A biased allosteric modulator is a molecular glue for β₂AR dimerization

G-protein-coupled receptors (GPCRs) represent the largest superfamily of transmembrane (TM) proteins, exhibiting complex signalling behaviours. GPCRs can interact with G proteins or other effectors, such as arrestins and kinases, mediating G-protein-independent pathways and giving rise to diverse physiological responses^5,6. Over 800 human GPCRs have been identified and classified into five families⁷: A, B1, B2, C and F. Although family C glutamate-like receptors are known to be obligate dimers⁸, mediated primarily by Venus flytrap extracellular domains, dimerization of other families is not well described.

Over the past few decades, many studies have suggested that family A rhodopsin-like receptors can also form dimers^{9,10,11,12,13}, yet defining their physiologically relevant quaternary structures and functional roles has been a long quest in the field. Recent cryogenic electron microscopy (cryo-EM) studies on rhodopsin¹, apelin receptor (APJR)^2,14,15,16 and GPR3 (refs. ^3,17,18) have provided direct evidence for the dimerization of these family A GPCRs. In contrast to family C receptors, family A receptors typically do not have a large extracellular domain and therefore dimerize primarily through TM interactions¹⁴. Despite extensive studies on potential family A dimers, their physiological relevance is still debated. The development of ligands that stabilize family A receptor dimers remains a challenge^19,20, and it is difficult to reliably reproduce these dimers and to comprehensively study their functional, biochemical, and biophysical properties.

We previously identified a class of negative allosteric modulators (NAMs) selective for the β₂-adrenergic receptor (β₂AR), a family A GPCR, that exhibited biased inhibition of β-arrestin recruitment with minimal impact on stimulatory G protein (G_s)-mediated cAMP production in cell-based assays⁴. Through comprehensive structure–activity relationship studies, we developed AP-7-168 (hereafter, AP)—a compound that has high potency and biased effects⁴. Activation of G_s after coupling with β₂AR triggers the downstream cAMP signalling pathway, leading to many physiological effects, including the relaxation of airway smooth muscle, which is the core mechanism underlying the use of β₂AR agonists in treating airway diseases such as asthma and chronic obstructive pulmonary disease. However, after prolonged agonist stimulation, β₂AR undergoes desensitization through arrestin-mediated inhibition of G_s coupling and receptor endocytosis, resulting in a diminished bronchodilatory effect^21,22. Our ex vivo tissue assays demonstrated that AP effectively maintained agonist-induced β₂AR activation and the subsequent bronchodilation while significantly delaying receptor desensitization⁴, highlighting its potential clinical value. Despite these promising findings, the molecular mechanism by which AP modulates β₂AR function remains unclear, posing a considerable obstacle to the further development and optimization of AP as a therapeutic agent.

Here, using cryo-EM combined with various biochemical and biophysical approaches, we demonstrate that AP functions as a molecular glue by stabilizing a β₂AR homodimer, which in turn alters the downstream signalling behaviours. These findings reveal an allosteric modulation mechanism in β₂AR that may have broad implications in drug development for family A GPCRs.

As depicted in Fig. 1a, AP and its parent compound difluorophenyl quinazoline (DFPQ)⁴ have the pharmacophore characterized by a quinazoline ring with 2- and 4-amino substitutions. AP features a 3,4-difluorophenyl group at the 2-amino position, a cyclohexane substituent at the 4-amino position and a bromine atom at the 6-position of the quinazoline ring. AP exhibited a biased NAM effect in β₂AR-expressing cells, being approximately 1,000-fold more effective at inhibiting β-arrestin-2 recruitment versus cyclic adenosine monophosphate (cAMP) production after stimulation with full β-agonist isoproterenol (ISO) at a near-saturating dose (1 µM)²³ (Fig. 1b). To elucidate the molecular mechanism underlying this biased inhibition, we sought to determine the structure of β₂AR bound to AP.

We initially prepared the cryo-EM sample using detergent-solubilized and size-exclusion chromatography (SEC)-purified β₂AR incubated with AP in the presence of carazolol, a β-antagonist. Given the small size (<50 kDa) of a monomeric receptor, we included nanobody 60 (Nb60), which is specific to an inactive conformation of β₂AR²⁴. Notably, the extracted particles from the dataset revealed two predominant two-dimensional (2D) class averages, with approximately 70% of monomers and 30% of dimers (Extended Data Fig. 1a). Subsequent 3D classification and reconstruction showed the dimer consists of two parallel protomers (Extended Data Fig. 1). Owing to the larger size and the C₂ symmetry of the dimer, we obtained a cryo-EM map at a resolution of 2.9 Å (Extended Data Fig. 1a). The map revealed two ligand-like densities located at the dimer interface near the centre of the membrane (Extended Data Fig. 1a). This finding suggests that AP may stabilize β₂AR dimerization.

We also managed to reconstruct a 4.3 Å resolution map for monomeric β₂AR bound to Nb60 in the same dataset (Extended Data Fig. 1a). Despite the modest resolution, a distinct non-protein density, comparably strong to that of TM helices, was observed near the AP-binding site identified in the dimer structure (Extended Data Fig. 2a,b). This indicates that AP can bind to monomeric β₂AR at a similar position as in the dimer. Surface plasmon resonance (SPR) measurements further confirmed the direct binding of AP to the monomeric receptor with a dissociation constant (K_D) of approximately 30 μM (Extended Data Fig. 2c).

Encouraged by the initial dimeric structure in detergent, we sought to produce a homogeneous dimer sample and solve its structure in a native-like environment. Although incubation of AP with purified monomeric β₂AR in detergent did not result in efficient dimerization as assessed by SEC (Fig. 1c), we achieved almost complete β₂AR dimerization by pretreating resuspended cell membrane with AP for around 30 min at room temperature before detergent extraction (Fig. 1c and Methods). This highlights the critical role of a continuous fluid membrane in facilitating TM contacts mediated by AP and maintained in detergent after solubilization. We reconstituted the dimer into lipid nanodiscs and solved its structure in complex with the agonist BI-167107 (hereafter, BI), without nanobodies (Extended Data Fig. 1b), in an attempt to capture an active-like conformation. Cryo-EM analysis confirmed the presence of predominantly dimeric species, with no monomer detected in 2D classification (Extended Data Fig. 1b). The final cryo-EM map achieved a resolution of 2.5 Å, with a local resolution of around 2.2 Å near the AP-binding site (Fig. 1d,e, Extended Data Figs. 1b and 3 and Extended Data Table 1).

The high-resolution map enabled modelling of two AP molecules in the dimer, revealing a novel ligand binding mode (Fig. 1e). The two AP molecules are packed against each other through π–π and van der Waals interactions and fit in an allosteric pocket formed by TM3, TM4 and TM5 of two β₂AR protomers (Figs. 1e and 2a–c). The two AP molecules also form extensive hydrophobic and van der Waals interactions with a series of non-polar residues in TM3 (Cys125^3.44, Val126^3.45, Val129^3.48), TM4 (Met156^4.48, Val157^4.49) and TM5 (Val206^5.45, Val210^5.49, Ile214^5.53) (where the superscripted notation refers to Ballesteros–Weinstein numbering) (Fig. 2a–c). Moreover, the 2-amino group in AP forms hydrogen bond interactions with Glu122^3.41 in TM3 (Fig. 2a). These specific interactions in our structural model rationalize previously reported structure–activity relationship and mutagenesis data⁴. For example, removing the bromine atom reduced AP’s activity, probably due to the loss of interactions with Val206^5.45 (Fig. 2a). The difluorophenyl substitution forms optimal contacts with Val126^3.45 and Val129^3.48 (Fig. 2c), whereas larger halogen atoms reduce ligand potency⁴. Similarly, the V129^3.48L mutation or swapping TM3 with β₁-adrenergic receptor (β₁AR), whereby isoleucine replaces the valine residue, introduces steric clashes with the difluorophenyl group, therefore significantly reducing the potency of AP. Moreover, the E122^3.41W mutation also leads to an almost complete loss of response to AP⁴.

Notably, the binding site of AP on β₂AR closely overlaps with a previously identified binding site for an unbiased NAM, AS408 (ref. ²⁵) (Fig. 2d,e). Structural alignment of one of the β₂AR protomers in the AP-bound dimer with the β₂AR–AS408 complex (Protein Data Bank (PDB): 6OBA) reveals a highly similar overall conformation, with an overall Cα root mean squared deviation (r.m.s.d.) of only 0.53 Å (Fig. 2e). Both AP and AS408 possess a bromide substitution at the C6 position of the quinazoline ring (Fig. 2d,e), and this bromide is positioned in close proximity to Val206^5.46 in both structures (Fig. 2e). However, there are notable differences in their binding orientations. The quinazoline rings of AP and AS408 are nearly parallel, but they are flipped 180° relative to each other (Fig. 2e). Consequently, in the AS408-bound structure, the 4-amino group of AS408 forms a hydrogen bond with Glu122^3.41, while in the AP-bound structure, it is the 2-amino group of AP that interacts with Glu122^3.41 (Fig. 2e). The shift in hydrogen bonding partners stabilizes Glu122^3.41 in a distinct conformation (Fig. 2e), which may modulate the interactions of β₂AR with different transducers. Moreover, the cyclohexane group in AP, which is not present in AS408, has a crucial role in ligand packing and hydrophobic contacts with the adjacent β₂AR protomer. These differences may explain why AS408 lacks the ability to stabilize receptor dimerization and the biased NAM effect of AP²⁵.

The AP-mediated β₂AR dimer interface is primarily built by TM3, TM4 and TM5, along with the loop regions ECL2 and ICL2 (Fig. 2f). In addition to the AP-mediated interactions around the central binding pocket (Fig. 2f–h), there are extensive intermolecular interactions between the two β₂AR protomers from both extracellular and intracellular sides. As shown in Fig. 2f–h, in the extracellular side, Leu167^4.59 and Met171^4.63 in TM4 and Trp173^4.65 in ECL2 of one protomer form extensive hydrophobic and van der Waals interactions with Gln197^5.36, Ala198^5.37, Ile201^5.40 and Ile205^5.44 in TM5 of the other protomer. On the intracellular side, Lys149^4.41 in TM4 forms a hydrogen-bonding interaction with the conserved Tyr132^3.51 (of the DRY motif) in TM3 of the adjacent protomer. Moreover, Phe133^3.52 in TM3 and Phe139^34.51, Lys140^34.51, Leu144^34.56 and Leu145^34.57 in the ICL2 loop create an extensive hydrophobic network between the two protomers. Moreover, we observed four prominent elongated lipid-like densities at the dimer interface (Extended Data Fig. 3a; interfacial lipids). Two of these densities are positioned toward extracellular leaflet and are oriented perpendicular to the membrane plane, inserting between TM4/ECL2 on one protomer and TM5 on the other, where they make van der Waals contacts with Leu167^4.59, Met171^4.63, Trp173^4.65 and Ala198^5.37 (Fig. 2f,g); the other two densities are located in the intracellular leaflet and run almost parallel to the membrane plane, bridging hydrophobic contacts near TM3/ICL2 and the cytoplasmic end of TM4/TM5, packing against Phe133^3.52, Phe139^34.51, Leu144^34.56 and Leu145^34.57 (Fig. 2f,h), which further stabilize the dimer conformation of β₂AR.

The β₂AR dimer interface captured in our cryo-EM structures differs from the potential interface observed in the lipid cubic phase crystal structure of β₂AR²⁶ (Extended Data Fig. 4a). The AP-stabilized β₂AR dimer interface also contrasts with those in the APJR dimer^2,14,15,16 (Extended Data Fig. 4b), the rhodopsin dimer¹ (Extended Data Fig. 4c) and the GPR3 dimer^3,17,18 (Extended Data Fig. 4d), while it is similar to a potential β₁AR dimer interface in the lipid cubic phase crystal structure²⁷ (Extended Data Fig. 4e). As previous biophysical studies have shown that the β₂AR could form dimers and higher-order oligomers by itself^10,28, it is possible that the new interface in our structures represents a native yet transient one for β₂AR.

To test whether the AP-stabilized β₂AR dimer retains the same architecture in a continuous membrane, we used single-molecule fluorescence resonance energy transfer (smFRET) microscopy to examine the dimer conformation and stability in liposome. As shown in Fig. 3a, the TM helices in the dimer structure resolved by cryo-EM form a parallelogram-like arrangement when viewed perpendicular to the membrane plane, with two diagonals of different lengths. Fluorescent donor and acceptor dyes labelled on residues near different vertices should yield distinct inter-dye distances and corresponding FRET values. TM5-labelled samples exhibited a homogeneous Gaussian distribution of FRET values centred at around 0.9, while helix 8 (H8)-labelled samples showed a distribution centred at around 0.5 (Fig. 3a), with the higher FRET consistent with a shorter distance between the fluorophore pairs. These results are consistent with the cryo-EM structure and indicate the conformational stability of the dimer. To assess long-term stability, we measured ensemble FRET changes using dye-labelled samples in detergent (Extended Data Fig. 5a). When dimers were diluted into AP-free buffer, the relative FRET values remained nearly unchanged over 24 h (Extended Data Fig. 5b,c). Moreover, incubation with G_s or activated β-arrestin could not dissociate the dimer (Extended Data Fig. 5d).

To establish cellular relevance of β₂AR dimerization, we used two complementary approaches in cells. First, we analysed the size of β₂AR nanoclusters with or without AP treatment. GPCRs are known to form mesoscale nanoclusters in cells, which act as signalling hubs to locally concentrate effectors and enhance signalling efficiency^29,30. The molecular mechanisms underlying nanoclustering are poorly understood but probably involve weak but specific interactions among receptors (Extended Data Fig. 6a). We hypothesized that AP stabilizes the dimer as a single unit that has increased interaction sites, facilitating prolonged contacts and the assembly of larger receptor clusters (Extended Data Fig. 6a). To visualize nanocluster distribution on cell membranes, we labelled β₂AR with a gold nanoparticle (AuNP)-conjugated antibody and prepared cryo-EM grids with cytoplasmic content removed by unroofing. As shown in Extended Data Fig. 6b,c, cryo-EM imaging revealed distinct AuNP clusters rather than random scatters. Statistics of AuNPs in each cluster demonstrated that AP significantly increased nanocluster size (Fig. 3b) as a result of AP-stabilized dimerization of β₂AR. The resulting enlarged nanoclusters may amplify signalling outcomes compared with isolated monomers or dimers combined.

We also performed cell-based bioluminescence resonance energy transfer (BRET) assays to quantify β₂AR dimerization by measuring distance-dependent BRET signals between Rluc-tagged and GFP-tagged receptors (Fig. 3c and Supplementary Figs. 2–4). Cells co-expressing differently tagged β₂AR exhibited AP-dependent dimerization with a –log₁₀-transformed half-maximum effective concentration (pEC₅₀) of 5.93 ± 0.18 (Fig. 3d). Control experiments confirmed the specificity of this effect, showing no fluorescence interference from AP, no dimerization stabilized by ISO alone and no expression-level-dependent self-homodimerization (Supplementary Fig. 2a–c). To further validate dimerization specificity, we tested a β₂AR mutant (V129^3.48L) that disrupts the AP-binding pocket due to steric clashes. As expected, this mutation completely abolished AP-mediated dimerization (Fig. 3e). Similar effects were observed when swapping TM3 from β₁AR into β₂AR (Supplementary Fig. 2d) or assessing AP-mediated β₁AR dimerization (Supplementary Fig. 2e). Together, these results demonstrate that AP robustly stabilizes β₂AR dimerization in native cellular environments.

To understand the structural basis of the biased NAM activity of AP, we first compared our dimer structure with previously resolved monomeric β₂AR structures. One protomer in the dimer structure aligns well with the previous crystal structure of β₂AR in an inactive conformation (PDB: 2RH1), with an overall Cα r.m.s.d. of 0.53 Å (Fig. 4a). Notably, the structure of the dimer in a nanodisc, resolved without nanobodies or engineered fusions, exhibits a slightly different ICL2 conformation from the crystal structure with an Cα r.m.s.d. of 1.5 Å (Fig. 4a). Phe139^34.51 on ICL2 participates in dimer contact with the same residue on the opposing protomer (Fig. 2f,h), resulting in an outward shift away from the central cavity involved in G-protein engagement (Fig. 4b). This shift is accompanied by other conformational changes in ICL2 and the intracellular end of TM6 (Fig. 4b and Extended Data Fig. 8d): Tyr141^34.53 forms hydrogen bond and π–cation interactions with Arg131^3.50 in TM3, which normally forms a hydrogen bond with His269^6.31 in TM6; Lys263^6.25 at the intracellular end of TM6 appears to replace Lys140^34.52 on ICL2 in neutralizing the helical dipole of TM3 (Extended Data Fig. 3b). These local conformational changes provide a plausible structural route by which AP binding and dimer formation could modulate transducer coupling.

Structural analysis of β₂AR bound to its transducers provided additional insights into how the dimer achieves biased inhibition. Alignment of β₂AR–G_s ^{(ref. 31}), β₁AR–β-arrestin-1 (ref. ³²) (recently reported β₂AR–β-arrestin-1/2)³³ and NTSR1–GRK2 (ref. ³⁴) structures with one protomer in the dimer (Fig. 4c) revealed significant steric clashes between β-arrestin-1 or GRK2 and the intracellular side of the other protomer (MolProbity³⁵ clashscore, 21.05 and 11.87, respectively), preventing their engagement with the receptor. By contrast, docked G_s appeared to be accommodated (clashscore, 6.55), and the outward movements of TM5 and TM6 required for G-protein binding are not restricted by the dimer contacts. However, dimerized β₂AR adopts an inactive-like conformation. Structural comparison with the G_s-bound receptor (Extended Data Fig. 8a) revealed that local conformational changes that break dimer symmetry are needed to allow for the loop-to-helix transition in ICL2 (Extended Data Fig. 8b,c), necessary for engagement with G_s in the canonical state.

To further confirm the functional consequences of the AP-stabilized dimerization, we conducted a series of biochemical assays to investigate the in vitro activities of the β₂AR monomer and dimer in isolation. First, we demonstrated that the dimer could form a stable complex with nucleotide-free G_s. This complex eluted as a single peak in SEC (Supplementary Fig. 5a), and the peak fractions contained the bands for the G_s heterotrimer on SDS–PAGE (Supplementary Fig. 5b). From these gels, we estimated the relative ratio of β₂AR to Gα to be 2:1 in the dimer–G_s complex. The docking model (Fig. 4c) depicts a single G_s bound to the dimer and G_s bound to one receptor protomer in this orientation would disfavour a second G_s binding to the second protomer in the same orientation. Moreover, preliminary cryo-EM analysis of the dimer–G_s complex showed 2D class averages with one G_s density associated with an oval-shaped micelle housing one dimer (Supplementary Fig. 5c).

We then used biolayer interferometry (BLI) to measure the binding affinities and kinetics of guanosine diphosphate (GDP)-bound G_s to monomeric or dimeric β₂AR, with and without AP (Fig. 4d and Supplementary Fig. 5d–f). All of the samples exhibited micromolar-level affinity for G_s–GDP (Supplementary Table 1). Despite adopting an inactive conformation in the structural models, the β₂AR dimer retains the ability to engage G_s, suggesting that the dimeric β₂AR can adopt a G_s-favourable conformation. The dimer showed a modest (around threefold) decrease in affinity for G_s (K_D = 3.11 ± 0.89 μM) compared with the monomer (K_D = 0.96 ± 0.30 μM). AP at 10 μM caused a slight reduction in the monomer–G_s binding affinity (K_D = 1.71 ± 0.48 μM) (Supplementary Table 1). Notably, dimer–G_s binding exhibited both a slower association rate (k_on) and dissociation rate (k_off) than monomer–G_s binding (Supplementary Table 1), while AP’s effect on the monomer only slightly reduced k_on (1.4-fold, P = 0.055). These results suggest that AP partially inhibits recruitment of G_s by β₂AR, with a more pronounced effect (2.7-fold, P < 0.001) when tightly bound within the dimer. The lower k_off observed in dimer–G_s dissociation may reflect a cooperative mechanism, whereby G_s engagement with one protomer promotes a G_s-favourable conformation in the adjacent protomer, thereby increasing G_s residence time during dissociation.

We next evaluated the potency of G_s in stabilizing the active state of β₂AR by allosterically modulating the orthosteric pocket to reduce antagonist binding. We titrated G_s(GDP) into β₂AR pre-bound with saturated [³H]DHA, a β-antagonist, and quantified the remaining bound [³H]DHA. The results showed that G_s modulates the orthosteric pocket in both dimeric and monomeric β₂AR, irrespective of AP presence (Fig. 4e). This suggests that AP does not significantly hinder G_s engagement at equilibrium. We then assessed the effects of AP on the function of β₂AR as a guanine nucleotide exchange factor, which facilitates GDP–GTP exchange in G_s and accelerates GTP hydrolysis. Using the GTPase-Glo assay to monitor GTP turnover, we found that AP significantly slows down the GTP turnover rate of monomeric β₂AR (Fig. 4f) but reaches its maximal inhibition on the monomer only at higher concentrations (Supplementary Fig. 6c), consistent with its double-digit micromolar binding affinity for monomeric β₂AR (Extended Data Fig. 2c). By contrast, we observed much stronger inhibition of GTP turnover rate for the dimeric β₂AR within the assay time window (Fig. 4f). This probably stems from two factors: (1) the slower association rate k_on of G_s with dimeric β₂AR as observed by BLI (Supplementary Fig. 5f) and (2) AP’s biasing of β₂AR toward an inactive-like state, potentially introducing additional rate-limiting steps in a full productive cycle.

To test whether AP impedes β-arrestin core engagement with β₂AR, we performed competition radioligand binding assays measuring β-arrestin-induced positive allosteric modulation of agonist binding. Using the sortase ligation strategy, we attached the synthetic phosphorylated vasopressin receptor 2 (V₂R) peptide (V₂Rpp) to monomeric or dimeric β₂AR (Fig. 4g and Supplementary Fig. 7a), ensuring full tail engagement of β-arrestin, a prerequisite for efficient core engagement in β₂AR. Monomeric β₂AR in nanodiscs displayed an approximately sevenfold increase in ISO affinity in the presence of β-arrestin-1 (Supplementary Fig. 7b), consistent with previous reports^36,37. AP reduced this increase in affinity to about twofold in the monomer and abolished it in the dimer (Fig. 4g), suggesting AP partially hinders β-arrestin core interactions on monomers and completely prevents it in dimers. These data align with structural predictions whereby steric clashes between β-arrestin and the adjacent protomer in the dimer prevent core engagement.

Previously, we showed that DFPQ inhibits GRK-mediated phosphorylation of β₂AR, a prerequisite for β-arrestin recruitment, in cells⁴. To identify which species is responsible for this inhibition, we measured GRK5 activity towards purified dimeric or monomeric β₂AR. AP-bound β₂AR dimers exhibited near-complete inhibition of GRK5 phosphorylation in both detergent and nanodiscs (Fig. 4h and Supplementary Fig. 8a), whereas GRK5 activity toward monomer in nanodiscs remained unaffected by AP (Fig. 4h and Supplementary Fig. 8a). The slight reduction in GRK5 activity observed for monomer in detergent at high AP concentrations (Fig. 4i) probably reflects the inefficient dimerization of β₂AR in detergent as we previously observed (Fig. 1c). However, the dose-dependent inhibition observed for monomer reconstituted in bicelles (Fig. 4i) is a result of AP-stabilized dimerization in membrane. The relatively large diameter of typical bicelles (around 40 nm) compared to nanodiscs (about 12 nm) increases the likelihood of incorporation of multiple receptors and provides a fluid membrane environment conducive to dimer formation. AP-dependent inhibition of GRK5 activity was further corroborated by a BRET-based assay monitoring AP effect on direct GRK5 recruitment to the β₂AR (Supplementary Fig. 8b,c). These findings are consistent with structural analyses indicating that steric clashes in the dimer hinder GRK engagement, further supporting the role of the dimer in preventing β-arrestin recruitment.

Modulation of protein dynamics often underpins how ligands achieve biased effects in highly dynamic proteins such as GPCRs. To provide further insights into the mechanism of AP’s biased NAM effect, we examined how AP tunes the dynamic behaviours of β₂AR. We used smFRET microscopy to monitor conformational changes and double electron–electron resonance (DEER) spectroscopy to measure residue-level distance distributions. Using the established protocols in our laboratory^38,39, we placed fluorescent dyes for smFRET or spin labels for DEER near the intracellular ends of TM4 and TM6 on the minimal cysteine construct of β₂AR. This enabled us to monitor TM6 outward movement—a hallmark of GPCR activation. Increased distances between TM4 and TM6 sites correspond to high-to-low FRET state transitions in smFRET experiments. As shown in Fig. 5a and described in the Methods, we used a two-affinity-tag purification strategy to produce an AP-stabilized β₂AR heterodimer, whereby only one protomer contained double-cysteine mutants for dye labelling. The recorded traces from the TM4–TM6 sensors of β₂AR predominantly fall into two FRET states (Fig. 5b), representing an active and inactive state. In the absence of an agonist, β₂AR was primarily in the inactive state; however, monomers without AP exhibited a significantly higher fraction of active species than dimers, which showed almost no active population (Fig. 5c,d). Adding AP to monomers substantially reduced the active state population from 26.5 ± 2.9% to 11.2 ± 3.2% (Fig. 5d), whereas adding ISO to dimers did not increase active-state species (Fig. 5c). These results indicate that AP stabilizes an inactive conformation of TM6, and ISO alone fails to drive TM6 transition into an active state in the dimer. This is consistent with the observed enhancement of antagonist binding in the dimer (Supplementary Fig. 6a,b). After adding G_s to the dimer or the monomer with AP, β₂AR transitioned predominantly to an active state, although dimers retained a higher fraction of inactive species (Fig. 5c,d), consistent with G_s engaging only one protomer of the dimer, while the other may remain in inactive or partially active conformation. This indicates the potential of G_s to overcome the AP-stabilized inactive-like conformation, driving receptor activation even in the dimeric form. Washing G_s away from dimers reversed the population to a primarily inactive state, resembling the condition without G_s coupling (Fig. 5c). DEER measurements on the β₂AR (Supplementary Fig. 9), with spin labels reporting on the TM4–TM6 conformation (Fig. 5e), revealed that AP causes changes that are consistent with those observed by smFRET. Specifically, AP biases ISO-bound receptors toward a more homogeneous inactive-like conformation (Fig. 5f), while having a minimal effect on receptors bound to the super-agonist BI (Fig. 5g). The addition of G_s caused a fraction of the fully active conformer to become populated, regardless of AP presence (Fig. 5h). Together, these findings demonstrate that G_s can couple to both dimers and monomers with AP, driving TM6 outward movement. Moreover, the inhibitory effect of AP on the TM6 outward movement of β₂AR monomer may also contribute to the level of biased signalling observed.

Our cryo-EM structures reveal that AP, a β-arrestin-biased NAM, stabilizes β₂AR dimerization, thereby modulating downstream signalling. AP shows strong bias toward β-arrestin inhibition in cell-based assays in which both the cAMP and β-arrestin pathways can be maximally stimulated, and its parent compound DFPQ was previously shown to produce sustained relaxation of contracted airway smooth muscle in both human airway smooth muscle cells and mouse airway models by preventing β₂AR desensitization⁴, suggesting its physiological relevance. We elucidate the mechanistic basis underlying this biased effect: AP robustly stabilizes β₂AR dimerization. Biochemical and biophysical analyses demonstrate that the dimer acts as a biased signalling species, strongly inhibiting GRK-mediated phosphorylation and β-arrestin recruitment while permitting G-protein coupling. Structural analysis indicates that steric hindrance imposed by the adjacent protomer prevents GRK and β-arrestin from adopting their canonical binding modes. These findings offer a novel perspective for achieving signalling bias: besides fine-tuning local conformational states, AP exploits the quaternary structure of receptor to selectively restrict transducer interactions (Extended Data Fig. 9).

The dimer interface observed in our AP-bound β₂AR complex displays unique features compared with other family A GPCR dimers. The AP-stabilized dimer predominantly involves interactions on TM3, TM4, TM5 and ICL2, a configuration that is distinct from the earlier β₂AR crystal structure²⁶ and other family A GPCR dimers, suggesting that AP may selectively stabilize one β₂AR dimer conformation from a broader ensemble. Importantly, the outward mobility of TM5 and TM6 is not restricted in our dimer configuration, which is essential for efficient transducer engagement. Notably, recent work⁴⁰ on the platelet-activating factor receptor using a cysteine cross-linking strategy reported that dimerization significantly biases signalling toward G-protein pathways while limiting β-arrestin recruitment; one of their proposed dimer models resembles our structure with a TM3–TM4–TM5 interface. Similarly, recent studies used a computational and mutagenesis approach to design stable TM4–TM5 dimers of CXCR4, which biased signalling towards G_i ^{(ref. 41}). Structural modelling suggested that distinct dimer geometries can differentially permit transducer engagement, with a closed dimer arrangement being sterically constrained for productive β-arrestin engagement relative to a more open form⁴¹.

In summary, our study deciphers the structural basis by which AP acts as a biased NAM for β₂AR and provides broader insights into how receptor dimerization impacts GPCR signalling. Such insights not only deepen our understanding of structure–function relationships in GPCR modulation but also highlight potential strategies for designing next-generation ligands that exploit oligomerization as a route to bias receptor signalling.

Expression and purification of β₂AR in Sf9 cells

The β₂AR construct PN1 was expressed and purified as previously described^39,42. In brief, Sf9 cells were infected with a PN1-containing baculovirus produced using the BestBac method. Cells were then collected and resuspended in chilled lysis buffer containing 10 mM HEPES, pH 7.4, 1 mM EDTA, 1 μM alprenolol and protease inhibitors (leupeptin and benzamidine). Lysed cells were then pelleted at 18,600 rpm for 20 min and dounced to homogeneity in chilled solubilization buffer containing 20 mM HEPES pH 7.4, 350 mM NaCl, 1% n-dodecyl β-D-maltoside (DDM), 0.1% cholesteryl hemisuccinate (CHS), 2 mM MgCl₂, 1 μM alprenolol, protease inhibitors and benzonase. After stirring for 90 min at 4 °C and centrifugation at 18,600 rpm for 30 min, 2 mM CaCl₂ was added to the soluble fraction, which was then applied to anti-Flag (DYKDDDDK) M1 immunoaffinity resin. The receptor was then washed (20 mM HEPES, pH 7.4, 350 mM NaCl, 0.1% DDM, 0.01% CHS, 2 mM CaCl₂, leupeptin and benzamidine), eluted (20 mM HEPES, pH 7.4, 350 mM NaCl, 0.1% DDM, 0.01% CHS, 5 mM EDTA and 200 μg ml⁻¹ Flag peptide), and further purified on Superdex 200 10/300 Increase gel filtration column equilibrated in NH buffer (20 mM, HEPES pH 7.4, 100 mM NaCl) plus 0.1% DDM and 0.01% CHS.

To produce homogeneous β₂AR dimer, Sf9 cells expressing PN1 were resuspended at room temperature in 20 mM HEPES, pH 7.4, 150 mM NaCl, 10% glycerol, 1 μM alprenolol, protease inhibitors and 10 μM AP. After incubation for 30 min at room temperature, membrane solubilization was initiated by adding 1% lauryl maltose neopentyl glycol (LMNG) and 0.1% CHS. The purification steps followed the same protocol as for the monomeric receptor, except that 0.01% LMNG replaced DDM in all buffers, and 10 μM AP was maintained throughout the purification.

Expression and purification of heteromeric Gα_sβ₁γ₂

As previously described^43,44, heterotrimeric G_s was expressed and purified from Trichoplusia ni Hi5 cells. In brief, two baculoviruses were generated using the BestBac method, one encoding the wild-type human Gα_s subunit and the other encoding the wild-type human β_1γ2 subunits containing a histidine tag on the N terminus of the β subunit. Cells were infected with both viruses for 48 h and collected by centrifugation. The pellet was then resuspended and stirred for 30 min at 4 °C in hypotonic buffer containing 10 mM HEPES pH 7.4, 100 μM MgCl₂, 5 mM β-mercaptoethanol, 20 μM GDP and protease inhibitors. Lysed cells were then pelleted at 18,600 rpm for 15 min and dounced to homogeneity in chilled NH buffer plus 1% sodium cholate, 0.05% DDM, 1 mM MgCl₂, 5 mM β-mercaptoethanol, 20 μM GDP and protease inhibitors. After solubilization for 1.5 h while stirring at 4 °C and centrifugation at 18,600 rpm for 35 min, 20 mM of imidazole was added to the soluble fraction, which was then allowed to batch-bind to washed nickel-chelated Sepharose for 2 h. Pelleted resin was then loaded into a narrow column, washed with buffers containing gradually declining amounts of cholate and eluted with NH buffer plus 0.05% DDM, 1 mM MgCl₂, 20 μM GDP, 100 μM tris(2-carboxyethyl)phosphine (TCEP) and 250 mM imidazole. Human rhinovirus 3C protease was added to cleave the histidine tag and the eluate was dialysed overnight at 4 C in 2 l of dialysis buffer (NH buffer plus 1 mM MgCl₂, 0.05% DDM, 20 μM GDP and 100 μM TCEP). The protein solution was run through a second nickel-chelated Sepharose column, washed with dialysis buffer supplemented with 20 mM imidazole and dephosphorylated for 30 min on ice with lambda protein phosphatase, calf intestinal phosphatase and Antarctic phosphatase with 1 mM manganese chloride. The heterotrimer was further purified from excess βγ subunits using ion-exchange chromatography on the MonoQ 10/100 GL column. The sample was loaded and washed with 20 mM HEPES, pH 7.4, 1 mM MgCl₂, 0.05% DDM, 100 μM TCEP and 20 μM GDP. Heterotrimeric G_s was then eluted with a linear salt gradient from 50 mM NaCl to 500 mM NaCl.

Expression and purification of Nb60

Nb60 was expressed and purified as previously described²⁴. In brief, Nb60 was expressed in Escherichia coli BL21(DE3) cells. The lysates were then purified on a nickel-chelated Sepharose column and subsequently on the Superdex 200 Increase 10/300 column in 20 mM HEPES, pH  7.4, and 150 mM NaCl.

Sample preparation for cryo-EM

For the sample in detergent, purified PN1 in 0.1% DDM/0.01% CHS was loaded onto anti-Flag M1 immunoaffinity resin equilibrated in the same detergent mixture supplemented with 2 mM CaCl₂. The receptor was exchanged into a synthetic triglucoside-based detergent, TTG-T10 (ref. ⁴⁵), during which the receptor bound to the resin was washed with increasing ratios of TTG-T10 to DDM, ultimately transitioning the receptor into 0.01% TTG-T10/0.001% CHS. Each detergent exchange buffer contained NH buffer with 2 mM CaCl₂, 1 μM carazolol and 10 μM AP. The receptor was subsequently eluted in NH buffer plus 0.01% TTG-T10, 0.001% CHS, 1 μM carazolol, 10 μM AP, 5 mM EDTA and 200 μg ml⁻¹ Flag peptide. After the detergent exchange, PN1 was incubated for 1 h with twofold molar excess of Nb60. Excess Nb60 was cleared on anti-Flag M1 immunoaffinity resin after washing with NH buffer plus 0.0006% TTG-T10, 0.00006% CHS, 1 μM carazolol, 10 μM AP and 2 mM CaCl₂. The complex was then eluted off the resin with NH buffer plus 0.0006% TTG-T10, 0.00006% CHS, 1 μM carazolol, 10 μM AP, 5 mM EDTA and 200 μg ml⁻¹ Flag peptide. The final sample was concentrated to over 10 mg ml⁻¹ and used immediately for cryo-EM grid preparation.

To prepare the AP-bound β₂AR dimer in lipid nanodiscs, the purified dimer in LMNG was reconstituted into nanodiscs following a previous protocol⁴² with modifications. Lipids were prepared by mixing 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (POPC, Avanti), 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine (POPS, Avanti) and cholesterol (Sigma-Aldrich) at a molar ratio of 7:2:1, followed by drying under argon and vacuum desiccation for 2 h. Lipids were resuspended in NH buffer containing 14 mM DDM at 20 mg ml⁻¹. The receptor was diluted to 10 μM, incubated with 50 μM BI-167107 (PubChem CID: 45483813, MedChemExpress) for 10 min on ice, and then combined with membrane scaffold protein (MSP) 1E3D1 and lipids at a molar ratio of 1:2.5:100. The mixture was incubated on ice for 1 h to allow nanodisc assembly. Detergent was removed by sequential addition of semi-wet Bio-Beads SM2 (60 mg ml⁻¹) three times over 3 h period, followed by overnight incubation at 4 °C. The next day, the Bio-Beads were removed, and empty nanodiscs were separated by M1 affinity purification. The dimeric β₂AR in nanodiscs were eluted in buffer containing 1 μM BI, 10 μM AP, 5 mM EDTA and 200 μg ml⁻¹ Flag peptide. The eluate was further polished by SEC in NH buffer plus 1 μM BI and 10 μM AP. Peak fractions were pooled and concentrated to 5 mg ml⁻¹ for grid freezing.

The cryo-EM grids were prepared using Vitrobot Mark IV (Thermo Fisher Scientific). Quantifoil R1.2/1.3 Au grids were glow-discharged with air for 90 s at 10 mA using Plasma Cleaner (PELCO EasiGlow). Aliquots of 3 μl protein sample were applied to the glow-discharged grids. After blotting with filter paper (Ted Pella) for 3.0 s, the grids were plunged into liquid ethane cooled with liquid nitrogen.

Cryo-EM data collection and processing

For detergent- and nanodisc-reconstituted β₂AR samples, a total of 7,139 and 17,997 micrograph stacks, respectively, were collected using a Titan Krios G3i operating at 300 kV on a Falcon 4i direct electron detector or on a K3 camera (Gatan) with a Quantum energy filter. Micrographs were recorded at video stack received a total electron dose of about 50 e⁻ Å⁻² over 40 frames.

Motion correction and contrast transfer function (CTF) estimation were performed using the patch motion and patch CTF in cryoSPARC⁴⁶. Particle picking was carried out using the blob picker, and extracted particles were binned 4× for initial 2D classification. 2D classes with recognizable structural features were manually selected, and monomeric and dimeric particles were grouped separately. Ab initio reconstruction was performed using around 10% of the selected particles to generate four initial 3D references. Iterative rounds of heterogeneous refinement were conducted until poor-quality classes accounted for less than 5% of input particles. For the detergent dataset, 48,805 dimeric and 25,201 monomeric particles were selected for non-uniform (NU) refinement⁴⁷. For the nanodisc dataset, 237,408 particles of dimeric species were selected. Particles were imported into RELION 4 (refs. ^48,49) for Bayesian polishing⁵⁰ and subsequently returned to cryoSPARC for further NU refinement. C₂ symmetry was imposed for the final NU refinement of dimer species. After handedness correction, local and CTF refinement were performed using a soft mask centred on the TM domains. The final resolutions were estimated using the gold-standard Fourier shell correlation 0.143 criterion. Local resolution estimation was performed in cryoSPARC.

Model building and refinement

Initial models of β₂AR (PDB: 2RH1) were docked into cryo-EM maps using UCSF Chimera⁵¹ and manually adjusted in COOT⁵² to fit the density, including placement of ligands. Real-space refinement was performed in PHENIX^53,54 with secondary structure and geometry restraints. Model validation was conducted using EMRinger⁵⁵ to assess side-chain density fitting. Structural figures were prepared using PyMOL and ChimeraX⁵⁶.

Mass photometry

Mass photometry measurements were performed using a Refeyn TwoMP instrument (Refeyn) and the AcquireMP software (v.2.3) according to an established protocol⁵⁷. Microscope coverslips (24 × 50 mm, #1.5 thickness; Corning) were cleaned with deionized water and isopropanol, then dried before use. Silicone gaskets were applied to the coverslips to form individual wells immediately before sample loading. The instrument was calibrated using NativeMark unstained protein standards (Thermo Fisher Scientific) according to the manufacturer’s instructions. Each measurement was conducted by first pipetting 10 μl of NH buffer into a well, followed by focal alignment and locking. Then, 1 μl of β₂AR sample at a final concentration of 20 nM after dilution was added, mixed gently and data were acquired for 60 s. At least 2,000 binding events were recorded per sample. Data processing and molecular mass determination were performed using the DiscoverMP software suite (Refeyn).

Cell unroofing and immunogold cryo-EM

Cell unroofing and immunogold labelling were performed as previously described^58,59,60, with modifications. Quantifoil R1.2/1.3 300-mesh gold EM grids were glow-discharged for 15 s, rinsed three times with 70% ethanol and washed four times with Dulbecco’s PBS (DPBS). The grids were incubated with 0.1 mg ml⁻¹ poly-D-lysine (Gibco) for 1 h at room temperature, followed by four washes in DPBS (Gibco). Laminin (Sigma-Aldrich, 15 μg ml⁻¹) was applied to the grids and incubated at 37 °C for 2 h, then washed again with DPBS. HEK293F cells in suspension were seeded onto prepared grids and cultured until reaching 60–70% confluency. The HEK293 cell line was obtained from and authenticated by ATCC, and was not routinely tested for mycoplasma contamination. β₂AR was overexpressed through baculovirus transduction according to the BacMam protocol (Thermo Fisher Scientific). Around 16 h after transfection, cells were rinsed with DPBS containing calcium and magnesium. Grids bearing adherent cells were held with tweezers and dipped into a hypotonic swell buffer (6 mM HEPES-KOH pH 7.4, 43.3 mM K-gluconate, 1.6 mM NaCl, 0.6 mM MgCl₂) for 30 s. An additional 6 μl of swell buffer was added to each grid, followed by blotting with Whatman Grade 5 filter paper (Sigma-Aldrich) to remove the apical membrane. Unroofed samples were blocked with 3% goat serum (Thermo Fisher Scientific) in DPBS containing protease inhibitors (Thermo Fisher Scientific) for 20 min at room temperature. Primary antibodies against β₂AR (Thermo Fisher Scientific) diluted in DPBS with 1% goat serum was applied for 1 h. After three washes in DPBS, grids were incubated with a goat anti-mouse secondary antibody conjugated to colloidal gold (Ted Pella) diluted in DPBS with 1% goat serum for 1 h, then washed again three times.

Grids were plunge-frozen in liquid ethane using a Leica EM GP2 system and stored in liquid nitrogen. Cryo-EM imaging was performed on a Glacios G2 operated at 200 kV, equipped with a Falcon 4i detector, at a nominal magnification of ×11,000.

Cell-based BRET

To evaluate the effects of AP on cAMP production and β-arrestin recruitment, HEK293 cells endogenously expressing β₂AR were transiently transfected with either the BRET-based intramolecular cAMP sensor CAMYEL or with plasmids encoding β-arrestin-2–GFP10 and β₂AR–RlucII. CAMYEL comprises both donor and acceptor fused to the cAMP-binding domain of EPAC and undergoes a conformational change after cAMP binding that alters the BRET signal⁶¹. At 48 h after transfection, cells were pre-incubated with increasing concentrations of AP (0.03–100 μM) for 30 min, followed by stimulation with 1 μM ISO (Sigma-Aldrich) for 30 min in the presence of 5 μM coelenterazine H (for CAMYEL) or deep blue coelenterazine (DBC) (Cayman Chemical) for the β-arrestin assay.

For BRET-based β₂AR dimerization studies, HEK293 cells were co-transfected with 10 ng per well of either β₂AR–RlucII, β₂AR(V129L)–RlucII, β₂AR_TM3–RlucII or β₁AR–RlucII (BRET donors), along with 20 ng per well of their respective GFP-tagged counterparts (BRET acceptors). Then, 48 h after transfection, cells were incubated with increasing concentrations of AP (0.03–100 μM) or ISO (0.1–100 μM) for 1 h (or a different incubation time according to the experiment goal), followed by addition of DBC substrate and incubation for 20 min. To control for potential AP autofluorescence effects on the BRET signal, HEK293 cells transfected with β₂AR–GFP were treated with AP (0.03–100 μM) for 30 min, and the induced GFP signal was directly measured and represented as fold change over basal.

For saturation studies of β₂AR homodimerization, cells were co-transfected with a fixed amount of β₂AR–RlucII (10 ng per well) and increasing amounts of GFP-tagged β₂AR plasmid (0–100 ng per well). BRET measurements were taken at 48 h after transfection after incubation with DBC for 30 min.

To evaluate GRK5 recruitment to the β₂AR, cells were co-transfected with β₂AR–Rluc and GRK5–GFP. Cells were either stimulated with increasing concentrations of ISO for 30 min followed by DBC incubation for 20 min or pretreated with increasing concentrations of AP (0.03–100 μM) for 1 h followed by 1 μM ISO stimulation for 30 min in the presence of the Rluc substrate DBC.

All BRET signals were recorded at 395 nm (donor emission) and 510 or 530 nm (acceptor emission) using an Infinite F500 plate reader (Tecan). Induced BRET changes were calculated by subtracting the basal BRET signal (in the absence of ligand) from the BRET signal measured after stimulation.

For the AP-mediated cAMP and β-arrestin assays, as well as GRK5 recruitment, results are presented as percent of the ISO-alone response. For dimerization studies and ISO-induced GRK5 recruitment, data are presented as the percentage of the maximal BRET signal. All dose–response curves were fitted using the log[agonist/inhibitor] versus response (three parameters) or log[agonist/inhibitor] versus response (four parameters) function in GraphPad Prism. Data are shown as the mean ± s.e.m., n = 3 or 4.

SPR analysis on AP binding

SPR measurements were performed using the Biacore T200 system. Monomeric β₂AR was captured onto a high-affinity streptavidin (SA) sensor chip (Cytiva) through a biotinylated anti-Flag M2 antibody (Sigma-Aldrich), yielding a final response of about 1,500 resonance units. The running buffer consisted of 20 mM HEPES (pH 7.5), 100 mM NaCl and 0.01% LMNG. AP was injected at increasing concentrations ranging from 0.6 to 10 μM at a flow rate of 30 μl min⁻¹. Association and dissociation phases were recorded for each injection. All sensorgrams were processed using double-referencing to correct for non-specific binding and instrument artifacts. This was accomplished by subtracting both (1) the response obtained from a blank injection of running buffer over the active surface (to account for bulk refractive index changes and injection artifacts); and (2) the response from the compound injection over a reference surface without immobilized protein (to correct for nonspecific binding to the surface and matrix effects). Sensorgrams were processed using Biacore Evaluation Software, and binding curves were fit using a steady-state affinity model.

BLI on G_s binding

BLI measurements were conducted at 30 °C with continuous shaking at 1,000 rpm using an Octet RED384 system (FortéBio). SA biosensor tips (Sartorius) were coated with 10 nM biotinylated anti-Flag M1 fragment antigen-binding region (Fab) in the NH buffer with 0.01% LMNG and 0.001% CHS for 300 s. Flag-tagged β₂AR was then captured by incubating the tips in 100 nM β₂AR with 10 μM ISO for 600 s. After receptor immobilization, the biosensors were transferred into wells containing a concentration series of G_s (30 nM to 10 μM) in the binding buffer containing 10 μM GDP and 0.1% BSA (Sigma-Aldrich) for 180 s (association phase), followed by transfer into buffer-only wells for 300 s (dissociation phase). Control channels lacking either G_s or immobilized β₂AR were used for double-reference subtraction. Association and dissociation kinetics were fitted with a single-exponential model to derive apparent k_on and k_off. Equilibrium binding responses were used to determine the dissociation constant K_D.

Radioligand-binding assay

For saturation binding studies, 50–100 femtomoles of monomeric or dimeric β₂AR reconstituted in nanodiscs, following the protocol described in the sample preparation for cryo-EM, were incubated with increasing concentration of [³H]DHA at room temperature for 1 h in a buffer containing 20 mM HEPES, 100 mM NaCl and 0.5% BSA. Non-specific binding of the radioligand was determined by adding 10 μM alprenolol in the same reaction system. For monomeric β₂AR, the assay was performed with or without AP. For competition binding studies, monomeric or dimeric β₂AR reconstituted in nanodiscs were incubated with 1 nM [³H]DHA and increasing concentrations of ISO or GDP-bound G_s in the same buffer as saturation binding. Nanodiscs were separated from excess [³H]DHA on Whatman GF/B filters using a Brandel 96-well harvester. The bound radioligand was read on the liquid scintillation counter (MicroBeta Jet, PerkinElmer). Data were analysed by GraphPad Prism 10.

GTP Turnover

The GTPase GLO assay was performed using a modified GTPase-Glo assay from Promega as previously described^38,39. In brief, 100 nM of monomeric PN1, reconstituted in nanodiscs with MSP1E3D1 following the protocol described in the sample preparation for cryo-EM, was incubated for 1 h at room temperature with 20 μM ISO and a range of concentrations of AP in NH buffer plus 0.2% DMSO and 20 μM GTP. Simultaneously, a 1 μM stock of heterotrimeric G_s protein was prepared in a buffer consisting of NH buffer plus 0.04% DDM, 200 μM TCEP, 20 mM MgCl₂ and 20 μM GDP. Equal volumes of PN1 and G_s were then mixed and incubated for 60 min. The final reaction consisted of 50 nM of ligand-bound PN1 and 500 nM G_s in NH buffer plus 0.1% DMSO, 0.02% DDM, 100 μM TCEP, 10 μM MgCl₂, 10 μM GTP and 10 μM GDP. An equal volume of GTPase-Glo reagent in NH buffer plus 0.02% DDM and 5 μM ATP was then added and incubated for 30 min. Detection reagent was subsequently added and incubated for 10 min. Luminescence was detected using the MicroBeta counter. For the time-course assay, the experimental setup remained identical except that 200 nM of AP-mediated dimeric PN1 was included as a condition and the PN1–G_s reactions occurred for 30, 60, 90 and 120 min.

Ligation of V₂Rpp to receptor and β-arrestin competition radioligand binding

β₂AR constructs modified with a C-terminal sortase recognition sequence (LPETGHH inserted after residue 365) were expressed in Sf9 cells and purified as described above for monomeric and AP-stabilized dimeric receptors. Sortase-mediated ligation of synthetic V₂Rpp to receptor was performed as previously described^36,37. For ligation reaction, 10 μM purified receptor was incubated in NH buffer supplemented with 0.01% LMNG, 0.001% CHS and 5 mM CaCl₂ with 50 μM synthetic GGG–V₂Rpp peptide and 2 μM evolved sortase A pentamutant (eSrtA)⁶². The mixture was incubated overnight at 4 °C. Unreacted receptor and eSrtA (bearing the C-terminal His tag) was removed by binding to nickel-chelated Sepharose resins. Labelled monomeric or dimeric β₂AR–V₂Rpp samples were reconstituted into nanodiscs following the protocol described in the sample preparation for cryo-EM.

The equilibrium competition radioligand binding assays were performed with β₂AR–V₂Rpp in nanodiscs in the presence of 2 nM [³H]DHA, increasing concentrations of ISO and 1 μM C-tail-truncated β-arrestin-1(382), prepared as previously described⁶³. Then, 10 μM AP was added where applicable. After incubation at room temperature for 1 h, the samples were collected and the radioactivity was measured as described in the previous section to calculate the inhibitory constant values of ISO.

GRK5 radiometric phosphorylation assays

To evaluate the effect of β₂AR dimerization on receptor phosphorylation, β₂AR monomers or dimers (1 μM), purified in LMNG micelles or reconstituted in nanodiscs, following the protocol described in the sample preparation for cryo-EM, were incubated for 5 min at 30 °C with purified C-terminally Strep-tagged GRK5 (50 nM) in a reaction buffer containing 20 mM Tris-HCl, pH 7.4, 5 mM MgCl₂, 30 mM NaCl, 0.5 mM EDTA, 100 μM [γ³²P]ATP (1,000 to 2,000 cpm pmol⁻¹) and 25 μM BI. The β₂AR samples in LMNG micelles were additionally supplemented with 20 μM C8-PIP₂ to increase efficiency of β₂AR phosphorylation in detergent. To evaluate the effect of AP on β₂AR phosphorylation, purified β₂AR monomers (1 μM) in LMNG micelles were reconstituted into bicelles with PIP₂ (ref. ⁶⁴) and the AP concentration was varied from 0 μM to 24 μM. The reactions were quenched with SDS sample buffer, and the samples were separated by SDS–PAGE. Gels were stained with Coomassie blue (Sigma-Aldrich), dried, exposed to autoradiography film and ³²P-labelled proteins were excised and counted to determine the amount of phosphate transferred. The reaction rates were normalized to phosphorylation of the β₂AR monomers (β₂AR monomers and dimers) or to phosphorylation in the absence of AP (AP effect).

Sample preparation for fluorescence measurements

Site-specific fluorophore labelling of β₂AR was performed using engineered cysteine mutants on a minimal cysteine background (Δ6), as previously described^38,39. For smFRET experiments, β₂ARΔ6 constructs were cloned into the pcDNA-Zeo-tetO vector and transfected into Expi293 cells stably expressing the tetracycline repressor (Thermo Fisher Scientific, A14635). Transfections were carried out using the Expifectamine kit according to the manufacturer’s protocol. Then, 2 days after transfection, receptor expression was induced with 4 μg ml⁻¹ doxycycline and 5 mM sodium butyrate in the presence of 1 μM alprenolol. Cells were collected 40 h after induction and immediately processed for purification.

For studies on the β₂AR dimer in liposomes, single-cysteine mutants were introduced at TM5 (R228C) or H8 (I334C). Homogeneous dimers were expressed in Sf9 cells and purified as described above. Labelling was performed by incubating 10 μM purified receptor with a fivefold molar excess of a premixed maleimide-conjugated dye pair: DY549P1 (Dyomics) and Alexa Fluor 647 (Thermo Fisher Scientific) at a 1:1.5 ratio. The reaction was incubated for 30 min at room temperature and quenched with 5 mM L-cysteine. Excess dye was removed by SEC (Superdex 200 Increase 10/300) in 20 mM HEPES (pH 7.4), 150 mM NaCl and 0.01% LMNG/0.001% CHS. Labelled dimers were reconstituted into liposomes consisting of POPC, 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (Avanti), and cholesterol at a molar ratio of 6:3:1 using an established protocol⁶⁵.

For studies on TM6 dynamics, double-cysteine mutants (N148C on TM4 and L266C on TM6) were introduced. Monomeric receptor was expressed in Expi293 cells and purified according to the Sf9 purification procedure. For the dimer, Expi293 cells were co-transfected at a 1:1 plasmid ratio with constructs encoding an 8×His-tagged β₂ARΔ6 (no Flag tag) and a Flag-tagged β₂ARΔ6 carrying the N148C/L266C mutations. Heterodimers were isolated by tandem affinity purification. The clarified lysates were first incubated with nickel-chelated Sepharose resins and washed with buffer containing 20 mM imidazole. Proteins were eluted with 250 mM imidazole, then supplemented with 2 mM CaCl₂ and subjected to anti-Flag M1 immunoaffinity purification. This two-step procedure enriched for heterodimers containing only one protomer with the double-cysteine mutations. Fluorophore labelling was performed as described above for the single-cysteine-mutant dimer sample.

Ensemble fluorescence measurements

Ensemble FRET experiments were conducted on the Fluoromax 4C spectrofluorometer (Horiba Scientific) with excitation and emission slit widths set to 5 nm and 3 nm, respectively. Emission spectra were recorded after excitation at 532 nm. AP-bound β₂AR dimers labelled with donor and acceptor fluorophores (I334C H8 sensor) were diluted 1,000-fold in NH buffer plus 0.01% LMNG without AP to a final concentration of 1 nM. Fluorescence spectra were collected at 1, 5, 30, 120 min, and 16 and 24 h after dilution. All spectra were normalized to donor intensity. To assess the effects of transducer binding, samples were incubated with 100 μM ISO and (1) 10 μM G_s (in the presence of apyrase) or (2) 20 μM β-arrestin-1(382), together with V₂Rpp and Fab30, which stabilizes the active V₂Rpp-bound β-arrestin-1 conformation. β-arrestin-1(382) and Fab30 were prepared as previously described⁶³. The samples were incubated for 1 h in the dark to allow full equilibration before measurement. All of the experiments were performed in triplicate.

smFRET microscopy

Flow chambers for smFRET experiments were assembled using mPEG-passivated glass coverslips (VWR), doped with biotin-PEG16 (Laysan Bio), as described previously^38,39. Before use, coverslips were incubated with 1 mg ml⁻¹ NeutrAvidin (Thermo Fisher Scientific), followed by 10 nM biotinylated anti-Flag M1 Fab. Labelled β₂AR samples were diluted to 100–500 pM in NH buffer plus 2 mM CaCl₂ and added to the chambers. After achieving optimal surface density, unbound receptor was washed out using imaging buffer supplemented with 100 μM cyclooctatetraene (Sigma-Aldrich) and an oxygen scavenging system (1% D-glucose, 1 mg ml⁻¹ glucose oxidase, 0.04 mg ml⁻¹ catalase).

Fluorescence imaging was performed on a custom-built, objective-based TIRF microscope as reported previously⁶⁶. The setup is built on a Zeiss Axiovert S100 TV platform with a ×100, 1.45 NA oil-immersion objective (Zeiss). Donor and acceptor fluorophores were excited with 532 nm and 637 nm lasers (OBIS LS 150 mW and LX 140 mW, Coherent). Emissions were separated by a 652 nm dichroic beamsplitter (Semrock), filtered through 580/60 nm and 731/137 nm band-pass filters, and split using an OptoSplit II beamsplitter (Cairn Research) onto an EMCCD camera (iXon DU897E, Andor). Data acquisition was controlled by μManager using custom BeanShell scripts, and videos were recorded as stacked TIFFs in frame-transfer mode at 100 ms exposure per frame. The laser power was tuned to balance a high signal-to-noise ratio with photobleaching timescales of tens of seconds. Each slide typically yielded 10–20 videos per channel. All imaging was performed at room temperature.

Fluorescence traces were analysed using custom Python scripts. Donor and acceptor channels were aligned using registration images, and individual molecules were identified as local intensity maxima within a five-pixel neighbourhood. Donor-only spots were excluded. For each fluorophore pair, intensities were background-corrected using a local circular region (35-pixel diameter). Donor leakage into the acceptor channel (~7%) was subtracted.

Donor excitation was used to monitor emission for 80 s, followed by direct acceptor excitation for 1 s to confirm fluorophore identity. Traces were selected for analysis on the basis of the following criteria: (1) signal-to-noise ratio ≥ 5; (2) single-step acceptor photobleaching before donor bleaching; (3) γ factor between 0.5 and 2.5; (4) anticorrelated donor and acceptor intensity fluctuations; and (5) single-step donor bleaching, if present.

FRET efficiency (E) was calculated as E = I_a/(I_a + γI_d), where I_a and I_d are the background-corrected acceptor and donor intensities, respectively. γ-correction was applied as described previously. For each trace, FRET values were binned into 30 intervals across the range [–0.25, 1.25] and normalized to the total number of datapoints. Ensemble FRET histograms were generated by averaging the normalized histograms from individual molecules and fit to a two-Gaussian distribution model.

Sample preparation for DEER

For DEER measurements, β₂ARΔ6-N148C/L266C was expressed and purified as described above in Sf9 cells. To exchange detergent from 0.1% DDM/0.01% CHS to 0.01% (w/v) LMNG/0.001% CHS, the receptor was extensively washed with a progressive gradient of DDM:LMNG buffer. In parallel, while the receptor was bound to the resin, alprenolol was removed by washing with saturating concentrations of the low-affinity antagonist atenolol. Owing to the fast dissociation kinetics of atenolol from the β₂AR, subsequent washes with ligand-free buffer yielded unliganded β₂AR for spin labelling. The Flag eluted receptor was labelled with the spin label reagent 3-(2-iodoacetamido)-proxyl in the presence of 100 μM TCEP in buffer containing 20 mM HEPES, pH 7.4, 150 mM NaCl and 0.01% LMNG/0.001% CHS. Twentyfold molar excess of 3-(2-iodoacetamido)-proxyl was added to 10 μM β₂ARΔ6 receptor for 3 h at room temperature. After quenching of the reaction with 5 mM final L-cysteine, the receptor was separated from the excess spin label by SEC (Superdex 200 10/300) in SEC buffer (20 mM HEPES, pH 7.4, 150 mM NaCl and 0.01% LMNG/0.001% CHS) prepared with D₂O. The sample was concentrated using a 50 kDa concentrator to a concentration of >25 μM. D8-glycerol was added as a cryoprotectant to 25% (v/v). Then, 13 μl of sample was added to a borosilicate capillary (1.4 mm (inner diameter) × 1.7 mm (outer diameter); VitroCom) and flash-frozen in liquid nitrogen.

DEER spectroscopy

DEER experiments were conducted as previously described³⁹ at Q-band (~33.68 GHz) using the Bruker Elexsys 580 spectrometer equipped with a SpinJet AWG, EN5107D2 resonator, variable-temperature cryogen-free cooling system (ColdEdge Technologies) and the 300W TWT amplifier (Applied Systems Engineering). All measurements were performed at 50 K. Dipolar evolution data were acquired using a dead-time-free 4-pulse DEER sequence with gaussian pulses⁶⁷ and with 16-step phase cycling.

The experimental parameters used for DEER data collection were: π/2, π_obs and π_pump pulse lengths of 40 ns; a frequency offset (Δv) of 90 MHz; d1 = 250 ns; d2 = 5,150 ns; shot repetition time = 2000 μs; shots per point = 4; and integration window = 40 ns. The optimal microwave power (that is, pulse amplitude) for the π/2, π_obs and π_pump pulses were determined using transient nutation experiments, in which pulse amplitudes were adjusted to maximize the inversion of the Hahn echo⁶⁸. Pump pulses were applied to the maximum intensity of the field swept echo detected absorption spectrum. Observe pulses were applied at a frequency of 90 MHz lower than the pump pulses.

DEER data were processed using DeerAnalysis 2022 (ref. ⁶⁹), which uses two fitting routines: neural network analysis (DEERNet⁷⁰, Spinach revision 5662) and Tikhonov regularization (DeerLab v.0.9.1)⁷¹. The consensus fit represents the mean of both methods, with reported 95% confidence intervals also incorporating errors from both methods. Time traces were normalized to signal intensity at t = 0, and distance distributions were area normalized. Custom Python scripts were used for plotting the dipolar evolution time traces and the distance distributions.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Custom scripts used for data analysis are publicly available at GitHub (https://github.com/vandorp/stim1_paper.git).

1. Zhao, D. Y. et al. Cryo-EM structure of the native rhodopsin dimer in nanodiscs. J. Biol. Chem. 294, 14215–14230 (2019).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

2. Yue, Y. et al. Structural insight into apelin receptor-G protein stoichiometry. Nat. Struct. Mol. Biol. 29, 688–697 (2022).

   Article  CAS  PubMed  Google Scholar 

3. Chang, H. et al. Structural basis of oligomerization-modulated activation and autoinhibition of orphan receptor GPR3. Cell Rep. 44, 115478 (2025).

   Article  CAS  PubMed  Google Scholar 

4. Ippolito, M. et al. Identification of a β-arrestin-biased negative allosteric modulator for the β₂-adrenergic receptor. Proc. Natl Acad. Sci. USA 120, e2302668120 (2023).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

5. Hilger, D. The role of structural dynamics in GPCR-mediated signaling. FEBS J. 288, 2461–2489 (2021).

   Article  CAS  PubMed  Google Scholar 

6. Wootten, D., Christopoulos, A., Marti-Solano, M., Babu, M. M. & Sexton, P. M. Mechanisms of signalling and biased agonism in G protein-coupled receptors. Nat. Rev. Mol. Cell Biol. https://doi.org/10.1038/s41580-018-0049-3 (2018).

7. Fredriksson, R., Lagerstrom, M. C., Lundin, L. G. & Schioth, H. B. The G-protein-coupled receptors in the human genome form five main families. Phylogenetic analysis, paralogon groups, and fingerprints. Mol. Pharmacol. 63, 1256–1272 (2003).

   Article  CAS  PubMed  Google Scholar 

8. Ellaithy, A., Gonzalez-Maeso, J., Logothetis, D. A. & Levitz, J. Structural and biophysical mechanisms of class C G protein-coupled receptor function. Trends Biochem. Sci 45, 1049–1064 (2020).

   Article  CAS  PubMed  PubMed Central  Google Scholar 

9. Zeng, F. Y. & Wess, J. Molecular aspects of muscarinic receptor dimerization. Neuropsychopharmacology 23, S19–S31 (2000).

   Article  CAS  PubMed  Google Scholar 

10. Angers, S. et al. Detection of β₂ adrenergic receptor dimerization in living cells using bioluminescence resonance energy transfer (BRET). Proc. Natl Acad. Sci. USA 97, 3684–3689 (2000).

    ADS  CAS  PubMed  PubMed Central  Google Scholar 

11. Hiller, C., Kuhhorn, J. & Gmeiner, P. Class A G-protein-coupled receptor (GPCR) dimers and bivalent ligands. J. Med. Chem. 56, 6542–6559 (2013).

    Article  CAS  PubMed  Google Scholar 

12. Hebert, T. E. et al. A peptide derived from a β₂-adrenergic receptor transmembrane domain inhibits both receptor dimerization and activation. J. Biol. Chem. 271, 16384–16392 (1996).

    Article  CAS  PubMed  Google Scholar 

13. Cvejic, S. & Devi, L. A. Dimerization of the delta opioid receptor: implication for a role in receptor internalization. J. Biol. Chem. 272, 26959–26964 (1997).

    Article  CAS  PubMed  Google Scholar 

14. Yue, Y. et al. Structural insights into the regulation of monomeric and dimeric apelin receptor. Nat. Commun. 16, 310 (2025).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

15. Ji, S.-Y. et al. Dynamic monomer-dimer transition in ligand-induced apelin receptor activation. Nat. Commun. 17, 4820 (2026).

16. Yue, Y. et al. Mechanistic insights into the versatile stoichiometry and biased signaling of the apelin receptor-arrestin complex. Nat. Commun. 16, 7403 (2025).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

17. Qiu, Z. et al. Allosteric modulation of dimeric GPR3 by ligands in the dimerization interface. eLife 14, RP107185 (2025).

18. Chen, G. et al. Mechanism and function of GPR3 regulated by a negative allosteric modulator. Nat. Commun. 16, 7988 (2025).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

19. Huang, B. S., St Onge, C. M., Ma, H. G. & Zhang, Y. Design of bivalent ligands targeting putative GPCR dimers. Drug Discov. Today 26, 189–199 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

20. Ullmann, T. et al. Homobivalent dopamine D₂ receptor ligands modulate the dynamic equilibrium of D₂ monomers and homo- and heterodimers. ACS Chem. Biol. 16, 371–379 (2021).

    Article  CAS  PubMed  Google Scholar 

21. Bhagat, R., Kalra, S., Swystun, V. A. & Cockcroft, D. W. Rapid onset of tolerance to the bronchoprotective effect of salmeterol. Chest 108, 1235–1239 (1995).

    Article  CAS  PubMed  Google Scholar 

22. Newnham, D. M., Grove, A., McDevitt, D. G. & Lipworth, B. J. Subsensitivity of bronchodilator and systemic beta 2 adrenoceptor responses after regular twice daily treatment with eformoterol dry powder in asthmatic patients. Thorax 50, 497–504 (1995).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

23. De Pascali, F. et al. β2-Adrenoceptor agonist profiling reveals biased signalling phenotypes for the β₂-adrenoceptor with possible implications for the treatment of asthma. Br. J. Pharmacol. 179, 4692–4708 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

24. Rasmussen, S. G. et al. Structure of a nanobody-stabilized active state of the β₂ adrenoceptor. Nature 469, 175–180 (2011).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

25. Liu, X. et al. An allosteric modulator binds to a conformational hub in the β₂ adrenergic receptor. Nat. Chem. Biol. https://doi.org/10.1038/s41589-020-0549-2 (2020).

26. Cherezov, V. et al. High-resolution crystal structure of an engineered human β₂-adrenergic G protein-coupled receptor. Science 318, 1258–1265 (2007).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

27. Huang, J., Chen, S., Zhang, J. J. & Huang, X. Y. Crystal structure of oligomeric β₁-adrenergic G protein-coupled receptors in ligand-free basal state. Nat. Struct. Mol. Biol. 20, 419–425 (2013).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

28. Fung, J. J. et al. Ligand-regulated oligomerization of β₂-adrenoceptors in a model lipid bilayer. EMBO J. 28, 3315–3328 (2009).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

29. Sungkaworn, T. et al. Single-molecule imaging reveals receptor–G protein interactions at cell surface hot spots. Nature 550, 543–547 (2017).

    Article  ADS  CAS  PubMed  Google Scholar 

30. Calebiro, D. & Koszegi, Z. The subcellular dynamics of GPCR signaling. Mol. Cell. Endocrinol. 483, 24–30 (2019).

    Article  CAS  PubMed  Google Scholar 

31. Rasmussen, S. G. et al. Crystal structure of the β₂ adrenergic receptor–Gs protein complex. Nature 477, 549–555 (2011).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

32. Lee, Y. et al. Molecular basis of β-arrestin coupling to formoterol-bound β₁-adrenoceptor. Nature 583, 862–866 (2020).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

33. Zhai, R. et al. Distinct C-tail dynamics between β-arrestin isoforms in forming GPCR tail- and core-engaged complexes. Preprint at LangTaoSha Preprint Server https://doi.org/10.65215/LTSpreprints.2026.04.11.000183 (2026).

34. Duan, J. et al. GPCR activation and GRK2 assembly by a biased intracellular agonist. Nature 620, 676–681 (2023).

    Article  ADS  CAS  PubMed  Google Scholar 

35. Chen, V. B. et al. MolProbity: all-atom structure validation for macromolecular crystallography. Acta Crystallogr. D 66, 12–21 (2010).

    Article  ADS  CAS  PubMed  Google Scholar 

36. Staus, D. P. et al. Sortase ligation enables homogeneous GPCR phosphorylation to reveal diversity in β-arrestin coupling. Proc. Natl Acad. Sci. USA 115, 3834–3839 (2018).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

37. Li, Y. et al. Chemical synthesis of a full-length G-protein-coupled receptor β₂-adrenergic receptor with defined modification patterns at the C-terminus. J. Am. Chem. Soc. 143, 17566–17576 (2021).

    Article  ADS  CAS  PubMed  Google Scholar 

38. Gregorio, G. G. et al. Single-molecule analysis of ligand efficacy in β2AR-G-protein activation. Nature 547, 68–73 (2017).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

39. Casiraghi, M. et al. Structure and dynamics determine G protein coupling specificity at a class A GPCR. Sci. Adv. 11, eadq3971 (2025).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

40. Liu, J. et al. Biased signaling due to oligomerization of the G protein-coupled platelet-activating factor receptor. Nat. Commun. 13, 6365 (2022).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

41. Paradis, J. S. et al. Computationally designed GPCR quaternary structures bias signaling pathway activation. Nat. Commun. 13, 6826 (2022).

    Article  ADS  PubMed  PubMed Central  Google Scholar 

42. Strohman, M. J. et al. Local membrane charge regulates β₂ adrenergic receptor coupling to G_i3. Nat. Commun. 10, 2234 (2019).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

43. Hilger, D. et al. Structural insights into differences in G protein activation by family A and family B GPCRs. Science https://doi.org/10.1126/science.aba3373 (2020).

44. Krishna Kumar, K. et al. Negative allosteric modulation of the glucagon receptor by RAMP2. Cell 186, 1465–1477 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

45. Ghani, L. et al. Tris(hydroxymethyl)aminomethane linker-bearing triazine-based triglucosides for solubilization and stabilization of membrane proteins. Bioconjug. Chem. 34, 739–747 (2023).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

46. Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: algorithms for rapid unsupervised cryo-EM structure determination. Nat. Methods 14, 290–296 (2017).

    Article  CAS  PubMed  Google Scholar 

47. Punjani, A., Zhang, H. & Fleet, D. J. Non-uniform refinement: adaptive regularization improves single-particle cryo-EM reconstruction. Nat. Methods 17, 1214–1221 (2020).

    Article  CAS  PubMed  Google Scholar 

48. Scheres, S. H. RELION: implementation of a Bayesian approach to cryo-EM structure determination. J. Struct. Biol. 180, 519–530 (2012).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

49. Kimanius, D., Dong, L., Sharov, G., Nakane, T. & Scheres, S. H. W. New tools for automated cryo-EM single-particle analysis in RELION-4.0. Biochem. J. 478, 4169–4185 (2021).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

50. Zivanov, J., Nakane, T. & Scheres, S. H. W. A Bayesian approach to beam-induced motion correction in cryo-EM single-particle analysis. IUCrJ 6, 5–17 (2019).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

51. Pettersen, E. F. et al. UCSF Chimera—a visualization system for exploratory research and analysis. J. Comput. Chem. 25, 1605–1612 (2004).

    Article  ADS  CAS  PubMed  Google Scholar 

52. Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. Features and development of Coot. Acta Crystallogr. D 66, 486–501 (2010).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

53. Adams, P. D. et al. PHENIX: a comprehensive Python-based system for macromolecular structure solution. Acta Crystallogr. D 66, 213–221 (2010).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

54. Afonine, P. V. et al. Real-space refinement in PHENIX for cryo-EM and crystallography. Acta Crystallogr. D 74, 531–544 (2018).

    Article  ADS  CAS  Google Scholar 

55. Barad, B. A. et al. EMRinger: side chain-directed model and map validation for 3D cryo-electron microscopy. Nat. Methods 12, 943–946 (2015).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

56. Pettersen, E. F. et al. UCSF ChimeraX: structure visualization for researchers, educators, and developers. Protein Sci. 30, 70–82 (2021).

    Article  CAS  PubMed  Google Scholar 

57. Sonn-Segev, A. et al. Quantifying the heterogeneity of macromolecular machines by mass photometry. Nat. Commun. 11, 1772 (2020).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

58. Peitsch, C. F., Beckmann, S. & Zuber, B. iMEM: isolation of plasma membrane for cryoelectron microscopy. Structure 24, 2198–2206 (2016).

    Article  CAS  PubMed  Google Scholar 

59. Zhang, Y. et al. Higher-order transient membrane protein structures. Proc. Natl Acad. Sci. USA 122, e2421275121 (2025).

    Article  CAS  PubMed  Google Scholar 

60. Zhang, Y. & MacKinnon, R. Higher-order transient structures and the principle of dynamic connectivity in membrane signaling. Proc. Natl Acad. Sci. USA 122, e2421280121 (2025).

    Article  CAS  PubMed  Google Scholar 

61. Jiang, L. I. et al. Use of a cAMP BRET sensor to characterize a novel regulation of cAMP by the sphingosine 1-phosphate/G13 pathway. J. Biol. Chem. 282, 10576–10584 (2007).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

62. Chen, I., Dorr, B. M. & Liu, D. R. A general strategy for the evolution of bond-forming enzymes using yeast display. Proc. Natl Acad. Sci. USA 108, 11399–11404 (2011).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

63. Janetzko, J. et al. Membrane phosphoinositides regulate GPCR-β-arrestin complex assembly and dynamics. Cell 185, 4560–4573 e4519 (2022).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

64. Komolov, K. E. et al. Structural and functional analysis of a β₂-adrenergic receptor complex with GRK5. Cell 169, 407–421 e416 (2017).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

65. Goddard, A. D., Dijkman, P. M., Adamson, R. J., dos Reis, R. I. & Watts, A. Reconstitution of membrane proteins: a GPCR as an example. Methods Enzymol. 556, 405–424 (2015).

    Article  CAS  PubMed  Google Scholar 

66. van Dorp, S. et al. Conformational dynamics of auto-inhibition in the ER calcium sensor STIM1. eLife https://doi.org/10.7554/eLife.66194 (2021).

67. Teucher, M. & Bordignon, E. Improved signal fidelity in 4-pulse DEER with Gaussian pulses. J. Magn. Reson. 296, 103–111 (2018).

    Article  ADS  CAS  PubMed  Google Scholar 

68. Teucher, M. et al. Strategies to identify and suppress crosstalk signals in double electron-electron resonance (DEER) experiments with gadolinium^III and nitroxide spin-labeled compounds. Magn. Reson. 1, 285–299 (2020).

    Article  Google Scholar 

69. Jeschke, G. et al. DeerAnalysis2006—a comprehensive software package for analyzing pulsed ELDOR data. Appl. Magn. Reson. 30, 473–498 (2006).

    Article  CAS  Google Scholar 

70. Worswick, S. G., Spencer, J. A., Jeschke, G. & Kuprov, I. Deep neural network processing of DEER data. Sci. Adv. 4, eaat5218 (2018).

    Article  ADS  CAS  PubMed  PubMed Central  Google Scholar 

71. Fabregas Ibanez, L., Jeschke, G. & Stoll, S. DeerLab: a comprehensive software package for analyzing dipolar electron paramagnetic resonance spectroscopy data. Magn. Reson. 1, 209–224 (2020).

    Article  Google Scholar 

Download references

Author notes

1. These authors contributed equally: Jiemin Shen, Teja Nikhil Peddada

Authors and Affiliations

1. Department of Molecular and Cellular Physiology, Stanford University School of Medicine, Stanford, CA, USA

   Jiemin Shen, Teja Nikhil Peddada, Haoqing Wang, Jun Xu & Brian K. Kobilka

2. Department of Biochemistry and Molecular Biology, Sidney Kimmel Medical College, Thomas Jefferson University, Philadelphia, PA, USA

   Konstantin E. Komolov, Francesco De Pascali & Jeffrey L. Benovic

3. Department of Biophysics, Medical College of Wisconsin, Milwaukee, WI, USA

   Alexander M. Garces & Michael T. Lerch

4. Department of Bionano Engineering, Center for Bionano Intelligence Education and Research, Hanyang University, Ansan, South Korea

   Muhammad Ehsan & Pil Seok Chae

5. Department of Medical Neuroscience, SUSTech Homeostatic Medicine Institute, School of Medicine, Institute for Biology Electron Microscopy, Southern University of Science and Technology, Shenzhen, China

   Jun Xu

Contributions

K.E.K. and F.D.P. contributed equally. Conceptualization: J.S., T.N.P., J.X., J.L.B. and B.K.K. Methodology: J.S., T.N.P., J.X., K.E.K., F.D.P., A.M.G. and M.E. Investigation: J.S., T.N.P., J.X., K.E.K., F.D.P., A.M.G. and H.W. Data curation and formal analysis: J.S., T.N.P., J.X., K.E.K., F.D.P. and A.M.G. Writing (original draft): J.S., J.X. and T.N.P. Writing (review and editing): all of the authors. Supervision and funding acquisition: B.K.K., J.L.B., M.T.L. and P.S.C.

Corresponding authors

Correspondence to Jun Xu or Brian K. Kobilka.

Competing interests

A patent on the reported compounds was submitted by J.L.B. and others in 2022. B.K.K. is a cofounder of and consultant for ConfometRx. The other authors declare no competing interests.

Peer review information

Nature thanks the anonymous reviewers for their contribution to the peer review of this work. Peer reviewer reports are available.

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions