Selective positive allosteric modulation of β₂AR using antibody fragments targeting the third intracellular loop

Abstract

GPCR therapeutics primarily target structured cavities formed by the seven transmembrane α-helices. However, amino acid sequence conservation in structured regions, especially among receptor subtypes, limits target selectivity. Here, we leverage the sequence divergence of the third intracellular loop (ICL3) of the receptor fold to derive a selective positive allosteric modulator for the β₂ adrenergic receptor (β₂AR). We repurpose the variable regions of a previously reported monoclonal antibody (Mab5) as single-chain (ScFv5) and single-domain (VhhL5) antibody fragments that enhance the maximal second messenger cyclic AMP signaling response downstream of the receptor. Despite ScFv5 binding a segment of ICL3 distinct to β₂AR, it also non-selectively binds and modulates cAMP signaling downstream of β₁AR and β₃AR. We find that this lack of specificity stems from multiple redundant interactions that facilitate ScFv5-ICL3 binding. In contrast, VhhL5, derived from the Mab5 light chain, selectively modulates β₂AR signaling over the β₁/β₃ subtypes. Mechanistically, VhhL5 enhances agonist-stimulated β₂AR-G protein coupling through releasing autoinhibitory ICL3 conformational states. In parallel, VhhL5 decreases β₂AR internalization promoting greater ligand-induced accumulation of cAMP. Our study demonstrates proof-of-concept for selective allosteric modulation of a GPCR by targeting a sequence divergent loop region within the receptor fold.

G protein-coupled receptor (GPCR) therapeutics primarily target structured and sequence conserved domains limiting target selectivity. Here, authors demonstrate the mechanism for selective modulation of a GPCR through disordered regions.

Introduction

GPCRs are a superfamily of integral membrane proteins that are targets of many FDA-approved drug therapies¹. A major challenge in GPCR drug discovery efforts is selective targeting of individual receptor subtypes². This challenge stems from sequence and structural similarities in the orthosteric ligand binding site in the extracellular face of the receptor^1,2. To circumvent this issue, drug discovery efforts increasingly focus on less-conserved allosteric sites that can either act independently (allosteric agonizts/antagonists) or modulate endogenous ligand activity (allosteric modulators)³.

Orthosteric ligands and allosteric modulators typically bind structured cavities/pockets formed in the assembly of the 7 transmembrane (TM) α-helical bundle of the receptor^1,4. In contrast, TM domains are linked by loop regions that often have subtype-specific amino acid sequences. The third intracellular loop (ICL3) is one such loop region with demonstrated roles in GPCR function, including modulating receptor-G protein selectivity^5,6, receptor autoregulation^6,7, G protein engagement⁸, and interacting with other cytosolic regions within the receptor⁹. These features of ICL3 make it a potential target for allosteric modulators^6,7,10,11. However, the inherent conformational heterogeneity of ICL3 limits the use of structure-based small-molecule discovery approaches to target this region. Here, we demonstrate a proof-of-concept to selectively modulate the β₂ adrenergic receptor (β₂AR) through its ICL3 using antibody fragment engineering.

Structural characterization of GPCRs has been facilitated by the use of antibody fragments that stabilize distinct inactive, intermediate and active conformational states of the receptor^12–14. Antibody fragments are able to stabilize these states through their larger receptor binding interface relative to small molecules or peptides¹⁵. While the most commonly observed antibody fragment interactions with GPCRs occur through interfaces involving the TM bundle, there are reported interactions with disordered GPCR domains, including intracellular loops^14,16,17. Antibody fragments that interface with ICL3 either occlude the effector binding site or stabilize inactive receptor conformational states, thereby serving as negative allosteric modulators (NAMs) of GPCR signaling^16,18,19. Here, we repurpose the variable regions of a previously reported monoclonal antibody (Mab5)²⁰ that interacts with the β₂AR ICL3 as intracellular targeted antibody fragments (intrabodies)¹².

In this study, we derive a single-chain variable fragment (ScFv5) and a single-domain antibody (VhhL5) that share complementarity-determining regions (CDRs) with the original Mab5²⁰. We find that ScFv5 interacts with all three β_xAR subtypes and subsequently enhances signaling downstream of each of them. By contrast, VhhL5, derived from the Mab5 light chain, acts as a selective positive allosteric modulator (PAM) of β₂AR. Our data demonstrate that the VhhL5 acts through disrupting receptor autoregulation and consequently enhancing receptor-G protein coupling and signaling.

Results

Positive allosteric modulation of β-adrenergic receptor signaling using a repurposed antibody fragment

The initial effort to solve the structure of the β₂AR relied on a monoclonal antibody, Mab5, that stabilizes the receptor’s ICL3 and surrounding TM domain^12,20. The first β₂AR crystal structure utilizes an antigen-binding fragment of Mab5, Fab5, that interacts with β₂AR without influencing agonist binding nor imparting conformational changes in the TM domain that accompany receptor activation²⁰. Despite this, Fab5 prevents G protein coupling to the receptor, presumably through steric competition. We hypothesized that in the absence of the constant regions of the Fab fragment, interactions between the variable chains of Fab5 and β₂AR will not hinder G protein accessibility to its binding site within the receptor. Hence, we attempted to minimize steric hindrance by creating a single-chain variable fragment derived from Mab5 (ScFv5)^21,22. ScFv5 was linked individually to β_xAR subtypes (Fig. 1a).

Fig. 1. β-adrenergic receptor (β_xAR) ICL3 targeted antibody fragment (ScFv5) augments agonist-stimulated cAMP accumulation.

a Schematic of β_xAR-ScFv fusion construct. ScFv5 is fused to the C-terminus of β_xAR via a SPASM linker²³ consisting of an mCitrine (mCit) FRET acceptor, an ER/K α-helical linker, and mCerulean (mCer) FRET donor. b Ratiometric FRET measurement of the interaction strength for the fusion of 4xGSG control (-) and ScFv5 to β_xAR subtypes. c–e Isoproterenol (Iso)-stimulated cAMP accumulation downstream of β₂AR (c), β₁AR (d), β₃AR (e) fused to ScFv5 (red) or a 4xGSG control ((-), black). Symbols in (b) represent individual biological replicates, and bars represent mean ± SD. Two-tailed paired t test for the effect of the fusion of ScFv5 to each β_xAR 4xGSG control. c–e are represented as the mean ± SD, replicates for the dose responses n = 3 (c), n = 3 (d), n = 4 (e). Expression quantification and comparative maximal cAMP response (Emax) and EC₅₀ data are provided in Supplementary Fig. 2.

We used a linker topology termed systematic protein affinity strength modulation (SPASM) sensor, which enables high local concentrations (~ 1–10 μM) of a bimolecular interacting pair with 1:1 stoichiometry²³. The GPCR SPASM sensor consists of a single polypeptide chain, (N-to-C terminus): GPCR - SPASM module - antibody fragment²⁴. The SPASM module consists of a FRET donor/acceptor pair (mCerulean/mCitrine) flanking an α-helical ER/K linker²⁵. The distinct biophysical properties of the ER/K α-helix enable ratiometric FRET measurements of the relative interaction strengths between the proteins flanking the SPASM module²³. The SPASM sensor involves a C-terminal fusion to the receptor, which has been shown to be important for regulating β₂AR activation⁹. We found that at matching surface expression (Supplementary Fig. 1a, b), C-terminal fusion to β₂AR does not influence signaling Emax (Supplementary Fig. 1c, d) but modestly increases EC₅₀ (Supplementary Fig. 1c, e).

All three β_xAR subtypes interact with ScFv5 as measured by FRET ratio (Fig. 1b). We used concentration-response curves to examine the effect of ScFv5 fusion on the signaling response of the three β_xAR subtypes (Fig. 1c–e and Supplementary Fig. 2a–c). ScFv5 fusions enhance the maximal cAMP response (Emax) for all three subtypes 18% β₁AR, 25% β₂AR, and 21% β₃AR; without impacting the potency (EC₅₀) of β₁AR and β₃AR. A small decrease in EC₅₀ is observed for β₂AR-ScFv5 fusion (1.6 nM) relative to the control (0.7 nM) (Supplementary Fig. 2b). The observed response is distinct to ScFv5, as fusion of a previously reported ScFv (G6) that binds vascular endothelial growth factor has a reduced interaction with β₂AR, and no effect on β₂AR signaling (Supplementary Fig. 2 d–f)²⁶. Together, these data demonstrate that ScFv5 serves as a non-selective positive allosteric modulator (PAM) for β_xAR subtypes.

Multiple redundant interactions underline allosteric activity of ScFv5

The structurally determined β₂AR-Fab5 interface contains intermolecular interactions involving both ICL3 and the cytosolic ends of TM5 and TM6¹². While the ICL3 sequence is distinct among β_xAR subtypes, residues in TM5 and TM6 are partially conserved, suggesting these interactions drive the observed non-selective modulation of the three receptor subtypes by ScFv5 (Fig. 2a). We postulated that by determining the necessary interactions in this broad interaction interface, we could derive a more β₂AR selective modulator. To this end, we performed alanine scanning mutagenesis of the heavy and light chains of ScFv5 (Fig. 2b). In the context of a SPASM sensor, none of the alanine scanning mutants we tested had a significant effect on either interaction strength (FRET ratio) (Fig. 2c) or cAMP at saturating concentrations of agonist relative to the wild-type (Fig. 2d), suggesting that ScFv5 modulation stems from multiple redundant interactions between the ScFv and the receptor.

Fig. 2. Heavy- and light- chain complementarity determining regions (CDRs) are independently sufficient to enhance β₂AR Iso-induced cAMP accumulation.

a Sequence alignment of the TM5-ICL3-TM6 segment of β_xAR. Shaded residues in the β₂AR ICL3 sequence form contacts (≤ 4 Å) with the light chain (green) and heavy chain (purple) of Fab5 (PDB: 2R4S). β₁AR and β₃AR residues with amino acid properties similar to β₂AR at matching positions are highlighted in comparison to the rest of the sequence. b Schematic for the residue and location of the alanine mutants in the light chain (green, LC) and heavy chain (purple, HC) of the ScFv5. c Ratiometric FRET measurement of the interaction strength for β₂AR-SPASM sensor fusions with 4xGSG control (n = 13), ScFv5 (n = 12), and ScFv5 alanine mutants(LC-1 n = 4, LC-2 n = 4, LC-3 n = 3, LC-4 n = 3, LC-5 n = 3, HC-1 n = 4, HC-2 n = 3, HC-3 n = 3, HC-4 n = 3, HC-5 n = 3). d Iso-stimulated cAMP accumulation for β₂AR-SPASM-ScFv5 and alanine scanning mutants. Data are relative to β₂AR-SPASM-4xGSG control 4xGSG control (n = 13), ScFv5 (n = 12), and ScFv5 alanine mutants(LC-1 n = 4, LC-2 n = 4, LC-3 n = 3, LC-4 n = 3, LC-5 n = 3, HC-1 n = 4, HC-2 n = 3, HC-3 n = 3, HC-4 n = 3, HC-5 n = 3). e Grafting of Fab5 CDRs onto a single-domain antibody (Vhh) scaffold. f Ratiometric FRET measurement of the interaction strength for β₂AR fused to 4xGSG control (n = 5), ScFv5 (n = 4), a heavy chain Vhh (VhhH5) (n = 4), or a light chain Vhh (VhhL5) (n = 4). g Iso induced cAMP accumulation normalized to β₂AR-SPASM-(-) for β₂AR-SPASM sensors fused to 4xGSG control (-) (n = 5), ScFv5 (n = 5), VhhH5 (n = 4), or VhhL5 (n = 4). Each symbol represents a biological replicate, and bars show mean ± SD, one-way ANOVA comparing each condition to: ScFv5 WT (c, d, f) and (-) (g). Expression quantification for (d) and (g) are provided in Supplementary Fig. 4.

Based on a 4 Å cut off, the light chain of Fab5 interacts with the most N-terminal region of the ICL3, while the heavy chain interface involves both N- and C- terminal residues of ICL3 that are resolved in the β₂AR-Fab5 complex structure (PDB: 2R4S). We speculated that presenting either the heavy or light interface in isolation would be sufficient for binding and modulating β₂AR. To test this, we created single-domain antibodies (Vhh) by grafting CDRs from ScFv5 heavy and light chains onto a camelid heavy chain framework (VhhH5 and VhhL5, respectively; Fig. 2e)²⁷. VhhL5 showed a stronger interaction with β₂AR compared to VhhH5 (Fig. 2f), albeit they show similar levels of cAMP responses to saturating concentrations of agonist (Fig. 2g). Both VhhL5 and VhhH5 show enhanced cAMP response comparable to the ScFv5. This finding is consistent with the premise that single-domain antibodies with grafted CDRs can retain binding function from the parent antibody²⁷.

Subtype-selective modulation of β2-AR activity by VhhL5

We measured the subtype selectivity of VhhL5 using β_xAR SPASM sensors (Fig. 3a). VhhL5 showed similar increases in FRET for all three β_xAR subtypes compared to their respective controls (Fig. 3b). While β₂AR showed a 53.5% increase in maximal response (Emax) in the presence of VhhL5 (Fig. 3c), no significant change was observed for β₁AR (3.7%) (Fig. 3d) or β₃AR (14.4%) (Fig. 3e). There was no significant change in EC₅₀ of β₂AR (Supplementary Fig. 5a) and β₃AR (Supplementary Fig. 5b) by fusion of VhhL5. On the other hand, a slight increase in potency was observed for β₁AR-VhhL5 (2.9 nM) compared to the control (6.2 nM) (Supplementary Fig. 3c). To test the specificity of β₂AR’s response to VhhL5, we tested a previously reported single-domain antibody (Nb6) that binds the ICL3 of Kappa Opioid Receptor (KOR)²⁸. KOR and β₂AR ICL3s do not share a conserved sequence (Supplementary Fig. 5d). β₂AR-Nb6 fusion has an increased FRET ratio compared to the control but is lower than the β₂AR-VhhL5 fusion (Supplementary Fig. 5e). Furthermore, the cAMP response elicited by VhhL5 appears to be specific to this antibody fragment, as Nb6 fusion has no effect on β₂AR signaling (Supplementary Fig. 3f–h).

Fig. 3. VhhL5 selectively enhances β₂AR signaling.

a Schematic of β_xAR SPASM sensor fused to VhhL5. b Ratiometric FRET measurement of the interaction strength for the fusion of 4xGSG control (-) and VhhL5 to β_xAR subtypes. β₁AR: (-) n = 12, VhhL5 n = 8; β₂AR: (-) n = 8, VhhL5 n = 9; β₃AR: (-) n = 11, VhhL5 n = 12. c–e Iso-dependent cAMP accumulation downstream of β₂AR (c), β₁AR (d), β₃AR (e) fused to VhhL5 (green) or 4 x GSG control ((-), black). Symbols in (b) represent individual biological replicates, and bars represent mean ± SD. Two-tailed paired t test for the effect of the fusion of VhhL5 to each β_xAR 4xGSG control fusion. Data in (c–e) are mean ± SD of n = 4 for all curves. Dose responses are normalized to the maximum (-) response. Expression quantification and comparative Emax and EC₅₀ data are provided in Supplementary Fig. 5.

We found that VhhL5 was poorly compatible with the recombinant expression and purification strategy used for other Vhh (Supplementary Fig. 6a). To optimize VhhL5 for recombinant expression and purification, we derived Vhhv3.1 that contains sub-regions of CDRs 1 and 3 of the Fab5 light chain that are in proximity to the ICL3 (PDB ID: 2R4S). While soluble purification was observed for Vhhv3.1 (Supplementary Fig. 6b), it did not interact with β₂AR (Supplementary Fig. 6c) nor did it modulate cAMP signaling (Supplementary Fig. 6d) in the context of a SPASM sensor. Thus, Vhhv3.1 was used as a negative control in further experimentation throughout this study. In our next iteration of optimization, we re-incorporated the CDR2 of Fab5 light chain. Vhhv4A maintained soluble purification (Supplementary Fig. 6e), appeared to interact with β₂AR in the context of a SPASM sensor (Supplementary Fig. 6f), and modulated cAMP signaling (Supplementary Fig. 6g). While these data demonstrate that Vhhv4A recapitulates receptor modulation of VhhL5, modulation demonstrated to be distinct to the β₂AR ICL3, they also highlight aspects of VhhL5 that cannot be explained by previous structural data, to the best of our knowledge, and merit further investigation into the interaction.

Structural Models of β₂AR bound to VhhL5

To identify the binding site of VhhL5, we performed scanning deletion mutagenesis of the β₂AR ICL3 (Fig. 4a). Deletion of the N-terminal ICL3 regions Ndel1 significantly decreased the FRET ratio, with no measurable effects for Ndel2, Middel and Cdel1 deletions (Fig. 4b). This indicates that the N-terminal region of ICL3 is the most important site for the VhhL5 interaction.

Fig. 4. VhhL5 targets the most N terminal part of the β₂AR’s ICL3.

a SPASM sensor used to test the effect of deletions in the ICL3 sequence on VhhL5 binding to β₂AR. b FRET intensity for SPASM fusion of wild-type β₂AR with (-) 4 x GSG (n = 7), Vhhv3.1 (n = 5), and VhhL5 (n = 10) compared to the effect of ICL3 scanning deletion mutagenesis on FRET intensity of the SPASM sensor (Ndel1 n = 7, Ndel2 n = 7, Middel n = 9, Cdel1 n = 9). Bars are color-coded based on the residues shown in (a). Each symbol represents a biological replicate, and bars show mean ± SD, one-way ANOVA comparing each condition to: WT (VhhL5) condition. c Steps in the computational modeling of β₂AR-full-ICL3-VhhL5 complex. d The sum of contact frequencies for VhhL5 on ICL3 residues from the active state model MD. e Heatmap of the sum of contact frequencies between the intracellular loops and VhhL5 CDRs from the active state model MD. f The sum of contact frequencies for VhhL5 on ICL3 residues from the inactive state model MD. g Heatmap of the sum of contact frequencies between the intracellular loops and VhhL5 CDRs from the inactive state model MD. Color scheme of (d) and (f) matches the different deletion blocks shown in (a).

While this interaction region is consistent with the interface of the Fab5 light chain (PDB ID: 2R4S), the differences in the framework region of VhhL5 and the Fab5 light chain, as well as the intrinsically disordered character of ICL3⁶, merited a further investigation into the VhhL5- β₂AR ICL3 interaction using a molecular modeling approach. To create initial binding poses, we used AlphaFold3 (AF3)²⁹ to generate structural models of β₂AR-Vhh complexes. Given the limited structural data on ICL3, we limited our AF3 model generation to a partial ICL3 sequence that just included the ends proximal of the transmembrane helices that are resolved in high-resolution structures (Fig. 4c). Principal Component Analysis (PCA) of all structural models identified five conformational clusters of interactions (Supplementary Fig. 7ac, b). Two of the five clusters show a greater proportion of models generated with a partial ICL3, including ICL3 residues resolved in high resolution structures, versus models generated in the absence of ICL3 (cluster 0 and 3; Supplementary Fig. 7c). Notably, in these two clusters, models with partial ICL3 were more abundant for VhhL5 than for Vhhv3.1 (Supplementary Fig. 7de), indicating that clusters 0 and 3 most likely represent specific VhhL5-ICL3 binding interactions (Supplementary Fig. 7e–g). As an alternative strategy, we generated β₂AR-VhhL5 models in the presence of miniGs. PCA identified 3 representative poses of VhhL5 interaction with the β₂AR ICL3 N-terminus (Supplementary Fig. 7h, i). From these 5 representative poses across the two AF3 modeling strategies, we incorporated representative conformations of the full ICL3 derived from our previously reported all-atom explicit MD simulation data set for the β₂AR ICL3⁶. After incorporating the full ICL3, we ended with four putative models of the VhhL5-receptor interaction to initiate molecular simulations (Supplementary Fig. 8).

Coarse grained MD (CGMD) simulations in membrane-mimicking multi-lipid bilayers strengthened the residue contacts between ICL3 and VhhL5 across all four starting complexes, attesting to the formation of dynamic complex (Fig. 4d–g and Supplementary Fig. 8). Three of the four models demonstrate persistent interactions with the ICL3 N-terminus (red and/or blue regions) (Fig. 4d–g and Supplementary Fig. 8b), consistent with the effects of Ndel1 in the β₂AR-VhhL5 interaction (Fig. 4b). The fourth model, in which the VhhL5 is positioned within the G protein binding cavity, shows interactions across the ICL3, including C-terminal residues that do not impact the overall binding strength of the β₂AR-VhhL5 interaction (Fig. 4b). Together, our computational analyses complement experimental measurements, while highlighting putative binding modes that facilitate allosteric modulation.

VhhL5 enhances β₂AR-Gs coupling by disrupting spatial distance constraints in ICL3

To investigate the mechanism of VhhL5 modulation of β₂AR signaling, we first examined allosteric effects independent of the SPASM fusion construct. Specifically, we co-expressed β₂AR-mCit with a VhhL5-mCer containing a plasma membrane tethering motif (-CAAX)³⁰ (Fig. 5a). The SPASM fusion and membrane tethered system may provide different accessibility, orientation, and concentration of VhhL5. Nonetheless, consistent with our observations using a β₂AR fusion construct, co-expression of β₂AR with VhhL5-mCer-CAAX substantially enhances (48%) the maximum agonist-stimulated cAMP response (Fig. 5b and Supplementary Fig. 9). The observed allosteric modulation of β₂AR by VhhL5 is not influenced by changes in receptor surface expression levels (Supplementary Fig. 10). Further, VhhL5 exhibits β₂AR selectivity, with no observable modulation at β_xAR subtypes (β₁AR and β₃AR), and two unrelated GPCRs (PTH₁R and D₁R) (Supplementary Fig. 11a).

Fig. 5. VhhL5 releases spatially constrained residues in ICL3 and increases cognate G_s protein coupling to β₂AR.

a Schematic of β₂AR co-expressed with VhhL5-CAAX motif. b HEK293 adherent cells transfected with β₂AR alone (black) or co-transfected with VhhL5-CAAX (green), measuring Iso-dependent cAMP accumulation. Data is represented in mean ± SD of n = 3. Statistical comparisons in (b) are performed by a two-tailed paired t test of the Emax. c Schematic of FRET-based reporter of β₂AR G protein coupling. FRET pair (mCitrine, yellow and mCerulean, light blue) are linked by an ER/K α helix, with the α5 helix of Gαs fused C-terminally to mCerulean. d β₂AR-Spep sensor vesiculated in GPMVs stimulated with varying concentrations of the agonist isoproterenol. Data is represented in mean ± SD of n = 3 independent GPMV preparations. Statistical comparisons in (d) are performed by two-tailed paired t test of the Emax. e Schematic of β2AR SPASM sensor fused to VhhL5. A β₂AR mutant previously demonstrated to release conformational constraints in ICL3, QDG(AAA), was used. f Iso-induced cAMP accumulation downstream of β₂AR wild-type (WT) or β₂AR QDG(AAA). Symbols for β₂AR SPASM sensors fused to 4xGSG (-) (black circles) (n = 5) or VhhL5 (green squares) (n = 5). Data is represented in mean ± SD of n = 5, representing independent biological replicates. Two-way ANOVA as used to analyze the statistical significance of the effect of QDG(AAA) mutation and the fusion of VhhL5 (n = 5). Expression quantification, Emax, and EC₅₀ data for (b) is provided in Supplementary Fig. 9.

We previously demonstrated that ICL3 of β₂AR autoregulates receptor activity by occluding receptor interactions with the G protein⁶. Agonist stimulation releases these autoregulatory constraints to displace ICL3 from the G protein binding cavity. VhhL5 does not significantly activate the receptor in the absence of agonist, suggesting the requirement for an agonist for its allosteric activity (Supplementary Fig. 11b). To determine whether VhhL5 can facilitate the release of ICL3 autoregulation in the presence of agonist, we assessed G protein recruitment (miniGs) and coupling (α5 helix of Gαs – S peptide) (Fig. 5c and Supplementary Fig. 12a). A modest but significant decrease (7%) was observed for miniGs recruitment quantified using a bioluminescence resonance energy transfer (BRET) assay in the presence of VhhL5 (Supplementary Fig. 12b, d). This decrease was not observed when miniGs recruitment was challenged with Vhhv3.1 (Supplementary Fig. 12c, e). In contrast, the β₂AR-S peptide interaction measured using both luminescence (Supplementary Fig. 12f) and FRET-based assays (Fig. 5c) shows significant increases at saturating agonist concentration (Supplementary Fig. 12g and Fig. 5d), suggesting enhanced G protein activation³¹. This is consistent with the observed increase in downstream cAMP signaling mediated by VhhL5 (Figs. 3c, 5b).

ICL3 occlusion of the G protein binding site is dependent on residues in the C-terminal region of ICL3 forming spatial distance constraints with other regions of the receptor⁶. Mutagenesis of spatially constrained residues ²⁵⁰QDG to AAA enhances cAMP signaling (Fig. 5e, f), consistent with previous observations⁶. VhhL5 is unable to further augment cAMP signaling downstream of the QDG(AAA) mutant (Fig. 5f). This effect is also observed in the context of ScFv5 (Supplementary Fig. 13a–d). We previously reported that GPCRs signal through non-cognate G proteins in the absence of ICL3⁶. VhhL5 has no measurable effect on G_q mediated IPx accumulation downstream of β₂AR (Supplementary Fig. 13e). This is consistent with our finding that the releasing autoregulatory constraints in β₂AR ICL3 alone are not sufficient to enhance non-cognate G protein coupling⁶. Together, these observations suggest that β₂AR ICL3 binders augment G_s coupling and subsequent cAMP signaling by releasing autoregulatory interactions in ICL3.

VhhL5 suppresses β₂AR internalization

Receptor internalization is an alternate mechanism to modulate GPCR signaling^32–34. To assess whether VhhL5 could influence β₂AR trafficking, isoproterenol-induced internalization was monitored for HEK293 cells expressing a flag tagged-β₂AR alone or co-expressed with VhhL5/Vhhv3.1-tRFP-CAAX (Fig. 6a). We observe substantially fewer internalized receptor punctae in the presence of VhhL5 (Fig. 6b, c) but not Vhhv3.1 (Supplementary Fig. 14a, b). GPCR endocytosis is often accompanied with enhanced extracellular signal-regulated kinase (ERK) phosphorylation³². Accordingly, VhhL5 diminishes phospho-ERK (pERK) signaling (Fig. 6d). β-arrestin plays an important role in β₂AR internalization, desensitization, and pERK signaling³⁵. In β-arrestin knockout cells, VhhL5 has no measurable influence on agonist-induced internalization (Supplementary Fig. 14c, d), and is still able to modulate β₂AR signaling (27% increase, Fig. 6f and Supplementary Fig. 14e–g), albeit to a lesser extent than wild-type cells (48%, Fig. 5b). Although VhhL5 clearly influences receptor internalization, these changes in β₂AR internalization alone cannot fully account for the enhanced cAMP response observed. Together, our findings suggest that VhhL5 binds to the ICL3 N-terminus to release C-terminal regulatory constraints that steer ICL3 away from the G protein binding site and facilitate G protein coupling in addition to hindering receptor trafficking (Fig. 7).

Fig. 6. VhhL5 suppress β₂AR trafficking and compartmentalized ERK signaling.

a Schematic of β₂AR internalization and downstream ERK signaling in the presence or absence of VhhL5. The following panels suggest that co-transfection of VhhL5-CAAX decreases receptor internalization and ERK signaling. b Representative images of β₂AR internalization in HEK293 cells expressing β₂AR alone or co-expressed with VhhL5-tRFP-CAAX, stimulated with isoproterenol (Iso, 10 μM). The first two columns illustrate the receptor’s fluorescence, while the far-right column represents tRFP fluorescence from VhhL5. c Number of puncta per cell for β₂AR internalization in HEK293 cells expressing β₂AR alone or co-expressed with VhhL5-tRFP-CAAX under basal or Iso conditions. Data is represented as mean ± SD of n = 3 biological replicates, two-way ANOVA used to compare the effect of iso stimulation and co-expression of VhhL5. d pERK signaling of HEK293 cells expressing β₂AR alone or co-expressed with VhhL5-tRFP-CAAX, stimulated with isoproterenol (Iso, 10 μM). e Schematic of co-transfection in β-arrestin knock-out (βarr1/2 KO) cells. f βarr1/2 KO adherent cells expressing β₂AR alone (black) or co-expressed with VhhL5-CAAX (green), measuring Iso-dependent cAMP accumulation. Data is represented as mean ± SD of n = 4. Statistical comparisons performed in (d) were a two-tailed t test and (f) two-tailed paired t test of the Emax. Expression quantification, Emax, and EC₅₀ data for (f) is provided in Supplementary Fig. 14.

Fig. 7. VhhL5 releases spatial constraints in ICL3 to enhance β₂AR signaling.

Left, ICL3 autoregulates receptor β₂AR activity through conformations that autoregulate receptor signaling and internalization. Right, VhhL5 releases spatially constrained residues in these ICL3 conformational states, enhancing G protein coupling and cAMP accumulation. As a parallel allosteric mechanism, VhhL5 suppress receptor internalization and downstream pERK signaling. Our data suggests a G protein biased mechanism for the VhhL5 positive allosteric mechanism on β₂AR.

Discussion

GPCRs contain variable loop, N- and C-termini regions that have not historically been a focus for drug discovery. This is partly due to their conformational heterogeneity that eludes structurally inspired approaches. In this study, we sought to target the third intracellular loop (ICL3), the largest and most sequence-diverse domain of the GPCR fold. Here, we demonstrate that a single-domain antibody directly influences β₂AR function through ICL3, thereby acting as a positive allosteric modulator (PAM).

GPCR structural biology has been accelerated by the use of single-domain antibodies (Vhh), commonly referred to as nanobodies. Vhh are demonstrated to mediate pharmacological inhibition or activation of GPCRs^36–39. However, Vhh that target intracellular domains described in the literature thus far predominantly function as either antagonists or negative allosteric modulators¹⁹. A notable exception is an intrabody PAM of the 5HT2A receptor that lacks a characterized binding site³⁸. VhhL5 developed for this study demonstrates an alternative mechanism of action for the pharmacological action of antibody fragments on GPCRs. Given the generality of ICL3-based autoregulation across GPCRs with diverse ICL3 sequences⁶, we propose that this mechanism can be applied to discover additional PAM-antibody fragments for other receptor subtypes.

We propose a model wherein antibody binding to ICL3 releases distal spatial constraints to facilitate receptor activation and enhanced G protein coupling. Our engineered antibody fragment (VhhL5) binds to the N-terminus of β₂AR ICL3 (Fig. 4b). The binding site overlaps with a structurally resolved segment of the β₂AR ICL3-Fab5 complex (PDB: 2R4S, Supplementary Fig. 3). VhhL5 engagement involves multiple intermolecular contacts (Fig. 2c, d) leading to a dynamically distributed structural interface (Fig. 4d, g). Our findings suggest that VhhL5 interactions with the β₂AR ICL3 N-terminus allosterically disrupt autoregulatory constraints at the ICL3 C-terminus (Fig. 5f) to facilitate receptor activation and enhance G protein coupling (Fig. 5d)^6,8,31. Our computational analyses suggest that the dynamic nature of the VhhL5-β₂AR ICL3 interaction could include other weak, transient contacts throughout this disordered region (Fig. 4 and Supplementary Fig. 8). While our FRET sensor measurements show VhhL5 interactions with all three β_xAR subtypes (Fig. 3b), the lack of measurable signaling effects on β₁ and β₃ (Fig. 3d, e and Supplementary Fig. 11a) suggest the sequence composition of the β₂AR ICL3 is important for selective allosteric modulation by VhhL5. Specificity to the β₂AR ICL3 is potentially rendered through a combination of distinct molecular contacts within of the VhhL5 interaction site, allosteric communication within the ICL3 region, or subtype-specific structural roles of ICL3⁸.

Thus far, bimolecular interactions with IDRs have focused on short linear motifs (SLiMs) that maintain disorder through binding^40,41 or molecular recognition features (MoRFs) that undergo a disorder-to-order transition through binding^41,42. Structural characterization of IDR-single domain antibody fragment interactions reports either the entrapment of SLiMs between two CDRs (CDR2 and CDR3)⁴³ or a headlock interaction between CDR3 and the antibody core⁴⁴. In contrast, the Fab5-β₂AR structure contains only a partial α-helix in the N-terminus ICL3 trapped between the heavy and light chains (Supplementary Fig. 3). VhhL5 cannot recapitulate this trapped interaction mode, highlighting a more dynamic and complex interface that nonetheless modulates receptor activity. Our data highlights the complexity of this interaction. First, receptor-Vhh binding, assessed by FRET measurements, does not correlate with modulation of downstream signaling. We observe this phenomenon across different Vhh fused to the same GPCR (Fig. 2, Supplementary Figs. 5, 6), as well as different GPCRs fused to the same Vhh (Fig. 3). Furthermore, our Vhh optimization (Supplementary Fig. 6) and computational interface characterization (Supplementary Fig. 8), highlights the importance of a distributed interaction across all three CDRs. While the VhhL5 binding site on the β₂AR ICL3 that we identified is consistent with the structural interface of the light chain of Fab5, structural data would point to an interaction centered on CDR1 and CDR3 contacts (Supplementary Fig. 3). Our modeling data is consistent with the importance of CDR1 and CDR3 (Fig. 4), indicating CDR2 as a potential site for future optimization of efficacy, affinity and selectivity. Our study provides foundational insights into how Vhh can modulate GPCR signaling through the ICL3 region. Future structural work will delineate the complex modes of interaction that make modulation through ICL3 possible.

Our data suggests that suppression of receptor internalization as a parallel mechanism for the PAM activity of VhhL5 (Fig. 6). In addition to augmenting cAMP signaling at the plasma membrane, blocking receptor internalization diminishes pERK signaling, a hallmark on endomembrane signaling in GPCRs. Furthermore, knockout of β-arrestin1/2, central regulators of β₂AR internalization that also contribute to pERK signaling at endomembrane compartments, only partially abolished cAMP augmentation by VhhL5 (Fig. 6f). These data support the role of VhhL5 as a partially G protein signaling biased modulator (Fig. 7). The precise role of VhhL5 in suppressing the β-arrestin pathway is unclear, although recent studies have highlighted synergies between receptor ICL3 and C-tail that comprise their respective binding sites⁹.

Our study establishes three design criteria for an ICL3 targeted positive allosteric modulator of G protein signaling: it must bind receptor subtype-selective sequences within this loop region, facilitate active signaling states of the receptor, and act without interfering with G protein coupling. Our data demonstrates proof of concept for all three of these criteria. First, targeting ICL3 enables specificity through the inherent sequence diversity of this region (Fig. 2a; Fig. 4b). Second, binding ICL3 augments receptor function by disrupting receptor autoregulation (Fig. 5f) and enhancing G protein coupling (Fig. 5d). Lastly, the VhhL5-ICL3 binding interface (Supplementary Fig. 8) does not occlude the G protein binding cavity (Fig. 5d, f). This design framework, paired with further structural characterization of the dynamic Vhh-ICL3 interaction, holds the potential for rationale design of ICL3-targeted allosteric modulators across the GPCR superfamily.

Methods

Reagents

(-)-Isoproterenol (+)-bitartrate salt (cat# I2760), ascorbic acid (cat# A0278), and X-tremeGENE HP DNA transfection reagent (cat# 6366546001), phenylmethylsulfonyl fluoride (PMSF) (cat# 52332), lysozyme (cat# L6876), DNAse I (cat # 04716728001), 100% sulfuric acid(cat# 339741) and N-ethylmaleimide (NEM) (cat# E3876) were purchased from Sigma Aldrich. Polyethyleneimine (PEI) 25 kDa linear polymer (cat# 23966) was purchased from Polysciences. Opti-MeM (cat# 31985062), 1-Step™ TMB ELISA Substrate Solutions (cat# 3402), and Expi293 expression medium (cat# A1435102) were purchased from Thermo Fisher. Ni-NTA agarose (cat# QIA30210) was purchased from Qiagen. Isopropyl b-D-Thiogalactopyranoside (IPTG) (cat# BP175510) was purchased from Fisher Scientific. Aprotinin (cat#14716) and Leupeptin (cat# 14026) were purchased from Cayman Chemicals. Pfu turbo polymerase (cat# CX26169) was purchased from Agilent. All reagents were reconstituted and stored according to the manufacturer’s specifications.

Plasmid construction

Human β₁AR, β₂AR, and β₃AR were cloned into pcDNA5/FRT following standard cloning procedures. SPASM sensor constructs were assembled as previously described^6,23,31,45 with Gly-Ser-Gly repeats between domains (Receptor-4×GSG-mCitrine-4×GSG-10 nm ER/K linker-4×GSG-mCerulean-4×GSG) with C-terminal fusion of antibody fragments. The ScFv5 sequence was derived from the heavy and light chain variable domain of Mab5, a monoclonal antibody used to resolve the structure of β₂AR. ScFv5 sequence is as follows (N-to-C terminus): 113 amino acid of Mab5 heavy chain, 5x GGGGS linker, 107 amino acids of Mab5 light chain (sequence in SupplementaryTable 1)⁴⁶. Mutagenesis of ScFv5 and β₂AR’s ICL3 were performed with Pfu turbo polymerase one-step site-directed mutagenesis⁴⁷. VhhL5-CAAX, assembled as VhhL5-2xGSG-mCerulean-CAAX motif from the last 12 amino acids of Kras4B (KKKKSKTKCVIM), was cloned into pCDNA5/FRT following standard cloning procedures. The β₂AR luciferase complementation reporter construct had the same topology as SPASM sensor constructs with LgBiT/SmBiT luciferase fragments (Promega cat #N2014) replacing the FRET acceptor/donor and with a fused fluorescent protein to track expression level (β₂AR-LgBiT-4×GSG-10 nm ER/K linker-4 × GSG-TagRFP-3×GSG-SmBiT-4 × GSG-Speptide). The Vhhv4A bacterial expression construct with a C-terminal 6xHis tag was cloned into pET-15b following standard cloning procedures. All ScFv and Vhh sequences can be found in Supplementary Table 1.

Antibody fragment expression and purification

500 mL of terrific broth with 100 µg mL^-1 carbenicillin was inoculated with SHuffle T7 Express lysY competent Escherichia coli (New England Biolabs, cat #C3030J) transformed with the Vhh pET-15b vector at A600  =  0.05. The culture was grown to A600  =  1.0 at 37 °C with shaking (180 rpm). Then, the culture was induced with 0.4 mM IPTG and incubated for 5 hours at 25 °C with shaking (180 rpm). The culture was pelleted at 3000 x g for 15 min. Then, the pellet was resuspended in 30 mL lysis buffer (20 mM HEPES pH 7.4, 150 mM NaCl, 10 mM imidazole, 10 mM MgCl2, 5 mM CaCl2, 1 mg mL⁻¹ lysozyme, 1 µg mL⁻¹ DNAse I, 1 mM DTT, 1 µg mL⁻¹ aprotinin/leupeptin, 0.1 µg mL⁻¹ PMSF, 1% v/v Triton X-100) and incubated for 30 min at 4 °C on an orbital shaker (100 rpm). The lysate was then sonicated for 10 min (10 s on, 10 s off) and clarified by centrifugation (18,000 x g, 20 min, 4 °C).

2 mL Ni-NTA agarose was equilibrated with 15 ml wash buffer (20 mM HEPES pH 7.4, 150 mM NaCl, 10 mM imidazole). Clarified lysate was incubated with the column for 30 minutes at 4 °C in a rocker. After incubation, the resin was poured into a 1 mm closed column and allowed to settle. The column was washed with 15 ml wash buffer, 15 ml high-salt wash buffer (20 mM HEPES pH 7.4, 500 mM NaCl, 10 mM imidazole), and 15 ml wash buffer. Four total 0.5 ml elution fractions were collected in 20 mM HEPES pH 7.4, 150 mM NaCL buffer containing 60 mM imidazole, 120 mM imidazole, 180 mM imidazole, and 240 mM imidazole. Fractions were analyzed for purity via SDS-PAGE before pooling the fractions eluted with 180 mM and 240 mM imidazole. The mix of elution fractions was concentrated using a 10,000 kDA molecular weight cut-off centrifugal filter (Amicon Ultra-15), and further purified over a Superdex 200 Increase 10/300 GL gel filtration column (GE Healthcare) in size exclusion buffer (20 mM HEPES pH 7.4, 400 mM NaCl). Purity of size exclusion eluate was confirmed via SDS–PAGE. Eluate was concentrated and rebuffered in storage buffer (20 mM HEPES pH 7.4, 150 mM NaCl, 10 mM KCl, 5 mM MgCl₂, 10% w/v Glycerol). Concentration was determined by 280 nm absorbance (Thermo Scientific Nanodrop One). Aliquots were flash frozen and stored at − 80 °C. SDS-PAGE expanded images for protein purified proteins can be found in Supplementary Fig.15-17.

Cell culture

Adherent cell lines HEK293T Flp-In T-Rex Cells (ThermoFisher, cat# R78007, referred throughout the manuscript as HEK293 WT) and HEK293 β-arrestin 1/2 knockout (HEK293 βarr1/2 KO)⁴⁸ were cultured in DMEM supplemented with 10% FBS, 4.5 g/liter D-glucose, 1% Glutamax, 20 mM HEPES, pH 7.5, at 37 °C in a humidified 5% CO₂ atmosphere. Suspension-adapted HEK cells, Expi293F cell line (kind gift from the Lefkowitz lab), were cultured in Expi293 expression medium in a shaking incubator (125 rpm) at 37 °C in a humidified 8% CO₂ atmosphere.

Surface expression assay

HEK293 cells were seeded at 30% confluency in tissue culture-treated 96-well plates, any empty well was filled with 50 µL of PBS. Each tested condition was transfected with eight technical replicates containing 50 ng of DNA, 0.3 µL of X-tremeGENE HP DNA transfection reagent and 10 µL Opti-MeM per well. 24 hr after transfection, the media was removed using a vacuum manifold without disrupting the adhered cell layer. Then, cells were fixed with 100 µL of 4% formaldehyde for 20 minutes under light orbital shaking at 25 °C. Cells were gently washed three times with 100 µL PBS. Blocking was carried out with 100 µL of complete DMEM growth media (methods: cell culture) for 1 hr under light orbital shaking at 37 °C. After blocking, all blocking media was removed with a vacuum manifold without disrupting the adhered cell layer. Then, added 100 µL per well of anti-HA HRP peroxidase antibody (Roche, cat# 12013819001) diluted 1:1000 dilution (0.1 U/mL per well) and incubated for 1 hr under light orbital shaking at 25 °C. This step was followed by three washes with 100 µL PBS. After removal of residual PBS, to quantify total protein expression, mCitrine’s fluorescence intensity was measured using a Tecan Spark plate reader, exiting at 488 nm and measuring emission at 525 nm. Following the measurement of total protein expression, 100 µL of 1-Step™ TMB ELISA Substrate Solution was added to each well and incubated for 15 minutes under light orbital shaking at 25 °C. After 15 minutes, the peroxidase reaction was stopped with 100 µL 1 M sulfuric acid, and the absorbance at 450 nm was measured and subtracted from the reference at 652 nm. All experiments were performed with eight technical replicates per biological replicate.

cAMP accumulation assay

For all cAMP experiments, HEK293 WT and HEK293 βarr1/2 KO were used during passages 10-27. 16-20 hours before transfection, cells were seeded at 30% confluency in tissue culture-treated six-well plates. Cells were transfected with 1 µg of DNA, 3 µL of X-tremeGENE HP DNA transfection reagent and 100 µL Opti-MeM. In order to provide comparative receptor expression, transfection times varied between 18 – 30 h depending on the construct.

After transfection, cells were harvested via gentle pipetting and centrifuged at 300x g for 3 min. Media was removed by vacuum manifold, and pellets were resuspended in 1 mL cAMP assay buffer (phosphate-buffered saline (PBS) with 0.5 mM ascorbic acid and 0.2% (w/v) glucose). Cells were washed once by repeating the previous steps. Cell density was measured (Thermo Fisher Scientific, Countess II) and diluted to 1.5 × 10⁶ cells per mL. Expression of transfected constructs was measured via fluorescence (Horiba Fluoromax 4). The metric used to assess the expression of the constructs was the ratio between the mCerulean peak emission (430 nm excitation, 475 nm emission) over optical density (430 nm excitation, 450 nm emission). For dose response curves, at least two transfections were performed for the control and experimental conditions, increasing the probability of matching expression. Alongside expression, the interaction strength facilitated by the SPASM sensor was measured as the FRET ratio of peak emission mCitrine over mCerulean (430 nm excitation, 525 nm / 475 nm emission). The best match expressions were used to perform the dose-response curves, and the FRET ratios of all transfections were recorded.

Resuspended cells were incubated in an opaque 384-well flat-bottom plate (Greiner Bio-One) with an equal volume of 2X the concentration of isoproterenol. After 10 min of incubation at room temperature, cells were lysed and processed for the cAMP-Glo Assay (Promega cat# V1502) following the manufacturer’s instructions. Luminescence was measured on a Tecan Spark plate reader (500 ms integration, one measurement per well). Each condition was performed on four independent wells, and the experiment was performed at least 3 times.

MiniGs recruitment BRET assay

BRET experiments were performed in HEK293 WT cells were cultured as previously described (methods: Cell culture). For BRET experiments, cells were plated into 6 well dishes at approximately 25% confluency and allowed to recover for 24 hrs prior to transfection. Cells were transfected using 0.5 μg of β2AR-rGFP DNA, 0.5 μg of rLuc-tagged miniG protein DNA. The conditions containing antobody fragments were transfected with an additional 0.5 μg of VhhL5/Vhhv3.1-CAAX (without fluorescent tag) DNA. Transfection reactions were prepared in 450 μL of Optimem I (ThermoFisher) with 30 μL of 1 mg/mL polyethylenimine (PEI, 25,000 kDa; PolySciences) and allowed to rest for 15 min at room temperature prior to addition to the cultured cells. The media containing yhtransfection reagent was removed from the cells after ~ 5-6 hrs and fresh media was replaced. Media was exchanged again 24 h post-transfection. Cells were harvested 48 hrs post-transfection by removing the media and resuspending the cells in 1 mL of BRET buffer (20 mM HEPES, pH 7.4, 135 mM NaCl, 5 mM KCl, 400 μM MgCl2, 1.8 mM CaCl2, and 5 mM glucose in water). Resuspended cells were spun at 350x g for 5 min at room temperature. The supernatant was aspirated, and the pellet resuspended again in 1 mL of BRET buffer before being spun again at 350x g for 3 min at room temperature. The supernatant was aspirated again, and the cells resuspended for a final time in 1 mL of BRET buffer. Cells were counted using a Countess II automated cell counter (ThermoFisher) and diluted to approximately 7.5*106 cells/mL. 90 μL of diluted cells were added to individual wells of a white, round-bottom 96 well plate so that there were three sets of wells for each condition to be tested. The plate was allowed to rest at room temperature for approximately 3-5 minutes while the luminescent substrate was prepared. Prolume purple (NanoLight Technology) was diluted to 5 μM in BRET buffer before 10 μL was added to each well using a repeater pipet (Eppendorf) to minimize time differences. An initial read of fluorescence (470 nm excitation, bandpass 9 nm; 505 nm emission, bandpass 20) using a Tecan Spark plate reader. Immediately following this reading, the mode was automatically switched to dual color luminescence and read continuously using light filters (magenta: ~450 nm shortpass/610 nm longpass and green: ~ 505–540 nm) for 3 min. Following this, the plate was ejected from the machine, and a multichannel pipet was used to add and mix 50 μL of 30 μM isoproterenol or buffer into the monitored wells. Then, the plate was read continuously using the same dual color luminescence protocol for 5 min. One technical replicate for each condition was read at a time for each biological repeat. BRET ratios were quantified as rGFP divided by Prolume purple luminescence for all wells. ΔBRET was measured as the BRET ratio difference between iso-stimulated and buffer conditions. To account day to day variability, all traces were normalized to the maximal response of the control condition (absence of Vhh) of the respective biological replicate. Data was analyzed via GraphPad PRISM to determine the area under the curve from the miniGs recruitment traces. Analyzed traces started at one time point before drug incubation and were analyzed until the completion of five minutes.

Giant plasma membrane vesicles (GPMVs) preparation

The β₂AR-luciferase complementation reporter was vesiculated (GPMV) as previously described, with modifications for suspension cell handling³¹. Cells were transfected between 3 and 5 million cells/mL in 30 mL cultures with 3 mL of Opti-MeM containing 30 μg of DNA and 160 μL PEI. Approximately 48 h post transfection, cells were harvested by centrifugation for 5 minutes at 350 x g, resuspended in 10 mL PBS, pelleted again for 5 min at 350 x g, resuspended in 10 mL GPMV buffer (1 mM HEPES pH 7.4, 150 mM NaCl, 2 mM CaCl₂), pelleted again for 5 minutes at 350 x g before a final resuspension in 40 mL GPMV buffer. Cell suspensions were divided in half into fresh 125 mL glass flasks and incubated with 2 mM NEM for 2 h at 28 °C with 140 rpm (19 mm orbit) shaking. After incubation, cell suspensions were recombined and centrifuged at 1000 x g for 2 min. Supernatant containing vesicles was set aside, and pellets were resuspended in 10 mL PBS. The remaining vesicles in the pellet were separated by centrifugation at 1000 x g for 2 min, and the resulting supernatant was combined with the one from the previous step. A final spin of 1000 x g for 2 min of the combined supernatants was performed to remove any remaining cell debris. The supernatant from this spin was transferred into a new 50 mL conical tube and spun for 40 minutes at 3220 x g, 4 °C. Pelleted vesicles from this spin were resuspended in 2 mL of luminescence assay buffer (20 mM HEPES pH 7.4, 150 mM NaCl, 10 mM KCl, 5 mM MgCl2) and spun again for 10 min at 3220 x g, 4 °C. The pelleted vesicles were resuspended in a final volume of 2 mL luminescence assay buffer or low salt buffer to obtain the final preparation of GPMVs.

β₂AR-fluorescence (FRET sensor) complementation reporter assay

The integrity of GPMVs was assessed by fluorescence spectroscopy using an excitation

of 430 nm (bandpass 8) and monitoring emission from 450-600 nm (bandpass 4) using

a Fluoromax-4 fluorometer (Horiba). Samples were diluted such that the maximum mCerulean emission were approximately 1 × 10⁶ counts per second. For FRET dose response curves, three to four technical replicates were prepared for each condition on ice in low salt assay buffer and warmed to 25 °C for 5 min prior to their spectra being collected using the above parameters. FRET ratios were calculated as the ratio of mCitrine fluorescence over mCerulean fluorescence during mCerulean excitation (430 nm).

β₂AR-luciferase complementation reporter assay

GPMV of β₂AR-luciferase SPASM sensors samples were plated in a 96-well clear- bottom plate and assayed for TagRFP fluorescence (Tecan Spark, excitation 521 nm, emission 585 nm, gain 150). Samples were diluted to 450 TagRFP counts, either with GPMVs alone or GPMVs with 2.2 µM purified Vhhv4A. 45 µL of each of these conditions (+/- Vhh) was transferred to an opaque 96-well plate. 45 µl of Nano-Glo Substrate (Promega cat# N1110) diluted 1:50 in luminescence assay buffer was added to each well in the new plate. After tapping the plate to collect liquid at the bottom of the well, a kinetic luminescence read was started (500 ms integration, continuous for 40 min) and the luminescence signal was tracked. After 2 minutes, the kinetic read was paused, and 10 µL of 1 mM of isoproterenol (final concentration of 100 μM) or buffer (unstimulated) was added to each well. The plate was tapped 2–3 times to mix, and the kinetic read was resumed. Each kinetic trace was normalized to the last point of the pre-drug equilibration. For each condition (+/- Vhh), the maximum luminescence signal of isoproterenol stimulation was subtracted from the luminescence signal of the corresponding time point of buffer treatment. The experiment was performed at least twice (technical replicates) for n = 7 GPMV preparations.

HTRF IP-One assay

Resuspended cells were added in equal parts with a 2x concentration of ligand (70 µl final) into an opaque 96-well U-bottom plate (Greiner Bio-One). Saturating amount of ligand (10 µM isoproterenol, 10 µM AVP) were added to each well and incubated at 37 °C for 2 h to stimulate InsP1 production. Reactions were quenched and processed for the InsP1 HTRF Assay (CisBio), following a protocol modified to achieve a higher signal to noise ratio. 15 µl of D2-conjugated InsP1 resuspended in lysis buffer (Cisbio) and 15 µl of terbium cryptate conjugated anti-InsP1 antibody resuspended in lysis buffer (Cisbio) were added to the stimulated cell mixture. Cell lysate was equilibrated for 1 h at ambient temperature with orbital shaking (500 rpm). Reactions were transferred (4 × 20 µl) to a 384-well plate for technical replicates. Fluorescence readings (Flexstation3, Molecular Devices) of acceptor D2 (excitation 340 nm, emission 665 nm, cut-off 630 nm) and donor terbium cryptate (excitation 343 nm, emission 620 nm, cut-off 570 nm) were acquired with a delay of 50 µs and an integration time of 300 µs. The FRET ratio for each reading was calculated as the ratio of acceptor emission to donor emission. InsP1 signal for a drug and transfection combination was calculated as the average FRET ratio of a given transfection condition without drug treatment, subtracted from the average FRET ratio of a given transfection condition with drug treatment.

Internalization assay

HEK293 WT or HEK293 βarr1/2 KO cells were seeded on 0.001% poly-L-lysine coated glass coverslips in 35 mm plates (GenClone) at 30% density. Cells were allowed to attach overnight and then transfected with 0.5 μg of FLAG-mNeonGreen-β2AR + /- 1 μg of VhhL5-tRFP-CAAX or 0.25 μg Vhhv3.1-tRFP-CAAX constructs. Polyethylenimine (PEI; PolySciences) was used as the transfection reagent at 4 μL PEI for 1 μg DNA. Internalization assay was carried out 24 h post-transfection for all conditions.

Cells were blocked in DMEM with 0.1% BSA for 15 min at 4 °C prior to labeling. Anti-FLAG M2 antibody (Millipore Sigma) was incubated with a secondary AlexaFlour 647 (Thermo Fisher) in DMEM with 0.1% BSA at 4 °C for 60 min. Next, surface β2AR receptors in HEK293 cells were labeled with Anti-FLAG+AlexaFlour complex at 4 °C for 60 min. Next, cells were stimulated with Iso (10 μM) for 15 min. Cells were fixed in 4% formaldehyde for 10 min at room temperature. Coverslips were washed in PBS, mounted on glass slides using Prolong Diamond Anti-Fade (ThermoFisher) and allowed to cure overnight. Coverslips were sealed the next day with valap (vaseline/lanolin/paraffin). Images were acquired using Nikon Eclipse Ti inverted epifluorescence microscope with 100X oil immersion objective (1.4 numerical aperture) equipped with an Evolve EMCCD camera (Photometrics) or PCO.edge sCMOS camera using the Nikon Elements software (Nikon). Maximum intensity projection (MIP) were created using Fiji (ImageJ) (Schindelin, J. et al., 2012, Methods) for images acquired as Z-stacks (0.6 μm step size). Internalization was quantified as the number of particles using Fiji, as described previously³⁴. At least three biological replicates were carried out for each condition, and 10-20 cells were analyzed per each condition for every biological replicate.

Phospho-ERK1/2 (pERK) assay

HEK293 WT were seeded in 35-mm plates at 30% confluency. Approximately 24 hours post-seeding, the media was exchanged to serum-free DMEM and cells were transfected with 0.65 μg of β2AR-mCit or 1 μg of β2AR-mCit + 0.6 μg of VhhL5-mCer-CAAX using PEI (4 μL PEI :1 μg DNA). The next day, cells were harvested in culturing media and cell density was determined using a hemocytometer (Countess II). Expression of receptor was monitored with a fluorescence spectrum on FluoroMax-4 spectrofluorometer (Horiba Scientific - excitation 488 nm, 8 nm bandpass; emission 500-650 nm, 4 nm bandpass). Next, cells were added to a white 384-well plate (Greiner Bio-One) at 1*10^4 cells/well density. The pERK assay was carried out following the manufacturer’s instructions (Revvity). Briefly, cells were incubated with 10 μM Iso for 15 min at room temperature. Lysis buffer (5 μL/well) was added to quench the reactions, and the plate was incubated at room temperature for 30 min with shaking (500 rpm). Next, a pre-mixed antibody mixture (5 μL/well) was added and incubated at room temperature overnight. Finally, pERK levels were measured using an HTRF-compatible plate reader (Molecular Devices - excitation 314 nm, cutoff 630 nm; emission 665 nm and 620 nm, cutoff 570 nm). pERK amount was calculated as the ratio between acceptor (665 nm) and donor (620 nm) values for each well. At least four technical replicates and three biological replicates were included for each condition.

Computational methods for generating structural models of β₂AR-ICL3-VhhL5

AlphaFold3-based structural modeling and clustering analysis

Our goal was to develop structural models of β₂AR with the full ICL3 bound to the VhhL5. We employed a computational workflow guided by FRET sensor measurements to select models consistent with experimental observations. The study integrates AF3²⁹ predictions to identify regions of ICL3 involved in antibody fragment binding with full-length ICL3 conformations obtained from our previous experimentally validated MD simulations^6,49, followed by CGMD simulations to efficiently optimize VhhL5–β₂AR interactions. From this ensemble, we selected structural models exhibiting the highest number of residue-level contacts between the N-terminal region of ICL3 and the CDR loops of VhhL5 (Fig. 4).

Step 1: Generate structural models of β₂AR complex with VhhL5 and Vhhv3.1: Experimental three-dimensional structures of β₂AR, as well as many other class A GPCRs, lack ICL3 due to its intrinsically disordered nature. To avoid low-confidence ICL3 predictions by AlphaFold3 (AF3), we therefore used truncated β₂AR sequences as inputs for model generation. Three ICL3 truncation variants were considered: (i) β₂AR–ΔICL3 (ICL3 deleted – see Supplementary Table 2 for the sequence details), (ii) β₂AR–p₁ICL3 (partial ICL3; Supplementary Table 2), and (iii) β₂AR–p₂ICL3 (a shorter partial ICL3 lacking additional C-terminal residues; Supplementary Table 2). For each antibody fragment (VhhL5 and Vhhv3.1), AF3 was used to generate 500 models per input sequence, yielding 1500 models per antibody fragment and 3000 models in total. The resulting models lack full ICl3 conformations and primarily differ in the predicted antibody fragment binding locations on β₂AR.

Step 2: For each structural model, we computed the average distances between residues in ICL1, ICL2, and the C-terminal region of TM5 and residues in the three antibody fragment CDR loops (residues highlighted in gray in Supplementary Table 2). This yielded nine distance measurements per model (three ICL regions × three CDR loops). Models with an average distance greater than 30 Å were excluded, resulting in 2989 retained models across both Vhh.

Step 3: To assess whether the remaining structural models formed distinct groups, we performed PCA on the 2989 retained models for both Vhh (VhhL5 and Vhhv3.1) using the nine distances between ICLs and CDRs. K-means clustering was then applied in the PC1 and PC2 space, which together captured 92% of the variance (Supplementary Fig. 7a), with the number of clusters varied from 1 to 10. Based on the elbow criterion (Supplementary Fig. 7b), five conformational clusters were identified as optimal (Supplementary Fig. 7a).

Step 4: In the next step, we compared the number of structural models in each cluster obtained from the ΔICL3 system with the minimum number of conformations observed in the p₁ICL3 and p₂ICL3 systems. This comparison was used to quantify the sensitivity of each cluster to the presence of partial ICL3, with larger differences indicating a greater dependence on ICL3-mediated interactions (Supplementary Fig. 7c). Clusters 0 and 3, marked with circles and crosses in (Supplementary Fig. 7a), showed more structural models with antibody fragment and ICL3 contacts than the other clusters. This approach was motivated by FRET sensor measurements indicating that VhhL5 binding is sensitive to residues in the N-terminal region of ICL3 (Fig. 4b).

Step 5: Because VhhL5 exhibits stronger coupling than Vhhv3.1 in the FRET experiments, and mutations in the N-terminal region of ICL3 (Ndel1; Fig. 4b) selectively impair VhhL5 coupling, we hypothesized that VhhL5 forms more frequent residue contacts with this region. To test this, we calculated the percentage of VhhL5 and Vhhv3.1 models in clusters 0 and 3 generated using the p₁ICL3 and p₂ICL3 constructs (Supplementary Fig. 7d). Consistent with the FRET results, VhhL5 showed a substantially higher fraction of ICL3-dependent models (78%) compared with Vhhv3.1 (38%), indicating a stronger association of VhhL5 with these conformational states. These results suggest that the stronger binder VhhL5 adopts a distinct binding mode in the presence of partial ICL3, reflecting a pronounced ICL3-dependent sensitivity specific to this antibody fragment. All pLDDT data is provided in for the main models in Supplementary Fig. 18.

Computational method for generating structural models of VhhL5 bound to β₂AR with full ICL3

In these AF3 models, VhhL5 consistently occupied the G-protein binding cavity. We hypothesized that AF3 preferentially placed Vhh in the cytosolic cavity due to the quantity of available 3D structures of GPCR:antibody complexes with the antibody fragment in the cavity. To explore alternative binding modes, we considered the possibility that VhhL5 functions as a positive allosteric modulator (PAM) and does not occlude the G protein binding cavity. We therefore implemented a modified computational protocol in which the β₂AR sequence included mini-G_s as input. This approach constrained the G protein binding cavity and promoted VhhL5 binding to alternative intracellular regions of the receptor.

The following section summarizes the computational workflow used to generate structural models of the β₂AR–full ICL3–VhhL5 complex. The protocol was applied to AF3 input sequences with and without G protein (mini-G_s sequences are listed in Supplementary Table 2). The AF3 models were used to identify regions of ICL3 involved in antibody fragment binding. These VhhL5 binding regions were combined with β₂AR conformations containing full-length ICL3 obtained from clustering of our previously reported extensive all-atom MD simulations of ICL3 that were experimentally validated⁶. The resulting β₂AR–full ICL3–VhhL5 models were further refined using CGMD simulations to optimize residue contacts between ICL3 and VhhL5. Structural models were selected based on the number of residue-level contacts between the N-terminal region of ICL3 and the CDR loops of VhhL5 (Fig. 4c).

Step 1: We generated 2500 AF3 structural models using sequences of β₂AR-p₂ICL3-mini-G_s and VhhL5. In these structural models, VhhL5 did not occupy the G protein binding cavity. We also generated 500 AF3 models using the sequence of β₂AR-p₂ICL3 and VhhL5 and in these models, VhhL5 occupied the G protein binding cavity in β₂AR.

Step 2: For each structural model, we calculated the minimum distance between residues in the antibody fragment CDR loops and residues in the N-terminus region of ICL3, which we previously identified as critical for VhhL5 coupling (Fig. 4b). For models predicted with mini-G_s included in the sequence (n = 2500), structures with distances > 20 Å to the Ndel2 motif were excluded, yielding 849 models that satisfied the selection criteria. For models generated without mini-G_s, VhhL5 occupied the G-protein–binding cavity (Supplementary Fig. 7b); therefore, a more stringent distance cutoff of > 10 Å was applied. Using this criterion, 369 of 500 models were retained.

Step 3: We performed principal component analysis on the remaining conformations using all Cα atom coordinates, followed by K-means clustering. This analysis yielded two conformational clusters for AF3 predictions generated without the mini-G_s protein sequence (Supplementary Fig. 7c) and three clusters for predictions generated with the mini-G_s sequence (Supplementary Fig. 7d). Representative conformations were extracted for each of the three clusters obtained in the presence of mini-Gs. Among these, only one orientation of VhhL5 relative to β₂AR positioned VhhL5 in proximity to the Ndel1 and Ndel2 regions of ICL3; this conformation was therefore selected for subsequent analysis.

Step 4: Preparing Starting structures for CGMD simulations: Our previous study combining extensive all-atom MD simulations (22 μs total sampling) with FRET sensor measurements of β₂AR containing the full ICL3 demonstrated that ICL3 adopts multiple open and closed conformations, thereby acting as a secondary autoregulatory determinant of β₂AR activity^6,49. From this MD ensemble, we derived 12 representative conformations of β₂AR with full ICL3 by selecting cluster centroids from 12 conformational clusters.

Starting structures for β₂AR+full ICL3+VhhL5 (predicted without mini-G_s sequence): We then positioned VhhL5 using two structural models generated without the mini-G_s sequence and overlaid each orientation onto the 12 β₂AR–full ICL3 conformations, yielding 24 candidate β₂AR–full ICL3–VhhL5 complexes. Because VhhL5 occupies the G-protein binding cavity in these models, severe steric clashes were observed upon inclusion of the full ICL3 loop in most cases. Only one of the 24 complexes was free of steric clashes and was retained for further analysis.

Starting structures for β₂AR+full ICL3+VhhL5 (predicted with mini-G_s sequence): Next, the 12 β₂AR–full ICL3 conformations were classified into active, intermediate, and inactive states based on the distance between residues 3.50 and 6.30 on TM3 and TM6, respectively. One representative conformation from each state was selected and overlaid with the single VhhL5 orientation that was in proximity to Ndel1 and Ndel2 regions identified in Step 3. Together with the original compatible complex, this procedure yielded four starting conformations (shown in Supplementary Fig. 8) for subsequent CGMD simulations.

Step 5: We performed 5 runs of CGMD on each of the 4 starting models. We performed 4μs CGMD simulations using the Martini 2.2p force field⁵⁰ within a cell membrane-mimicking multi-lipid bilayer. To capture the influence of lipid composition on GPCR conformational ensembles, we constructed a mixed-lipid bilayer that closely reflects the native cell membrane. The outer leaflet consisted of POPC, DOPC, POPE, DOPE, POSM, PNG3, and cholesterol in a ratio of 20:20:5:5:15:10:25, while the inner leaflet contained POPC, DOPC, POPE, DOPE, POPS, DOPS, PIP2, and cholesterol in a ratio of 5:5:20:20:8:7:10:25^51,52. For each initial conformation, we generated five independent lipid distributions. All simulations were neutralized by adding 0.15 M of NaCl_. For each starting β₂AR+full ICL3+Vhhl5 system, we first equilibrated each lipid configuration and then performed 4μs of CGMD simulations with Martini2.2 forcefield (⁵⁰). We used the NPT ensemble for the production run where the temperature was set to 310 K, and the pressure was set to 1 bar and maintained using the Parrinello–Rahman barostat. The simulation box size was 130 × 130 × 130 ų, and the total number of atoms varied between systems. This is because we used 5 different starting configurations or arrangements of the multi-lipid bilayer. The Coulombic terms were calculated using particle mesh Ewald (PME) and a real-space cutoff of 1.1 nm. We used the aggregated 4 μs x 5 trajectories for contact analysis. The convergence of the simulations was determined by monitoring the time evolution of the root mean squared deviation (RMSD) of the β2AR-VhhL5 complex with time (see Supplementary Fig. 19). The RMSD shift of ~ 1.0 nm is due to the shift in the position of VhhL5 closer to the ICL3 of β2AR.

Step 6 We calculated the sum of all the residue contact frequencies from the CGMD trajectories between residues in the three CDR loops and the three ICL loops of β₂AR. Supplementary Fig. 8 shows the distribution of the sum of contact frequencies calculated from CGMD to the ICL3 residues for all the four conformations for which CGMD were performed. We selected those CGMD ensembles that showed a greater number of contacts between CDR loops of VhhL5 and Ndel1 and Ndel2 regions of ICL3 and a minimum number of contacts with Middel and Cdel1 regions of ICL3 compatible with the FRET experimental findings. We then picked the best CGMD snapshot from each of the two CGMD ensembles that showed the maximum number of contacts to Ndel1 and Ndel2 regions. These structures are shown in Supplementary Fig. 8. The contacts made between the CDR loops and IC loops of β₂AR are shown in Fig. 4e,g.

Residue contact analysis from CGMD simulations

After completing the CG simulations, trajectories were preprocessed using GROMACS to recenter the protein and correct for periodic boundary conditions (PBC), followed by concatenation of the trajectories. Contact analysis was performed using the GetContacts tool (getcontact.io⁵³), which we adapted for coarse-grained (CG) simulations. In CG models, van der Waals interactions dominate protein–protein interactions; in Martini 2.2p, these interactions are defined by Lennard-Jones (LJ) potentials with predefined σ and ε values for each a pair of particles. To determine the cutoff distance between particles, we used the relation $r_1+r_1=2^{1/6}{\sigma }_0$ where $r_1$ and $r_2$ correspond to the effective radii of the first and second particles, respectively. In the Martini 2.2p force field, particles are categorized as backbone (BB) or side-chain (SC) beads, leading to three distinct LJ interaction types: BB–BB, BB–SC, and SC–SC. The corresponding σ₀ values are 4.7 Å for BB–BB interactions and 4.3 Å for both BB–SC and SC–SC interactions.

Because GetContacts uses a tolerance parameter to define contacts, we adjusted this parameter by increasing its default value from 0.5 to 1.66. After computing contacts, we analyzed residue–residue interactions and calculated contact frequencies between ICL3 residues and the VhhL5 for each initial structure individually (see Fig. 4d). Figure 4d shows the distribution of contact frequencies, which is consistent with the experimental truncation data presented in Fig. 4b. In addition, we quantified which complementarity-determining regions (CDRs) form the most contacts with the receptor and ICL3 for each initial conformation separately (Supplementary Fig. 8).

Statistical analysis

All data graphs and statistical analysis were performed using GraphPad Prism (10.2.0). To address potential differences in the levels of accessibility, orientation, and concentration on the modulation of antibody fragments, all statistical comparisons for FRET and cAMP are performed using matched ER/K linked or membrane-tethered configurations. In general, a two-tailed paired or unpaired t test was performed when comparing two conditions. For comparisons of more than two conditions, analysis of variances (ANOVA) was used. Single-level comparisons were performed with one-way ANOVA followed by a Tukey’s post hoc test. Two-way ANOVA was used for two-level comparisons (example, effect QDG mutation and the effect of the fusion of VhhL5) followed by a Tukey’s post hoc test. Data is presented as mean +/− standard deviation (SD) with the number of biological replicates indicated in the figure legends. All data statistics provided by GraphPad Prism are in the source data file with its respective data sources.

Reporting summary

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

Author contributions

E.S.R., F.S. and S.S. designed the study. F.S. designed and cloned Mab5-derived antibody fragments. E.S.R. performed the secondary messenger accumulation assay (cAMP, IPx), BRET recruitment assay, surface expression experiments, and data analysis of this study. N.V., M.S., N.M. and S.B. performed the computational approach (AF3 and MD simulations), M.R. performed the GPMV G protein coupling experiments, and N.P. performed microscopy and pERK data acquisition and analysis. E.S.R., F.S., M.S., N.V. and S.S. wrote the paper.

Peer review

Peer review information

Nature Communications thanks Carol Robinson, who co-reviewed with Xingyu Qiu; Reza Vafabakhsh and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

All data pertaining to this study are presented in the main figures and supplementary information. Referenced PDB: 2R4S. Source data are provided in this paper.

Code availability

The source code used in this study are deposited in the Zenodo repository under the accession code 19637340⁵⁴.

Competing interests

F.S. and S.S. are co-founders of Oxbow Therapeutics. S.S. is a co-inventor on US patent US12098184B2. All other authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-72785-y.