In silico design of novel probes for the atypical opioid receptor MRGPRX2.

The primate-exclusive MRGPRX2 G protein-coupled receptor (GPCR) has been suggested to modulate pain and itch. Despite putative peptide and small-molecule MRGPRX2 agonists, selective nanomolar-potency probes have not yet been reported. To identify a MRGPRX2 probe, we first screened 5,695 small molecules and found that many opioid compounds activated MRGPRX2, including (-)- and (+)-morphine, hydrocodone, sinomenine, dextromethorphan, and the prodynorphin-derived peptides dynorphin A, dynorphin B, and α- and β-neoendorphin. We used these to select for mutagenesis-validated homology models and docked almost 4 million small molecules. From this docking, we predicted ZINC-3573-a potent MRGPRX2-selective agonist, showing little activity against 315 other GPCRs and 97 representative kinases-along with an essentially inactive enantiomer. ZINC-3573 activates endogenous MRGPRX2 in a human mast cell line, inducing degranulation and calcium release. MRGPRX2 is a unique atypical opioid-like receptor important for modulating mast cell degranulation, which can now be specifically modulated with ZINC-3573.

G protein-coupled receptors (GPCRs) are seven transmembrane receptors that transduce extracellular signals into biological responses via heterotrimeric G-proteins and β-arrestins[1]. GPCRs are involved in nearly every known biological system and, unsurprisingly, GPCR-targeting small molecules represent the largest target class for FDA-approved drugs[2]. Despite their therapeutic utility, only 10% of GPCRs function as therapeutic targets for FDA approved drugs[2], while ~120 of 394 non-olfactory GPCRs represent “orphan receptors” or understudied GPCRs (“oGPCRs”[3]) without useful probes and, frequently, without validated endogenous ligands[4,5]. The process of discovering specific and potent probes for oGPCRs yields useful research tools and can also illuminate previously unrealized drug interactions, potentially establishing new drug targets.

To identify oGPCR ligands, we have developed and tested new physical and computational approaches for screening these receptors. The first physical method, dubbed “PRESTO-Tango,” involves high-throughput, massively parallel screening of potential modulators (including small molecules, bioactive peptides, and other reagents) for their ability to stimulate β-arrestin recruitment at most non-olfactory receptors in the GPCRome[6]. The second orthogonal and complementary approach relies on the principle that over-expressed G proteins facilitate a partially active state for most GPCRs enabling the discovery of both allosteric and orthosteric modulators for oGPCRs using engineered yeast[3]. Encouraged by the success of other structure-based drug design methods in GPCRs[7-9], our in silico approach leverages the physical screens to develop comparative structural models of the receptors, and then computationally screens a much wider chemical space—typically several million commercially available molecules—to find specific ligands for the oGPCRs.

We apply this strategy to the Mas-related G protein-coupled receptor X2 (MRGPRX2) -- a primate-exclusive[10,11] class A orphan GPCR expressed in mast cells and small diameter neurons in the dorsal root and trigeminal ganglia[12-16]. Several unrelated peptides are reported to activate MRGPRX2 including Cortistatin-14[16], Substance P[13], and PAMP(9–20)[14], whether any proposed peptides are endogenous MRGPRX2 agonists is unknown. MRGPRX2 remains an oGPCR in part because no convincing rodent ortholog has been validated (see Discussion) and because MRGPRX2-selective nanomolar potency probes are unavailable. Although several selective agonists are reported for MRGPRX2[17,18], the compounds are not easily obtained and have not been validated for specificity or potency. The identification of demonstrably selective, potent MRGPRX2 agonist probes represents an essential step toward illumination of its function in vitro and in vivo.

Here, we describe how an integrated approach combining our PRESTO –Tango method and our modeling and docking platforms led to the identification of MRGPRX2 agonists, including exogenous and endogenous opioids and a selective MRGPRX2 probe ZINC-3573. We confirmed ZINC-3573’s selectivity for MRGPRX2 via testing at 315 class A GPCRs using PRESTO-Tango, binding assays performed by the National Institute of Mental Health Psychoactive Drug Screening Program (NIMH-PDSP), and by testing the parent scaffold using a commercial (DiscoverX) kinome screen. Using ZINC-3573, we show MRGPRX2 activation induces intracellular calcium release and degranulation in a human mast cell line. We also demonstrate that MRGPRX2 represents a novel Gαq-coupled opioid-like receptor that could mediate some peripheral side effects of commonly prescribed opiate medications. This discovery of the specific MRGPRX2 agonist ZINC-3573, matched with an inactive enantiomer, provides the community with a pair of chemical probes by which the in vivo function of this fascinating target may be investigated with exquisite specificity and control.

RESULTS

Identification of MRGPRX2 agonists

We initially attempted to replicate prior reports of potential MRGPRX2 agonists to determine if any might prove suitable as leads for probe development. Unfortunately, we could not replicate the activities of most reported MRGPRX2 agonists when tested at the highest concentrations possible for our assays (Fig 1, Supplementary Results, Supplementary Table 1). Thus, of the many putative MRGPRX2-activating peptides and peptide-like compounds, we could replicate activities only for substance P, cortistatin-14 and PAMP (9–20) (Fig 1a,b). Mastoparan, octreotide, leuprolide, and kallidin were inactive or only marginally so (e.g. Gαq EC50 = 11.5 μM for kallidin) at human MRGPRX2 (Fig 1c, Supplementary Table 1).The putative MRGPRX2 agonist mastoparan[19] was active in cells with and without MRGPRX2 expression suggesting non-specific activity (Fig 1d., Supplementary Table 1).

Figure 1

Validation of MRGPRX2 and MRGPRB2 Agonists

A–D) Average concentration response curves in the FLIPR intracellular calcium release assay (n=3 in triplicate) in MRGPRX2-inducible cells, designated +X2 or –X2 for +/−tetracycline-induced receptor expression, for previously published MRGPRX2 agonists. Y axis is fold change calcium release over basal. E) Concentration response curves in FLIPR intraceullar calcium release assay for stable cell lines expressing either the proposed orthologues human MRGPRX2 (solid lines) or mouse MRGPRB2 (dotted lines) with X2-activating peptides Cortistatin-14 and Substance P and small molecule agonist TAN-67. Y axis is fold change calcium release over basal (n=3 in quadruplicate) F) Concentration response curves for Cortistatin-14, Substance P, and small molecule agonist TAN-67 with MRGPRX2-Tango (solid lines) and MRGPRB2-Tango (dotted lines) in the PRESTO-Tango arrestin recruitment assay (n=3 in triplicate). Y axis is fold change response over basal luminescent signal. All error bars represent SEM.

Of the more than one dozen non-peptide compounds reported to activate MRGPRX2, we could only replicate four (TAN-67[20], compound 48/80[19], cetrorelix[19] and the proposed selective agonist complanadine A[18]) and even these had low potencies or have other known receptor targets (Fig 1c,d, Supplementary Table 1). Notably, we could not validate several recently reported secretagogue agonists for MRGPRX2 including the THIQ motif-containing octreotide, rocuronium, ciprofloxacin, atracurium, moxifloxacin, and levofloxacin[19] even when tested up to 100 μM (Fig 1c,d, Supplementary Table 1). Likewise, although the proposed nanomolar MRGPRX2-selective agonist complanadine A[18] was active in our assays, we measured a Gαq EC50 of 18 μM and could not obtain a β-arrestin EC50 due to apparent cytotoxicity above 1 μM using PRESTO-Tango, a luciferase reporter-based β-arrestin screening platform[6] (Fig 1c,d, see Supplementary Table 1 for a full list of validated compounds).

As MRGPRX2 is expressed only in primates, finding a rodent analog would enable the use of genetic techniques to probe for the receptor’s functional roles, with the caveat that GPCRs may often be knocked down without incurring phenotypes that recapitulate their roles in pharmacology. The suggested MRGPRX2 rodent ortholog MRGPRB2[19] only shares 52% sequence identity with MRGPRX2 (Supplementary Fig 1a) and we could not substantiate any proposed shared ligands except for Cortistatin-14, which had high micromolar activity at MRGPRB2 (Fig 1e, f, Supplementary Fig 1b). Thus, although there have been prior studies of MRGPRX2 and/or MRGPRB2’s pharmacology, many remain insufficiently robust for facile replication; and no independent studies have emerged that replicate these prior findings.

Accordingly, we chose an unbiased approach to identify reliable MRGPRX2 probes. We recently proposed parallel screening strategies[3,6] using small libraries of drugs and drug-like compounds as fruitful initial approaches to discover active compounds at oGPCRs[6]. Here we screened 5,695 unique compounds for agonist activity at three members of the MRGPRX family (MRGPRX1, MRGPRX2, and MRGPRX4) using the PRESTO-Tango platform (see Online Methods). Our strategy was to screen against three MRGPRX family receptors in parallel to find active compounds with selectivity within the family.

The screening revealed MRGPRX1 had the fewest number of actives (39), followed by MRGPRX4 (54), and MRGPRX2 (81) (Fig 2a). We were most interested in compounds showing selective MRGPRX2 agonism without apparent activity at other MRGPRX receptors. Among the 67 compounds that activated MRGPRX2, and neither MRGPRX1 nor MRGPRX4, were five opioid-related ligands ADL-5859, sinomenine, dextromethorphan, dextromethorphan’s metabolite dextrorphan, and the previously reported MRGPRX2 ligand TAN-67[20], a delta opioid receptor agonist.

Figure 2

PRESTO-Tango Screening of MRGPRX2 reveals new agonists

A) Venn diagram depicting pooled PRESTO-Tango screening actives for MRGPRX1, MRGPRX2, and MRGPRX4 for all hits with greater than 2-fold activation over basal (known false-positives and duplicates excluded). B-C) Average concentration response curves (n=3 in triplicate wells for all, except ADL5859 n=1) for the five compounds from the screening show low micromolar activation of MRGPRX2. D) Concentration response curve for previously published MRGPRX2 peptide agonists at the MRGPRX2-Tango construct. Y axes shown as % of TAN-67 activity. All error bars represent SEM.

Confirmatory concentration response curves using the PRESTO-Tango platform indicated the five opioid-like compounds had low micromolar potency (Fig 2b, c). To confirm the MRGPRX2-Tango construct performed similarly to the unmodified wild-type (WT) receptor, we tested previously reported MRGPRX2 agonists TAN-67 [20], cortistatin-14 [16], substance P [13], and compound 48/80 [19] and found all activated the MRGPRX2-Tango receptor at similar reported potencies for Gαq assays at the WT receptor (Fig 2d).

To confirm MRGPRX2 Gαq-mediated functional activity of agonists, we used a tetracycline-inducible WT MRGPRX2 stable HEK-T cell line to test for intracellular calcium release. Dextromethorphan, dextrorphan, sinomenine, and TAN-67 promoted intracellular calcium release when MRGPRX2 expression was induced by tetracycline (1 μg/ml; Supplementary Fig 2a) but not in the absence of tetracycline (e.g, MRGPRX2 is not expressed, Supplementary Fig 2b).

Preference for dextro-enantiomers and N-methyl scaffolds

As several dextrorotatory opiate ligands activated MRGPRX2, we initially investigated ligand stereochemistry, aware that classical opioid receptors prefer levorotary morphinans and benzomorphans[21]. We assayed levorphanol and levallorphan, enantiomers of the screening hits dextrorphan and dextromethorphan, respectively, for activity at MRGPRX2 in β-arrestin recruitment and calcium mobilization assays. Levorphanol was approximately ten-fold less potent at MRGPRX2 than dextrorphan and levallorphan was completely inactive up to 100 μM (Fig 3a). Likewise the dextrorotary morphinan sinomenine had comparable potency to dextrorphan and dextromethorphan at MRGPRX2 (Supplementary Table 2), more potent than sinomenine’s purported activity at the opioid receptors[22] (Supplementary Fig. 3). MRGPRX2 also preferred dextrorotary benzomorphan compounds, as (−)-cis-normetazocine was ten-fold less potent compared to (+)-cis-normetazocine (Fig 3b, Supplementary Table 2).

Figure 3

MRGPRX2 is activated by many opioid scaffolds

A–D) Average concentration response curves (n=3 in triplicate) for structurally related compounds (morphinans, benzomorphans, morphine and analogues, and codeine and analogues, respectively) in intracellular calcium release assay where Y axis is fold change calcium release over baseline. Error bars represent SEM. E) Summary of major findings in structure activity relationships for opioid scaffolds at MRGPRX2, including size of N-substituent, stereochemistry of chiral centers, and bulk on the 3-position of the morphinan.

To determine if other opioids activated MRGPRX2 with similar stereochemical preferences, we tested a panel of 11 morphine and codeine analogues for activity in PRESTO-Tango and calcium mobilization assays, including the most common metabolites of morphine and codeine. Both (+)- and (−)-morphine enantiomers activated MRGPRX2 with comparable potency in PRESTO-Tango and calcium assays, as did (+)- and (−)codeine and thebaine (Fig 3c,d, Supplementary Table 2; structures in Fig 3e). For the metabolites, we observed differential effects of substitutions on C-3 and C-6 positions in the morphine scaffold, as replacement of morphine’s 3-hydroxy with a 3-methoxy (e.g. codeine and thebaine) had little effect on affinity or efficacy at MRGPRX2 whereas replacement of the morphine’s 3-hydroxy with gluconuride to yield morphine-3β-gluconuride eliminated MRGPRX2 agonist activity (Fig 3c,d, Supplementary Table 2; structures in Fig 3e). In the C-6 position, gluconuride or acetyl modifications to morphine and codeine all retained agonist activity, albeit with loss in potency compared to (−)-morphine and (−)-codeine, respectively (Fig 3c,d Supplementary Table 2). These preliminary opioid structure-activity relationships (SARs) demonstrate that larger modifications on C-6 are more tolerated than on C-3.

We then explored the effect of substituents on the cationic nitrogen in the morphinans on MRGPRX2 activation. N-methyl substituted scaffolds, as in codeine, morphine, and metazocine, conferred receptor agonism while the N-unsubstituted norcodeine was inactive up to 100 μM (Fig 3b–d, Supplementary Table 2). Morphinans and benzomorphans with substituents larger than N-methyl, such as the N-allyl of levallorphan and N-allyl-normetazocine, and the N-cyclopropyl of naltrexone and cyclazocine, were inactive at MRGPRX2 (Fig 3a,b, Supplementary Table 2). The inactive morphinans and benzomorphans showed no antagonist activity against MRGPRX2, suggesting a tight SAR around substitution of this position.

As mentioned, TAN-67 is a δ-opioid receptor agonist known to activate MRGPRX2[20]. The active enantiomer at the δ-opioid receptor is (−)-TAN-67 whereas (+)-TAN-67 displays pro-nociceptive activity in vivo through an unidentified receptor[23]. We tested (+)-TAN-67 for MRGPRX2 activity in PRESTO-Tango and calcium mobilization assays and found (+)-TAN-67 was more potent than racemic (+/−)-TAN-67 (Gαq EC50 = 290 nM vs 740 nM) (Supplementary Fig 4), demonstrating that MRGPRX2 prefers dextro-enantiomers at most chemical scaffolds with some exceptions. Neither (−)- nor (+)-naloxone antagonized (+/−)-TAN-67 agonist activity (Supp Fig 5a, b) and no other canonical opioid antagonists, including naltrindole, naltrexone, and β-chlornaltrexamine, inhibited (+/−)-TAN-67 activity. (Supp Fig 5 c–e). These SARs suggests that MRGPRX2, despite its distant sequence relationship to the four classic opioid receptors (transmembrane region sequence identity no greater than 26%), is an opioid-responding oGPCR with important differences in ligand recognition patterns.

Prodynorphin-derived peptides activate MRGPRX2

A key question is whether MRGPRX2 is also activated by endogenous opioid ligands. Accordingly, we tested a panel of 20 endogenous opioid peptides for activity in PRESTO-Tango and intracellular calcium assays. MRGPRX2 was preferentially activated by prodynorphin-derived peptides (Fig 4a, b) and only minimally activated, if at all, by other opioid peptides (Fig 4c, d). The full length dynorphin A (1–17) and several truncated prodynorphin peptides activated MRGPRX2 (Fig 4a). C terminal amino acid truncation of dynorphin peptides reduced agonist potency until dynorphin A (1–7) and dynorphin A (1–6) fragments, which were completely inactive (Fig 4a, Supplementary Fig 6a).

Figure 4

MRGPRX2 is preferentially activated by prodynorphin-derived peptides

A–B) Averaged concentration response curves (n=2 or 3 in triplicate, see Supplementary Table 2) for pro-dynorphin-derived peptides in MRGPRX2 Tet-On cells where Y axis is fold change calcium release over baseline. Error bars represent SEM. C, D) Average concentration response curves (n=2 or 3 in triplicate, see Supplementary Table 2) depicting non-prodynorphin-derived peptides with minimal activity compared to dynorphins in A and B. Error bars represent SEM.

To determine if the C terminal portion of dynorphin A was sufficient for MRGPRX2 activation, we assayed dynorphin A (13–17) and dynorphin A (6–17) and found that dynorphin A (6–17) retained minimal activity (EC50 = 39.6 μM) and dynorphin A (13–17) was inactive (Fig 4b), suggesting both the N terminal YGGF motif and C terminal cationic tail are important for MRGPRX2 activation (Supplementary Fig 6a). Dynorphin B (1–13) and α- and β-neoendorphin also activated MRGPRX2, with dynorphin B and α-neoendorphin having almost identical μM potency (Fig 4b). Intriguingly, EC50 values for the dynorphin peptides were right-shifted in the PRESTO-Tango assay, indicating perhaps modest G protein bias, similar to observations made for dynorphin peptides at the κ-opioid receptor[24].

To determine whether more closely-related receptors might share peptide ligands with MRGPRX2, we used the GPCRdb to identify class A receptors with greater than or equal to 30% sequence similarity to MRGPRX2 in the class A binding pocket (Supplementary Fig 6b). We identified 14 receptors, 11 of which had established probes (10 ligands total). 3 receptors were oGPCRs with no known ligands. We tested the 10 established ligands for activity and found only one ligand, somatostatin-14, activated MRGPRX2 (Supplementary Fig 5b,d; EC50 = 380 nM). This is consistent with previous reports that somatostatin-14, somatostatin-28, and Cortistatin-14 are MRGPRX2 agonists[16,20]. Interestingly, the canonical opioid receptors and somatostatin receptors are closely related in binding site-sequence space, but not in small molecule-chemical space[25]. Our data, along with previously proposed peptide ligands, suggest that although MRGPRX2 can be activated by many peptide ligands, this cannot be predicted from sequence similarity alone.

MRGPRX2 is an opioid-like oGPCR that responds to endogenous pro-dynorphin-derived opioid peptides, binds to many well-known synthetic opioid agonists, but which differs from classic opioid receptors in its unique preference for dextro-morphinans, dextro-benzomorphans and its inability to be antagonized by potent and classic opioid receptor antagonists. Although none of these compounds are suitable as a selective probe or tool, the chemical matter identified was useful to launch an in silico campaign to identify novel chemotypes active at MRPGRX2.

Structure-based docking predicts MRGPRX2-selective ligands

We next turned to our computational approach using a 1000-fold larger compound library than that used in the physical screen—the over 3.7 million commercially available “lead-like” molecules then in the ZINC database[26]. Our strategy was to calculate many 3D MRGPRX2 models, select those few that recapitulated the ligand recognition patterns we observed experimentally, and use these to template a final model to screen the full, larger compound library for novel molecules unrelated to opioid ligands but selective for MRGPRX2.

The general approach for modeling and docking has been recently described[3], and we summarize it here (see Methods for more details). We first sought GPCRs of known structure with MRGPRX2 sequence similarity; intriguingly, this led to the κ-opioid receptor, which only shares 23.3% sequence identity in the transmembrane regions with MRGPRX2. We calculated an initial MRGPRX2 structure using Modeller[27] and expanded on this using elastic network modeling to increase the amount of sampled backbone conformations[28]. From the initial model, we calculated 360 further structures using the Elastic Network Modeling program ENM; half of these had an extracellular disulfide bond imposed and half did not. Each of these was expanded three-fold with Modeller (i.e., using the ENM models as templates for Modeller). Overall, 1,080 models were calculated.

We docked the 5,695 unique compounds from physical screening against all 1080 models (i.e., over 6 million compound-receptor docking calculations, and over 1012 individual configurations of the library molecules in the receptor). We looked for models that best enriched the discovered opioid agonists over the vast number of inactive decoy molecules. The top-enriching models were inspected visually for binding poses that captured sensible ion-pairing interactions with the conserved aminergic group of the opioid agonists. This prioritized a model that had a background-corrected logAUC for enrichment [29,30] of 8.42—which happened to be the top-enriching model (this is not always the case, as sometimes the top-enriching model does not have sensible geometries[31], though usually the model chosen is among the best enriching models). Residues within 6 Å of the dextromethorphan pose were minimized around the docked ligand with PLOP[32], and the 5,695 molecules re-docked. Two rounds of docking and PLOP-based refinement led to the final modeled receptor conformation, which predicts the opioid agonists making ionic interactions with Glu1644.60 and Asp1845.36 (Fig 5, showing dextromethorphan). The exact rotamer state of these residues allowed some flexibility; i.e., several different rotamers of the Asp/Glu led to productive placement of the known agonists (Supplementary Fig 7e).

Figure 5

In silico MRGPRX2 homology modeling predicts a selective agonist

Workflow depicting MRGPRX2 homology model construction (top left) followed by identification of a putative binding site (top middle) that was confirmed by testing the mutations E164Q and D184N (top right, average dose response, n=3 in triplicate, shown with dextromethorphan (DXM). Then, ~3.7 million molecules were docked to predict the agonist ZINC-9232 (bottom left). Further iteration and testing revealed the tool and selective compounds (S)- and (R)-ZINC-5373 (bottom middle and right). Docking pose of selective compound is supported with mutation experiments in PRESTO-Tango (bottom right, average dose response curve, average of n=4 in triplicate). Error bars on graphs shown is representative of SEM.

To test the predicted ion pair between the receptor and the ligands, we made a series of mutations at Glu1644.60 and Asp1845.36. E164Q and D184N substitutions retained steric properties of the WT residues but removed their negative charges and both resulted in a loss of activity for dextromethorphan, (−)-morphine, and related opioids in the PRESTO-Tango assay (Fig 5, Supplementary Fig 7a). We then tested the importance of the length of the acidic residue via E164D and D184E mutations and observed that E164D increased the Emax and reduced the potency of both dextrorphan and morphine (Supplementary Fig 7 a,c) while D184E ablated activity of all opioids tested (Fig 5, Supplementary Fig 7a–d). The substitutions support the importance of the proposed ionic interaction and the putative binding site predicted by the in silico model, suggesting both Glu1644.60 and Asp1845.36 are necessary for MRGPRX2 opioid activation.

As an aside, dynorphin A (1–13) activity was lost with the D184N but not the E164Q mutation, suggesting only one of these residues is required for the opioid peptide interaction (Supplementary Fig 7d). We modeled the putative MRGPRX2-dynorphin binding site by overlaying the MRGPRX2 homology model with the previously published κ-opioid receptor docked with dynorphin A (1–13) [33]. The N-terminal Tyr1 of dynorphin A (1–13) is accommodated within a negatively charged aromatic pocket, whereas Arg7 and Phe4 appear to interact with Asp1845.36 (Supplementary Fig 8 a,b,c). The predicted orientation of dynorphin A (1–13) at Asp1845.36 and Glu1644.60 shows a clear preference for charged interactions between dynorphin A and Asp1845.36, a result supported by the mutagenesis results (Supplementary Fig 8d).

With experimental support for the MRGPRX2 comparative model, we proceeded to virtually screen the ZINC library for new ligands. Over 3.7 million commercially available molecules were docked, each in an average of 926 orientations and 558 conformations; overall 1.9 trillion docked complexes were sampled and scored (for details, see Methods). We selected, purchased, and assayed 20 compounds from the 0.13% of top-ranked compounds for activity at MRGPRX2. As there is little scoring difference among the top-ranked molecules, we chose molecules following our usual strategy[34] and prioritized those with diverse, non-opioid chemotypes. We deprioritized molecules with apparently high internal energies, which the docking scoring function does not well account for[34], and prioritized those that made ionic interactions with the Asp1845.36/Glu1644.60 pair, similar to the modeled opioids. Of the 20 ZINC compounds tested, one, ZINC-72469232 (1, here on referred to as ZINC-9232), had substantial activity below 10 μM in PRESTO-Tango and calcium release assays (Fig 5). To improve potency, 22 analogs of ZINC-9232 topologically similar to the lead were obtained and tested. Of these 22, 7 were active, with the most potent ZINC-72453573 (2, from here on referred to as ZINC-3573) having sub micromolar potency (EC50 = 760 nM, Fig 5, Supplementary Fig 9). We then assessed whether ZINC-3573 interacts with the predicted residues Asp1845.36 and Glu1644.60 using the E164Q and D184N mutations and found that both eliminated ZINC-3573 agonism at MRGPRX2 (Fig 5), consistent with the modeling.

ZINC-3573 as a chemical probe for MRGPRX2

To confirm MRGPRX2 selectivity, we tested ZINC-9232 and ZINC-3573 activity at 315 other human GPCRs using our PRESTO-Tango GPCRome assay. ZINC-9232 and ZINC-3573 showed minimal agonist efficacy at receptors other than MRGPRX2 at 10 μM (Supplementary Fig 10 a,b). ZINC-9232 was also screened against a panel of 97 representative human kinases using the DiscoverX KINOMEscan diversity panel; only three kinases were modestly inhibited, with IC50 values between 20–30 μM (Supplementary Fig 11).

These results encouraged us to synthesize both enantiomers of the more potent ZINC-3573, originally supplied as a racemic mixture, in an effort to create a pair of differentially active molecules that could jointly be used as a chemical probe pair (Supplementary Scheme 1, Supplementary Info.). The R-isomer (3) retained an EC50 of 740 nM in PRESTO-Tango and a similar EC50 value of 1 μM in the calcium mobilization assay (Supplementary Fig 9). The S-isomer (4) had little activity below concentrations of 100 μM (Supplementary Fig 9). The separate activity of the enantiomers makes them highly attractive probes, as one can distinguish MRGPRX2 activity (owing to the R-isomer) from general, non-specific activity owing to the scaffold and its physical properties (due to either isomer). Each enantiomer was further tested using the PRESTO-Tango GPCRome screening platform (Supplementary Fig 10 c,d), which indicated 21 receptors may be activated at 2-fold or higher by (R)- and (S)-ZINC-3573; however, subsequent concentration-response studies showed that no other receptor was significantly activated by either compound (Supplementary Fig 11).

MRGPRX2 agonists induce degranulation in human mast cells

MRGPRX2 has been implicated in IgE-independent inflammatory responses in mast cells using peptide ligands and other non-specific agonists[13,15,19], prompting us to investigate whether MRGPRX2-selective agonists induce degranulation and mobilize intracellular calcium in the LAD2 human mast cell line. ZINC-9232 and ZINC-3573 induced intracellular calcium release and degranulation in LAD2 mast cells at comparable potencies to assays performed in MRGPRX2-expressing HEK-T cells (Fig 6a,b). For the stereochemical pair, only (R)-ZINC-3573 promoted degranulation in LAD2 mast cells (Fig 6b), consistent with the HEK-T in vitro activity. As expected, MRGPRX2 siRNA significantly reduced (R/S)-ZINC-3573-induced degranulation (Fig 6c, p=0.01).

Figure 6

MRGPRX2 mediates intracellular calcium release and degranulation in the LAD2 human mast cell line

A) Average concentration response for MRGPRX2-selective agonists ZINC-6932 and ZINC-3573 in the calcium mobilization assay (n=3 in triplicate). B) Average concentration response (n=3 in triplicate) for MRGPRX2 probes ZINC-3573 (R) and ZINC-3573 (S) in the β-hexosaminidase degranulation assay. C). Bar graph depicting fold change %degranulation (baseline = average DMSO of all plates) induced by EC80 concentration of drug following 25 nM MRGPRX2 siRNA transfection; NT = Non Targeting pool, siRNA 2+3 = MRGPRX2 siRNA pool. Statistics = Two-way ANOVA with a Sidak post-hoc test (p <0.05 = *, p=0.031, 0.031, 0.033, and 0.017 for TAN, Dyn, Morphine, and ZINC-3573, respectively). n=3 for all except ZINC-3573, n=2. All replicates in triplicate wells. D) Concentration response curves for MRGPRX2-activating opioid ligands and the structurally related, κ-opioid receptor ligand (−)-cyclazocine in the calcium mobilization assay (n=3 in triplicate). E) Average concentration response in intracellular calcium release with MRGPRX2-activating (+/−)-TAN-67 and canonical opioid receptor ligands DADLE, DAMGO, Salvinorin A, and BW373U86 (n=3 in triplicate). F) Bar graph depicting baseline normalized percent degranulation induced for each 10 μM (+/−)-TAN-67, DMSO, and calcium ionophore ionomycin in the presence of absence of biotin-labeled IgE antibodies. (Streptavidin in all wells, average of n=2 in triplicate.) G). Average concentration response curves for MRGPRX2-activating opiates (n=3 in triplicate.) in the β-hexosaminidase degranulation assay. H) Degranulation concentration response with canonical opioid receptor agonists and MRGPRX2-activating (+/−)-TAN-67, (n=3 in triplicate). All error bars demonstrate SEM.

We then considered whether the opioid ligands might activate endogenously expressed MRGPRX2 to induce degranulation in mast cells. This is important, in part, because of the well-known but enigmatic ‘histamine flush’ associated with many opioids, which is not due to engagement of classical opioid receptors. We found (+/−)TAN-67, (+)-morphine, (−)-morphine and dynorphin A (1–13) produced robust intracellular calcium release in the mast cells while the structurally similar κ-opioid receptor agonist (−)-cyclazocine, which does not activate MRGPRX2, did not (Fig 6d). Correspondingly, structurally unrelated agonists for the κ, δ, and μ opioid receptors (Salvinorin A, BW373U86 and DAMGO, respectively) were inactive (Fig 6e). To determine whether MRGPRX2-activating opioids induce degranulation, we first tested 10 μM (+/−)-TAN-67 in the presence or absence of biotin-labeled IgE antibodies and found that with or without IgE-biotin antibody activation via streptavidin, the MRGPRX2-ligand (+/−)-TAN-67 produced 70% degranulation, indicating MRGPRX2-mediated degranulation is IgE-independent (Fig 6f). MRGPRX2 agonists (+/−)-TAN-67, (+)-morphine, (−)-morphine, and dynorphin A (1–13) all promoted degranulation in LAD2 cells (Fig 6g) while the classical opioid agonists salvinorin A, BW373U86 and DAMGO did not (Fig 6h). 10 μM naloxone pretreatment did not influence the EC50 or Emax values of MRGPRX2 opioid agonists (Supplementary Fig 12a,b), whereas MRGPRX2 siRNA significantly attenuated (+/−)-TAN-67, dynorphin A (1–13), and (−)-morphine-induced degranulation (Fig 6c, Supplementary Fig 12c,d). These data suggest MRGPRX2, not the canonical opioid receptors, mediates opioid-induced degranulation in human mast cells.

DISCUSSION

Our MRGPRX2 results underscore two major findings. First, the oGPCR MRGPRX2 responds to opioid drugs and endogenous pro-dynorphin peptides at potentially physiologically relevant concentrations and mediates opioid-induced degranulation in a human mast cell line. Although this receptor is not in the opioid receptor family by sequence similarity, its agonism by opioid drugs and peptides qualifies it as an atypical opioid receptor that responds to morphinan-based opioids and transmitters. A second key result is the structure-based discovery of the selective, sub-micromolar MRGPRX2 agonist (R)-ZINC-3573. This agonist promotes degranulation in mast cells, has no measurable agonism at over 315 other GPCRs, and the parent scaffold has little activity against 97 representative kinases. The inactive enantiomer (S)-ZINC-3573 and the active (R)-ZINC3573 are an effective and internally controlled probe-pair for investigating the biology of this intriguing primate-exclusive receptor. The probes will be made available to the community from Sigma-Aldrich (#SML1699 and #SML1700).

SAR results show MRGPRX2 is distinct from the canonical opioid receptors. MRGPRX2 prefers dextro-enantiomers and N-methyl substituted opioid scaffolds whereas opioid receptors prefer levo-enantiomer opioids[35,21] and tolerate a wider array of N-substituents. In further contradistinction to the canonical opioid receptors, MRGPRX2 is Gαq- rather than Gαi-coupled. Thus, MRGPRX2 may be classified as an atypical opioid recognizing receptor, arguably more an example of convergent evolution within the GPCR family than the divergent evolution that relates the four canonical opioid receptors.

MRGPRX2 activation by opiates may be relevant physiologically and therapeutically. We found MRGPRX2, not the canonical opioid receptors, mediates morphine and dynorphin A (1–13)-induced mast cell degranulation. Morphine and structurally similar analgesics induce mast cell histamine release[36,37] in humans, resulting in pruritus [38], vasodilation, and hypotension that is poorly reversed by naloxone [39,36]; these effects are not seen for opioid analgesics lacking MRGPRX2 activity, such as fentanyl[37]. MRGPRX2 may also be involved in the efficacy of sinomenine, which is used to treat rheumatoid arthritis due to its histamine-releasing properties[40]. Dextromethorphan’s potency at MRGPRX2 is also greater or equal to reported Ki values for μ, κ, δ, NMDA, and σ2 receptors[21], suggesting MRGPRX2 could contribute to dextromethorphan’s side effect profile which includes itch at high doses[41]. Definitive studies in non-human primates are needed to address these hypotheses.

The micromolar potency of pro-dynorphin peptides at MRGPRX2 suggests in vivo receptor activation might require close proximity to high local concentrations of dynorphin. Local synaptic concentrations of neuropeptides can reach millimolar range[42] and dynorphin is expressed in MRGPRX2-expressing regions[43-45]. At such concentrations, dynorphin could activate synaptic GPCRs or those expressed in mast cells, which can localize to nerve terminals[46]. Thus, it is conceivable that MRGPRX2 is exposed to activating concentrations of dynorphin in vivo.

Structural modeling enabled the discovery of a new scaffold, represented by ZINC-9232 and ZINC-3573, unrelated to classical opioids topologically and structurally. The unusual specificity of ZINC-3573 against essentially the entire GPCRome and the relevant kinome, and the availability of an inactive enantiomer makes this molecule a uniquely useful MRGPRX2 probe. Methodologically, we previously used the combination of an initial physical compound screen followed by a much larger docking screen to discover probes for oGPCRs[3]. Identification of ZINC-3573 by a similar approach against a wholly different receptor family suggests this approach has broad utility in structure-based drug design.

The modelled MRGPRX2 provided intriguing insights to this receptor’s unique opioid pharmacology. MRGPRX2 shares no more than 26% sequence identity with the μ-, κ-, δ- or nociceptin-opioid receptors, and its modeled orthosteric binding differs from that of the canonical opioid receptors. The cationic nitrogen of the morphinans is recognized by an aspartate in the opioid receptors, located in transmembrane (TM) helix 3, position 3.32. The residue for cationic nitrogen recognition in MRPRX2 appears instead to be in TM5, Asp1845.36 (see Supplementary Fig. 5c). Many other recognition residues do not overlap in sequence or structural placement. Indeed, the entire region surrounding Asp1845.36 and Glu1644.60 of the MRGPRX2 putative binding site is closed off from ligand contacts in the classical opioid binding site by Tyr3.33 – this is a smaller Thr110 in MRGPRX2. These seeming mismatches reflect that different receptor environments can recognize related ligand chemistry and the capacity for ligands to interact with receptors from different evolutionary families[25,47] and different G protein coupling.

Certain caveats merit airing. We have not determined the structure of MRGPRX2 in complex with any of the ligands discussed here, and the modeling that we have used to interpret activity and to predict new molecules—however successfully—must remain tentative. Until the opioid actions at this receptor can be probed in vivo, so, too, must the physiological implications. We note that a prior study also suggested that MRGPRX2 is a novel “morphine receptor,” that mediates some of morphine’s analgesic activity[48]. Whereas this report certainly shares some similarity with our observations, we find, in contrast, that MRGPRX2 is more potently activated by (+)-, rather than (−)-, morphine, suggesting MRGRPX2 is unlikely to confer analgesia, as (+)-morphine is devoid of such activity[35,49].

MRGPRX2 is a novel Gαq-coupled opioid-like receptor activated by endogenous prodynorphin-derived peptides and opioid compounds, including FDA-approved drugs and their metabolites. The discovery of selective and relatively potent MRGPRX2 agonist (R)-ZINC-3573 and its inactive (S)-isomer provides researchers with a chemical probe pair to specifically modulate this receptor, illuminating its role in pathological reactions such as itch and potentially revealing a path for therapeutic design.

ONLINE METHODS

Cell Lines

HTLA cells were a gift from Dr. Richard Axel (Columbia University) and are maintained at low passage batches. FLP-IN/T-REX HEK-293-T cells were purchased from Invitrogen for ease of inducible stable cell line generation and were certified as HEK-T and mycoplasma-free by Invitrogen. FLP-IN/T-REX cells were made into inducible stable cell lines according to the manufacturer’s instruction. The LAD2 human mast cell line was obtained from cells’ original source lab of Dr. Arnold Kirshenbaum and Dr. Dean Metcalfe [50] at the Laboratory of Infectious Diseases at the National Institutes of Health in Bethesda, MD and were maintained as per the their instructions. No further characterization of HTLA or LAD2 cells was performed, as the cells came from the source labs.

Chemical Compounds

The nine libraries used in screening were the NCC-1 (NIH Clinical Collection), NCC-2 (NIH Clinical Collection), NIMH Library, Tocris, Prestwick, LOPAC, Selleck, Spectrum, and the Roth Lab Collection (in-house). Opioid ligands were obtained from either Sigma Aldrich or synthesized.

Constructs

MRGPRX-Tango plasmids for MRGPRX1, MRGPRX2, or MRGPRX4 were made as was previously described[6] MRGPRX2 constructs were also “de-Tangoized” to form the WT-MRGPRX2 construct by inserting a STOP codon before the 3’ TEV site for transient transfection. MRGPRB2 insert was obtained from Dharmacon cDNA and subcloned into the PRESTO-Tango backbone.

PRESTO-Tango Assay Screening

HTLA cells (HEK-T cells stably expressing a β-arrestin2-TEV fusion protein and a tTa-dependent luciferase reporter) were maintained in DMEM (Corning) containing 10% FBS, 2 μg/mL puromycin and 100 μg/mL hygromycin B in a humidified atmosphere at 37°C with 5% CO2. In brief, cells were plated to 50% confluency and transfected with a codon-optimized MRGPRX-Tango construct using the calcium phosphate method [51]. The next day, transfected cells were transferred to glass-bottomed, poly-L-lysine-coated white 384-well plates at a density of 20,000 cells/well. The following day, cells were treated with 10 μM concentrations of small molecules (in quadruplicate) diluted in drug buffer (1X HBSS with 20 mM HEPES and 0.3% Bovine Serum Albumin, pH 7.4) and incubated for 18–24 hours. After drug incubation, medium was removed and 20 μL of Bright Glo (Promega) (diluted 20-fold) was added to each well and incubated 15 minutes at room temperature. Luminescence was measured on a TriLux luminescence counter and analyzed as relative luminescent units (RLU) using GraphPad Prism. Compounds that increased the luminescence signal by two fold or greater were considered “actives” and were analyzed further using concentration-response curves. More detailed discussion of the PRESTO-Tango assay can be found in[6]. Total number of dose response replicates for each compound can be found in Supplemental Table 2.

Intracellular Calcium Mobilization Assay

We generated a MRGPRX2-expressing tetracycline-inducible stable cell line containing the MRGPRX2 codon-optimized receptor sequence with a N-terminal FLAG tag using the FLP-IN/T-REX Core Kit (Invitrogen) following the manufacturer’s instructions. MRGPRX2 stable cells were maintained in DMEM containing 10% FBS, 100 μg/mL hygromycin B, and 15 μg/mL blasticidin. For the calcium mobilization assay, cells were plated into glass bottomed, poly-L-lysine coated, black 384 well plates at a density of 20,000 cells/well in medium containing 1% dialyzed FBS, 1 μg/mL tetracycline, 100 IU/mL penicillin and 100 μg/ml streptomycin and incubated 24 hours. Following tetracycline induction, medium was removed and cells were loaded with 20 μl/well of 1X FLIPR Calcium dye (Molecular Devices) and 2.5 mM probenic acid, for 1 hour in a humidified environment at 37°C with 5% CO2. For mast cell calcium experiments, cells were seeded at a density of 50,000 cells/well in Tyrode’s buffer containing 1X calcium dye and incubated for 1 hour before treatment and analysis. After dye loading, baseline was measured for 10 seconds before drug treatment, and then cells were treated with 10 μL of 3x concentrated drug in drug buffer (1X HBSS with 20 mM HEPES and 2.5 mM probenic acid, pH 7.4) in 16 point concentration response curves from 10 μM to 0.003 nM. Fluorescence was measured for an additional 120 seconds and data was analyzed using GraphPad Prism. Total number of dose response replicates for each compound can be found in Supplemental Table 2.

β-hexosaminidase Assay

LAD2 cells were maintained in Stem-Pro34 medium (GIBCO) supplemented with StemPro34 Supplement (GIBCO), 2 mM L-Glutamine, 100 U/mL penicillin/streptomycin, and recombinant human stem cell factor (100 ng/ml) (Peprotech) as described in[52]. For the assay, adapted from [53], cells were incubated in fresh medium for 18 hours prior to the experiment. If testing IgE activation, cells were also incubated with biotin-labeled IgE during this time. Then, cells were washed twice in Tyrode’s buffer (20 mM HEPES with 134 mM NaCl, 5 mM KCl, 1.8 mM CaCl2, 1 mM MgCl2, 5.5 mM glucose, and 0.3% bovine serum albumin, pH 7.4) and seeded at 10,000 cells per well of V-bottom clear 96 well plates in 90 μL of Tyrode’s buffer. Drugs were diluted at single concentrations or in 8-point concentration-response curves at a 10X working concentration in Tyrode’s buffer and 10 μL drug, or streptavidin for biotin-labeled IgE samples, and were added in triplicate to the 90 μL mast cell suspension. Drugs and cells were incubated for 30 minutes in a humidified incubator without CO2 at 37°C. After drug incubation, plates were centrifuged at 250g for 5 minutes and 30 μl supernatant was added to new 96 well plates containing 10 μL of 10 μM 4-nitrophenyl N-acetyl-β-D-glucosaminide (NAG) diluted in 0.1M Citrate Buffer (49.5% 0.1M citric acid and 50.5% 0.1M sodium citrate, pH 4.5). The remaining supernatant was discarded and cells were lysed by adding 100 μl of Tyrode’s buffer with 0.1% Triton-X-100 to the pellets and mixed. Thirty μL of lysate was added to 10 μL of 10 μM NAG. All plates were incubated for 90 minutes in a humidified incubator without CO2 at 37°C. Lastly, 100 μL of bicarbonate buffer (88 mM Na2CO3, 40 mM NaHCO3, pH 10) was added to stop the reaction, inducing a color change that was measured on an absorbance plate reader (BMG Labtech POLARstar Omega) at 405 nm. Results were analyzed by dividing the released fraction by the total fraction (released + lysed) and multiplying by 100 to get a percentage degranulation. To correct for spontaneous degranulation, average percent degranulation for vehicle-treated wells was subtracted from drug-treated wells. Total number of dose response replicates for each compound is reflected in figures and figure legends.

siRNA Knockdown and qPCR Analysis

Using TransIT 20/20 transfection reagent (Mirus), 25 nM of each siRNA (Dharmacon # EQ-005666-00-0002, set of 4 distinct MRGPRX2 siRNAs) or the non-targeting siRNA control (Dharmacon # D-001910-10-05) was prepared in 250 μL of OPTI-MEM with 5 μL TransIT per reaction. After 30 minutes, the transfection mixture was added dropwise to 1.2 million mast cells/well of a 6-well plate and allowed to incubate 72 hours. After 72 hours, 200,000 cells were removed for use in the β-hexosaminidase assay and 1 million cells were assayed using qRT-PCR analysis with the High-Capacity cDNA Reverse Transcription Kit (Life Technologies) following the manufacturer’s instructions and TaqMan Universal Master Mix II with UNG (Life Technologies) and TaqMan Gene Expression Assay for MRGPRX2 (Life Technologies #4331182, Hs00365019_s1) following the manufacturer’s suggested protocol.

Significance calculations of siRNA experiments

To assess significance, we used GraphPad Prism to perform a Two-way ANOVA and Sidak post-hoc with an alpha value of 0.05. Total number of replicates is reflected in figure legends and figures.

Homology Modeling

The alignment for the construction of the MRGPRX2 models was generated using PROMALS3D[54], and homology models were built with MODELLER-9v8[27] using the crystal structure of the κ-opioid receptor (PDB code 4DJH) as the template. The homology models were aligned against the entire MRGPRX and opioid receptor family. The alignment was manually edited to remove the amino and carboxy termini that extended past the template structure, to remove the engineered T4 lysozyme, and to create different alignments of the flexible and non-conserved second extracellular loop (EL2); the final resulting sequence alignment between the template opioid structure and the MRGPRX receptors is shown in Supp. Fig 5c. Three models were built from each of 180 elastic network models (ENMs), produced by the program 3K-ENM[28], for a total of 540 models built during each iterative round of model refinement. Although EL2 is significantly shorter in MRGPRX2 than in other GPCRs, and it lacks the conserved disulfide bond between TM3 and EL2, an additional 540 models were also built forcing a putative disulfide between Cys168 in EL2 and Cys180 in TM5.

Model evaluation

Models were ranked on the basis of prioritizing active opioids (dextromethorphan) over the rest of the inactive PRESTO-Tango library that was used in the β-arrestin screen, as well as over property-matched decoy small molecules. We further insisted that in its docked pose, dextromethorphan engaged in an ion pair between its charged nitrogen and an anionic side chain in MRGPRX2. The entire PRESTO-Tango library was docked to analogous 4DJH ligand binding site in the modeled MRGPRX2 receptors for several rounds of iterative binding site refinement. In each round, the top-ranked models were examined for a binding pose that made hydrophobic and electrostatic interactions with the receptor, including the key positive-negative charge coordination. Residues within 6 Å of the dextromethorphan pose were then minimized around the docked ligand with PLOP [55]. The PRESTO-Tango library was then re-docked into this optimized binding site for each model. This refinement continued for several cycles until the top-ranked models all converged to the same dextromethorphan pose, with the top-scoring model chosen as the final one. Structural models (PDB files) of characteristic MRGPRX2-modelled complexes (with dextromethorphan and ZINC-9232) are shown in the Supplementary Data.

Virtual screens

We used DOCK 3.6 to screen the ZINC database (http://zinc15.docking.org/) (Results). The flexible ligand sampling algorithm in DOCK 3.6 superimposes atoms of the docked molecule onto binding site matching spheres, which represent favorable positions for individual ligand atoms. Forty-five matching spheres were used, with each respectively starting from the previous refinement round’s best pose of dextromethorphan. The degree of ligand sampling is determined by the bin size, bin size overlap, and distance tolerance, set at 0.4 Å, 0.1 Å and 1.5 Å, respectively, for both the matching spheres and the docked molecules. The complementarity of each ligand pose was scored as the sum of the receptor–ligand electrostatic and van der Waals’ interaction energies and corrected for context-dependent ligand desolvation using QNIFFT point-charge Poisson-Boltzmann electrostatics models[56]. Partial charges from the united-atom AMBER[57] force field were used for all receptor atoms; ligand charges and initial solvation energies were calculated using AMSOL [58,59] (http://comp.chem.umn.edu/amsol/). The best scoring conformation of each docked molecule was then subjected to 100 steps of rigid-body minimization.

Selection of potential ligands for testing

We docked the approximately 3.7 million commercially available molecules of the lead-like subset of the ZINC database to the final MRGPRX2 models. The full hit list was automatically filtered to remove molecules that possess high internal energy, non-physical conformations, which are not well modeled by our scoring function. The reported rankings reflect this filtering. From the top 0.13% (~5,000 molecules) of the docked ranking list, 20 compounds were chosen for testing, based on complementarity to the binding site and presence of predicted charge interactions with Glu1644.60 and Asp1845.36, mimicking those predicted for dextromethorphan.

Chemistry General Procedures

HPLC spectra for all compounds were acquired using an Agilent 1200 Series system with DAD detector. Chromatography was performed on a 2.1×150 mm Zorbax 300SB-C18 5 μm column with water containing 0.1% formic acid as solvent A and acetonitrile containing 0.1% formic acid as solvent B at a flow rate of 0.4 mL/min. The gradient program was as follows: 1% B (0 1 min), 1 99% B (1 4 min), and 99% B (4 8 min). High-resolution mass spectra (HRMS) data were acquired in positive ion mode using an Agilent G1969A API-TOF with an electrospray ionization (ESI) source. Nuclear Magnetic Resonance (NMR) spectra were acquired on eith a Bruker DRX-600 spectrometer (600 MHz 1H, 150 MHz 13C). Chemical shifts are reported in ppm (δ). HPLC was used to establish the purity of target compounds. All final compounds had > 95% purity using the HPLC methods described above.

7-Chloro-5-phenylpyrazolo[1,5-a]pyrimidine

The title compound was synthesized according to a known literature procedure.[60] To a solution of 3-aminopyrazole (2.0 g, 24 mmol) in acetic acid (15 mL) was added ethylbenzoylacetate (4.7 mL, 27 mmol). The resulting reaction mixture was refluxed for 3 h before being cooled to rt and concentrated. The solid residue was diluted with EtOAc and filtered to afford 5-phenylpyrazolo[1,5-a]pyrimidin-7(4H)-one (4.19 g, 83%) as white solid. This intermediate (1.0 g, 4.7 mmol) was dissolved in POCl3 (5.0 mL, 55 mmol). To the resulting solution was added pyridine (0.25 mL, 3.1 mmol) at rt. After being stirred for 3 days at rt, the reaction mixture was diluted with Et2O (50 mL) and filtered. The solid was rinsed with Et2O (2 x). The combined Et2O solution was cooled to 0 °C and treated with ice and water. The organic layer was washed with water (2 x) and brine (2 x). The organic layer was dried (Na2SO4) and concentrated to give the tittle compound (0.82 g, 76%) as pale yellow solid. 1H NMR (600 MHz, CDCl3) δ 8.23 (d, J = 2.3 Hz, 1H), 8.12 – 8.04 (m, 2H), 7.58 – 7.48 (m, 3H), 7.45 (s, 1H), 6.84 (d, J = 2.3 Hz, 1H). 13C NMR (151 MHz, CDCl3) δ 156.2, 149.8, 146.1, 139.2, 136.5, 131.0, 129.2 (2C), 127.5 (2C), 105.8, 98.8. HRMS (ESI-TOF) m/z: [M+H]+ calcd for C12H9ClN3, 230.0485; found: 230.0484.

(R)-N,N-Dimethyl-1-(5-phenylpyrazolo[1,5-a]pyrimidin-7-yl)pyrrolidin-3-amine ((R)-ZINC-3573)

To a solution of 7-chloro-5-phenylpyrazolo[1,5-a]pyrimidine (0.069 g, 0.3 mmol) in dioxane (2 mL) was added DIEA (0.19 mL, 0.6 mmol), followed by (R)-(+)-3-(dimethylamino)pyrrolidine (0.038 g, 0.33 mmol). The resulting solution was stirred for 16 h at rt before being concentrated and purified by silica gel column to provide the title compound (0.070 g, 76%). 1H NMR (600 MHz, CDCl3) δ 8.02 – 7.97 (m, 2H), 7.95 (d, J = 2.3 Hz, 1H), 7.51 – 7.40 (m, 3H), 6.48 (d, J = 2.3 Hz, 1H), 6.12 (s, 1H), 4.30 (t, J = 8.8 Hz, 1H), 4.23 (t, J = 9.8 Hz, 1H), 4.02 (q, J = 9.4 Hz, 1H), 3.83 (t, J = 9.5 Hz, 1H), 2.90 – 2.82 (m, 1H), 2.36 (s, 6H), 2.29 – 2.21 (m, 1H), 2.01 – 1.91 (m, 1H). 13C NMR (151 MHz, CDCl3) δ 157.5, 151.8, 147.9, 143.8, 139.2, 129.6, 128.7 (2C), 127.3 (2C), 94.9, 86.2, 65.2, 55.0, 50.1, 44.5 (2C), 30.1. HRMS (ESI-TOF) m/z: [M+H]+ calcd for C18H22N5, 308.1875; found: 308.1871.

(S)-N,N-Dimethyl-1-(5-phenylpyrazolo[1,5-a]pyrimidin-7-yl)pyrrolidin-3-amine ((S)-ZINC-3573)

The title compound (0.075 g, 81%) was prepared using the same synthetic procedure for the preparation of (R)-ZINC-3573 with (S)-(+)-3-(dimethylamino)pyrrolidine (0.038 g, 0.33 mmol). 1H NMR (600 MHz, CDCl3) δ 8.05 – 7.97 (m, 2H), 7.95 (d, J = 2.3 Hz, 1H), 7.52 – 7.39 (m, 3H), 6.48 (d, J = 2.3 Hz, 1H), 6.12 (s, 1H), 4.30 (t, J = 8.8 Hz, 1H), 4.23 (t, J = 9.8 Hz, 1H), 4.02 (q, J = 9.3 Hz, 1H), 3.83 (t, J = 9.5 Hz, 1H), 2.91 – 2.81 (m, 1H), 2.36 (s, 6H), 2.30 – 2.21 (m, 1H), 2.02 – 1.91 (m, 1H). 13C NMR (151 MHz, CDCl3) δ 157.5, 151.8, 147.9, 143.8, 139.2, 129.6, 128.7 (2C), 127.3 (2C), 94.9, 86.2, 65.2, 55.0, 50.1, 44.5 (2C), 30.1. HRMS (ESI-TOF) m/z: [M+H]+ calcd for C18H22N5, 308.1875; found: 308.1872.