Structure‐Guided Design of G‐Protein‐Coupled Receptor Polypharmacology

Abstract Many diseases are polygenic and can only be treated efficiently with drugs that modulate multiple targets. However, rational design of compounds with multi‐target profiles is rarely pursued because it is considered too difficult, in particular if the drug must enter the central nervous system. Here, a structure‐based strategy to identify dual‐target ligands of G‐protein‐coupled receptors is presented. We use this approach to design compounds that both antagonize the A2A adenosine receptor and activate the D2 dopamine receptor, which have excellent potential as antiparkinson drugs. Atomic resolution models of the receptors guided generation of a chemical library with compounds designed to occupy orthosteric and secondary binding pockets in both targets. Structure‐based virtual screens identified ten compounds, of which three had affinity for both targets. One of these scaffolds was optimized to nanomolar dual‐target activity and showed the predicted pharmacodynamic effect in a rat model of Parkinsonism.


Introduction
Despite major efforts from the pharmaceutical industry, effective therapies for many central nervous system (CNS) diseases are still lacking. [1,2] Ac ommon property of CNS drugs (e.g.a ntipsychotics) is that these compounds interact with multiple targets and that this is essential for their therapeutic effect. [3,4] Thef act that multi-target profiles may be required for treatment of complex diseases contrasts with the philosophy of modern drug discovery,w hich focuses on ligands with selectivity for asingle target. However,drugs that modulate several nodes in anetwork of targets often provide synergistic therapeutic effects,f ewer side effects,a nd are more cost-effective compared to combination therapy based on single-target compounds. [5][6][7][8][9] Thepotential of polypharmacology has been recognized for more than ad ecade,b ut further progress is limited by difficulties to rationally design such compounds. [4,6,[8][9][10] We undertook the challenge to design ligand polypharmacology relevant for Parkinsonsd isease,an eurological disorder that has proven very difficult for traditional drug development. [1,11] In Parkinsonsd isease,p rogressive degeneration of dopaminergic neurons leads to motor dysfunction that initially is treated effectively with the dopamine precursor l-DOPA. However,long-term use of l-DOPAleads to agradual loss of drug efficacyand side effects such as motor fluctuations and dyskinesia. [1] An attractive alternative to targeting only the dopamine receptors is to consider the network of G-protein-coupled receptors (GPCRs) in the basal ganglia controlling movement, which includes the A 2A adenosine receptor. [12] Ac ompound with the ability to interact with both the A 2A adenosine receptor (A 2A AR) and the D 2 dopamine receptor (D 2 R) could delay progression of the disease and treat the symptoms.A ntagonism of the A 2A AR is not only symptomatic,b ut also neuroprotective in animal models of Parkinsonism. [13,14] This complements the strictly symptomatic benefits of D 2 agonists.M oreover, A 2A AR antagonists alleviate dyskinetic side effects of longterm l-DOPAt reatment. [15] Thed ual-target approach is supported by the fact that combined treatment with aD 2 agonist and A 2A antagonist has synergistic therapeutic effects. [16] Rational design of drugs targeting GPCRs is currently being accelerated by breakthroughs in structural biology, [17] providing opportunities to design drugs with novel properties. [18] In this study,wedeveloped astructure-based approach to design GPCR polypharmacology,w hich was employed to identify ac ompound that antagonizes the A 2A AR and activates the D 2 R. Structure-based virtual screening was used to predict dual-target compounds that were synthesized and evaluated experimentally.One scaffold with affinity for both targets was optimized, leading to potent dual-target ligands with functional activity tailored for treatment of Parkinsons disease.Our results suggest ageneral strategy to design dualtarget ligands of GPCRs.

Results and Discussion
Structure-Guided Design of aVirtual Library Theb inding sites of the A 2A AR and D 2 Rw ere first inspected to assess strategies to design dual-target ligands of these two GPCRs.Several crystal structures of the A 2A AR in the inactive conformation, which is the relevant state for development of antiparkinson drugs,w ere analyzed. [19] No experimental structure of the D 2 Rw as available and therefore ah omology model of the agonist-bound receptor was used. Ther ecently determined structures of the D 2 Rc onfirmed the accuracyofthe binding site model. [20,21] Structural alignment of the targets revealed large differences between the orthosteric sites that recognize the endogenous ligands adenosine and dopamine.This was evident both based on the overall shapes of the orthosteric sites and the lack of sequence identity in this region (Figure 1a). Only one out of 12 residues,t he GPCR family-conserved Trp 6.48 (Ballesteros-Weinstein residue numbering scheme is shown as superscripts [22] ), was the same in both sites and key residues for ligand recognition (A 2A AR/Asn253 6.55 and D 2 R/Asp114 3.32 ) were different. [19,23] Thed isparate nature of the targets was further supported by comparing known A 2A AR and D 2 R ligands from the ChEMBL bioactivity database. [24] Whereas the vast majority of D 2 Rligands are cations,compounds that bind to the A 2A AR are neutral (Supplementary Table 1). Identifying compounds with dual A 2A /D 2 activity hence appeared very challenging.
Many class AG PCRs have secondary binding pockets, which are formed by the extracellular entrance to the orthosteric site.S econdary pockets are targets of allosteric modulators and subtype selective ligands can be obtained by forming interactions in these less conserved regions. [25,26] We hypothesized that the secondary binding pockets could also be targeted to achieve polypharmacology and analysed these sites in both receptors.W ei dentified that Glu169 EL2 in as econdary binding pocket at the extracellular interface of the A 2A AR could potentially act as counter ion to ap ositive charge,w hich is one of the main characteristics of D 2 R ligands.T his was supported by crystal structures of the A 2A AR in complex with compounds that extended towards this region. [27,28] Based on the observation that the secondary binding pocket of A 2A AR and the orthosteric site of D 2 R could both accommodate cations,w es earched for as tarting point for ligand design among D 2 Ra gonists.A mong the few privileged structures that activate the D 2 R, N-methyl-2aminoindane was selected (compound 1). [29] Molecular docking calculations positioned N-methyl-2-aminoindane in the orthosteric site of the D 2 Rm odel and its charged nitrogen formed as alt bridge to Asp114 3.32 ,a ni nteraction that is conserved among biogenic amine GPCRs (Supplementary Figure 1). [23] Thep redicted binding mode suggested that the core scaffold could be expanded into as econdary pocket in the D 2 Rb yl inking ab uilding block to the amino moiety ( Figure 1b). Similarly,N -methyl-2-aminoindane docked to as econdary binding pocket in the A 2A AR crystal structure revealed that abuilding block fused to the core scaffold could access the orthosteric binding site (Figure 1b and Supplementary Figure 1). Structural analysis hence supported that Figure 1. Design of virtual chemical library.a )Sequence alignment of the orthosteric binding sites of the A 2A AR and D 2 R. Only one out of 12 residues (marked yellow) is the same in both pockets. b) Avirtual chemical library was constructed guided by the receptor binding sites. Compounds were designed to target both the orthosteric binding pocket (OBP) and asecondary binding pocket (SBP). Key binding site residues are shown as circles (negativelyc harged Glu169 and Asp114 in red, and Asn253 in yellow). c) An N-methyl-2-aminoindane (1)scaffold was used as the core fragmentofthe virtual library,w hich was connected to building blocks using two reactions (amide coupling and Buchwald-Hartwig amination).
dual-target ligands of the A 2A AR and D 2 Rcould be obtained by targeting the orthosteric and secondary pockets.
In the next step,avirtual chemical library with potential dual-target ligands was designed. Thel ibrary was generated by linking building blocks to the core scaffold with robust reactions that enabled rapid synthesis.T he receptor binding sites guided identification of two linkers that connected the N-methyl-2-aminoindane to building blocks.L inkers of optimal length to access both the orthosteric and secondary binding pockets with either at erminal carboxyl or amino group were selected, which allowed facile connection of building blocks by amide coupling or Buchwald-Hartwig amination ( Figure 1c). Importantly,t he resulting amide and amine moieties from these reactions would be positioned within hydrogen bond distance of Asn253 6.55 in the A 2A AR, which fulfilled ak ey interaction for antagonists in the orthosteric site.T he final virtual library was created by connecting the linker groups to building blocks from two sources.The first set of building blocks was based on utilizing known single-target A 2A AR ligands from the ChEMBL bioactivity database. [24] As direct connection of potent A 2A AR ligands led to compounds that lacked drug-like properties (e.g.due to high molecular weight), we developed an alternative approach. In silico retrosynthesis was used to deconstruct A 2A AR ligands into building blocks that could be used as starting material for synthesis.Reacting the resulting building blocks with the two linkers resulted in alibrary with 87 compounds.Asecond library was created by identifying commercial building blocks in the ZINC15 database [30] that could be fused to the same linkers,y ielding an additional 10 448 products ( Figure 1c). Thec ompounds in the virtual library had drug-like properties with median molecular weight and cLogD of 368 Da and 1.1, respectively (Supplementary Figure 2).

Structure-Based Virtual Screening for Dual-Target Ligands
Structure-based docking screens were used to identify the most promising compounds in the virtual library.E ach compound was first docked to ac rystal structure of the A 2A AR using the molecular docking program DOCK3.6. Thousands of conformations and orientations of the compounds were explored in the receptor binding site,f ollowed by ranking of these using the binding energy scores. [31] The500 top-ranked compounds from the library based on commercial building blocks and the compounds based on the ChEMBL database were inspected visually.T en compounds (2-11)were selected for synthesis based on favourable interactions with key residue Asn253 6.55 in the orthosteric binding site of the A 2A AR and positioning of the aminoindane moiety in the extracellular vestibule.D ocking to the D 2 Rs upported that the N-methyl-2-aminoindane moiety could maintain the interaction with Asp114 3.32 in the orthosteric binding site and that the linker allowed the building block to reach into the secondary binding pockets.Six of the selected compounds originated from the library based on commercial building blocks and the remaining four were derived from A 2A AR ligands using the retrosynthesis approach (Table 1a nd  Supplementary Table 2).

Compound Synthesis
As planned in the design of the virtual library,t he syntheses of compounds 2-11 was achieved by short routes of two to four steps (Scheme 1). In brief,c ompounds 2 and 5-7 were obtained by alkylation of N-methyl-2-aminoindane with preformed derivatives of the N-pentanamide linker and the second building block. Compound 8 was prepared by athreestep version of this route (Supplementary Scheme 1). Thefive compounds based on an N-butylamine linker (3-4 and 9-11) were obtained through routes in which the key steps consisted of alkylation of N-methyl-2-aminoindane by the linker, followed by attachment of the second building block by Nalkylation or Buchwald-Hartwig amination (Supplementary Scheme 2). Detailed synthesis procedures are available in the Supplementary Information.

Biological Assays for Dual-Target Ligands
Compounds 2-11 were evaluated in competition binding assays (Table 1a nd Supplementary Table 2). K i values were determined for the three compounds (2, 3 and 4)that showed significant radioligand displacement at 10 mm for both the   (Table 1). In the predicted binding modes,t he discovered ligands formed interactions with the key binding site residues Asn253 6.55 (A 2A AR) and Asp114 3.32 (D 2 R) and explored secondary pockets in the extracellular vestibule ( Figure 2). In the A 2A AR, the linker moieties of compounds 2 and 4 extended towards the extracellular loops and transmembrane helix (TM) 6/7, and the aminoindane moiety was in the vicinity of Glu169 EL2 .T he aminoindane moiety of compound 3 interacted primarily with as econdary pocket formed by TM2/7 and was predicted to hydrogen bond with Ser67 2.65 .All compounds formed asalt bridge to Asp114 3.32 in the D 2 Rand extended towards acommon secondary binding pocket formed by extracellular loop 2a nd TM2/7. Compound 2 was considered to be the most promising starting point for hit-to-lead optimization as it showed the best affinity for the A 2A AR and submicromolar affinity for the D 2 R ( Supplementary Figure 3). Functional experiments measuring G-protein-mediated changes in intracellular cAMP also confirmed that compound 2 activated the D 2 R (EC 50 = 9.7 mm,E max = 93 %). Retrospective analysis of how this virtual library compound was generated demonstrated the power of using in silico retrosynthesis to identify building blocks and obtain drug-like compounds.Whereas the A 2A AR ligand that it was based on was drug-like (molecular weight of 434 Da), [34] the retrosynthesis approach deconstructed it into as maller 2-amino-4-methyl-benzothiazole building block. This commercially available compound could be connected to the aminoindane scaffold in as ingle chemical reaction, yielding ad ual-target ligand with lower molecular weight (393 Da) than the original A 2A AR ligand. In contrast, direct linking of the A 2A AR ligand would have resulted in ac ompound with am olecular weight of 635 Da, less favorable physicochemical properties,a nd am ore elaborate synthetic route (Supplementary Figure 4).
To assess if binding to the secondary pockets improved affinity,c ompounds representing the moieties that anchored the ligands in the orthosteric sites (12 and 13)were tested in binding assays.A sa nticipated, the benzothiazole-based scaffold 12 showed binding to the A 2A AR (K i = 8.1 mm), but not to the D 2 R. Conversely,t he aminoindane-based scaffold 13 was aD 2 Rligand (K i = 5.0 mm), but showed no activity for the A 2A AR. Theinteractions with the secondary pocket hence improved binding to the A 2A AR and D 2 Rb y7 -a nd 6-fold, respectively.W ea lso noted that benzothiazole is as ubstructure of the compound Tozadenant, which was evaluated as an antiparkinson drug. [35] Ar ecent crystal structure of the A 2A AR in complex with this drug candidate (PDB code: 5OLO 33 ), which was released after the discovery of compound 2,c onfirmed our predicted binding mode of the scaffold (Supplementary Figure 5). Molecular dynamics (MD) simulation refinement of the predicted complex with compound 2 also showed that the interaction with key residue Asn253 6.55 and as trong salt bridge between the positively charged aminoindane moiety and Glu169 EL2 were formed (Supplementary Figure 6).

Research Articles Optimization of Dual-Target Activity
Structure-guided design of analogs to compound 2 was performed to further improve affinity and optimize functional potency. By focusing on commercially available building blocks,a nalogs could rapidly be synthesized to obtain structure-activity relationships.D etailed synthetic procedures are described in the Supplementary Information.
As eries of analogs with modifications on the benzothiazole moiety were first synthesized (14-20,T able 2a nd Supplementary Table 3). Whereas substituents in the 5position led to loss of activity (14-17,S upplementary Table 3), compounds with small substituents in the 4-position of the benzothiazole moiety had improved affinities for the A 2A AR (18-20,T able 2). Compound 20 showed an affinity of 190 nm at the A 2A AR, corresponding to a > 6-fold improvement over 2,a nd a K i value of 340 nm at the D 2 R (Supplementary Figure 3). Although compounds 18-20 had submicromolar affinities for both targets,f unctional assays revealed that they were weak D 2 Ragonists.The aminoindane group was modified to improve potency and efficacy at the D 2 R. Replacement of the N-methyl with ethyl or propyl substituents (21)(22)(23)(24)(25)(26)(27)m aintained D 2 Ra ffinity and increased functional potency. Theb enzothiazole moiety was then further optimized and we identified substituents at the 7position that resulted in high A 2A AR affinity (28)(29)(30). Compound 30 had an affinity of 160 nm for the A 2A AR with nanomolar inhibitor potencyi nf unctional assays (K b = 720 nm)a nd was also ap otent D 2 Ra gonist (K i = 370 nm, EC 50 = 180 nm with E max = 77 %) (Supplementary Figure 3 and Supplementary Figure 7). Models of compound 30 bound to the A 2A AR and D 2 Rshowed that hydrogen bonds with the key residues Asn253 6.55 and Asp114 3.32 ,aswell as interactions with the secondary binding site,were maintained (Figure 3a). Physicochemical properties relevant for CNS drugs were calculated for the dual-target ligands and compared to reference A 2A AR and D 2 Rc ompounds (Supplementary  Table 4). Notably,t he most potent dual-target compounds have properties similar to some approved drugs and are within the recommended property ranges for blood-brain barrier permeability (molecular weight < 500, polar surface area < 90 2 ,n umber of hydrogen bond donors < 3, and cLogD = 2-4). [36] Modification of the linker was explored to test if it could be shortened and thereby reduce the size and lipophilicity of the scaffold. N-butanamide and N-propanamide linkers generally reduced affinity,b ut several compounds in the series maintained submicromolar K i values for at least one of the targets and low micromolar for the other (31-41, Table 2  and Supplementary Table 5). Forexample,compound 37 had affinities of 300 and 1300 nm for the A 2A AR and D 2 R, respectively,a nd had lower molecular weight and cLogD values (Table 2a nd Figure 3b). Several of the compounds were D 2 Ra gonists,b ut functional potencyw as reduced to between 1.2 and 31 mm.T hese results supported that the optimal linker length had been selected in the library design and that physicochemical properties could be further optimized with maintained dual-target activity.  [32] ) and D 2 R( homology model) are shown as blue and grey cartoons, respectively.Key binding site residues and ligands are shown as sticks.  T he A 2A AR (PDB code:5OLO [33] ,MD-refined binding modes) and D 2 R(homologym odel) are shown as blue and grey cartoons, respectively.Key binding site residues and the ligands are shown as sticks.

Selectivity of Dual-Target Compounds
Ap otential concern in development of ligands with polypharmacological profiles is that such compounds may be promiscuous rather than interacting specifically with the targets. [6] To assess the selectivity properties of the dual-target ligands,wetested six compounds (18, 20, 22, 26, 27,and 30)in binding assays at the A 1 adenosine receptor (A 1 AR), D 3 dopamine receptor (D 3 R), D 4 dopamine receptor (D 4 R), and H 1 histamine receptor (H 1 R) (Supplementary Table 6). Thec ompounds generally showed high affinity for the D 3 R, which was expected considering the high sequence similarity to the D 2 R. Compared to the two targets,c omparable or substantially weaker ligand binding affinities were observed for the A 1 AR, D 4 Rand H 1 R. Notably,compound 30 showed high affinity for the A 2A AR and D 2 R, but displayed weak or no significant binding to the A 1 AR, D 4 Ra nd H 1 R( K i > 10 mm).

Blood-Brain Barrier Penetration
Thep ermeability of compound 30 across aC aco-2 cell monolayer was first determined to assess its likelihood to cross the blood-brain barrier. As 30 displayed medium to high permeability (P app AB:3 .8 10 À6 cm s À1 )a nd am oderate efflux ratio (ER:8 .0), its brain exposure was determined in rats.I ntraperitoneal (IP) administration of 30 (24 mg kg À1 ) resulted in ab rain-to-plasma ratio that peaked at 0.79 AE 0.13 after 30 min (Figure 4a). Forc omparison, CNS active drugs such as morphine and risperidone have brain-to-plasma ratios of 0.69 and 0.95, respectively. [37] At 30 min compound 30 reached at otal concentration of close to 3 mm in the brain, that is,several fold higher than its affinity at the A 2A AR and D 2 R( Supplementary Figure 8). Taken together these results demonstrated that 30 achieves as ufficiently high brain exposure to justify the evaluation of its activity in an animal model for Parkinsonsdisease.

Evaluation in Rodent Model of Parkinsonism
To assess if the designed dual-target ligands could elicit antiparkinsonian effects,w ea dministrated compound 30 intraperitoneally to unilaterally 6-OHDAl esioned rats and performed ar otation test. 6-OHDAl esioned rats display an innate tendencyt or otate ipsilaterally due to the unequal dopaminergic innervation. D 2 Ragonists induce contralateral rotational behaviour in unilaterally dopamine depleted rats, and this was confirmed by administration of apomorphine (Supplementary Figure 9). [38] Rats treated with the dualtarget compound 30 (24 mg kg À1 )h ad as ignificantly higher number of contralateral rotations compared to the control group (p < 0.01, Figure 4b). Using an independent set of rats, we assessed if the effect was mediated through the dopamine receptors by performing the experiments in the presence of the D 2 antagonist raclopride (Figure 4c). Administration of compound 30 again resulted in as ignificant increase of contralateral rotations (p < 0.05). This effect was reversed if compound 30 was administered in combination with the D 2 antagonist raclopride,c onfirming the involvement of dopamine receptors (Figure 4c). These experiments showed that dual-target compound 30 is taken up from the intraperitoneal space and is able to cross the blood-brain barrier to reach its targets in the CNS,w here it elicits the desired antiparkinsonian effect.

Strategies to Design Drugs with Polypharmacological Profiles
Polypharmacology poses am ajor challenge for the traditional drug discovery approach, and the optimal compound design strategy depends on the nature of the targets.I n favorable cases,t he targets of interest recognize similar compounds and multi-target activity can be found among previously identified single-target ligands or by identifying common pharmacophoric features. [39] Fore xample,B esnard et al. combined aligand-based method with machine learning to optimize an acetylcholineesterase inhibitor scaffold for activity at both D 2 Ra nd acetylcholinesterase. [40] Similarly, Keiser et al. used chemical similarity methods to identify polypharmacology of approved drugs. [41] However,l igandbased approaches are generally limited to targets with overlapping pharmacophore features.F or disparate targets, one possible strategy is to design bivalent compounds,which consist of two single-target ligands connected by al inker. Although these are useful as chemical probes to study GPCR dimers, [25,42,43] this approach often leads to compounds with high molecular weight that are unlikely to possess drug-like properties and cross the blood-brain barrier. Structure-based modelling provides an additional route to design drugs with multi-target profiles.C rystal structures of GPCRs have revealed druggable secondary binding pockets,w hich can be targeted either by allosteric or bitopic compounds.B itopic (also termed dualsteric) ligands that extend into secondary binding pockets have primarily been used to attain subtype selectivity and biased signalling. [25,26,44] Here,wedemonstrate that such compounds can also be used to design polypharmacology.Bitopic compounds hence provide ageneral approach to tune selectivity,e fficacy,a nd multi-target activity to achieve maximal therapeutic effect and minimal side effects. As the presence of secondary pockets appears to be ageneral property of class AG PCRs (Supplementary Figure 10), the same strategy can be applied to many targets in this large family of therapeutically relevant proteins.

Conclusion
Although the importance of polypharmacology in treatment of complex diseases is well-established, [3-6, 8-10, 45] the pharmaceutical industry has stayed clear of multi-target drugs because it has been considered to be too challenging.H ere, we present the first example of structure-guided design of dual-target activity at disparate GPCR targets.W eshow that it is possible to rapidly develop compounds with complex polypharmacology that display in vivo activity even for targets located in the CNS.W ee xpect that the rapidly increasing access to structural and bioactivity data for GPCRs [46] will make it possible to apply the same approach to design drugs with interaction profiles tailored for treatment of other complex diseases.