Advances in G Protein-Coupled Receptor High-throughput Screening

Abstract

G protein-coupled receptors (GPCRs) detect compounds on the cell surface and are the starting point of a number of medically relevant signaling cascades. Indeed, over 30% of FDA approved drugs target GPCRs, making them a primary target for drug discovery. Computational and experimental high-throughput screening (HTS) approaches of clinically relevant GPCRs are a first-line drug discovery effort in biomedical research. In this opinion, we review recent advances in GPCR HTS. We focus primarily on cell-based assays, and highlight recent advances in in vitro assays using purified receptors, and computational approaches for GPCR HTS. To date, GPCR HTS has led to the identification of new and repurposing of existing drugs, and the deorphanization of GPCRs with unknown ligands. As automation equipment becomes more common, GPCR HTS will move beyond a drug discovery tool to a key technology to probe basic biological processes that will have an outsized impact on personalized medicine.

Introduction

G protein-coupled receptors (GPCRs) are the target of more than 30% of FDA approved drugs, placing them as the premier protein family for approved pharmaceuticals [1]. Only ~130 of the 360 non-sensory GPCRs are targeted by drugs [2], leaving > 200 GPCRs as potential therapeutic targets. Computational and experimental high-throughput screening (HTS) of clinically relevant GPCRs with compounds and peptides is a first-line drug discovery tool [3]. One of the earliest examples of GPCR HTS for drug discovery is the screening of > 500,000 compounds for agonists of chemokine receptor 5 using a radioligand binding assay to arrive at maravirnoc, now an antiviral HIV treatment [4].

Today, experimental GPCR HTS approaches primarily rely on cells to report on changes in the intracellular concentration of secondary messengers upon GPCR ligand binding. These changes result from the activation of one or more G_α subunit families upon GPCR stimulation. Widely used assays detect GPCR-dependent activation of G_αs, G_αi and G_αq. Activation of G_αs upregulates cAMP production, activation of G_αi inhibits cAMP production and activation of G_αq ultimately results in calcium (Ca²⁺) accumulation. The ability of most GPCRs to recruit β-arrestin to initiate receptor desensitization has also been leveraged to report on GPCR ligand binding. For GPCRs with a known G_α coupling subunit family, the correct cell-based assay can be chosen. However, for the ~100 non-sensory GPCRs that lack known ligands and for which the G_α subunits they couple to are unknown—i.e. orphan GPCRs [5]—multiple cell-based assays that activate different signaling pathways must be performed for ligand identification. Finally, GPCR HTS is accessible to laboratories beyond those with deep cell-based assay expertise. Radioligand binding assays as well as functional assays for G_αi, G_αs and G_αq-coupled receptors for over 175 GPCR targets can be purchased for screening or the screening can itself be outsourced (e.g. Eurofins DiscoverX, ThermoFisher, Promega).

GPCR cell-based assays require heterologous GPCR expression in an engineered cell to report on ligand binding, which has its own set of technical challenges. First, functional expression of human GPCRs outside their endogenous tissue sometimes requires addition of expression tags, which may affect ligand binding. Next, signal transduction from the GPCR to the synthetic machinery in the engineered cell may be inefficient; for example, the engineered cell may express the correct G_α subunit but the incorrect isoform. Third, most GPCR cell-based assays take place in mammalian cells that express endogenous GPCRs. For example, HEK293, a widely used cell line for GPCR HTS, expresses 75 endogenous GPCRs [6]. Given that all GPCRs couple via the same four G_α protein families, leading to concentration changes of only a handful of secondary messengers, the potential for cross talk is significant. That is, the measured reporter output may be due to activation of an endogenous GPCR rather than the desired heterologous one. Finally, as is the case with all cell-based assays, compounds may result in GPCR independent reporter output changes leading to false negatives and false positives. For example, compounds may inhibit the reporter activity or emit fluorescence at the same wavelength as the reporter [7]. Therefore, GPCR HTS should be followed by lower-throughput biochemical assays (e.g. pull down assay) to confirm hits before validation in the GPCR endogenous tissue.

Here, we review advances in GPCR HTS over the last two years. We primarily focus on cell-based assays and we classify them by the intracellular proteins to which they couple (Figure 1). We also highlight key advances in in vitro GPCR HTS as well as computational approaches that reduce the number of compounds to be tested experimentally. Taken together, it is an exciting time for GPCR drug discovery and deorphanization.

Figure 1. High-throughput GPCR cell-based assays.

A. GPCR activation of G_αs upregulates adenylate cyclase (AC) resulting in increased cAMP levels. cAMP activates protein kinase A, which phosphorylates the transcription factor CREB ultimately resulting in reporter gene expression. B. GPCR activation of G_αi downregulates AC resulting in a decrease of cAMP levels. C. GPCR activation of G_αq upregulates phospholipase C (PLC), which breaks down PIP₂ (phosphatidylinositol biphosphate) into IP₃ (inositol triphosphate) and DAG (diacylglycerol). IP₃ goes on to activate Ca²⁺ channels in the endoplasmic reticulum increasing the Ca²⁺ concentration in the cytosol, which is detected via chemical dyes or protein sensors. D. In β-arrestin recruitment-based assays, GPCR activation recruits β-arrestin linked to a protease (yellow) that cleaves the transcription factor (TF) linked to the GPCR via a protease cleavable linker (yellow). Transcription factor release ultimately results in reporter gene expression. E. In Saccharomyces cerevisiae, human GPCRs can couple to yeast or human/yeast G_α chimeras and via G_β activate the yeast mating pathway ultimately leading to expression of a reporter gene. Gold circles: agonist.

1. Cell-based GPCR high-throughput assays

1.1. Secondary messenger- based assays: cyclic adenosine monophosphate (cAMP)

Upon GPCR ligand binding, G_αs upregulates adenylate cyclase, which in turn increases the levels of cAMP. Activated by cAMP, protein kinase A phosphorylates the CREB transcription factor, leading to transcription activation of a reporter gene downstream of a minimal promoter containing CRE binding sites (Figure 1A). Using β-galactosidase as the reporter, Chen et al. applied this assay to screen the β2-adrenergic receptor (β2AR) against ~7,000 chemicals in neuroepithelial cells and confirmed that β2AR is activated by higenamine, a treatment for heart failure [8]. Importantly, depending on the ligand, some GPCRs couple to different G_α subunit families leading to biased signaling [9]. The β2AR is known to activate G_αs, G_αi or β-arrestin [10], thus a β2AR ligand identified via cAMP accumulation preferentially activates G_αs signaling. Additional assays that couple β2AR via G_αi or β-arrestin should be used to identify β2AR ligands that preferentially signal via those pathways. Alternative cAMP reporter systems include an immunoassay to directly measure cAMP levels (Eurofins) and an assay that measures the decrease in ATP levels as protein kinase A consumes ATP to phosphorylate transcription factors (Promega).

Having a single reporter output (e.g. luminescence, fluorescence) requires spatial separation of cells expressing different GPCRs during compound screening. Testing a single GPCR against a single chemical allows identification of both high and low affinity binders. Nevertheless, the screening throughput is limited to that achievable using 96- or 384-well plates in a couple of days. By linking activation of each GPCR to a different RNA barcode reporter, cells expressing different GPCRs can be pooled together and tested against a single chemical en masse, thus significantly increasing the throughput of the screen (Figure 2). This strategy was used to deorphanize G protein-coupled olfactory receptors (ORs), which couple to G_α olfactory (G_αolf) leading to cAMP accumulation. Specifically, Jones et al. screened ~39 murine ORs against 181 odorants and successfully deorphanized 15 ORs [11●●]. Only if a specific OR was activated, the assigned barcode was transcribed and quantified via RNA sequencing. Of note, the many GPCRs to one compound screening strategy effectively has all GPCRs competing for the same ligand, leading to the isolation of the highest affinity binders. Although for GPCR deorphanization this may be a desired outcome, for pharmaceutical applications low affinity binders may be a better starting scaffold in terms of pharmacological profile or more amenable for a medicinal chemistry campaign.

Figure 2. Cell-based screening approaches.

A. One-to-one GPCR ligand screening. Cell lines expressing different GPCRs are spatially separated in different wells and their ligand binding ability assessed independently. This approach is used when GPCRs are linked to the same reporter, e.g. fluorescence or luminescence reporters. B. Many-to-one GPCR ligand screening. Each GPCR is linked to a different DNA barcode. Cell lines expressing different GPCRs are pooled and en masse screened against a single ligand. Activation of each GPCR leads to different transcription levels of the barcode that can be quantified using RNA sequencing.

GPCRs can bind both peptides and small molecules. Typically, peptide libraries are more expensive to synthesize than small molecule libraries, and thus likely less often studied. Instead of relying on a large library of purified peptides, Yaginuma et al. engineered Saccharomyces cerevisiae to secrete mutants of the known glucagon-like peptide-1 receptor (hGLP1R) agonist, Ex4. The engineered yeast were encapsulated together with a mammalian reporter cell line expressing hGLP1R that secreted the β-galactosidase reporter. Droplets containing activated hGLP1R were highly fluorescent, making them easily identified and isolated via fluorescence microscopy. This assay identified one novel peptide for hGLP1R out of the 400 peptides screened [12●●]. With improved throughput, specifically in the droplet screening, ligand screening could be increased to explore the entire mutational space for a given peptide.

Opposite of G_αs, the activation of G_αi results in a decrease of cAMP. To measure this, an adenylate cyclase activator, such as forskolin, is first added to increase cAMP levels before GPCR-dependent G_αi activation, allowing an observable decline in cAMP-dependent signal (Figure 1B). Using this strategy, Myers et al. screened the apelin receptor, a drug target for heart failure, against a Bristol Myers Squibb proprietary library and identified aryl hydroxy pyrimidinones as agonists for this receptor [13].

1.2. Calcium-based assays

Activation of G_αq ultimately results in an increase in intracellular calcium levels (Figure 1C). Calcium dyes, such as Fluo-4, can be used to report on calcium fluctuation. Calcium dye fluorescence is often read using a Fluorescent Imaging Plate Reader (FLIPR^®, Molecular Devices) assay. Smith et al. used a FLIPR^® assay to screen a 360,000 member small molecule library to identify 25 agonists, four positive allosteric modulators and 41 antagonists of the muscarinic acetylcholine receptor M₄ [14]. The main drawbacks of dye-based assays are their cost ($25/96 well plate) and additional assay time (~1 hour), which limit their HTS applications. Genetically encoded calcium biosensors (GECIs) are a less expensive alternative to calcium dyes. One of the most widely used GECIs is GCaMP, a protein fusion of calmodulin, green fluorescent protein (GFP) and the calmodulin substrate M13 peptide. Binding of Ca²⁺ displaces M13 resulting in GFP fluorescence [15].

In 2018, Mella et al. introduced genetically encoded sensors for Ca²⁺ and cAMP that can be used simultaneously, thus enabling GPCR activation of multiple G_α subunit families in the same experiment [16●]. In these sensors, a plasma membrane localized fluorescent protein is linked to a peptide (calmodulin for Ca²⁺ and thyroid A-kinase anchoring protein for cAMP) that undergoes a conformational change upon chemical binding resulting in increased fluorescence. Mella et al. screened the endothelin B receptor against 1,200 chemicals, and identified two new ligands that only signaled via G_αq (Ca²⁺ accumulation) and one ligand that signals via both G_αs and G_αq (cAMP and Ca²⁺ accumulation).

Before raising calcium levels, G_αq activation increases in DAG and IP₃ concentrations, with IP₃ ultimately metabolizing to inositol monophosphate (IP₁). Although an assay that detects IP₁ [17], and several fluorescent-based DAG sensors exist [18], they have not been used for GPCR HTS.

1.3. β-arrestin recruitment-based assays

Activated GPCRs eventually undergo desensitization via recruitment of β-arrestin, a process that has been exploited to report on GPCR activation. A GPCR fused to a transcription factor via a protease cleavage site recruits β-arrestin linked to a protease, resulting in cleavage of the transcription factor and expression of the reporter gene [19] (Figure 1D). In 2015, this assay was expanded to > 300 GPCRs by appending the C-terminus of the vasopressin receptor to promote efficient β-arrestin recruitment [20]. Fan et al. used this expanded assay to screen 320 GPCRs in parallel against three chemicals and identified selective agonists for the D₂ dopamine receptor, a key neuropsychiatric drug target [21]. Li et al. improved the reliability of the β-arrestin recruitment assay by integrating the reporter gene and β-arrestin into the genome [22]. This assay was used to screen the D₂ dopamine receptor against 1,000 natural products and identified wilfortrine as an agonist for the first time [22].

A new application of GPCR HTS is the screening of human microbial metabolomes to identify microbial metabolites that activate GPCRs. Colosimo et al. screened the fractionated metabolomes of seven bacteria that make up the simplified human gut microbiome against 241 GPCRs and identified eight GPCRs that showed reproducible β-arrestin recruitment and passed their hit threshold level. A following bioassay-guided fractionation sought to determine the activating compound for hydroxycarboxylic acid, neurotransmitter and histamine receptors. Testing of the fractions of interest with the GPCR-based assay resulted in the identification of nine microbial compounds activating 15 GPCRs, including five orphan GPCRs [23●]. In a different study, Chen et al. focused on 144 bacteria isolated from 11 inflammatory bowel disease patients, and screened the monoculture supernatants against 314 human GPCRs. This screen revealed a high level of interaction between commensal bacteria and the aminergic receptors in the human gut. Specifically, follow up studies in germ-free mice showed that histamine and phenethylamine produced by Morganella morganii play important roles in colonic motility and efficacy of anti-depressants, respectively. In addition, this study also found potential hits for 17 orphan GPCRs, with further investigation identifying L-phenylalanine as an agonist for GPR56/AGRG1 and GPR97/AGRG3, indicating a potential connection between microbiota metabolites and digestion or satiety [24●●].

Reporters used in β-arrestin recruitment-based assays include luciferase, GFP, split luciferase, β-lactamase and RNA barcodes. A luciferase signal can be read rapidly in plate readers; however, it requires mammalian cell lysis, thus limiting the reporter to end-point detection [25]. In comparison, GFP has a long maturation time and requires a high-throughput cell imaging system for signal acquisition when using mammalian cells. Split luciferase enables continuous monitoring of GPCR activation [26]; however, it has not yet been used for GPCR HTS applications. In a split TEV β-arrestin recruitment-based assay, Galinski et al. used RNA barcodes to assess 19 aminergic receptors against five known ligands [27].

1.4. Non-mammalian cell-based assays

To our knowledge, the only application of GPCR HTS outside of mammalian cells has been in the yeast S. cerevisiae. Indeed, S. cerevisiae is an ideal host to screen human GPCRs. In yeast, human GPCRs can couple to the yeast mating pathway directly via an endogenous G_αs, GPA1, or a GPA1/human G_α chimera [28] leading to activation of growth, fluorescent or luminescent reporters (Figure 1E). The ability to use growth reporters in yeast enables the screening of millions of GPCR variants against one chemical en masse, which has been exploited to engineer GPCRs exclusively activated by inert drugs (DREADDs), a widely used chemogenetic tool in neuroscience [29]. Yeast show advantage over mammalian cells due to the faster growth rate (3 hrs vs 12 hrs for mammalian cells) and its robustness, as it can be stored at 4°C for a month without the need for passage. Certainly, the use of yeast for human GPCR HTS also has some drawbacks, such as the lack of proper GPCR glycosylation or phosphorylation, which may result in the non-functional expression of GPCRs. As with any GPCR HTS, the ligand must be validated in the GPCR endogenous tissue for activity. Taken together, yeast has the potential to provide a more robust and faster platform for human GPCR screening.

In 2019, Yasi et al. screened the human serotonin 4 receptor (5-HTR₄), the pharmacological target for irritable bowel syndrome with constipation, against 1,206 compounds and ultimately identified three 5-HTR₄ agonists leading to an increase in human colon cell motility [30●●]. The use of a luciferase reporter was pivotal as a GFP reporter lead only to a three-fold increase in signal after activation and a limit of detection in the μM range [31]. Swapping GFP for luciferase improved the assay dynamic range to more than 30-fold and lowered the limit of detection to the low nM range. Further, the use of luciferase reduced the total assay time from four to 2.5 hours, and enabled a screening throughput of one compound per second. Of note, 5-HTR₄ coupled to GPA1; however, this may not be true of other GPCRs and thus yeast/human G_α chimeras may need to be used.

Olfactory receptors have also been shown to couple to the yeast mating pathway [32, 33]. Yasi et al. screened seven ORs against 57 chemicals resulting in the deorphanization of two receptors [34]. Sarria et al. used the olfactory receptor OR1G1 to detect medium-chain fatty acids produced by Escherichia coli directly in the culture broth, opening the doors to use GPCR-based assays for the high-throughput screening of microbially produced compounds [35]. OR-based assays in yeast are currently linked to GFP expression that requires a flow cytometer for read out, limiting the throughput to one compound per minute.

2. In vitro GPCR high-throughput assays

The low stability of purified GPCRs, makes their screening outside cells or membrane systems challenging, and GPCRs engineered for stability [36] are used in in vitro GPCR assays. Among the most widely used in vitro GPCR HTS assays are fluorescence polarization and affinity mass spectrometry (MS). Fluorescence polarization can identify GPCRs bound to fluorescently labelled ligands based on their molecular tumbling speed (Figure 3A). Heine et al. screened neurotensin receptor type 1 (NTS1) against 1,272 chemicals by displacing a fluorescently labeled NTS1 peptide ligand [37]. Eight of the 15 hits were confirmed via surface plasmon resonance. As expected, the availability of fluorescently labelled substrates is a key limitation of this approach. In affinity mass spectrometry, a GPCR is incubated with hundreds of compounds simultaneously. While unbound compounds are washed off, the GPCR ligands are eluted using protein denaturation and analyzed via mass spectrometry [38] (Figure 3B). Lu et al. expanded the number of compounds that can be screened via affinity MS from hundreds to tens of thousands by introducing an enrichment step of eluting the GPCR ligands and incubating them again with a purified receptor [39●]. A thermostable adenosine receptor was screened either as a purified receptor or embedded in a membrane against ~20,000 chemicals, and the assay identified 70% and 50% of ligands, respectively, previously identified using non-enrichment affinity MS. Although some compounds are not identified via the enrichment assay, the throughput is highly increased. Of note, thermostabilized GPCRs could affect ligand binding.

Figure 3. High-throughput in vitro GPCR screening.

A. In fluorescence polarization, the fluorescently labelled ligand has a fast tumbling rate leading to a decrease in fluorescence. Upon ligand binding, the tumbling speed of the ligand slows down increasing its fluorescence. B. In affinity mass spectrometry, a GPCR is incubated with hundreds of ligands simultaneously. Ligands bound to the GPCR are eluted and analyzed via liquid chromatography/mass spectrometry. The number of the ligands can be narrowed down by re-incubating eluted ligands with fresh GPCR.

3. In silico GPCR-based high-throughput assays

Computational methods can reduce the number of potential GPCR ligands (biased and non-biased) and allosteric regulators needed to test experimentally. Foster et al. used computational approaches rooted in GPCR-peptide ligand coevolution to predict 120 potential GPCR peptide ligands, and 21 GPCRs likely to respond to these peptides [40●]. As GPCRs can couple to multiple signaling pathways and their response may appear different depending on the pathway they activate, the GPCR ligand interactions were assessed using three different assays: mass spectrometry distribution, receptor internalization and β-arrestin recruitment. Ultimately, this work led to the deorphanization of five GPCRs.

Another in silico approach, computational docking, takes advantage of ultra-large chemical libraries, such as ZINC and ChEMBL, expanding the chemical diversity of screened potential GPCR ligands from thousands to hundreds of millions [41]. Virtual docking has been successfully applied to over a dozen classes of GPCRs [41]. Lyu et al. virtually screened 138 million chemicals against the D₄ dopamine receptor, the pharmacological target for schizophrenia [42]. The top 549 chemicals predicted by virtual docking were synthesized with a radiolabel, and, using a radioactivity binding assay, 22% of the hits were validated. In another example, Stein et al. virtually screened over 150 million chemicals against melatonin receptor 1 (MT1) [43●●]. After manual inspection the top 40 chemicals were synthesized, and 15 of them showed activity for either melatonin receptor 1 or 2 (61% similar) using a cAMP-dependent or β-arrestin-dependent assay. As computational docking significantly reduces the number of compounds to be tested, lower-throughput approaches are relevant. Notably, virtual docking is most effective when a crystal structure for the desired GPCR is available [44]. This currently limits the virtual docking approach, as 85% of non-olfactory GPCRs do not have a structure.

4. Future Outlook

The complexity of GPCR signaling, activating different pathways sometimes simultaneously, leads to the lack of a universal GPCR cell-based assay. Therefore, before embarking on GPCR HTS, there needs to be a deep understanding of the GPCR signaling pathway to be probed in order to choose the appropriate cell-based assay and understand its limitations. Such an understanding is not essential for in vitro assays, as GPCRs are evaluated in purified form or as part of membranes. Independent of the method used for GPCR ligand identification, the ligand needs to be confirmed using biochemical binding assays and ultimately validated in the GPCR endogenous tissue.

As automation equipment becomes more widely available in academic laboratories, GPCR HTS will be used to probe basic biological processes. For example, different GPCR isoforms tend to be expressed in different cell types. By applying GPCR HTS, we can begin to understand the different cell types that are activated by different ligands. Further, different patient populations tend to have different GPCR alleles, and GPCR HTS could help elucidate the extent to which different GPCR alleles are activated. With the potential of having each patient’s genome sequenced in the near future, we can envision GPCR HTS as a common occurrence to generate lifesaving therapies.

Bioinformatics and virtual docking rapidly scale down the number of ligands to be evaluated experimentally and can rely on lower-throughput biochemical binding assays. However, the necessity for the specific GPCR crystal structure for docking currently limits this approach to well-studied GPCRs. Given that there are only 64 unique crystal structures known (https://gpcrdb.org/structure/statistics), virtual docking approaches are not currently suitable to identify ligands for the majority of GPCRs. In the future, machine learning may be able to address this problem by combining bioinformatics, virtual docking on GPCR models and limited experimental GPCR HTS data to generate ligand predictions.