Accelerating GPCR drug discovery through computation and experiment integrated with direct detection of ligand binding events

Abstract

We present a perspective on an integrated workflow for GPCR drug discovery that combines computational modeling, functional cellular assays, and FLOWER, a label-free ultra-sensitive optical biosensing system for quantitating direct ligand receptor binding. We use bitter and sweet taste receptors (TAS2Rs and T1R2/T1R3) as examples of how FLOWER resolves receptor activation mechanisms left ambiguous after in silico screening and cell assays. This workflow offers a generalizable strategy for accelerating the discovery of selective and mechanism informed GPCR targeting therapeutics.

Introduction

G protein–coupled receptors (GPCRs) are the largest and most functionally diverse family of membrane proteins in the human genome, targeted by over 30% of FDA-approved drugs¹. Despite their importance, drug discovery for many GPCRs, particularly those outside of class A, remains challenging due to conformational complexity, limited structural information², and the difficulty of experimentally measuring ligand binding kinetics and mechanisms under near-native conditions³.

Over the past several years, major methodological advances have begun to reshape early-stage GPCR drug discovery. These include improved structure prediction tools such as AlphaFold2⁴ and GPCR-specific modeling frameworks, the rapid expansion of high-resolution cryo-electron microscopy (cryo-EM) structures⁵ for active and inactive GPCR states⁶, and enhanced computational docking⁷ and molecular dynamics approaches⁸ for predicting binding energetics and conformational transitions. On the experimental side, cell-based assays incorporating mini-G proteins⁹, conformational biosensors¹⁰, and engineered signaling reporters¹¹ provide mechanistically informative functional readouts relevant to efficacy, bias, and selectivity. Finally, sensitive biophysical methods including biolayer interferometry (BLI)¹², and emerging optical biosensing platforms¹³ now allow direct measurement of ligand-protein binding interactions with increasing precision. In parallel, these developments enable earlier and more informed decision-making in GPCR ligand optimization.

In this Perspective, we describe an integrated workflow that combines in silico computational modeling of huge data bases ( > 100 million), functional cellular assays on thousands of ligands, followed by direct ligand-protein kinetic binding measurements via the FLOWER (Frequency Locked Optical Whispering Evanescent Resonator) technique^14–16, which together accelerate and refine discovery efforts (Fig. 1). Unlike conventional workflows in which computational predictions are validated only indirectly through cell-based outcomes, the strategy described here directly links in silico modeling with in vitro measurements obtained from FLOWER. In our current implementation, FLOWER provides direct, label-free quantification of receptor–ligand binding affinity under near-physiological conditions, enabling computational predictions of binding strength and selectivity to be tested against experimentally determined equilibrium constants (K_d) without the use of tags.

Fig. 1. Overview Workflow Diagram.

Iterative GPCR ligand discovery workflow integrating computational modeling programs (such as GEnSeMBLE, DarwinDock, NAMD/CHARMM36m or for “hidden pockets” for allosteric modulators, AI-based programs such as Allosite and PASSer) followed by high-throughput screening (HTS) and other assays in cell-based systems. To obtain binding affinity, FLOWER is utilized for the promising high throughput screening candidates. Atypical FLOWER- or cell-based data require investigation with advanced molecular dynamics or other approaches to further understand mechanism of action. When rank order of potencies and affinities agree, additional modifications of the scaffold may reveal more potent or efficacious compounds. To prioritize candidates for experimental validation, candidates from modeling showing both stable predicted binding poses and favorable ΔG* values are advanced for experimental testing (FLOWER, followed by cell-based studies if the affinity meets criteria), while those with unfavorable energy profiles are deprioritized. The choice of the library, and the number of potential candidates from a virtual screening that are further studied, is highly dependent on the ligand-protein pairing that is being considered, the desired pharmacologic uniqueness, the computational method, and resource availability. Curated libraries of ~20 billion compounds can be parsed based on multiple factors leading to higher hit rates. Generative AI-based methods can find hidden pockets and design compounds that would be amenable to binding, opening additional avenues for generating libraries⁴¹. Color coding reflects workflow modules: blue (computational), peach (FLOWER), green (cell studies), gray (outcome states: inputs, decisions, and milestones).

We highlight the workflow using case studies of two structurally and functionally distinct GPCR families: the bitter taste receptors (TAS2Rs), a class of orphan-like receptors with atypical activation features, and the sweet taste receptor (T1R2/T1R3), a class C GPCR heterodimer with multiple allosteric and orthosteric binding sites. These systems illustrate how the workflow can resolve mechanistic ambiguities, characterize ligand bias, and guide rational lead optimization, building on methodologies we have previously established in separate studies. While focused on chemosensory GPCRs, the approach is generalizable across the broader GPCR landscape. The goal of this Perspective is to outline how these complementary methodologies can be integrated into a cohesive, iterative workflow to accelerate early-stage GPCR ligand discovery.

Computational Predictions

Activation and G Protein Coupling Features in Predicted Active-State Bitter Taste Receptor (TAS2R)

We recently used the GEnSeMBLE method to predict the active-state structures of three bitter taste receptors: TAS2R4¹⁷, TAS2R5^18,19, and TAS2R14²⁰. GEnSeMBLE offers advantages over approaches that generate a single predicted structure, such as traditional homology modeling, AlphaFold-based predictions, or RosettaCM, because it systematically samples transmembrane helix orientations and produces ensembles of active-state conformations. GEnSeMBLE may be particularly useful for this application because TAS2Rs have limited structural templates and atypical sequence motifs, making extensive conformational sampling specifically important for identifying plausible activation states.

TASRs exhibit distinct conserved sequence motifs that differ from canonical class A GPCRs. Our predicted models revealed a unique hydrogen-bonding network involving the conserved residues N1.50, R2.50, and S7.50 (Ballesteros–Weinstein numbering). This triad replaces the typical N1.50, D2.50, N7.49 motif found in class A receptors, which usually coordinates either a sodium ion or protonated water molecule within transmembrane domain 2 (TMD2). Additionally, in most class A GPCRs, the inactive state is stabilized by an “ionic lock”— a salt bridge typically formed between R3.50 and E6.30. TAS2Rs lack this conserved salt bridge, suggesting these receptors may exist in a pre-activated conformation even in the absence of ligand. Another hallmark of activation in class A GPCRs is the conformational switch of W6.48, which toggles between gauche+ and trans rotamers to toggle between inactive and active states. In TAS2Rs, W6.48 is replaced by a conserved tyrosine (Y6.48), whose functional role in activation remains unclear.

These differences may explain the propensity for most TAS2Rs to display low potency for tastants (agonists) and to be broadly tuned to activation by apparently structurally diverse ligands. To elucidate the mechanistic basis of agonist potency at TAS2R5, we employed a multi-pronged approach that combined live cell–based functional assays, direct ligand-binding measurements using FLOWER, and atomistic molecular dynamics (MD) simulations following DarwinDock binding site predictions. DarwinDock was selected because it performs exhaustive side-chain optimization and scoring within the predicted binding pocket, enabling identification of energetically favorable ligand poses in GPCR systems where binding site geometry is uncertain²⁰. In live cell studies, the functional outcome (increased [Ca²⁺]_i) provides for the potency (i.e., EC₅₀) and minimal and maximal responses. However, agonist potency will vary with receptor expression (sigmoid curves shifting to the left with increased expression) as can the maximal response (efficacy). We considered that a true measurement of affinity would be a necessary component for characterizing new agonists, and thus turned to FLOWER to obtain direct measurements. So called “hidden pockets” for allosteric modulators may require different computational methods since they may be distant from the presumed binding pocket for the orthosteric site. Bioptic ligands, which bind to both the orthosteric and allosteric sites may require yet another docking and screening approach. Careful consideration of the kinetics, dose-responses, and other classic pharmacologic studies in cell-based assays can reveal atypical properties suggesting certain mechanisms. Importantly, unlike orthosteric radiolabels which bind to (or are displaced from) the orthosteric binding pocket, FLOWER will detect and quantitate compound binding regardless of location on the target protein. This has been particularly helpful under atypical binding behavior.

Mechanism of detecting ligand-receptor interactions using FLOWER

Direct and quantitative characterization of protein–ligand interactions is essential for verifying computational predictions, identifying unusual binding actions, and confirming promising lead compounds targeting GPCRs from cell-based systems. One method that enables such measurements with high sensitivity and specificity is FLOWER^{13–16,21–31}. FLOWER is a label-free optical biosensing platform that tracks refractive index changes induced by molecular binding on the surface of ultra-high-Q whispering gallery mode (WGM) optical resonators (Fig. 2a).

Fig. 2. Representative characterization of a microtoroid optical resonator and its use in measuring molecular binding kinetics using FLOWER.

a Scanning electron microscope (SEM) image of a silica microtoroid resonator. The inset shows a finite-element simulation of the whispering gallery mode electric field distribution, which is highly confined at the periphery of the resonator, enabling sensitive evanescent detection. Reproduced with permission from¹⁴. b Real-time binding curves of Rubusoside to T1R2/T1R3, as measured by FLOWER, showing concentration dependent binding. Reproduced with permission from³⁹.

WGM resonators confine light by circulating it along a curved dielectric surface^32–34, where it repeatedly interacts with the surrounding medium through a tightly confined evanescent field. This architecture allows FLOWER to detect minute refractive index changes in real time³⁵ without the need for fluorescent or radioactive labels making it well-suited for studying weak or transient interactions under near-native conditions. Although FLOWER can be implemented using various cavity geometries, including microspheres, microdisks, and integrated photonic ring resonators, we typically employ ultra-high-Q silica microtoroid resonators due to their exceptional quality factors ( > 10⁸), small mode volumes that enhance light–matter interactions, strong and accessible evanescent fields for surface-sensitive detection, and compact on-chip format. These attributes enhance detection sensitivity and allow robust, reproducible measurements of ligand binding events. In a typical experiment, laser light is evanescently coupled into the microtoroid^36,37, where it circulates hundreds of thousands of times—akin to a whisper echoing within a circular gallery. The system actively locks the laser frequency to the resonant frequency of the microtoroid using a feedback control loop. When a molecule binds to the resonator surface, the local refractive index changes slightly, shifting the resonant frequency. The feedback system continuously tracks this shift in real time, allowing precise and dynamic quantification of molecular interactions. From the magnitude and kinetics of these frequency shifts, FLOWER can extract key binding parameters. As shown in Fig. 2b, the concentration-dependent peak resonance shifts can be used to estimate apparent binding constants by plotting the maximal response versus ligand concentration and fitting to a sigmoidal (Hill-type) binding model.

Compared to conventional techniques such as surface plasmon resonance (SPR) and biolayer interferometry (BLI)¹², FLOWER offers several distinct advantages for GPCR drug discovery:

• Label-free detection, avoiding the need for bulky or perturbative tags³⁸;

• Ultra-high sensitivity, capable of detecting interactions at zeptomolar concentrations¹⁵ or even single molecules¹⁶;

• Compatibility with complex biological samples, including detergent-solubilized receptors and crude membrane preparations¹⁵;

• Preservation of receptor conformation, supporting analysis of GPCRs embedded in reconstituted lipid bilayers, nanodiscs, or other native-like environments³⁹.

These features are particularly advantageous in early-stage drug discovery, where receptor materials are often limited, interactions can be low-affinity or transient. Here, maintaining membrane protein integrity is crucial for biologically meaningful measurements. In the FLOWER assay, the receptors are maintained in a membrane-like lipid environment throughout the measurement. Purified GPCRs are first solubilized in mild detergents that preserve transmembrane structure and then reconstituted into supported lipid bilayers on the microtoroid surface through gradual detergent removal. This appears to mimic the physiological cell membrane, ensuring that the hydrophobic transmembrane regions of the receptor are exposed to a lipid environment rather than the solid interface. In previous studies, we have also successfully incorporated GPCRs into nanodiscs³⁹, which further confirmed that receptor folding and ligand-binding activity are preserved in both membrane-mimetic formats. As shown in previous work with other GPCRs (e.g., κ-opioid receptor¹⁵), receptors reconstituted by this approach maintain ligand-specific binding and kinetic behavior, confirming functional integrity. The FLOWER platform measures label-free binding via resonance shifts, and its ultra-high sensitivity (zeptomolar-level) allows reliable measurement of receptor–ligand interactions even at very low surface densities. Because detection depends on refractive-index perturbation near the resonator rather than on optical labels or ensemble averaging, the assay is compatible with monomeric or multimeric receptor complexes, including GPCR dimers or receptor–G-protein assemblies, provided they are incorporated within the SLB.

Application to agonist discovery at bitter taste receptors (TAS2Rs)

For studies with TAS2R5, we studied multiple potential agonists and noted a significant variance from the computationally derived predicted binding energy and the potency of the agonists. As shown in Fig. 3a, the compounds denoted T5-1, T5-8, and T5-14 have different EC₅₀ values for stimulating Ca²⁺ in human airway smooth muscle (HASM) cells. The signal transduction coefficients (log [Rmax/EC₅₀]) differed by over 2-orders of magnitude between T5-8 and T5-14. When plotted as the computationally derived binding energy vs signal transduction coefficient (Fig. 3b), most compounds displayed a consistent relationship between the two (black symbols). However, a few (blue symbols, particularly T5-14) had greater computationally derived binding energies than would be predicted based on signaling in cells. In FLOWER studies, the results followed that of the computationally derived energies. This indicated that the compound, such as T5-14 indeed bound to the receptor with high affinity, but some portion of that binding did not contribute to activation. This might be because of the energy of binding at the orthosteric sites plus the energy from “null” binding (or even a pose consistent with surmountable antagonism). This discrepancy is appreciated by plotting the effects of the agonist on the outputs from FLOWER and Ca²⁺ from live cells (Fig. 3c). Binding measured by FLOWER with different cell signaling results can occur with antagonists, inverse agonists, partial agonists, and allosteric modulators. We had eliminated these as possibilities in cell-based experiments. We also considered that TAS2Rs may have a cryptic transition state representing the correct energy “baseline” from which to calculate the net change in energy. To assess this possibility, we utilized metadynamics (MetaD) simulations. We identified a key transition underlying TAS2R5 activation which involves ligand-induced opening of the Gα subunit, at the interface between the α-helical (AH) and Ras-like domains. This conformational change drives the transition from a pre-coupled TAS2R5–G protein complex to the fully active state leading to functional signaling. Crucially, the magnitude of the energetic barrier associated with this domain opening varies dramatically between the agonists, correlating with their functional potency. For the most efficacious agonist, T5-8, Gα opening occurred with a strongly favorable free energy change (ΔG = −11.1 kcal/mol) and a low activation barrier, ΔG* = 3.4 kcal/mol. For the moderately active T5-1, the AH–Ras cleft expanded from 22 Å to 31 Å with a more modest energy gain of −4.8 kcal/mol and a higher barrier, ΔG* = 4.8 kcal/mol. For the least potent agonist, T5-14, the cleft still opened, but the associated energy gain was minimal (ΔG = −1.4 kcal/mol and ΔG* = 6.2 kcal/mol), showing very inefficient activation despite stable binding in the pre-coupled complex (Fig. 4).

Fig. 3. Characterization of TAS2R5 agonists.

a Concentration–response curves for selected TAS2R5 agonists (T5-1, T5-8, and T5-14) measured in human airway smooth muscle (HASM) cells, showing differences in potency (EC50) and maximal response. b Relationship between computationally predicted binding energies and signal transduction coefficients. Most compounds exhibit a consistent correlation (black symbols), whereas others (blue symbols), particularly T5-14, display stronger predicted binding than would be expected based on cellular signaling. Reproduced with permission from¹⁸. c Comparison of FLOWER binding measurements with cellular Ca²⁺ responses for representative agonists. FLOWER results follow the computationally derived binding energies, confirming high-affinity binding of T5-14 despite its low functional potency in cells. This incongruity highlights cases where receptor binding and downstream signaling are discrepant and indicate an atypical binding pose for this agonist which was found using more advanced molecular dynamics.

Fig. 4. Ligand-dependent Gα opening in TAS2R5 revealed by metadynamics (metaD) simulations.

Precoupled and fully activated TAS2R5 structures from metaD simulations are shown for (a) the apo form and (b) the T5-8–bound form, illustrating the closed and open conformations of the Gα subunit. The metaD free-energy profiles for Gα opening are plotted as a function of the distance between the AH domain (center of mass of Cα atoms of residues 62–182) and the Ras-like domain (center of mass of Cα atoms of residues 42–58, 272–280, and 326–335) for the TAS2R5–Gi complex with different ligands: (c–f) Free energies of the three agonists and the unliganded (Apo) state. Reproduced with permission from²⁰.

Not only do these findings reveal the well-recognized axiom that agonist potency is not governed solely by binding affinity, but indicates its dependence on the energetics of G protein activation, particularly the ability to drive domain separation within Gα to promote GDP exposure to solvent and subsequent signaling. Importantly, the pursuit of this mechanism was made possible by obtaining a direct measurement of agonist-receptor binding energy by FLOWER which prompted the additional computational studies. This has led to a much more predictive method of deriving agonist action that will promote discovery of novel TAS2R5 agonists. MetaD could be used to further analyze the top hits from docking prior to cell based assays, potentially decreasing the number of compounds to be studied.

Application to agonist discovery at the sweet taste receptor (T1R2/T1R3)

Detection of sweet tastants has been particularly challenging because it involves binding to a heterodimer of two seven-transmembrane proteins, T1R2 and T1R3 (Fig. 5).

Fig. 5. Illustration of the human sweet taste receptor (T1R2/T1R3) highlighting multiple ligand binding sites.

The heterodimeric sweet taste receptor comprises T1R2 and T1R3 subunits, each containing extracellular Venus flytrap domains (VFD2 and VFD3), cysteine-rich domains (CRD2 and CRD3), and transmembrane domains (TMD2 and TMD3). Steviol glycosides, including RebM, RebD, RebC, RebB, and Rubu, engage different binding pockets. Additional sweeteners and allosteric modulators, such as aspartame, neotame, NHDC, cyclamate, and S819, show site-selective engagement across VFDs and TMDs. In addition to small molecules, several sweet-tasting proteins (e.g., thaumatin, brazzein, monellin) also activate T1R2/T1R3 through protein–protein interactions, further highlighting the structural diversity and complexity of ligand recognition by this receptor. This multi-site binding model provides a structural framework for understanding the complex pharmacological responses elicited by diverse sweet compounds. Reproduced with permission from³⁹.

To investigate the structural basis of ligand recognition, we employed DarwinDock molecular docking to characterize the interactions between steviol glycosides (“artificial sweeteners”) and the T1R2/T1R3 heterodimer³⁹. These simulations identified four distinct binding sites: two in the Venus flytrap domain of T1R2, one in T1R3, and one in the transmembrane domain. This multi-site binding architecture suggested that receptor activation is driven by interactions, which may act in a cumulative fashion, across several receptor domains. Because computational docking can predict low-affinity or transient binding interactions that may be challenging to detect with conventional methods, their experimental validation necessitated techniques that offer high sensitivity and specificity. FLOWER provides such a platform by enabling direct, real-time measurement of ligand binding to purified receptors in near-native environments.

In these experiments, FLOWER enabled high-sensitivity quantification of ligand binding kinetics across a concentration range spanning several orders of magnitude. We observed that the equilibrium dissociation constant (K_d) for RebM, Rubu, RebC, and RebD inversely correlated with their sweetness potency as drawn from the literature, with RebD deviating from this trend, as it binds more strongly than would be expected based on its reported sweetness. This suggests that binding affinity alone is inadequate to explain the complex interactions that ultimately lead to functional signaling. As in dynamic mass redistribution (DMR) assays⁴⁰, we chose to analyze the peak binding response from the FLOWER resonance shift as it most robustly captures the maximal ligand–receptor engagement while avoiding variability introduced by slower conformational changes or surface rebinding. This peak-based metric provides a reproducible and physiologically relevant readout of binding affinity. Analysis of Hill slopes showed values consistently below unity for all ligands, indicative of multiple, non-equivalent binding sites on the receptor heterodimer (Fig. 6). Depletion of G protein compared to receptor can also cause shallow curves, but this was not the case in these studies. Indeed, by comparing binding with and without the C-terminal 20-residue peptide (C20) of the Gα protein, we found, as expected, that this surrogate G protein binding significantly enhanced binding affinity. However the shift was in a domain-specific manner. For example, C20 coupling to TMD2 enhanced RebM and RebC binding at VFD2, while coupling to TMD3 enhanced Rubusoside binding at VFD3, consistent with the binding energy calculation. These trends are further visualized in the normalized binding curves (Fig. 6) which reveal C20-induced shifts in binding affinity and Hill slope that depend on specific ligand–domain configurations. For example, in the TMD2 open case, C20 coupling shifted the RebM curve leftward and increased its steepness, while Rubusoside displayed a leftward shift accompanied by a steeper Hill slope when C20 was bound in the TMD3 open case. Here, “TMD2 open” describes a receptor conformation that increases accessibility to the VFD2-linked binding site, making it easier for certain ligands (e.g., RebM and RebC) to bind tightly, whereas “TMD3 open” reflects increased accessibility to the VFD3-linked binding site, which preferentially affects ligands such as Rubusoside. In the FLOWER assay, the orientation of the T1R2/T1R3 sweet receptor on the microtoroid surface is controlled by the antibody used for immobilization. Each subunit of the heterodimer carries an epitope tag near one of the transmembrane domains: a FLAG tag adjacent to TMD2 and a Rho tag adjacent to TMD3. When the microtoroid surface is functionalized with anti-FLAG antibody, the receptor binds through the FLAG tag near TMD2, anchoring that side of the receptor to the surface. This configuration leaves the TMD3 side exposed to the solution and accessible for interactions with C20 or other binding partners. Conversely, when anti-Rho antibody is used instead, the receptor attaches through the Rho tag near TMD3, anchoring the TMD3 side and leaving the TMD2 region exposed. By switching the immobilization antibody, the experiment selectively “opens” either the TMD2 or TMD3 interface, allowing domain-specific effects of G-protein coupling or ligand binding to be measured.

Fig. 6. Normalized binding response curves obtained via FLOWER of sweet ligands binding to the human sweet taste receptor T1R2/T1R3 under different G protein coupling and domain-opening conditions.

a RebC binding curves in the presence or absence of the last 20 amino acids of the C-terminal tail of Gas (C20 peptide) and TMD2 or TMD3 domain opening. b Rubusoside binding with or without C20 peptide for TMD2 or TMD3 open configurations. c RebM binding under similar varied conditions. d RebD binding with and without C20 peptide under TMD3 open configuration. Reproduced with permission from³⁹.

This domain-specific modulation supports the hypothesis that receptor activation and ligand specificity are governed by allosteric interactions and conformational stabilization through G protein engagement—mechanistic insights uniquely enabled by FLOWER.

Integration into the Expanded Workflow

As part of the expanded GPCR drug discovery workflow, FLOWER acts as a final assessment of the binding predicted by in silico prediction and the assessment of cellular functional assays. FLOWER’s ability to monitor binding in real time provides access to association and dissociation rates, as well as equilibrium binding constants, enabling detailed characterization of ligand binding kinetics. Any atypical features of the fit data indicate the need for further exploration of the mechanism of action even when cell-based assays appear consistent with the desired pharmacologic properties. This level of resolution supports structure–activity relationship (SAR) analysis by revealing how specific chemical modifications influence both binding affinity and interaction dynamics. Such kinetic insights are critical for rational lead optimization, particularly in guiding R-group modifications to enhance potency, residence time, and target selectivity. Optimization is envisioned as an iterative design–test cycle guided by both computational modeling and experimental data, in which structural analogs are generated by modifying R-groups on the ligand scaffold to systematically explore chemical features that influence receptor binding. Computational prediction and cell-based assays prioritize these R-group variants, which can be refined by cellular assays with the final best cases measured with FLOWER, all together providing quantitative binding affinities and kinetic information that identify modifications that improve target engagement. Compounds showing improved activation in cell-based functional assays would then be evaluated by FLOWER to determine whether the R-group changes enhance activity and receptor selectivity, thereby guiding further refinement. Based on our published work to date, an initial round of cell-based assays on the top computationally ranked ligands has been sufficient to identify candidates with good activity which can then be followed by FLOWER measurements to confirm measurable binding and suitable affinity. FLOWER can be incorporated at several steps in the workflow. We show it after the cell-based high-throughput screening because the latter rapidly eliminates candidates from large libraries and is currently less labor intensive. In addition to screening existing ligand libraries, this workflow can also incorporate generative AI models to propose de novo ligand structures based on the predicted receptor binding pocket(s). Together, these capabilities position FLOWER to support both early validation and future iterative optimization as more ligands are explored.

While FLOWER offers a uniquely sensitive, label-free means of monitoring receptor–ligand interactions in real time, several limitations should be acknowledged. FLOWER quantifies physical binding events rather than downstream signaling and therefore does not directly report functional outcomes such as G-protein activation, second-messenger production, or β-arrestin recruitment. Nevertheless, functional correlations can be inferred through experimental design. For example, incorporation of mini-G proteins allows assessment of coupling-dependent conformational changes that alter ligand-binding kinetics and affinities, while competition assays on the same toroid enable discrimination between agonists and antagonists based on their relative binding behaviors. The method also currently requires purified receptor proteins reconstituted into stabilizing lipid environments such as nanodiscs or supported bilayers, which may limit its applicability to receptors that are difficult to express or purify. While the lipid environment of a cell cannot be perfectly matched with the FLOWER method, we note that the lipid content of the cell membrane can differ by cell type anyway so the absolute values may be shifted compared to the affinity measured using a specific cell expressing the receptor (if it could be measured). Furthermore, if known low affinity and high affinity ligands are utilized as controls with each experiment, a rank-order of affinity (or other outputs) can be established which places into perspective the properties of tested ligands. Ultimately, signaling assays in cells and physiologic studies in the relevant tissues or intact organisms remain essential to establish whether ligand-induced receptor engagement produces the desired therapeutic outcomes.

Future Directions

GPCR drug discovery is entering a transformative phase, enabled by advances in structural biology, computational modeling, and high-sensitivity experimental platforms. As our understanding of GPCR conformational dynamics and signaling bias expands, there is a growing need for integrated workflows that iteratively combine prediction, measurement, and validation. Technologies like FLOWER will play a central role in this evolution. Its ability to provide real-time, label-free, and ultra-sensitive kinetic measurements makes it particularly well-suited for early studies especially when receptor quantities are limiting or interactions are weak and transient. On the other hand, cell-based studies are essential to distinguish cases that may have high activation barriers or multimodal binding. Future development of FLOWER may include miniaturization, parallelization, and integration with microfluidic platforms, enabling high-throughput assays at unprecedented resolution. Coupling FLOWER with AI-driven compound design and automated feedback loops can further accelerate discovery. The integration of computational prediction, kinetic validation via FLOWER, and downstream functional assays forms a closed-loop pipeline that optimizes both ligand potency and specificity in real time. Overall, FLOWER exemplifies a new class of biophysical tools that bridge traditional divides between computation and experiment, as well as early discovery and preclinical validation. As workflows become more integrated, scalable, and physiologically relevant, we anticipate a substantial acceleration in the development of GPCR-targeted therapeutics across diverse disease areas.

Author contributions

All authors jointly conceived the Perspective. J.S. wrote the sections related to the FLOWER sensing platform; S.L. and D.K. contributed the content on cell-based assays; S.-K.K. and W.A.G. authored the sections on computational modeling. All authors contributed to the Introduction and Future Directions sections, and critically revised and approved the final manuscript.

Data availability

This Perspective does not report any new data. All previously published data referenced in this work are cited appropriately and available from the corresponding publications. No custom code was generated or used for this work.

Competing interests

J.S. owns a financial stake in Femtorays Technologies, which develops label-free molecular sensors. The other authors declare no competing interests.