 µL of CellTiter Blue reagent was added to each well and incubated for 3 h at 37°C. Fluorescence was then measured with a BMG Clariostar using excitation/emission wavelengths of 560/590 nm. Data were normalized to vehicle control-treated cells (.1% DMSO).

Data processing and data analysis were carried out using Graphpad Prism version 6 for Mac. Figures are presented as mean ± SEM. Statistical significance of any difference observed between samples was calculated through one-way ANOVA followed by Dunnett’s multiple comparisons test with a single pooled variance (control, unless otherwise specified) or, when comparing only two sets of data, using an unpaired t-test with Welch’s correction. When comparing data obtained from multiple independent experiments (N), each dataset was normalized to the mean value obtained for the vehicle control before pooling data. A non-linear regression (curve fit) with four parameters was applied to determine any dose-response correlation and the presence of any inhibition trend during the compound testing (curve-fit constraints were applied to normalized data). Y = Bottom + (Top-Bottom)/(1 + 10^((LogEC₅₀-X)*Hill Slope)) Where: X: agonist (or antagonist) concentrations expressed as logarithmic scale. Y: response measurements (raw or normalized value) This formula was also used for inhibition curve fitting by substituting the value of EC₅₀ with IC₅₀.

We generated a molecular model of human P2X4 based on the crystal structure of zebrafish P2X4 bound to ATP (Hattori and Gouaux, 2012). Three docking algorithms were validated—later employed in the virtual screening (PLANTS, FlexX and Glide SP)—by docking ATP in the model. In each simulation the docked pose of ATP was similar to that observed for ATP in the crystal structure (superposed, shown in grey lines in Supplementary Figure S1); in the dock obtained using PLANTS (Supplementary Figure S1A) the adenine ring was translated (positioned less deep in the pocket), with FlexX (Supplementary Figure S1B) the phosphate chain conformation was different (the ß and gamma phosphates in the docking occupied the a and ß positions in the model derived from the crystal structure), and with Glide SP (Supplementary Figure S1C) there was the greatest similarity. A protein-ligand interaction fingerprint (PLIF) showed comparable interactions in the zebrafish P2X4 crystal structure (4DW1) and our three docking simulations (Supplementary Figure S1D). In our docking simulations we observed some additional interactions with Leu-188 (surface contact), Lys-190 (sidechain hydrogen bond acceptor and ionic attraction), Leu-214 (sidechain hydrogen bond donor and surface contact) and Ser-289 (sidechain and backbone hydrogen bond acceptor). Our results suggested that each algorithm was capable of replicating the binding pose of ATP in the orthosteric pocket, but as Glide performed the best, this was used as the main algorithm for subsequent virtual screening.

A library of commercially available drug-like compounds (Specs, 2022) with approximately 300,000 molecules was docked in silico into the orthosteric pocket of our human P2X4 model, selecting a total of 17 compounds for biological assay (Table 1).

The antagonist activity of the 17 compounds (applied at 10 µM) was assayed initially on 1321N1 astrocytoma cells stably expressing human P2X4 receptors by measuring the ATP-induced (applied at 1.2 µM) increase in Fluo-4 fluorescence due to calcium influx via P2X4, using 2 µM BX430 as a positive control (Figure 1A). Compound 7 was excluded from biological assay due to very low solubility. None of the compounds showed a statistically significant reduction in response, suggesting no antagonist activity at human P2X4. The same compounds were then assayed at a single concentration of 30 µM (we initially expected that, having designed the compounds to bind to P2X4, they would likely be less potent at P2X7, and need to be applied at a higher concentration to observe an effect) on 1321N1 astrocytoma cells stably expressing human P2X7. The ATP-induced rate of uptake of YoPro-1 dye was measured upon application of at 300 µM ATP, using .1 µM A740003 as a positive control (Figure 1B). Several compounds showed at least 25% reduction in response (Table 1) and, strikingly, one compound (GP-25; (4-bromo-2-(3-(methoxycarbonyl)-2-methyl-5-oxo-4,5-dihydro-1H-indeno(1,2-b)pyridine-4-yl)phenoxy)acetic acid) inhibited the response by approximately 74.6%, and was taken forward for further analysis.

Concentration-response curves for both human and rat P2X7 receptors were constructed using the ATP-induced YoPro-1 uptake assay, and EC₅₀ values of 183.5 and 138.3 µM were determined for human and rat P2X7, respectively (Figure 2A). Concentration-inhibition curves were constructed for GP-25 at both human and rat P2X7 using the YoPro-1 uptake assay (Figure 2B; 300 µM ATP), and IC₅₀ values of 8.7 μM at human P2X7 and 24.4 μM at rat P2X7 were determined, demonstrating moderate selectivity of GP-25 for human P2X7 over the rat orthologue. To confirm antagonist action in an independent assay, whole-cell patch clamp experiments were performed on human embryonic kidney (HEK-293) cells stably expressing rat P2X7-GFP (Figure 2C), and a concentration-dependent inhibition was observed when GP-25 was pre-applied for 2 min (300 µM ATP application). A statistically significant reduction in P2X7 activity was observed upon application of 25 µM GP-25 in both the YoPro-1 and patch clamp assays confirming GP-25 activity in the low micromolar range (Figure 2D). In order to confirm GP-25 as a competitive (orthosteric) antagonist, BzATP concentration-response curves (ranging between 300 μM and .03 µM) were constructed for rat P2X7-GFP in the YoPro-1 uptake assay in the presence of increasing concentrations of GP-25 (Figure 3A). Maximal responses were observed in all except the highest GP-25 concentration (100 µM). Schild analysis, plotting the logarithm of the ratio of apparent EC₅₀ values ((antagonist/control) − 1) against the logarithm of GP25 concentration (Figure 3B) gave a straight line with a slope not significantly different than 1 (at p = .05 as determined by one-sample t-test against the theorical value 1; 3 biological replicates), indicative of competitive (orthosteric) antagonism.

A small library of 16 commercially available compounds was selected according to the presence of the same scaffold as GP-25, (70% similarity, Tanimoto index) and/or presence of at least one functional group we hypothesized could contribute to GP-25 activity (e.g. carboxylic acid, halogen substituent to the aromatic ring, ester, etc.) (Compounds 18–33; Table 1). When human P2X7-expressing cells were incubated with 30 µM of compounds (Figure 4A), several GP-25 analogues caused a statistically significant reduction in ATP-induced dye uptake (300 μM ATP, approx. EC₇₀ measured in our experiments), including GP-47, GP-50, Compound 23, Compound 28 and GP-66. None of these compounds displayed antagonist activity at rat P2X7 (Figure 4B and summarized in Table 1). Although the number of compounds tested was limited by their availability via a commercial source, the numerous ‘hits’ allowed for a basic structure-relationship (SAR) analysis. 13 out of 16 compounds presented either scaffold I or scaffold II—Scaffold I being the one shared among most of the compounds that showed moderate or good activity (Figure 4C). A methoxyl acetate as substituent (R1) on the phenyl ring (Scaffold I) appears to be important for compound activity, as substitution with a methoxyl group completely abolished the activity (as for Compound 19; Figure 4A and Table 1). Methoxyl acetate as either an ortho- or para-substituent (R1) was tolerated, as were halogens in the meta-position in the phenyl ring (R1), both moieties improving activity in Scaffold I. The introduction of a furan linker between the dihydro-1H-indeno(1,2-b)pyridine-3-carboxylate and the phenyl moiety (scaffold very similar to that of Scaffold I; Figure 4C) was permitted without loss of activity provided that there was of a halogen in the ortho- and a nitro-group in the para-position of the phenyl moiety (GP-50). However, we observed that GP-50 was less soluble than GP-47 or GP-25. Compounds bearing Scaffold II (Compound 13, 14, 18, 25, 26, 27) did not show activity over 25% at 30 μM, indicating that the substitution of the carboxylate to a cyclic ketone is not favored.

Finally, GP-25 and GP-47 were tested in a cytotoxicity assay showing that a 24-h incubation did not significantly reduce cell viability in HEK-293 wild-type (Figure 4D). As GP-50 showed low solubility, it was excluded from the cell viability assay.

To investigate potential binding modes of GP-25 to P2X4 and P2X7, docking of the ligands was performed in the orthosteric pockets of our human P2X4 model (GP-25 was re-docked), and of a model of human P2X7 built from the cryoEM structure of rat P2X7 bound to ATP (PDB ID: 6U9W; McCarthy et al., 2019). Both GP-25 and GP-47 contain a chiral center and were purchased as mixture of enantiomers (no info on % composition provided by the supplier). Our docking investigated the binding mode of both enantiomers (S and R, Figure 5) in the orthosteric pocket using Glide Extra Precision (XP).

The carboxyl groups of the S-enantiomer of GP-25 form extensive interactions with many of the polar residues in a similar fashion to that of the gamma phosphate of ATP (Lys-67, Lys-69, Arg-295 and Lys-313 (human P2X4 numbering); upper part of Figure 5A). However, docking the R-enantiomer of GP-25 showed a flipped pose with the carboxyl moiety pointing toward the bottom of the pocket (lower part of Figure 5B), lacking the interactions with residues that coordinate the gamma phosphate in ATP. Conversely, for human P2X7, we were able to obtain (for both GP-25 enantiomers), poses where the carboxylic group made strong multiple interactions with polar residues known to interact with ATP in the crystal structure, specifically coordinating the gamma phosphate (Figure 5C, D), including Lys-311, Lys-64, Lys-66 and Arg-294 (human P2X7 numbering). Additionally, the hetero-tricyclic scaffold of GP-25 tightly occupies the cavity where the adenine ring of ATP is found in the cryoEM structure, unlike in human P2X4 (compare Figure 5C, D with Figure 5A, B). Docking of GP-47 into the orthosteric pocket of human P2X7 (Figure 5E, F) gave rise to poses where some interactions between the carboxylic acid moieties and key positively charged amino-acids (e.g. Lys-311 in the S-enantiomer; Figure 5E) were present, and the adenine pocket at least partially occupied by the tricyclic ring. The combination of lack of key interactions with conserved phosphate-coordinating amino-acids and lack of occupation of the adenine pocket for one of the two enantiomers may explain why GP-25 displays little antagonist activity at human P2X4 compared to human P2X7.