G protein-coupled receptors: structure- and function-based drug discovery

As one of the most successful therapeutic target families, G protein-coupled receptors (GPCRs) have experienced a transformation from random ligand screening to knowledge-driven drug design. We are eye-witnessing tremendous progresses made recently in the understanding of their structure–function relationships that facilitated drug development at an unprecedented pace. This article intends to provide a comprehensive overview of this important field to a broader readership that shares some common interests in drug discovery.

Class A Class A GPCRs, the so called “rhodopsin-like family” consisting of 719 members, are divided into several subgroups: aminergic, peptide, protein, lipid, melatonin, nucleotide, steroid, alicarboxylic acid, sensory, and orphan. 17 They have a conventional transmembrane domain (TMD) that forms ligand-binding pocket and additional eight helices with a palmitoylated cysteine at the C terminal. 18 , 19 Given the wide range of their physiological functions, this class of receptors is the most targeted therapeutically among all other classes. By manually curating Drugs@FDA original New Drug Application (NDA) and Biologic License Application (BLA) database (data extracted from August 2017 to June 2020) and cross-referencing with Drugbank, 20 IUPHAR and ChemBL databases, we were able to find the approved drugs associated with this class. Over 500 GPCR drugs target class A and many of them act at >1 receptor: 75% are made against aminergic receptors and 10% for peptidic ligand receptors with indications ranging from analgesics, allergies, cardiovascular diseases, hypertension, pulmonary diseases, depression, migraine, glaucoma, Parkinson’s disease to schizophrenia, cancer-related fatigue, etc. Approximately 500 novel drug candidates are in clinical trials. Of them, 134 are for peptide-activated GPCRs, while small molecules still occupy the majority. It is noted that 6% of class A members are sensory and alicarboxylic acid receptors that have broad untapped therapeutic potentials (Table S1 ). Chemokine, prostanoid and melanocortin receptors constitute >8% clinical trial targets in this class. In the past 3 years, about 20 NDAs were approved targeting mostly peptide and aminergic receptors (Table 1 ). Siponimod and ozanimod provide alternatives to fingolimod (approved in 2010) for treating relapsing forms of multiple sclerosis by modulating sphingosine-1-phosphate receptor. Two radiolabeled ligands, gallium 68 dotatoc and lutetium 177 dotatate, have been approved for neuroendocrine tumor and pancreatic gastrointestinal cancer diagnosis, respectively. Pitolisant, a selective inverse agonist of histamine receptor, is used to treat narcolepsy-related daytime sleepiness, while lemborexant, an orexin receptor antagonist, is used for insomnia management. Gilteritinib (ASP2215) is a small molecule inhibitor of tyrosine kinase. However, it also antagonizes serotonin receptors without any reported pharmacological consequences. Revefenacin is a long-acting antagonist of muscarinic acetylcholine receptors (mAChRs) indicated for chronic obstructive pulmonary disease. Amisulpride, trialed for antiemetic and schizophrenia, was finally approved for antiemetic in 2020. This molecule is acting as an antagonist against dopamine and serotonin receptors. Fosnetupitant, a prodrug of netupitant, was approved for chemotherapy-induced nausea and vomiting. Cysteamine treats radiation sickness via modifying action of neuropeptide Y receptor. Cannabidiol is one the active constituents of the Cannabis plant and was trialed for schizophrenia, graft versus host disease, and anticonvulsant. It was eventually approved in 2018 for the treatment of severe forms of epilepsy—Lennox–Gastaut syndrome and Dravet syndrome. Meanwhile, fostamatinib, indicated for chronic immune thrombocytopenia, targets >300 receptors and enzymes, including adenosine receptor A3. Table 1 Newly approved drugs targeting class A GPCRs in the past 3 years Drug Brand name (manufacturer) Indication Target GPCR class FDA approval date Gilteritinib Xospata (Astellas) Relapsed or refractory acute myeloid leukemia Serotonin receptors A, aminergic, 5-hydroxytryptamine 11/28/2018 Lasmiditan Reyvow (Eli Lilly) Migraine HTR1F A, aminergic, 5-hydroxytryptamine 10/11/2019 Revefenacin Yupelri (Mylan Ireland) Chronic obstructive pulmonary disease CHRM1-CHRM5 A, aminergic, acethylcholine 11/09/2018 Lumateperone Caplyta (Intra-Cellular) Schizophrenia HTR2A, DRD1, DRD2 A, aminergic, dopamine, 5-hydroxytryptamine 12/20/2019 Amisulpride Barhemsys (Acacia Pharma) Surgery-induced nausea and vomiting prevention DRD2, DRD3, HTR7, HTR2A A, aminergic, dopamine, 5-hydroxytryptamine 02/26/2020 Pitolisant Wakix (Harmony) Narcolepsy excessive daytime sleepiness HRH3 A, aminergic, histamine 08/14/2019 Angiotensin II Giapreza (La Jolla Pharma) Septic vasoconstrictor for adults AGTR1 A, peptide, angiotensin 12/21/2017 Macimorelin Macrilen (Novo Nordisk) Diagnosis of adult growth hormone deficiency GHSR A, peptide, ghrelin 12/20/2017 Elagolix Orilissa (Abbvie) Endometriosis-associated moderate-to-severe pain GNRHR A, peptide, gonadotropin 07/23/2018 Cysteamine Procysbi (Horizon Pharma) Radiation sickness NPY2R A, peptide, neuropeptide Y 02/14/2020 Lemborexant Dayvigo (Eisai) Insomnia HCRTR1 A, peptide, orexin 12/20/2019 Gallium 68 dotatoc NA (UIHC-PET Imaging Center) Diagnostic agent for neuroendocrine tumors SSTR2 A, peptide, somatostatin 08/21/2019 Lutetium 177 dotatate Lutathera (AAA USA) Gastro

Class C (glutamate) contains 22 receptors, which are further divided into 5 subfamilies including 1 calcium-sensing receptor (CaSR), 2 gamma-aminobutyric acid (GABA) type B receptors (GABA B1 and GABA B2 ), 3 taste 1 receptors (TS1R1–3), 8 metabotropic glutamate receptors (mGluR1–8), and 8 orphan GPCRs. 48 The distinctive features of glutamate subfamily are their large ECD and obligated constitutive dimer for receptor activation. 49 The structural information of ECD indicates the roles of conserved venus fly trap (VFT) and cysteine-rich domain (CRD) on the ligand-binding site. Two conserved disulfide bonds between VFT domains stabilize the homodimers or heterodimers of class F GPCRs. 50 The cryo-EM structures of the first full-length mGluR5 51 and more recently the GABA B Rs further revealed their assembly mechanism and overall architecture. 52 – 55 To date, 16 drugs have been approved by the FDA targeting 8 class C GPCRs. As archetypal receptors, mGluRs mediate the stimulus of agonists such as glutamate and their malfunction are implicated in various diseases, including cancer, schizophrenia, depression, and movement disorders. Acamprosate, an antagonist of mGluR5, was launched in 2004 as an anti-neoplastic agent. 56 In fact, mGluRs have been vigorously pursued as therapeutic targets and there are 15 drug candidates undergoing clinical trials at present for pain, migraine, Parkinson’s disease, Fragile X syndrome, etc. Although allosteric modulators of class C have attracted significant development efforts involving 8 clinical trial stage compounds [2 positive (PAM) and 6 negative (NAM) allosteric modulators], the only success is cinacalcet, a small molecule PAM of CaSR approved in 2004 for hyperparathyroidism and calcimimetics. 57 Only one class F GPCR (smoothed receptor SMO) has been validated as a drug target whose small molecule antagonists were approved as anti-neoplastic agents. 58 Other 10 members of this class are all Frizzled receptors (FZD1–10), which mediate Wnt signaling and are essential for embryonic development and adult organisms. FZDs together with cognate Hedgehog and Wnt signal are associated with a variety of diseases such as cancer, fibrosis, and neurodegeneration. 59 They share a conserved CRD in the extracellular part and ECD structures of SMO and FZD2/4/5/7/8 were determined. 60 However, only SMO, FZD4, and FZD5 have TMD structures. 61 – 63 Lack of full-length structures and complexity in signaling pathways impeded drug discovery initiatives. 60 Linking of Wnt with extracellular CRD would activate downstream signaling, while the dimerization process and the interaction between CRD and TMD remain elusive. 64 It is known that the downstream effectors of Wnt signaling consist of β-catenin, planar cell polarity, and Ca 2+ pathways, whereas receptor activation involves in Wnt, Norrin, FZD, LDL receptor-related protein 5/6, and many other co-factors. 64 Key breakthrough is thus required to advance our knowledge of these receptors.

Studies of SARs are critical to the identification of drug-like molecules, especially when the crystal or cryo-EM structure of a drug target is not available. Given that many 3D GPCR structures have been solved in the past decade, most approved drugs were discovered without relevant structural information. Two examples are reviewed below to show the importance of SAR analysis. Orexin-1 and orexin-2 receptors (also known as hypocretin receptors, OX 1 R and OX 2 R) are class A GPCRs for which two endogenous peptide ligands were identified, orexin A and orexin B (also known as hypocretin 1 and hypocretin 2). The orexin signaling system plays a crucial role in regulating the sleep/wake cycle—both OX 1 R and OX 2 R are involved while the precise contribution of each has yet to be defined. Therefore, dual antagonists were developed as potential treatment for insomnia. 78 Suvorexant (belsomra), the first-in-class dual orexin receptor antagonist, was launched in 2014. 79 The second, lemborexant/E2006 (dayvigo) developed by Eisai, was approved by the FDA in 2020. 80 It started from hit compound 1 ( 6 , Fig. 3 ) with modest binding affinity to OX 2 R ( K i = 8.7 µM) and no affinity for OX 1 R. 81 The first round of SAR studies revealed that changing the ketone group to an amide led to a remarkable enhancement (~1000-fold) of binding affinity at both OX 1 R and OX 2 R (compound 2 ). Substitution of the aniline group with a 2-amino-5-cyano pyridine (compound 3 ) maintained OX 2 R affinity and reduced OX 1 R activity, but physicochemical properties were improved compared to compound 2 . 81 Further SAR studies focused on the modification of all three aromatic substitutions in compound 3 . 82 Changing the di -OMe-phenyl substituent to a pyrimidine group resulted in a significant loss of binding affinity, as shown with compound 4 , but an improved overall profile due to reduced lipophilicity and enhanced solubility. Then replacing the cyano group to a fluorine regained the binding affinity for both receptors (compound 5 ), and finally adding a second fluorine to the benzene group significantly improved OX 1 R affinity and led to lemborexant. 82 Clearly, slight structural modifications may cause significant change of compound activity, and SAR studies coupled with optimization of physicochemical properties are useful steps to obtain druggable candidates. Fig. 3 SAR studies that led to the discovery of the dual orexin receptor antagonist lemborexant CGRP is a 37-amino acid neuropeptide and its receptor is implicated in migraine. 83 The benzodiazapinone compound 7 was identified as a hit compound with modest CGRP receptor binding affinity ( K i = 4.8 µM, Fig. 4 ). 84 Replacing the right-hand spirohydantoin structure with piperidyldihydroquinazolinone, a privileged structure for CGRP receptor antagonists, 85 an affinity boost of 100-fold was gained. 84 Further optimization of the benzodiazepinone core resulted in the caprolactam compound 9 , which showed a K i of 25 nM. 86 Changing the piperidyldihydroquinazolinone moiety to a piperidylazabenzimidazolone led to compound 10 , with a binding affinity of 11 nM. 87 Then by changing the N-substituent on the caprolactam and adding di -fluro substitutions on the lower benzene ring delivered compound MK-0974 ( 11 , K i = 0.77 nM, Fig. 4 ), which entered clinical trials. Compound 12 (BMS-846372) shares the same piperidylazabenzimidazolone and the lower diflurobenzene substructures with 11 but differs from the latter with a carbamate core structure and a pyridine-fused-cyclopentane in replacement of the caprolactam. 88 Compound 11 displayed high binding affinity while suffered from poor physicochemical properties, such as low solubility. To improve this, a hydroxyl group was attached to the cycloheptane ring and it was discovered that the ( S )-isomer 13 was more potent than the ( R )-OH compound 14 . The –OH was finally replaced with an -NH 2 group, which led to the clinical compound rimegepant. 89 The latter was further developed for better safety and efficacy profiles and obtained regulatory approval by the FDA in 2020. Fig. 4 SAR studies that resulted in the discovery of CGRP antagonists The above examples demonstrate that, starting from a modest affinity hit compound, systematic SAR studies could successfully lead to very potent GPCR ligands that qualify as clinical candidates. Slight modifications of chemical structures sometimes cause remarkable changes of binding affinity or potency, which could not always be accurately predicted by conventional methods, such as docking. Therefore, SAR studies will continue to play a critical role in drug discovery.

The explosion of 3D GPCR structures and computational simulations has revealed the dynamic conformations between inactive, intermediate, and active states of GPCRs. The detailed structural information illustrated that cholesterol, ion, lipids, and water also participate in receptor activation. 99 , 119 The flexibility of receptor-binding pocket endows the complex pharmacological mechanisms of ligand recognition and signal transduction. Biased signaling, allosteric modulation, and polypharmacology are helping us better understand how GPCRs bind to numerous ligands and how they transmit diverse signals to elicit physiological functions. Polypharmacology Ligand binding to multiple targets leads to antagonism, additive, or synergism pharmacological responses that could be positive or negative based on the mechanism of action. The paradigm of one drug vs. multiple targets has outpaced the time and cost associated with the conventional therapy. 120 Polypharmacology thus emerges to study acceptable degree of specificity toward multiple targets, interconnected signaling pathways that result in clinical benefit or cross-reactivity that may cause adverse events. 121 , 122 T2DM, obesity, cancer, and Alzheimer’s disease are major indications for GPCR modulators. 4 These polygenic diseases are not completely treatable by a single agent, while desirable efficacies may be achieved for certain respiratory conditions, central nervous system (CNS) disorders, and cardiovascular diseases through modulators directed against β 2 AR, DRD2, and AGTR1, 123 respectively. It was shown that 5-hydroxytryptamine receptor 2 (5-HT 2 ) binds to selective inverse (ritanserin) and highly promiscuous (ergotamine) agonists but the interaction with ergotamine is broad. 121 This feature allows the development of pan serotonin receptor modulators to treat different diseases. 124 – 127 For instance, zolmitriptan as an anti-migraine drug is also used for hyperesthesia via binding to off-target site, 120 and lorcaserin (Belviq) is used to treat obesity while its therapeutic potential for depression, schizophrenia, and drug addiction is being investigated. 128 , 129 However, off-target activity, hallucinations, 130 and cardiac valvulopathy related to 5-HT 2A and 5-HT 2B modulation 129 should be carefully monitored. Atypical antipsychotics are mainly targeting both dopamine and serotonin receptors, usually as antagonist for DRD2 and antagonist or inverse agonist for 5-HT 2A. 131 Exemplified by clozapine 120 and aripiprazole, 132 haloperidol, amoxapine, and asenapine 4 display a diverse spectrum of receptor interaction. Additionally, carazolol, a member of aminergic division exerts its effects by interacting with multiple adrenergic receptors as inverse agonist or allosteric antagonist. 19 , 129 Istradefylline combined with L-DOPA/dopamine simultaneously target A 2A R, DRD1 and DRD2 in animal model of Parkinson’s disease. 131 Amitryptyline, a tricyclic compound targeting muscarinic and histamine H1 receptors, 133 is used to treat depression and non-selective muscarinic receptor antagonists are trialed for bladder dysfunction. 4 Lorazepam, indicated for anxiety due to interaction with GABA A R, is also an allosteric modulator of the proton-sensing GPCR (GPR68) 134 and has been repurposed to treat pancreatic cancer. 2 6’-Guanidinonaltrindole (6’-GNTI) is an agonist with higher selectivity for δ/κ-opioid receptor heterodimer but not homodimer. Importantly, 6’-GNTI is an analgesic that offers additional benefit. In cardiovascular diseases, β blockers decrease catecholamine-induced heart rate elevation via interaction with valsartan (AT1R-mediated signaling). 135 It is of note that mono-, dual-, and tri-agonists for the glucagon family of receptors (GLP-1R, GCGR, and GIPR) have been developed and trialed for weight loss and glucose control (Table 3 ). Successful outcome will determine whether unimolecular polypharmacology is a practical approach to translate safety and efficacy of multiple agents into a single molecule. 136

In recent years, studies on allosteric GPCR modulators have gained unprecedented momentum. 158 – 161 An allosteric modulator is a ligand binding to a position other than the orthosteric site but can modify responses of a receptor to stimulus. Allosteric modulators that enhance agonist-mediated response are called PAMs, while those attenuate the response are called NAMs. This phenomenon is very common such that the Allosteric Database 2019 (ASD, http://mdl.shsmu.edu.cn/ASD ) 162 records 37520 allosteric modulations on 118 GPCR members, covering all four classes. Allosteric modulation is advantageous in terms of (i) using highly druggable pockets. In some cases, it is easier to design ligands at an allosteric site than the orthosteric site, such as class B GPCRs with orthosteric pockets wide open. For example, both PAM 163 and NAMs 107 binding to the same position at the TMD of GLP-1R were reported; (ii) improving selectivity. The orthosteric site and cognate ligand are often highly conserved, making it hard to discover very selective orthosteric binders. Meanwhile, non-conserved allosteric sites would be a better choice evidenced by discovery of many subtype selective allosteric modulators of acetylcholine 102 , 164 and cannabinoid receptors 165 , 166 ; (iii) introducing signal bias. Allosteric modulators with biased signaling were developed for prostaglandin F2α receptor 167 and chemokine receptor CXCR4. 168 Albeit still as an emerging concept, allosteric modulators have exhibited a great potential with some compounds being marketed or in clinical trials. 160 However, developing allosteric modulators of GPCRs remains challenging—molecules recorded in the ASD largely concentrate on two subfamilies, the mGluRs (8 members, 17,115 modulations), and mAChRs (5 members, 7666 modulations), accounting for nearly 2/3 of the total number. Some individual receptors also contribute a significant proportion, such as CB1 (1948 modulations), GABA B (1286 modulations), and follicle-stimulating hormone receptor (1233 modulations). Excluding these “easy cases,” allosteric modulators are few in number. Furthermore, the structural diversity of the allosteric modulators is quite low, for many derivatives would be included soon after a parent compound is identified. The difficulty in developing allosteric modulators is partly due to the limitation of detecting allosteric behavior: Not every newly discovered active compound could be tested for its effect on binding affinity or EC 50 of an orthosteric agonist, therefore some allosteric modulators were not correctly identified. For instance, BPTU in P2RY1, the first GPCR NAM solved in complex structure (PDB code: 4XNV), 169 was not considered allosteric until the structure was obtained. To make things worse, NAMs may weaken the binding of an endogenous ligand thus behaving like a competitor, such as NDT9513727 in C5AR1 170 (PDB code: 5O9H). 171 The most effective way to identify the binding site of an allosteric modulator on a GPCR is solving the complex structure. Crystallography is an effective technique, while rapidly deployment of cryo-EM has started to deliver its promise (PDB codes: 6OIK 172 and 6U1N 173 ). To date, 17 GPCRs have reported structures in complex with allosteric modulators. Detailed analysis of complex structures before October 2018 was reported previously, 161 and here we focus on insights provided by newly published results. The most unusual allosteric-binding sites on GPCRs are at the lipidic interface embedded in cell membrane. Five different positions were identified by crystal structures (Fig. 9 ): UP12, UP34, LOW34, LOW345, and LOW67. Four of them were recently reviewed. 161 The LOW34 site was reported in 2019 for ORG27569 in CB1 (PDB code: 6KQI 166 ; Fig. 10a ). Fig. 9 Schematic diagram of allosteric sites at the lipidic surface identified by complex structures. The binding sites are manually labeled on the crystal structure of β 2 AR (PDB code: 6OBA 180 ). Solid line, allosteric site at front side; dashed line, allosteric site at back side. UP, upper part aka close to the extracellular end; LOW, lower part aka close to the cytoplasmic end; numbers, main interacting transmembrane helices Fig. 10 Binding sites of allosteric modulators in GPCRs reported after October 2018, in comparison with related ligands. a NAM ORG27569 in CB1 (PDB code: 6KQI 166 ) in comparison with cholesterol (PDB code: 5XRA 179 ); b NAM AS408 (PDB code: 6OBA 180 ) and PAM Cmpd-6FA (PDB code: 6N48 181 ) in β 2 AR, in comparison with NDT9513727 in C5AR1 (PDB code: 6C1Q 182 ) and PAM AP8 (PDB code: 5TZY 183 ); c NAM maraviroc in CCR5 (PDB code: 4MBS 187 ) in comparison with chemokine analog antagonist [5P7]CCL5 (PDB code: 5UIW 188 ) and HIV envelope glycoprotein gp120 (PDB code: 6MEO 189 ); d PAM TT-OAD2 in GLP-1R (PDB code: 6ORV 190 ) in comparison with GLP-1 (PDB code: 5VAI 191 ) ORG27569 attracted much attention for its distinctive function: increasing the binding of orthosteric agon

As two general types of computer-aided drug design techniques (Table 6 ), 205 SBDD and ligand-based drug design, exploit the structural information of protein targets and the knowledge of known ligands, respectively (Table 6 ). SBDD, on the basis of crystal/cryo-EM/NMR structures or homology models, first identifies key sites and important interactions responsible for target functions, then screens large virtual library/designed agents that disrupt or enhance such interactions to modulate relevant biological processes and/or signaling pathways by molecular docking, and finally discovers active leads with desired pharmacological properties. Clearly, the past decade is a golden age for SBDD on GPCR. With the year of 2011 (when LPC crystallization, 206 fusion proteins, 8 and other key techniques collaborated to launch the outbreak of GPCR structure determination including the landmark β 2 AR-G s ) for watershed, 207 SBDD of GPCR evolves two distinct stages: rhodopsin-based homology model and truly authentic structure of individual receptors. Boosted by the fast-increasing number of high-quality GPCR structures, improved accuracy of combinational computational approaches, and better understanding of activation mechanism and pharmacology, SBDD is developing rapidly with fruitful scientific reports and increasing GPCR-targeted drugs contributed by this approach. Considering the length of time required for a drug to be available on the market (10–15 years) and the chance of applying structural biology information to hit discovery and lead optimization in the first 2–3 years of a drug discovery program, it is probably too early to see the approval of GPCR-targeted drugs being developed with the aid of a structure, and such situation is likely to change as the tremendous efforts from both academia and industry start to bear the fruits of successes. The following is a brief account of recent advances in three main aspects of SBDD (chemical space, receptor dynamics, and pose evaluation) in the context of their application in GPCR pharmacology. Optimized virtual library Despite the vast chemical space (>10 63 drug-like molecules), only a nominal fraction has been explored by SBDD, where both the compound availability and insufficient diversity limited the number of screened ligands. To overcome these problems, ultra-large 208 – 210 and focused libraries 211 – 213 were employed. Lyu and colleagues 208 presented an excellent model of “bigger is better” in virtual drug screening. Based on the 130 well-characterized reactions, they generated 170 million make-on-demand compounds ( http://zinc15.docking.org/ ), the resulting library is remarkably diverse with >10.7 million scaffolds unavailable before. By docking 138 million molecules against DRD4, they discovered 81 new chemotypes (24% hit rate), 30 of them showed submicromolar activity, including a 180-pM subtype-selective and G i -biased DRD4 agonist. This ultra-large library docking study provides important information: (i) hit rate fell almost monotonically with docking score; (ii) hit rate vs. score curve of DRD4 predicted that 1 from every 873 compounds may have a minimum affinity of 1 μM; and (iii) human visual evaluation improved the selected compound with higher affinities, efficacies, and potencies but not the hit rate. A follow-up study on MT 1 by docking >150 million “lead-like” molecules 209 identified 15 active leads (39% hit rate) with potencies ranging from 470 pM to 6 μM. Alternatively, focused library 210 – 215 including scaffold library, natural products, dark chemical matter (i.e., chemicals that have never shown bioactivity tested in over 100 assays), and fragment- and lead-like libraries were introduced in virtual screening (VS) for dozens of receptors. Focused on compound library of traditional Chinese medicine (TCM), Liu et al. found that salvianolic acids A and C antagonized the activity of both P2RY1 and P2RY12 purinoceptors in the low µM range, while salvianolic acid B antagonized the P2RY12 purinoceptor. Remarkably, these three salvianolic acids are major active components of the broadly used hemorheologic TCM Danshen ( Salvia militorrhiza ). Taking NTSR1 as an example, Ranganathan et al. found that the fragment library tended to have higher hit rate than that of the lead-like library (19%) but the affinities were ∼100-fold weaker. Collectively, these results demonstrate the importance and advantages of ultra-large and tailored libraries in discovering potent GPCR modulators.

Correctly selecting and ranking poses of docked compounds in the ligand-binding pockets have been a challenge for SBDD, especially for GPCR that is embedded in the cell membrane with significant conformational adaptability. To address this problem, many physics-based scoring functions 223 – 226 integrated with some user-friendly computer programs (e.g., Dock, GOLD, AutoDock, Glide, and rDock) were routinely adopted in SBDD. Recently, precise computational approaches including free energy calculation methods like molecular mechanics/Poisson–Boltzmann surface area (MM/PB(GB)SA), 227 , 228 free energy perturbation (FEP), 229 quantum mechanical/MM calculations, 230 and fragment molecular orbital 231 , 232 have been employed with improved performance. Compared to the empirical scoring functions, MM/PB(GB)SA and FEP are physically more rigorous free-energy calculation methods with an increased computational cost and have been adopted in the studies of DNA–ligand, protein–ligand, and protein–protein interactions. 233 , 234 By introducing of the minimization-based MM/GBSA refining and rescoring of docked poses, Zhou et al. identified seven 5-HT 2B antagonists with novel chemical scaffolds and the most potent one has an IC 50 of 27.3 nM in a cellular assay. 227 Lenselink et al. used FEP to design A 2A R antagonists and identified a highly potent molecule with K i of 1.2 nM. 232 However, computational investigation across 20 class A crystal structures and 934 known ligands demonstrated that the correlations between predicted binding free energy by MM/PBSA and experimental data varied significantly. The observed variations exist between individual receptors and are highly system specific, 235 indicating that successful application of MM/PBSA may require additional efforts in validation of experimental data and optimization of simulation/calculation parameters. Alternatively, protein–ligand interaction fingerprints exacted from available crystal structures 225 , 236 fueled docking score with protein–ligand-binding mode information and resulted in improved VS virtual hit rates for β 2 AR (53%) and HRH1 (73%) with up to nM affinities and potencies. 225 , 236 Collectively, innovation in VS and pose evaluation, along with evolution of computational hardware, has significantly advanced SBDD and is expected to lift the discovery efficiency to a new height, since GPCRs have multiple downstream signaling pathways responsible for distinct functions or consequences. The high degree of sequence and pocket similarities between different subtypes demands for novel ligands with superior specificity and selectivity. In this regard, allosteric and biased modulators may offer additional pharmacological benefits.

Recently, Suomivuori et al. performed extensive MD simulations to identify two major signaling conformations that couple effectively to arrestin or G protein, respectively. They then designed ligands via minor chemical modification resulting in strong arrestin-biased or G q -biased signal transduction. 240 Meanwhile, McCorvy and colleagues discovered that specific ECL2–ligand contacts are associated with β-arrestin recruitment, whereas blockage of TM5 interaction reduces the G i/o signaling. An arrestin-biased DRD2 modulator was thus made exhibiting a calculated bias factor of 20 relative to quinpirole. 241 In addition, Mannel et al. conducted a VS of a tailored virtual library bearing 2,3-dichlorophenylpiperazine for DRD2 and found that 18 compounds occupy both orthosteric and allosteric sites, and 4 of them stimulated β-arrestin recruitment (EC 50 = 320 nM, E max = 16%) without detectable G protein signaling. 214

As GPCRs represent the most prominent family of therapeutic targets, 132 innumerable efforts have been made in both industry and academia to screen for novel ligands that can modulate the activity of a specified GPCR and serve as lead compounds for drug development. A diverse array of experimental technologies suitable for assaying protein–ligand interactions have been directly applied or tailored to GPCR-targeted ligand screening, and they can be classified into three main categories: binding-based, stability-based, and cell signaling-based assays (Table 7 ). Binding-based assays monitor the physical interactions between a GPCR protein typically in a purified recombinant form with individual test compounds. Cell signaling-based assays measure downstream effectors (e.g., cAMP, Ca 2+ , IP1/IP3) of specific intracellular signaling pathways known to be mediated by GPCR, which reflect the functional outcome of ligand binding to the receptor. Stability-based assays assess the variation of thermal stability for a purified protein when treated by test compounds. These different techniques vary in the ligand screening throughput and binding characteristics (Table 7 ). In the lead discovery stage, both binding- and signaling/activity-based assays are implemented in a parallel or sequential manner, as the multipronged use of complementary techniques would reduce the overall false-positive and false-negative rates. 246 Table 6 Comparison of the advantages and disadvantages of various computer-aided drug design approaches Approach Advantage Disadvantage Ligand-based drug design (LBDD) Quantitative structure–activity relationship 314 , 315 (QSAR) Understanding interactions between functional groups, convenient, does not require the structural information of a target Descriptor selection, false correlations, enough known ligands Pharmacophore modeling 316 , 317 Effective model, convenient, does not require the structural information of a target Known ligands, less novelty; missing conformations Structure-based drug design (SBDD) Virtual screening of ultra-large libraries 224 – 226 Novel scaffolds and chemotypes, higher chances of finding potent ligands Substantial computational resources, compound synthesis Virtual screening of focused libraries 227 – 229 Less demand for computational resources, compound easy to purchase, specific scaffolds or origins Reduced diversity, less novelty Ensemble docking 235 – 237 Considering protein flexibility, improved enrichment factors, rescue of false-negative ligands Increased computational burden, pose evaluation, false positive Energy-based pose evaluation 248 – 252 Improved scoring and ranking ability Substantial computational burden, method validation target dependency Here we summarize about 20 experimental screening technologies adapted to GPCR ligand discovery (Table 7 ) and highlight the most recent development of binding-based approaches. Notably, an update of assays assessing GPCR activation and signaling has been provided in a previous review 247 and will not be elaborated here. The structure-based VS is covered in the above section. Structural elucidation technologies (e.g., X-ray crystallography, single-particle cryo-EM, NMR, and HDX-MS) not suitable for high-throughput screening (HTS) are also excluded. DNA-encoded library (DEL) Impressive technological advances have been made for binding-based ligand screening over the past decade. Specifically, DEL has emerged as a powerful approach to drug discovery. 248 – 251 Created by split and pool synthesis, DEL usually contains hundreds of thousands to billions of distinct small molecule–DNA conjugates. A majority of DEL-based HTS reported to date involve incubation of an immobilized target protein with the library before the protein–ligand complexes are isolated. Encoding DNA tags associated with the immobilized target are then amplified and sequenced to assign relevant chemical structures. 250 , 251 Although DEL was predominantly applied to ligand screening against soluble proteins such as enzymes, successful adaptation of this technique to GPCRs was reported in a few cases. 252 – 255 Lefkowitz group reported the discovery of a NAM for β 2 AR by screening a DEL of 190 million. 252 This NAM not only has a unique chemotype but also exhibits low μM affinity and inhibits cAMP production as well as β-arrestin recruitment. Later on, the same team discovered the first small molecule PAM for β 2 AR through HTS of >500 million DEL compounds. 253 Both NAM and PAM demonstrated high selectivity. Of note is that the NAM was found using unliganded β 2 AR, whereas the PAM was unmasked via intentional application of β 2 AR with its orthosteric site occupied by an agonist thereby shifting the receptor to the active state. 252 , 253 These two studies elegantly demonstrated a proof-of-concept strategy for binding-based screening of allosteric modulators targeting different conformational states. 253 The power of DEL in the discovery of alloste

Recent scientific and technological advancements in GPCR biology have provided an enormous amount of information that will benefit our current and future efforts in rational drug design. Integration and refinement of massive data by artificial intelligence is a clear direction to guide both virtual and experimental screening of efficacious therapeutic agents with new scaffolds and of novel chemotypes for all classes of GPCRs. However, as described in this review, factors that influence GPCR drug discovery include, but not limited to, therapeutic target, chemical diversity, mechanism of signaling, ligand-binding site, mode of action, clinical indication, polypharmacology, etc. Future opportunities may arise from: (i) de-orphanization of orphan GPCRs to provide novel targets; (ii) new indication for drug intervention via discovery and/or repurposing efforts; (iii) development of lead compounds targeting classes B2 and F GPCRs to address unmet medical needs; and (iv) validation of polypharmacology may lead to improved drug therapies.