Inhibits Human P2X7 Receptor with New Scaffold

P2X7R orthologs have distinct allosteric binding sites

To characterize the molecular differences between key mammalian P2X7R orthologs, we obtained cryo-EM structures of the full-length wild-type mP2X7 (2.5 Å) and hP2X7 (2.5 Å) receptors in the apo closed state conformation (Fig. 1A, Supplementary Figs. 3 and 4 and Supplementary Table 1). The overall architecture of this P2X receptor (P2XR) subtype is consistent across human, mouse, and rat (PDB: 8TR5) orthologs (mean RMSD = 0.7 Å between Cα carbons)35. Conserved features include their trimeric architecture, the 310-helices forming a closed gate, a partially hydrated sodium ion present in the closed pore, palmitoylation of residues in the C-cys anchor, and zinc and guanosine nucleotide-binding sites in the cytoplasmic ballast (Fig. 1 and Supplementary Fig. 2A, 5)35,36. However, we also identified ortholog-specific features in the hP2X7R, including two cholesterol hemisuccinate (CHS) binding sites per protomer at the interface between transmembrane helix 1 (TM1) and transmembrane helix 2 (TM2) on what would be the extracellular leaflet of the membrane bilayer (Fig. 1B–D). The two CHS molecules are stacked on top of each other such that the inner molecule is closer to the TM1/TM2 interface, and the outer molecule is positioned between TM1 and the inner molecule (Fig. 1C, D). The inner CHS molecule is coordinated by hydrophobic interactions with residues F328 and L333 on TM2; F38, V41, C42, L45, and Y51 on TM1; and W265 on the lower body domain (Fig. 1C and Supplementary Fig. 2A). Two oxygen atoms on the hemisuccinate group of the inner CHS molecule also form hydrogen bonds with the side chain of R264 and backbone nitrogen of W265 (distances of 2.5 Å and 2.9 Å, respectively) (Fig. 1C). The outer CHS molecule is coordinated by hydrophobic interactions with residues F38, C42, and V46 on TM1; residue F266 on the lower body domain; and the inner CHS molecule (Fig. 1D and Supplementary Fig. 2A). There are also hydrogen bonds between the hemisuccinate group of the outer CHS molecule and the side chains of N261, H268, and K49 (distances of 2.8 Å, 3.4 Å, and 3.3 Å, respectively) (Fig. 1D). Although all three ortholog reconstructions are of similar resolution (~2.5 Å), there are no densities in either the rP2X7R or the mP2X7R reconstructions to support modeling of CHS molecules, suggesting that these bound CHS molecules are specific to the hP2X7R.

Comparison of these three structures also reveals molecular differences in the unoccupied classical allosteric ligand-binding sites (Fig. 1A, E–G)34,35. Across the three orthologs, the classical allosteric pocket comprises multiple conserved residues, including F88, D92, F103, M105, K110, F293, K297, Y295, K297, Y298, E305, and I310 (Fig. 1E–G and Supplementary Fig. 6). However, the identity of three residues within the classical allosteric pocket are different between human, mouse, and rat orthologs as seen in sequence alignments and visualized in their respective structures (Fig. 1E–G and Supplementary Figs. 6 and 7). First, F95 in hP2X7Rs and mP2X7Rs correlates to L95 in rP2X7Rs (Fig. 1E–G Supplementary Figs. 6 and 7). This residue occupies the same general location and rotameric orientation in all three orthologs, but the phenylalanine residue found in human and mouse P2X7Rs occupies much more space within the classical allosteric pocket35. Second, F108 in hP2X7Rs correlates to Y108 in both mP2X7Rs and rP2X7Rs (Fig. 1E–G and Supplementary Figs. 6 and 7). Again, this residue occupies the same position and rotameric conformation in all three structures, likely playing a similar role (Fig. 1E–G)35. Finally, V312 in hP2X7Rs correlates to A312 in both mP2X7Rs and rP2X7Rs (Fig. 1E–G and Supplementary Figs. 6 and 7). The larger valine side chain in the human ortholog occupies more space within the classical allosteric pocket than the alanine in rat or mouse P2X7Rs (Fig. 1E–G and Supplementary Figs. 6 and 7)35. Interestingly, although not apparent from sequence alignments, our structure of the hP2X7R shows that the larger side chain of V312 forces the adjacent, conserved residue Y295 to adopt an alternative rotameric conformation compared to its conformation in both mP2X7R and rP2X7R, condensing the size of the pocket in hP2X7R (Fig. 1E–G and Supplementary Fig. 6 and 7). Thus, the classical allosteric ligand-binding site in the human ortholog has a different size and shape, which impacts the binding of allosteric antagonists.

P2X7R orthologs have distinct orthosteric ATP-binding sites

Because pharmacological tools that activate human, mouse, and rat P2X7Rs are distinct, we examined the molecular determinants of agonism in the human ortholog to facilitate the development of higher-affinity agonists30,35,36,37. In agreement with previous measurements, two-electrode voltage clamp (TEVC) recordings of human, mouse, and rat P2X7Rs in the absence of divalent cations revealed half maximal effective concentrations (EC50) for ATP of 89 ± 8.3 µM, 70 ± 17 µM, and 34 ± 8.4 µM, respectively (Fig. 2A and Supplementary Fig. 8)35,36,38. The modestly lower apparent affinity of ATP for the hP2X7R was borne out by direct measurements of kinetics and equilibrium binding affinities using bio-layer interferometry (BLI). The rate constant for association (ka) of 7.4 ± 1.3 × 104 M−1 s−1, and rate constant for dissociation (kd) of 4.8 ± 1.2 × 10−2 s−1, are ~27% and ~6% slower than those for ATP binding to the rP2X7R (Fig. 2B)35. This corresponds to ATP binding to the hP2X7R with an equilibrium dissociation constant (KD = 650 ± 120 nM) that is ~20% higher the rP2X7R (KD = 540 ± 230 nM) (Fig. 2B)35.

To gain molecular insight into the binding of ATP to hP2X7Rs, we determined the cryo-EM structure of the full-length wild-type hP2X7R in the ATP-bound open state at 3.0 Å resolution (Fig. 2C–E and Supplementary Figs. 3 and 4 and Supplementary Table 1). As expected, the ATP-bound structure of the hP2X7R has a global architecture similar to the rP2X7R in the ATP-bound (RMSD = 0.8 Å at Cα carbons, PDB: 6U9W) and BzATP-bound (RMSD = 1.0 Å at Cα carbons, PDB: 8TRJ) open state conformations (Supplementary Fig. 8)35,36. The minimum pore radius of the hP2X7R in the ATP-bound open state is 2.5 Å, the same as the rP2X7R in the ATP-bound open state, and large enough to pass partially hydrated sodium ions such as those present in the closed pores of human, mouse, and rat P2X7Rs (Supplementary Fig. 9)36,39. Although there are structural similarities within the extracellular domains, the pore and cytoplasmic domains of the hP2X7R in the ATP-bound open state are rotated in comparison to the rP2X7R in the ATP-bound open state (Supplementary Fig. 8). Relative to the rP2X7R, TM1 of the hP2X7R is rotated up in-plane by ~7° and TM2 is rotated up in-plane by ~6° with the hinges at the extracellular ends of TM1 and TM2, respectively (Supplementary Fig. 8C). These differences result in lateral displacements of 5.2 Å (distance between Cα carbons of residue 24) at the start of TM1 and 4.8 Å (distance between Cα carbons of residue 358) at the end of TM2 (Supplementary Fig. 8C). The rotation of TM2 is further propagated into a global rotation of the cytoplasmic ballast by ~10° in the hP2X7R relative to the rP2X7R (Supplementary Fig. 8C). This twist in the pore and cytoplasmic domain, much like the tightening of a spring, decreases the overall height of the hP2X7R by ~4 Å compared to the rP2X7R (distance between Cα carbons of residues 514 and 521), likely representing an inherent flexibility of the transmembrane and cytoplasmic domains in the ATP-bound open state conformation.

The orthosteric ATP-binding site of the hP2X7R is clearly visualized at 3.0 Å resolution (Fig. 2C, D). In the ATP-bound open state of the hP2X7R, ATP is coordinated by the seven residues that are conserved in the orthosteric pocket of all P2XR subtypes: K64, K66, T189, K193, N292, R294, and K311 (Fig. 2E and Supplementary Fig. 6). In addition, the hP2X7R-specific residues, L191, I214, and Y288, form hydrophobic interactions with the ribose group and the side chain of Q143 interacts with the 2′-hydroxy on the ribose group (3.3 Å) (Fig. 2E)26. Of these subtype-specific residues, I214 and Y288 in hP2X7Rs are not conserved across P2X7R orthologs (Supplementary Fig. 6). Residue Y288 in hP2X7Rs corresponds to a valine in mP2X7Rs and a phenylalanine in rP2X7Rs, causing differences that might affect the pharmacology of ATP binding (Supplementary Fig. 6). Residue I214 in hP2X7Rs and rP2X7Rs corresponds to a glycine in mP2X7Rs, introducing flexibility in the mouse ortholog that could explain some pharmacological differences (Supplementary Fig. 6)30. Indeed, it has previously been shown that R125, Q143, and I214 in the rP2X7R are the critical determinants of efficacy, potency, and full agonism for BzATP compared to ATP35. In our structure of the hP2X7R in the ATP-bound open state, residues R125, Q143, and I214 are in similar positions and rotameric conformations to the open state structure of the rP2X7R, including the flexible side chain of R125, which is solvent-facing and stubbed at the Cβ carbon35. The molecular details gleaned from the hP2X7R structure in the ATP-bound open state conformation will be crucial for structure-based drug design of high-affinity hP2X7R agonists.

The classical allosteric pocket of the hP2X7R can fit larger cage alkyls

We sought to understand how an existing allosteric P2X7R antagonist binds to the human ortholog to determine its suitability as a starting candidate for structure-based drug design. We initially selected the adamantane-containing compound UB-ALT-P30 because of its simplicity, ease of synthesis, and potential for further functionalization40,41. UB-ALT-P30 consists of an adamantyl and a 2-chlorophenyl moiety connected by a hydrazide linker (Fig. 3A and Supplementary Fig. 1). In calcium influx assays, UB-ALT-P30 has a half-maximal inhibitory concentration (IC50) of 17.8 ± 8.3 nM for the hP2X7R, 116 ± 50 nM for the rP2X7R, and 148 ± 27 nM for the mP2X7R, establishing its greater potency for the hP2X7R (Fig. 3B). Moreover, using a high concentration at which we can expect 100% inhibition of the hP2X7R (10 μM), UB-ALT-P30 is highly selective for the hP2X7R compared to the human P2X1 receptor (hP2X1R, 36 ± 8% inhibition), the human P2X2 receptor (hP2X2R, 29 ± 7%), and the human P2X4 receptor (hP2X4R, 44 ± 14%) (Fig. 3C). The ligand showed some inhibitory potency at the human P2X3 receptor (hP2X3R, estimated IC50 of 707 ± 185 nM), which means that it is still 40-fold more potent at the hP2X7R versus the hP2X3R. The kinetics of UB-ALT-P30 binding to the hP2X7R were measured by TEVC, resulting in an on-rate of 0.0105 nM x min−1 and an off-rate of 0.29 min−1 (Supplementary Fig. 10). The high potency and moderate selectivity for the hP2X7R render UB-ALT-P30 a good initial compound for ligand development and optimization.

The high-resolution cryo-EM structure of UB-ALT-P30 bound to the hP2X7R (2.8 Å) reveals receptor-ligand interactions and provides key insights to optimize ligand design (Fig. 3D, E, Supplementary Fig. 3 and 4 and Supplementary Table 1). UB-ALT-P30 binds to the classical allosteric ligand-binding site, at the interface of upper body domains from neighboring protomers, in a shallow binding pose (Fig. 1A and Supplementary Fig. 2)34. The binding of ligands to the classical allosteric pocket is thought to prevent receptor movements necessary for transition to the ATP-bound open state33,34. UB-ALT-P30 binds to the hP2X7R in a similar pose as other adamantane-containing allosteric antagonists, generating a pruned RMSD of 0.6 Å compared to the structure of AZD9056 bound to the rP2X7R (PDB code: 8TR8) (Supplementary Fig. 11)34. However, important ortholog-specific interactions are also apparent.

The adamantyl moiety of UB-ALT-P30 is predominantly coordinated by hydrophobic interactions with the side chains of residues F95, F103, M105, F293, Y295, and V312 in the hP2X7R, which we confirmed with three replicas of 500 ns molecular dynamics (MD) simulations (Fig. 3E and Supplementary Fig. 12A, C). Of these residues, F95 is specific to the hP2X7R ortholog and located on a dynamic loop in the upper body domain (residues 88–100) that is positioned differently in rat and human adamantane-based antagonist-bound structures (Fig. 3E and Supplementary Figs. 1, 2, 6, 11)34. In the structure of AZD9056 bound to the rP2X7R, the corresponding L95 is rotated towards the ligand, making extensive hydrophobic interactions34. In contrast, F95 in the hP2X7R points away from UB-ALT-P30, forming weak hydrophobic interactions and creating empty space below the molecule (Fig. 3E, F, G)34. Another ortholog-specific residue, V312, fills a hydrophobic cavity at one side of the adamantyl moiety in the hP2X7R (Fig. 3E and Supplementary Fig. 6). One additional hydrophobic cavity in the hP2X7R exists on the opposite side of the adamantyl moiety from V312, where three water molecules are coordinated by the backbone carbonyls of A296 (2.8 Å), R294 (3.2 Å), A91 (2.6 Å), and T94 (3.0 Å), as well as the backbone nitrogen of Y295 (3.3 Å). These cavities create ~58 Å3 of empty hydrophobic space around the polycyclic core of UB-ALT-P30, presenting an opportunity to optimize this ligand by replacing the adamantyl with more precisely fitting scaffolds (Fig. 3F, G).

Additional receptor-ligand interactions in the hP2X7R can be observed closer to the extracellular surface of the classical allosteric pocket. The hydrazide linker of UB-ALT-P30 forms hydrogen bonds with the backbone carbonyl of D92 (2.8 Å) as well as the side chain hydroxyl of Y298 (3.4 Å) (Fig. 3E and Supplementary Fig. 12A, C). The chlorophenyl moiety is predominantly coordinated by hydrophobic interactions with the side chains of residues F88, M105, F108, and I310, as well as a cation-π interaction with the side chain ammonium of K297 (Fig. 3E and Supplementary Fig. 12A, C). Residues F108 in the hP2X7R and Y108 in the rP2X7R appear to play the same role, forming edge-to-face interactions with allosteric ligands34. Together, the empirical knowledge gained from identifying ligand-receptor interactions and empty space around the adamantyl moiety of UB-ALT-P30 in the classical allosteric pocket of the hP2X7R can be used to design ligand scaffolds with higher potency and greater selectivity (Fig. 3F, G)42.

UB-MBX-46 is a P2X7R antagonist with an improved scaffold

Given our detailed understanding of the molecular interactions between UB-ALT-P30 and the hP2X7R, we employed structure-based drug design to develop a more potent and selective antagonist. Notably, the adamantyl scaffold present in UB-ALT-P30 is also found in several clinically approved drugs and numerous preclinical and clinical candidates, including some P2X7R antagonists26. While the presence of other polycyclic scaffolds in approved drugs is rather limited, certain polycyclic hydrocarbons have outperformed adamantane for specific targets43,44,45,46. Therefore, we synthesized six compounds with alternative polycyclic cores of different sizes and shapes to replace the adamantyl scaffold of UB-ALT-P30 (Table 1). These six compounds were synthesized from the corresponding polycyclic carboxylic acids shown in Supplementary Fig. 13 following procedures reported for the synthesis of UB-ALT-P3040,45,46. All compounds were characterized by their spectroscopic data, melting point, exact mass, and elemental analysis or HPLC/UV (Supplementary Figs. 13–17 and Supplementary Methods).

The relative binding free energies (RBFEs) for each analog (relative to UB-ALT-P30) at the hP2X7R were determined in silico based on the corresponding perturbative transformation calculated using thermodynamic integration coupled with MD simulations (TI/MD) in phospholipid bilayers via Amber22 (Table 1 and Supplementary Fig. 18 and 19)47,48,49,50. Of all our compounds, only UB-MBX-46 and UB-MBX-P2 resulted in negative RBFEs (ΔΔGb,TI/MD), indicating that the perturbative transformations of these two compounds were energetically favorable (Table 1). The inhibitory potency of each compound was experimentally tested using ethidium bromide accumulation assays (Table 1) and generally correlated with the computational predictions (UB-MBX-46 being the most potent compound, followed by UB-MBX-P2). Similarly, compounds with positive ΔΔGb, TI/MD values were less potent, including UB-MBX-P1 and UB-ALT-P37 (Table 1). These data indicate that UB-MBX-46 is the most promising compound for further validation. The polycyclic tetracyclo [4.4.0.03,9.04,8] decane scaffold of UB-MBX-46 features two cyclopentane rings in a “frozen” envelope conformation and two unique cyclohexane rings in boat conformation. This scaffold is larger than adamantane and cubane (UB-ALT-P38), yet smaller than the pentacyclic moiety of UB-MBX-P1, and is unique since it has scarcely been used in medicinal chemistry (Table 1)46.

UB-MBX-46 is a potent and selective antagonist for the hP2X7R

To pharmacologically characterize the improved scaffold of UB-MBX-46, the compound was tested on different P2X7R orthologs and P2XR subtypes (Fig. 4A, Supplementary Fig. 1). In calcium influx assays, UB-MBX-46 has an inhibitory potency of 0.514 ± 0.035 nM for the hP2X7R and is ~35x more potent on the hP2X7R, ~3x more potent on the rP2X7R, and ~33x more potent on the mP2X7R than the starting compound UB-ALT-P30 (Fig. 4B). Furthermore, UB-MBX-46 is more potent at the hP2X7R than either the rP2X7R (~80-fold less potent) or the mP2X7R (~9-fold less potent) (Fig. 4B). The selectivity of UB-MBX-46 for other P2XR subtypes was also tested using calcium influx assays. Using a high concentration at which we can expect 100% inhibition of the hP2X7R (10 µM), UB-MBX-46 is much more selective for the hP2X7R compared to the hP2X1R (13 ± 10% inhibition), the hP2X2R (15 ± 9%), the hP2X3R (29 ± 16%), or the hP2X4R (41 ± 19%) (Fig. 4C). Thus, UB-MBX-46 is even more selective for the hP2X7R compared to other P2XR subtypes than UB-ALT-P30 (Figs. 3C and 4C). Finally, the kinetics of UB-MBX-46 binding to the hP2X7R were measured by TEVC, resulting in an on-rate of 0.0065 nM x min-1 and an off-rate of 0.07 min−1 (Fig. 4H, I). However, the virtually irreversible binding of UB-MBX-46 to the receptor within 10 min of wash-out does not correspond to the graphically determined Koff for UB-MBX-46, so better approximations (Koff of 0.001 min-1) were obtained based on extrapolated IC50 values (Supplementary Fig. 10).

We next investigated the molecular basis of the improved functional characteristics of UB-MBX-46 by determining the cryo-EM structure of the hP2X7R in the UB-MBX-46-bound inhibited state at 2.5 Å resolution (Fig. 4D, E; Supplementary Figs. 3 and 4 and Supplementary Table 1). Similar to the adamantane derivative, UB-MBX-46 binds to the classical allosteric ligand-binding site in a shallow binding pose (Fig. 1A and Supplementary Fig. 2)34. The two ligand-bound structures are very comparable, with a pruned RMSD of 0.5 Å however, the longer and larger scaffold of UB-MBX-46 relative to UB-ALT-P30 (hydrophobic volumes of 175 Å3 vs 133 Å3) binds deeper into the classical allosteric pocket and extends closer to the extracellular surface of the receptor (Figs. 3E, F and 4E, F). The hydrazide linker and chlorophenyl moieties of UB-MBX-46 are ~0.9 Å closer to the extracellular surface than UB-ALT-P30 (average measurements between equivalent chlorine atoms, nitrogen atoms on the hydrazide linker, and closest carbon atoms to the linker of the polycyclic moieties). Yet, due to the longer length of UB-MBX-46 and its larger polycyclic moiety, the ligand also extends deeper into the classical allosteric pocket by ~1.3 Å. As a result, there is only ~48 Å3 of empty hydrophobic space surrounding the polycyclic core of UB-MBX-46; 10 Å3 less empty space than around UB-ALT-P30 (Figs. 3F, G and 4F, G)42.

The cryo-EM structures and MD simulations show that many of the receptor-ligand interactions associated with UB-ALT-P30 in the classical allosteric pocket of the hP2X7R are present with UB-MBX-46. Similar to the adamantyl moiety in UB-ALT-P30, the polycyclic core of UB-MBX-46 is predominantly coordinated by hydrophobic interactions with the side chains of residues F95, F103, M105, F293, Y295, and V312 (Fig. 4E and Supplementary Fig. 12B, D). However, due to the ligand’s depth in the pocket, the larger polycyclic core of UB-MBX-46 is more able to form hydrophobic interactions with the human-specific residue F95. The hydrazide linker forms hydrogen bonding interactions with the backbone carbonyl of D92 (2.7 Å) as well as the side chain hydroxyl of Y298 (3.4 Å and 3.1 Å), creating one more hydrogen bond that is absent in the interactions with UB-ALT-P30 (Figs. 3E and 4E and Supplementary Fig. 12). Finally, the chlorophenyl moiety is predominantly coordinated by hydrophobic interactions with the side chains of residues F88, M105, F108, W167, and I310 as well as a cation-π interaction with the side chain ammonium of K297 (Fig. 4E and Supplementary Fig. 12B, D). UB-MBX-46 appears to better occupy the classical allosteric pocket in the hP2X7R due to the improved ligand-receptor interactions compared to UB-ALT-P30. Indeed, mutating the ortholog specific residues in the mP2X7R (residues 108 and 312) and the rP2X7R (residues 95, 108, and 312) influences the potency of UB-MBX-46 (Supplementary Fig. 20). Further, from TI/MD calculations, UB-MBX-46 also has a lower desolvation penalty to bind to the receptor, indicating a preference for the hydrophobic pocket of the receptor over the solvent phase (Fig. 4F). Thus, more favored hydrophobic interactions surrounding the polycyclic core, more favorable hydrogen bonding interactions, and a lower desolvation penalty contribute to the higher potency of UB-MBX-46.

The P2X7R is a promising therapeutic target for numerous pathological inflammatory diseases, but pharmacological differences between receptor orthologs coupled with a lack of structural information have hampered efforts to develop a selective allosteric antagonist against the hP2X7R using structure-based drug design. The structural and functional data that we present here provide insights into the molecular differences between full-length wild-type rat, mouse, and human P2X7Rs —the three most relevant orthologs for drug development—in apo closed and ATP-bound open state conformations. The structure of the hP2X7R in complex with the known adamantane-based allosteric antagonist UB-ALT-P30 reveals ligand-receptor interactions within the classical allosteric pocket. We leveraged these data to optimize an antagonist scaffold tailored to the classical allosteric ligand-binding site in the human ortholog using structure-based drug design. After synthesizing and characterizing six antagonists with different polycyclic cores, we identified one (UB-MBX-46) that was highly potent and selective for the hP2X7R. The high-resolution cryo-EM structure confirmed that it binds to the classical allosteric binding site with optimized ligand-receptor interactions. With subnanomolar potency and high subtype selectivity, UB-MBX-46 is a promising compound with significant therapeutic potential for treating P2X7R-associated diseases.