The T-REx™-293 Cell Line was obtained from Invitrogen (CA, United States). T75 and T175 mammalian cell culture flasks were purchased from Fisher Scientific (Loughborough, United Kingdom). All cell culture reagents, including Hank’s balanced salt solution (HBSS), phosphate-buffered saline (PBS), and foetal calf serum (FCS), were purchased from Sigma Aldrich (Gillingham, United Kingdom), except for blasticidin, which was obtained from Gibco™ (MA, United States), and Zeocin™. Polyethylenimine (PEI) (25 kDa) was obtained from Polysciences Inc. (PA, United States), and the culture plates were obtained from Greiner Bio-One (code 655098 Kremsmünster, Austria).

HisTrap FF crude 5-mL columns were obtained from GE Healthcare (IL, United States). Vivaspin protein concentrators were obtained from Sartorius (Göttingen, Germany). Slide-A-Lyzer Dialysis Cassettes, NuPAGE LDS Sample Buffer, NuPAGE 4%–12% Bis-Tris 15 × 1.0 mm well gels, NuPAGE MOPS SDS Running Buffer, PageRuler Prestained Protein Ladder, were all obtained from Thermo Fisher (MA, United States).

Salmeterol was obtained from Tocris (Bristol, U.K). Formoterol hemifumarate was obtained from APExBIO (TX, United States), and BI-167-107 was obtained from Boehringer Ingelheim (Ingelheim, Germany). Compound 26 was a gift from Novartis. (±)-Epinephrine hydrochloride, noradrenaline, salbutamol hemisulfate, and isoprenaline hydrochloride were purchased from Sigma-Aldrich (Gillingham, United Kingdom). Dobutamine hydrochloride was obtained from Merck Life Sciences, UK. Isoxsuprine hydrochloride, ritodrine hydrochloride, and tulobuterol were obtained from CliniSciences Limited. Nano-Glo luciferase substrate was obtained from Promega (WI, United States). All other chemicals were purchased from Sigma-Aldrich (Gillingham, United Kingdom).

BMG PHERAstar FSX plate reader (BMG Labtech, Offenburg, Germany), fitted with BRET1 plus optic module (ex. 475/30 nm, em. 535/30 nm) and MARS software, was purchased from BMG Labtech (Offenburg, Germany). GraphPad Prism 9 was purchased from GraphPad Software (San Diego, United States). Microsoft Excel™ XP was purchased from Microsoft (Washington, United States).

The construct pcDNA4TO-TwinStrep (TS)-SNAP-β₂AR was generated through the amplification of the SNAP and β₂AR sequences from the pSNAPf-ADRB2 plasmid (NEB) and inserted into pcDNA4TO-TS using Gibson assembly (Heydenreich et al., 2017). pcDNA4TO-TS-SNAP-β₂AR-nLuc was generated by Dr. Brad Hoare through the amplification of pcDNA4TO-TS-SNAP-β₂AR and nanoLuc, with the insertion of nanoLuc into pcDNA4TO-TS-SNAP-β₂AR via Gibson assembly. Both constructs used a signal peptide based on the 5HT_3A receptor to increase protein folding and expression. The CASE G_s (or G_s-CASE) protein constructs were designed and optimised by the Schulte Lab (Schihada et al., 2021) and were obtained from Addgene. Mammalian Venus-fused mini-G_s constructs were a kind gift from Nevin Lambert (Wan et al., 2018). For the bacterial expression of Venus-mini-G_s and mini-G_s, protein encoding DNA sequences were amplified from the corresponding mammalian constructs and inserted into the pJ411 vector containing MKK-HIS10-TEV N-terminal tag (Sun et al., 2015) via Gibson assembly, yielding the constructs MKK-HIS10-TEV-mini-G_s and MKK-HIS10-TEV-Venus-mini-G_s.

pcDNA4TO-TS-SNAP-β₂AR or pcDNA4TO-TS-SNAP-β₂AR-nLuc was stably transfected into T-REx™-293 cells (Invitrogen) using PEI. A stable mixed population was selected by resistance to 5 μg/mL blasticidin and 20 μg/mL zeocin. Stable cell lines were maintained in high-glucose DMEM (Sigma D6429) with 10% FBS, 5 μg/μL blasticidin, and 20 μg/μL zeocin at 37°C in a humidified atmosphere of 5% CO_2. When ∼70% confluent, TS-SNAP-β₂AR or TS-SNAP-β₂AR-nLuc expression was induced with 1 μg/mL tetracycline. Cells were left to express for 50 h before harvesting for assays. The T-REx™-293 pcDNA4TO-TS-SNAP-β₂AR-CASE G_s stable cell line was generated by stably transfecting the CASE G_s constructs into the T-REx™-293 pcDNA4TO-TS-SNAP-β₂AR using PEI. A mixed population stable cell line was generated by selection with 500 μg/mL G418, and then a single colony population was generated via FACS.

For membrane preparation, all steps were conducted at 4°C to avoid tissue degradation. Cell pellets were thawed and re-suspended using ice-cold buffer containing 10 mM HEPES and 10 mM EDTA (pH 7.4). The suspension was homogenised using an electrical homogeniser (ULTRA-TURRAX, IKA-Werke GmbH, Germany) and subsequently centrifuged at 1,200 × g for 5 min. The pellet obtained, containing cell nucleus and other heavy organelles, was discarded, and the supernatant was centrifuged for 30 min at 48,000 × g at 4°C (Beckman Avanti J-251 Ultra-centrifuge; Beckman Coulter). The supernatant was discarded, and the pellet was re-suspended in the same buffer (10 mM HEPES and 10 mM EDTA; pH 7.4) and centrifuged again for 30 min as described above. Finally, the supernatant was discarded, and the pellet was re-suspended in ice-cold 10 mM HEPES and 0.1 mM EDTA (pH 7.4). Protein concentration determination was carried out using the bicinchoninic acid assay kit (Sigma-Aldrich) with BSA as the standard. The final membrane suspension was aliquoted and maintained at −80°C until required for the assays.

TS-SNAP-β₂AR or TS-SNAP-β₂AR-nLuc was solubilised from stably transfected T-REx^TM-293 cell membranes, as described previously (Harwood et al., 2024). Solubilisation was carried out using 1% DDM (w/v) in 20 mM HEPES, 5% (v/v) glycerol, and 150 mM NaCl, pH 8, at 4°C for 2–3 h. Samples were clarified by ultracentrifugation at 4°C for 1 h at 100,000 × g.

His-TEV-Venus-mini-G_s and His-TEV-mini-G_s were expressed in NiCo21(DE3) E. coli, cultured in Terrific Broth (Gibco). 1L cultures were induced with 1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) at OD = 0.6 and incubated for a further 20 h at 20°C and 225 RPM. Pellets from 1L cultures were thawed on ice, and re-suspended in 50 mL lysis buffer (20 mM HEPES, pH 7.5, 500 mM NaCl, 40 mM imidazole, 10% glycerol, 8 mM β-mercaptoethanol (BME), 1 μM guanosine diphosphate (GDP), complete protease inhibitors (Roche), DNase I, and lysozyme) using a Dounce homogeniser. Lysis occurred on ice via sonication, using a Vibra-Cell probe sonicator with 5 × 10-s pulses, 30 s apart. The lysate was loaded onto the HisTrap FF crude 5-mL column, using ÄKTA™ start protein purification system at a flow rate of 5 mL/min. The system and column had been equilibrated with 10 column volumes (CV) of buffer A (20 mM HEPES, 500 mM NaCl, 40 mM imidazole, 10% glycerol, 8 mM BME, and 1 μM GDP). Unbound protein was washed out with 10 CV of buffer A. Bound protein was then eluted over an 8 CV gradient of 0% to 100% buffer B at a flow rate of 5 mL/min (Buffer B = 20 mM HEPES, 500 mM NaCl, 400 mM imidazole, 10% glycerol, 8 mM BME, and 1 μM GDP). The presence of His-TEV-Venus-mini-G_s and His-TEV-mini-G_s was confirmed by SDS-PAGE analysis and InstantBlue staining for protein. Pooled elution fractions were then concentrated using 10,000 or 30,000 molecular weight cutoff (MWCO) Vivaspin protein concentrators by centrifugation at 3000 × g and 4°C for 15-min intervals over 2–3 h. Protein was exchanged into assay buffer using Slide-A-Lyzer 10,000 or 30,000 MWCO dialysis cassettes for untagged and Venus-tagged mini-G_s protein samples, respectively. Dialysis occurred overnight at 4°C under constant stirring. The assay buffer consisted of 20 mM HEPES, 150 mM NaCl, 10% glycerol, 8 mM BME, and 1 μM GDP. The purified mini-G_s protein was flash-frozen using liquid nitrogen and stored at −80°C.

The assay buffer, consisting of HBSS (Sigma H8264) containing 10 mM HEPES, 0.1% BSA, and 0.1% ascorbic acid, pH 7.4, was used in all NanoBRET assays. For recruitment assays, varying concentrations of β₂AR agonists were used to recruit Venus-mini-G_s to the TS-SNAP-β₂AR. Assays were run in 50 μL volumes in white 384-well OptiPlate (Revvity). Receptor, ligand, 0.3 μM mini-G_s proteins, and 10 μM furimazine were added to the plate and incubated for 60 min at room temperature before reading on PHERAstar FSX using the BRET1 module. For kinetic assays, in which the affinity of Venus-mini-G_s for the agonist-bound TS-SNAP-β₂AR-nLuc receptors was measured over time, assays were run in 50 μL volumes in white 384-well OptiPlate. Varying concentrations (10–300 nM) of Venus-mini-G_s were added to assay plates. TS-SNAP-β₂AR membranes were pre-incubated with saturating concentrations (100x EC₅₀) of selected β₂AR agonists and furimazine for 15 min prior to addition to the plate. TS-SNAP-β₂AR membranes were added to the plate offline and mixed with the Venus-mini-G_s on a plate shaker (MixMate, Eppendorf) at 600 RPM for 10 s. The mixture was then immediately read on PHERAstar FSX as described above, with readings taken over a period of 240 min.

For G_s-CASE activation assays, a single population of T-REx™-293 stably expressing pcDNA4TO-TS-SNAP-β₂AR and CASE G_s was plated at 50,000 cells/well in 96-well plates, in a volume of 100 μL, and induced for 48 h with 1 μg/mL tetracycline at 37°C and 5% CO₂. Plates were washed once with 100 μL/well assay buffer (HBSS containing 10 mM HEPES, 0.1% BSA, and 0.1% ascorbic acid) prior to the addition of 90 μL/well of assay buffer containing 10 μM furimazine, diluted in assay buffer, to achieve a final concentration of 8 μM. The plates were incubated at 37°C and 5% CO₂ for 20 min. A white back seal was placed on the underside of the plate, and luminescence was read on a PHERAstar FSX using the BRET1 module for 3 min to establish a baseline BRET signal. The plate reader was then paused, and 10 μL of ×10 ligand dilutions were added accordingly. Readings were taken over a period of 30 min.

The previously described ordinary differential model (ODE) of the cubic ternary complex model (Weiss et al., 1996), with additional reactions to simulate the G protein activation cycle, was used (Woodroffe et al., 2009; Bridge et al., 2018). The model, encoded in COPASI (Hoops et al., 2006), includes ligand binding, receptor activation, G protein binding, and the G protein cycle, whereby the model output is activated G protein Gα_GTP and receptor occupancy (Bridge et al., 2018). Prior to the addition of the ligand, we first compute the system for 10⁶ s. To enable the simulation of the data, the cooperativity factor β (see Supplementary Figure 7; Supplementary Table 3) was varied, and simulations were performed. Steady state was reached after 5 min, and outputs are shown after 10 min.

All non-linear regression and statistical analyses were performed using GraphPad Prism 9. Multiple replicates were combined, such as TR-FRET equilibrium binding curves and mini-G_s equilibrium recruitment curves, as shown in Supplementary Material. Data points for each replicate were normalised to the maximum value obtained for each ligand in each experiment. Competition ligand-binding data were fitted to a one-site model (Equation 1). Y = B o t t o m + T o p − B o t t o m 1 + 10 x − L o g I C 50   , (1) where Y is the binding of tracer, x = Log [ligand], IC₅₀ is the concentration of the competing ligand that displaces 50% of radioligand-specific binding.

CASE G_s activation data from individual experiments were fitted to sigmoidal (variable slope) curves using a “four-parameter logistic equation” (Equation 2): Y = B o t t o m + T o p − B o t t o m 1 + 10 l o g E C 50 − X ∗ H i l l s l o p e , (2) where Bottom is the plateaus of the agonist concentration response curve and Top is the basal response (fixed to 1). LogEC₅₀ is the concentration of the agonist that produces a half-maximal effect, and the Hillslope is the unitless slope factor or Hillslope, which was fixed to −1.

Mini-G_s association data were fitted to a global fitting model (Equation 3) using GraphPad Prism 9.2 to simultaneously calculate k _on and k _off using the following equations, where k _obs equals the observed rate of association and L is the concentration of mini-G_s. K d = k off k on , L = Hotnm * 1 e − 9 , K ob = k on * L + k off , Occupancy = L / L + K d , Y max = Occupancy * B max , drift = B max * ⁡ exp − drift * X , Y = Y max * 1 − exp − 1 * k ob * X * drift . (3)

Saturation binding curves for Venus-mini-G_s binding to the agonist TS-SNAP-β₂AR-nLuc were fitted to a one-site specific binding model according to Equation 4. The final K _d values were taken as an average of K _d values from individual specific curve fits. Y = B m a x ∗ X   K d + X , (4) where Y is the specific binding, K _d is the equilibrium dissociation constant of the labelled ligand (in this case, Venus-mini-G_s), and x represents [Venus-mini-G_s] in nM.

Pearson’s correlation coefficient was used to investigate correlations between mini-G_s recruitment, CASE-G_s activation, mini-G_s binding K _d, k _on and k _off values, and literature pK _i/d. Deming regression was applied to determine the line of best fit while accounting for errors in observations on both the x- and y-axes. All statistical analyses were performed in GraphPad Prism 9, and p < 0.05 was considered statistically significant.

To produce a suitable dataset for analysis, we chose 12 β₂AR agonists anticipated to have a diverse range of efficacies, affinities, and ligand binding kinetics. We first characterised the efficacy of these compounds in activating the heterotrimeric G_s protein using a NanoBRET-based biosensor (Schihada et al., 2021; Harwood et al., 2024). In this assay format, G_s protein activation results in a decrease in the NanoBRET signal as the nLuc-labelled α-subunit of the G_s protein dissociates from the Venus-labelled γ-subunit. These experiments are summarised in Figures 1A–C and Table 1.

The Gs-CASE assay functions as a non-amplified system, showing very distinct differences in measurable efficacy between full and partial agonists. The concentration-response curves for formoterol (full) and tulobuterol (partial agonist) are shown in Figure 1A. A broad range of potencies was observed for the 12 tested ligands, with pEC₅₀ values ranging from 6.49 ± 0.48 for dobutamine to 8.69 ± 0.18 for isoprenaline (see Figure 1B; Table 1). Figure 1C shows a range of efficacy values for each agonist, represented by E_max (maximal decrease in basal BRET) values, with the lowest efficacy agonists being tulobuterol and isoxsuprine and the highest being formoterol and Compound 26.

In order to investigate the mechanism underlying the differences in efficacy, we expressed and purified fluorescently labelled mini-G_s proteins from E. coli (Supplementary Figure 1); our aim was to probe the affinity and binding kinetics of Venus-mini-G_s protein for the agonist-bound β₂AR–nLuc complex using NanoBRET. Figure 2A shows that all 12 agonists recruited Venus-mini-G_s protein to β₂AR-nLuc in HEK cell membranes in a concentration-dependent manner, with varying E_max and pEC_5o values (Table 2). Moreover, Figure 2B reveals a strong correlation (R² = 0.80, p = 0.0001) between E_max values for mini-G_s recruitment and E_max values for G_s-CASE activation, further validating these assays as effective tools for investigating β₂AR–G_s interactions.

We established a kinetic NanoBRET binding assay to measure Venus-mini-G_s protein recruitment to β₂AR-nLuc in membrane preparations. To achieve this, we pre-incubated receptor-containing membranes with a saturating concentration (×100 EC₅₀) of each β₂AR agonist, as characterised above. The pre-incubated membranes were then added to a plate containing various concentrations of Venus-mini-G_s protein, and we measured the association between these two proteins using NanoBRET (Figure 3; Table 3). Both association and dissociation rates (k _on and k _off) of Venus-mini-G_s for agonist β₂AR-nLuc could be obtained by analysing the observed association kinetics (Table 3). These studies showed that the full agonists, isoprenaline (k _on = 3.00 ± 0.1 × 10⁵ M⁻¹ min⁻¹) and adrenaline (k _on = 3.06 ± 0.15 × 10⁵ M⁻¹ min⁻¹), induce faster recruitment of the mini-G_s protein than the partial agonists, ritodrine (k _on = 6.13 ± 0.75 × 10⁴ M⁻¹ min⁻¹) and isoxsuprine (k _on = 4.97 ± 0.29 × 10⁴ M⁻¹ min⁻¹). k _off values were similar for all ligands, with all values within the range of 0.0070–0.0113 min⁻¹. We also conducted these mini-G_s kinetics studies on β₂AR-nLuc extracted into DDM detergent micelles, using 6 of the 12 ligands (Supplementary Figure 2; Supplementary Table 1) and observed similar results.

To probe the binding affinity of the Venus mini-G_s protein to the agonist β₂AR–nLuc complex, we added ligands in excess (×100 reported pEC₅₀ determined in the mini-G_s recruitment assay, see above) and incubated with the membrane fraction expressing β₂AR-nLuc for 15 min prior to the addition of Venus-labelled mini-G_s (Figure 4). The resulting affinity (pK _d) values are summarised in Table 3, which ranged from 24 nM for the full agonist isoprenaline to 193 nM for the partial agonist isoxsuprine. These data also showed a difference in the maximum amount of mini-G_s protein (E_max) recruited over the limited concentration range studied (300–10 nM), with full agonists exhibiting higher recruitment compared to partial agonists.

Finally, we performed Pearson’s correlation analysis between both the association rates (k _on) and dissociation rates (k _off) and the affinity (pK_d) values for Venus-mini-Gs binding agonist β₂AR–nLuc complexes vs. agonist efficacy, comparing both G_s-CASE and mini-G_s assay E_max values (Figure 5). This analysis showed a strong correlation between ligand efficacy (E_max) measured in both assay formats and mini-G_s association rates (k _on) (R² = 0.78, p < 0.0001 and R²= 0.99, p < 0.0001 respectively; see Figures 5A, D) and between ligand efficacy (E_max) and mini-G_s affinity (pK_d) (R² 0.70, p = 0.0007 and R² = 0.93, p < 0.0001, respectively; see Figures 5C, F). This suggests that the differences in agonist efficacy can be explained by agonist β₂AR complexes’ ability to recruit the G_s protein. No correlation was observed between ligand efficacy (E_max) measured in either assay formats and mini-G_s dissociation rates (k _off) (R² = 0.06, p = 0.45 and R² = 0.16, p = 0.20, respectively; see Figures 5B, E).

We also performed this same correlation analysis between these mini-G_s kinetics values obtained in detergent micelles and G_s efficacy data obtained in the Gs-CASE assay and found a similar trend (Supplementary Figure 4).

Previous studies have suggested that for some GPCRs, there is a relationship between ligand efficacy and the dissociation rates of ligand binding (Sykes et al., 2009a; Guo et al., 2016). To investigate the correlations between ligand residence time and efficacy, we analyzed existing kinetic data. This analysis revealed a broad range of measured k _off values, with adrenaline exhibiting the fastest dissociation rate and Compound 26 showing the slowest.

The relationships between agonist efficacy, as determined by E_max values obtained from the Gs-CASE and mini-G_s recruitment assays, and literature ligand binding association (k _on) and dissociation rates (k _off) were determined using Pearson’s correlation analysis (see Figure 6). This analysis showed no statistically significant correlation between ligand k _off values and the efficacy values determined for 6 of the 12 β₂AR agonists. Moreover, we also conducted kinetic TR-FRET-based ligand binding studies on 6 of the 12 β₂AR agonists in detergent micelles (Supplementary Figure 5) and found no statistically significant correlation (R² = 0.26, p = 0.29) between relative ligand residence times (IC₅₀ 1 min/IC₅₀ equilibrium) and their efficacy (Supplementary Figure 6; Supplementary Table 2).