Adrenaline-activated structure of the β₂-adrenoceptor stabilized by an engineered nanobody

Abstract

G protein-coupled receptors (GPCRs) are integral membrane proteins that play an essential role in human physiology, yet the molecular processes through which they bind to their endogenous agonists and activate effector proteins remain poorly understood. Thus far, it has not been possible to capture an active-state GPCR bound to its native neurotransmitter. Crystal structures of agonist-bound GPCRs have relied on the use of either exceptionally high-affinity agonists^1,2 or receptor stabilization by mutagenesis^3-5. Many natural agonists such as adrenaline, which activates the β₂-adrenoceptor (β₂AR), bind with relatively low affinity, and they are often chemically unstable. Using directed evolution, we engineered a high-affinity camelid antibody fragment that stabilizes the active state of the β₂AR, and used this to obtain crystal structures of the activated receptor bound to multiple ligands. Here, we present structures of active-state β₂AR bound to three chemically distinct agonists: the ultra high-affinity agonist BI167107, the high-affinity catecholamine agonist hydroxybenzyl isoproterenol, and the low-affinity endogenous agonist adrenaline. The crystal structures reveal a highly conserved overall ligand recognition and activation mode despite diverse ligand chemical structures and affinities that range from 100 nM to ~80 pM. The adrenaline-bound receptor structure is overall similar to the others, but shows substantial rearrangements in extracellular loop three and the extracellular tip of transmembrane helix 6. These structures also reveal a water-mediated hydrogen bond between two conserved tyrosines, which appears to stabilize the active state of the β₂AR and related GPCRs.

G-protein coupled receptors (GPCRs) relay extracellular signals across a cell membrane via conformational change following binding of an extracellular agonist. GPCR activation by endogenous agonists remains poorly understood due to the paucity of active receptor structures in complex with agonists. While a number of GPCRs have been crystallized in recent years, only the β₂-adrenoceptor (β₂AR) and rhodopsin have been crystallized in fully active states^1,6, and in each case structures are available only for complexes with a single agonist. Due to the conformational plasticity and biochemical instability of agonist-bound receptors⁷, the few agonist-bound structures of GPCRs solved thus far have relied on either the use of covalent¹ or extremely high affinity agonists², crystallographic chaperones to trap active states (a G protein⁸ or antibody fragment⁹), or thermostabilizing mutations³. The last approach has only yielded structures of agonist-occupied receptor in partially active^3,4 or inactive⁵ conformations.

To better understand how diverse agonists can activate a single receptor, we developed a strategy for stabilizing active-state structures of the β₂AR bound to low affinity agonists including the natural agonist adrenaline. Here, we describe the directed evolution of Nb80, a conformationally-selective single-domain camelid antibody fragment (nanobody) that was used to obtain the first active-state structure of the β₂AR⁹. Comparison with the structure of β₂AR in complex with the G protein G_s confirmed that Nb80 stabilizes a physiologically relevant active state⁸. However, the β₂AR-Nb80 structure was of modest resolution (3.5 Å) and crystals could only be obtained with the high-affinity agonist BI167107; crystallization trials with catecholamine agonists were unsuccessful despite extensive screening. We reasoned that improving the affinity of Nb80 for agonist-bound β₂AR would decrease receptor conformational heterogeneity and enable crystallization of the receptor bound to low-affinity agonists. However, directed evolution of conformationally-selective GPCR-binding proteins has never been described, likely due to the challenges involved in biochemical manipulation of integral membrane proteins. We used yeast surface display together with a conformationally-specific selection strategy to improve the binding affinity of Nb80 while maintaining its conformational selectivity. The resulting high-affinity variants retain their specificity for the active state of the receptor characteristic of the original Nb80. Using the high-affinity variant Nb6B9, we determined a high-resolution (2.8 Å) active-state structure of the β₂AR bound to BI167107, and also determined the structures of the β₂AR bound to two catechol-containing agonists: hydroxybenzyl isoproterenol (HBI) and adrenaline, an endogenous low-affinity agonist of the β₂AR, at 3.1 Å and 3.2 Å resolution, respectively.

To assess the feasibility of engineering Nb80, we displayed Nb80 on the surface of the yeast strain EBY100 as an amino-terminal fusion to the yeast cell wall protein Aga2p (Fig. 1a). Yeast displaying Nb80 were stained with purified, detergent-solubilized, biotinylated β₂AR following pre-incubation of receptor with the agonist BI167107 or the inverse agonist carazolol. Nb80-displaying yeast specifically bound to β₂AR with an overwhelming preference for agonist-occupied receptor (Fig. 1b), with an EC₅₀ of 140 nM (Supplementary Fig. 1a). Next, we constructed a library of Nb80 mutants in which residues at the receptor-binding surface were randomized with conservative substitutions (Supplementary Fig. 2). The library was subjected to six rounds of selection (Fig. 1c). First, the library was positively selected with decreasing concentrations of BI167107-bound β₂AR. Prior to positive selection in rounds 2-5, the library was negatively selected against binding to inverse agonist-occupied β₂AR in order to remove variants that had lost conformational specificity. For the final round of selection, we enriched variants with the slowest dissociation rates. Receptor re-binding was blocked by the addition of a large excess of soluble Nb80 following the initial receptor-binding step (Supplementary Fig. 3). This selection strategy resulted in a progressive increase in binding affinity for agonist-occupied receptor without a similar increase in binding to inverse agonist-occupied receptor (Fig. 1d).

Figure 1. Conformational selection of nanobodies and characterization of high affinity Nb6B9.

a, Schematic representation of yeast display of Nb80. Nb80 is fused to the amino terminus of Aga2p, which attaches to the yeast cell wall through a covalent interaction with Aga1p. b, Staining of Nb80-expressing yeast with β₂AR bound to the agonist BI167107 (left) or the inverse agonist carazolol (right). c, Flow-chart summary of conformational selection process. d, Histogram overlays assessing β₂AR staining of the library at each round of selection. The left panel shows staining with 1 μM BI167107-occupied receptor, while the right panel shows staining with 1 μM carazolol-occupied receptor. e, Representative single-cycle kinetics SPR sensorgram of wild-type Nb80 (top) and engineered Nb6B9 binding immobilized β₂AR bound to BI167107. f, ³H-dihydroalprenolol (³H-DHA) competition binding shows a comparable increase in β₂AR affinity for adrenaline in the presence of Nb6B9 as with G protein G_s. ³HDHA affinity is largely unchanged in the presence of Nb6B9 (Supplementary Table 2). Data and error bars represent the mean ± standard error of the mean from three experiments.

Nanobody 6B9 (Nb6B9) was chosen from 23 variants screened from the final round of selection (Supplementary Fig. 4) as it represented one of the highest affinity binders tested, contained mutations that reached consensus among all sequenced clones, and was the most prevalent sequence observed. We expressed and purified Nb6B9 and Nb80, and then used surface plasmon resonance to measure binding kinetics and affinities. Nb6B9 bound to BI167107-occupied β₂AR with an affinity of 6.4 nM, a near 10-fold improvement over Nb80 (Fig. 1e). This increase in affinity resulted from a 13-fold reduction in the dissociation rate. Competition binding experiments revealed that the β₂AR bound adrenaline with a high affinity in the presence of 100 nM Nb6B9, which is comparable to the affinity observed in the presence of the G protein G_s (Fig. 1f).

We used the lipidic mesophase method¹⁰ to crystallize complexes of Nb6B9 with β₂AR bound to three different ligands, shown in Fig. 2a. Crystals grown with β₂AR bound to the high affinity agonist BI167107 showed strong diffraction, and a structure was obtained to 2.8 Å resolution (Supplementary Table 1). This represents a significant improvement over the previous 3.5 Å structure of β₂AR bound to the same ligand⁹. This higher resolution structure showed few differences from the Nb80 complex structure (Supplementary Fig. 5). Mutations in Nb6B9 appear to increase shape complementarity to active β₂AR (Supplementary Fig. 6). Many water molecules were clearly resolved for the first time, particularly in the extracellular region of the receptor (Fig. 2b, c). On the intracellular side of the receptor, a water molecule was found to mediate a hydrogen bond between Tyr326^7.53 of the NPxxY motif and the highly conserved Tyr219^5.58 on the intracellular side of transmembrane helix 5 (TM5), similar to a water seen in a recent structure of metarhodopsin II¹¹. Electron density suggestive of a water molecule was also seen in HBI- and adrenaline-bound β₂AR structures, despite their slightly lower resolution. The water-mediated hydrogen bond between Tyr219^5.58 and Tyr326^7.53 is possible only in the active conformation of the receptor (Fig. 2c), and the observed water-mediated hydrogen bond may therefore contribute to active state stability in the β₂AR and other GPCRs, serving as an active-state counterpart to the “ionic lock” which stabilizes the inactive state¹². In support of this notion, mutation of the corresponding Tyr223^5.58 to phenylalanine in rhodopsin decreases the stability of the meta II state¹³ and greatly reduces activation of transducin¹⁴. Moreover, mutation of Tyr227^5.58 to alanine resulted in the largest increase in thermostability for the inactive-state thermostabilized β₁AR¹⁵.

Figure 2. Structure of the activated β₂AR in complex with three agonists.

a, Chemical structures of the three ligands used for crystallization trials. b, All three active-state structures, showing remarkable similarity in overall receptor conformation. c, The 2.8 Å resolution structure of BI167107-bound β₂AR reveals active-state water molecules: a bridging water molecule participates in a polar network at the ligand binding site (top) and a second water molecule mediates a hydrogen bond between two highly conserved tyrosines. Such an interaction is possible in the active state (orange) but not the inactive state (gray).

While BI167107 exhibits many features typical of β₂AR agonists, it lacks the catechol moiety of the endogenous agonists adrenaline and noradrenaline. Hence, it is conceivable that these agonists stabilize a different conformation of the activated receptor-binding pocket. To assess this possibility, we pursued crystallographic study of complexes of Nb6B9 with β₂AR bound to the low-affinity endogenous agonist adrenaline and the high-affinity catecholamine agonist hydroxybenzyl isoproterenol (HBI). In each case, crystals could be grown in nearly identical conditions to those for the BI167107 complex, with clear electron density to identify the position and orientation of each ligand (Supplementary Fig. 7).

Despite the chemical diversity of these ligands, the structures of β₂AR bound to the catecholamine agonists and BI167107 show highly similar overall structures (Fig. 3a, b). A notable exception is a shift in the position of Asn293^6.55, which was previously determined to hydrogen bond with the amide carbonyl on the head-group of BI167107. The smaller catechol ring of adrenaline and HBI precludes hydrogen bonding with Asn293^6.55 in the receptor conformation observed in the BI167107-bound structure. In order to maintain the corresponding hydrogen bond between Asn293 and the meta hydroxyl moiety on the catechol ring, the receptor undergoes a 1.2 Å shift in the extracellular side of TM6, which bends toward the ligand (Fig. 3c). This shift alters the hydrogen-bonding network in this region and thereby causes a change in the conformation of His296^6.58. For adrenaline-bound β₂AR, the TM6 conformational change is further propagated towards the extracellular side of the receptor, leading to a conformational rearrangement in extracellular loop 3 (Fig. 3d). This change also alters the extracellular surface of the receptor, with adrenaline-bound β₂AR having a contracted extracellular vestibule (Supplementary Fig. 8).

Figure 3. Comparison of agonist binding modes.

a, Comparison of BI167107-bound receptor (orange) with HBI-bound receptor (green) shows highly conserved agonist binding mode. b, Similarly, adrenaline-bound (cyan) and HBI-bound (green) receptor structures are highly similar. c, An analogous comparison of BI167107-bound β₂AR (orange) with adrenaline-bound receptor (cyan) shows the similar polar networks for the two ligands (black dotted lines) with a notable difference in the hydrogen bonding of Asn293^6.55 to the amide proton in BI167107 (red dotted line) or the meta hydroxyl of adrenaline (blue dotted line) d, Due to this difference, Asn293^6.55 and TM6 shift inward in the adrenaline-bound structure,, leading to a cascade of changes culminating in a rearrangement of ECL3.

The relatively subtle differences in receptor conformation observed for the different co-crystallized agonists suggest that the activation mechanism of the β₂AR is highly similar for all agonists. Much like BI167107, the catechol headgroups of adrenaline and HBI engage β₂AR residues previously characterized to be important for agonist binding and receptor activation (Supplementary Fig. 9). Consistent with prior mutagenesis studies, Ser203^5.42 and Ser207^5.46 make hydrogen bonds with the catecholamine phenoxy moeities¹⁶ (Fig. 3a - c), and the conformation of these residues is nearly identical to that observed for the BI167107 headgroup. Ser204^5.43, which was previously thought to directly contact the para-hydroxy group of the catecholamine^16,17, engages the catecholamine headgroup indirectly in an extended polar network with Tyr308^7.35 and Asn293^6.55. Unlike BI167107-bound β₂AR, however, this polar interaction network is extended by inclusion of His296^6.58 in the catecholamine-bound receptor, suggesting that conformational rearrangement of His296^6.58 may stabilize the slightly smaller orthosteric binding pocket observed for catecholamine agonists. The agonist-induced rearrangements in the central portion of the transmembrane segments and intracellular surface are virtually identical in all three agonist-bound structures (Fig. 3b, Supplementary Fig. 10). Structures of β₂AR bound to BI167107 and the catecholamine agonists all show highly similar activation-related changes in the residues that connect the orthosteric ligand-binding pocket to the intracellular surface, suggesting that the mechanism for allosteric coupling between the orthosteric binding site and the G-protein coupling domain is likely a conserved feature of β₂AR activation. Therefore, different agonists stabilize the same conformational rearrangements in the receptor through different chemical interactions.

In contrast to the striking similarity in receptor conformation for all three agonists crystallized here, far more substantial conformational differences are seen relative to a previously reported structure of the thermostabilized turkey β₁ adrenoceptor (β₁AR) bound to the catecholamine agonist isoproterenol⁵. Likely due to the thermostabilization procedure, the overall receptor conformation of isoprenaline-bound β₁AR closely resembles that of the antagonist-bound, inactive β₁AR¹⁸, as well as that of covalent agonist-bound, inactive β₂AR¹ (Supplementary Fig. 11). Thus, a comparison of the structural changes between inactive β₁AR and active β₂AR, each bound to catecholamines, offers new insight into how agonists bind adrenoceptors both in the low-affinity state and the high-affinity, G protein-coupled state. Within the binding pocket, isoprenaline makes a hydrogen bond with Ser211^5.42 and the β-hydroxylamine moiety engages conserved residues Asp^3.32 and Asn^7.39 in a highly similar manner to the active state structures (Fig 4a, b). Similar interactions can occur in both active and inactive states, likely accounting for the fact that the β-hydroxylamine moiety is an important feature of both β-adrenoceptor agonists and antagonists/inverse agonists. However, the isoprenaline catechol headgroup engages a limited network of polar contacts in the inactive β₁AR structure, while adrenaline bound to active β₂AR engages an extensive polar network linking the orthosteric site to the extracellular loops (Fig. 4a). As a consequence of structural changes stabilized by this polar network, the catechol headgroup of adrenaline is nearly completely enclosed within the orthosteric binding pocket of activated β₂AR (Fig. 4c). In comparison, isoprenaline bound to inactive β₁AR is highly exposed to the extracellular solvent and is slightly displaced towards the extracellular side of the receptor (Fig. 4d). Thus, for the β adrenoceptors, the combination of a more extensive polar network and a smaller binding pocket likely accounts for the enhanced agonist affinity in the presence of either G_s or G protein mimetic nanobodies. Moreover, such differences between active and inactive structures highlight the importance of active-state GPCR crystal structures in understanding the structural basis for agonist activity.

Figure 4. Activation by catecholamine agonists.

For the first time, structures of catecholamine-bound adrenergic receptors in active and inactive conformations can be compared. a, The structure of β₂AR in an active conformation bound to the agonist adrenaline reveals an extended polar contact network linking the orthosteric site to extracellular loops 2 and 3, while the structure of thermostabilized β₁AR in an inactive conformation bound to a similar catecholamine agonist, isoprenaline, shows a far more limited polar network. c, Likewise, a surface view of the active state structure shows a substantial contraction of the binding site compared with d, the inactive β₁AR structure.

In conclusion, the use of new approaches in combinatorial biology has led to the development of Nb6B9, an exceptionally high-affinity GPCR-stabilizing nanobody. This molecule exhibits enhanced affinity for β₂AR relative to wild-type Nb80, and it allowed crystallization of β₂AR in complex with three different agonists with diverse chemical structures and a wide range of affinities. The use of such high-affinity crystallization chaperones may be generally useful in determination of active-state structures of GPCRs bound to low-affinity agonists. The crystallographic studies presented here reveal subtle, ligand-specific differences in receptor conformation superimposed on the backdrop of an overall conserved agonist binding mode and activation mechanism, offering new insight into how chemically diverse agonists can activate a single receptor.

ONLINE METHODS

β₂-adrenergic receptor expression and purification

Human β₂AR bearing an amino-terminal FLAG epitope tag and truncated after residue 365 was expressed in Sf9 insect cells using the BestBac baculovirus system (Expression Systems; Davis, CA). Cells were infected at a density of 4 × 10⁶ cells/mL, then incubated for two days at 27 °C. Receptor was extracted as described previously¹⁹. Receptor was first purified by FLAG affinity chromatography, then labeled with a 10-fold molar excess of biotin-PEG₁₁-maleimide (Thermo Scientific; Rockford, IL), which reacts with the endogenous residue Cys265. Following a one-hour incubation at room temperature, unlabeled receptor was blocked with 2 mM iodoacetamide for 15 minutes and then purified by alprenolol sepharose chromatography to isolate only functional receptor. Alprenolol sepharose eluate was concentrated on FLAG affinity resin, and then washed with ligand-free buffer for 30 minutes at room temperature to eliminate bound alprenolol. Detergent was gradually exchanged from dodecyl maltoside (DDM) to lauryl maltose neopentyl glycol (MNG) by washing in buffer containing decreasing amounts of DDM and MNG at a fixed concentration of 0.1% (w/w). Receptor was eluted, aliquoted, and frozen in 20% glycerol.

Display and functional evaluation of Nb80 on yeast

Nb80 was cloned into the carboxy-terminal Aga2 yeast-display vector pYAL²⁰ and transformed and displayed on yeast as previously described²¹. Induced yeast displaying Nb80 were washed with PBE buffer (phosphate buffered saline with 0.5 mM EDTA and 0.5% BSA) supplemented with 0.02% MNG detergent (PBEM buffer) and stained with varying concentrations of biotinylated receptor bound to either BI167107 or carazolol for one hour at 4°C. The yeast were then washed with PBEM buffer and stained with Alexa 647-conjugated streptavidin for 15 minutes at 4°C. Mean cell fluorescence was measured using the FL-4 channel of an Accuri C6 flow cytometer.

Affinity maturation of Nb80

The affinity maturation library was prepared by assembly PCR with oligonucleotide primers (Supplementary Table 3) containing degenerate codons at 15 distinct positions (Supplementary Fig. 2). The PCR product was further amplified with primers containing 50 basepairs of homology to pYal. Mutagenic nanobody DNA and linearized pYAL vector were co-electroporated into EBY100 yeast to yield a library of 0.8 × 10⁸ transformants.

For the first round of selection, 1.0×10⁹ yeast induced with SGCAA medium were washed with PBEM buffer and then resuspended in 5 mL of PBEM buffer containing 200 nM biotinylated β₂AR bound to BI167107. After one hour of incubation at 4 °C with rotation, yeast were washed with PBEM buffer, and then stained with Alexa647-conjugated streptavidin in PBEM buffer for 15 minutes at 4°C. Yeast were washed again with PBEM buffer and magnetically labeled with 250 μL anti-647 microbeads (Miltenyi) in 4.75 mL PBEM buffer for 15 minutes at 4°C. Yeast were washed a final time and labeled yeast were isolated by magnetic selection with an LS column (Miltenyi) pre-equilibrated with PBEM buffer. Magnetically-sorted yeast were resuspended in SDCAA medium and cultured at 30°C.

Rounds 2 – 6 were selected in a similar manner, with the following modifications. Prior to positive selection with agonist-occupied β₂AR, negative selection with antagonist-bound receptor was performed to select for clones that maintained a preference for the active state of the β₂AR. Briefly, 1.0 × 10⁸ yeast were washed with PBEM buffer and resuspended in 500 μL PBEM buffer containing 1 μM biotinylated β₂AR bound to carazolol. Yeast were incubated at 4 °C for one hour, then labeled with Alexa647 or PE-conjugated streptavidin, and magnetically labeled with 50 μL of the respective anti-fluorophore microbeads (Miltenyi) in 450 μL PBEM buffer. Magnetically-labeled yeast were applied to an LS column and the depleted flow-through was collected for subsequent positive selection. In this manner, yeast clones binding the ‘inactive,’ antagonist-occupied receptor were discarded. Positive selections on ‘active,’ agonist-occupied receptor were performed as for round one, but in a staining volume of 500 μL and with successively decreasing concentrations of BI167107-bound β₂AR: 200 nM receptor for rounds 2 and 3, 20 nM receptor for round 4, and 1 nM receptor for round 5. For round 6, positive selection was performed by a kinetic selection strategy to select for clones with the slowest off-rates (Supplementary Fig. 3). Briefly, yeast were stained with 200 nM biotinylated BI167107-bound β₂AR for one hour, washed with PBEM, and then resuspended in 500 μL PBEM containing 15 μM Nb80. The cells were incubated at 25 °C for 155 minutes, after which they were washed with ice-cold PBEM and stained with fluorescent streptavidin. Enrichment with magnetic separation for rounds 2 – 6 was performed as for round one, but with 50 μL anti-fluorophore magnetic microbeads with 450 μL PBEM buffer. Subsequent to round 6, post-sorted yeast were plated onto SDCAA-agar plates, colonies were picked and cultured in liquid SDCAA medium, and the plasmids encoding the nanobodies were isolated with a ZymoPrep Yeast Plasmid Miniprep II kit (Zymo Research) and sequenced.

Expression and characterization of Nb80 and Nb6B9

Nanobodies were cloned into the periplasmic expression vector pET26b, containing an amino terminal signal sequence and a carboxy-terminal 8×histidine tag and transformed into BL21(DE3) Rosetta2 E. coli (Novagen). Cells were induced in Terrific Broth at an OD₆₀₀ of 0.8 with 1 mM IPTG and incubated with shaking at 22 °C for 24 hours. Periplasmic protein was obtained by osmotic shock and the nanobodies were purified using nickel–nitrilotriacetic acid (Ni-NTA) chromatography⁹. For crystallography, Nb6B9 was digested overnight with 1:50 (w/w) carboxypeptidase A (Sigma) to remove the His tag, then purified by size exclusion chromatography over a Sephadex S200 size exclusion column.

Surface plasmon resonance experiments were conducted with a Biacore T100 at 25 °C. Protein concentrations were determined by 280 nm absorbance with a Nanodrop2000 spectrometer (Thermo Scientific). Biotinylated BI167107-bound β₂AR was immobilized on an SA sensorchip (GE) at an R_max of approximately 40 response units (RU). Biotinylated tiotropium-bound M₂ muscarinic receptor was immobilized with an RU value matching that of the reference surface to control for nonspecific binding. Measurements were made using serial dilutions of Nb80 or Nb6B9 in HBSM (10 mM HEPES pH 7.4, 150 mM NaCl, 0.01% MNG) using single-cycle kinetics. All data were analyzed with the Biacore T100 evaluation software version 2.0 with a 1:1 Langmuir binding model.

Radioligand binding assays were performed using purified β₂AR reconstituted into HDL particles comprised of apolipoprotein A1 and a 3:2 (mol:mol) mixture of POPC:POPG lipid²². Binding experiments with G-protein were performed as previously described⁸. Binding reactions were 500 μL in volume, and contained 50 fmol functional receptor, 0.5 nM ³H dihydroalprenolol (³H-DHA), 100 mM NaCl, 20 mM HEPES pH 7.5, 0.1% bovine serum albumin, and ligands and nanobodies as indicated. Reactions were mixed, then incubated for four hours at room temperature prior to filtration with a Brandel 48-well harvester onto a filter pre-treated with 0.1% polyethylenimine. Radioactivity was measured by liquid scintillation counting. All measurements were performed in triplicate, and are presented as means ± SEM.

Purification and crystallization of β₂AR-Nb6B9 complexes

Human β₂AR fused to an amino-terminal T4 lysozyme²³ was expressed and purified as described above. Following purification by alprenolol sepharose, the receptor was washed extensively with 30 μM of the low affinity antagonist atenolol while bound to FLAG affinity resin to fully displace alprenolol, then washed and eluted in buffer devoid of ligand to produce a homogeneously unliganded preparation. The receptor was then incubated for 30 minutes at room temperature with a stoichiometric excess of ligand (HBI or BI167107). A 1.3-fold molar excess of Nb6B9 was then added, and the sample was dialyzed overnight into a buffer consisting of 100 mM sodium chloride, 20 mM HEPES pH 7.5, 0.01% lauryl maltose neopentyl glycol detergent, and 0.001% cholesteryl hemisuccinate. In each case, ligand was included in the dialysis buffer at 100 nM concentration or higher. The sample was then concentrated using a 50 kDa spin concentrator and purified over a Sephadex S200 size exclusion column in the same buffer as for dialysis, and the β₂AR-Nb6B9-ligand ternary complex was isolated. In the case of adrenaline, the low affinity and chemical instability of the ligand precluded overnight dialysis, so 100 μM adrenaline was added to receptor for 30 minutes at room temperature, then a 1.3-fold molar excess of Nb6B9 added and the sample was incubated for 30 minutes at room temperature. Following incubation, the sample was concentrated and immediately purified by size exclusion as above.

Following purification, samples were concentrated to A₂₈₀ = 55 using a 50 kDa concentrator to minimize the detergent concentration in the final sample, then aliquoted into thin-walled PCR tubes at 8 μL per aliquot. Aliquots were flash frozen in liquid nitrogen and stored at -80 °C for crystallization trials. For crystallization, samples were thawed and reconstituted into lipidic cubic phase with a 1:1 mass:mass ratio of lipid. The lipid stock consisted of a 10:1 mix by mass of 7.7 monoacylglycerol (generously provided by Martin Caffrey) with cholesterol (Sigma). Samples were reconstituted by the two syringe mixing method¹⁰ and then dispensed into glass sandwich plates using a GryphonLCP robot (Art Robbins Instruments). In the case of the β₂AR-adrenaline complex, 1 mM fresh adrenaline was mixed with receptor prior to reconstitution. Crystals were grown using 30 nL protein/lipid drops with 600 nL overlay solution, which consisted of 18 – 24 % PEG400, 100 mM MES pH 6.2 to pH 6.7, and 40 – 100 mM ammonium phosphate dibasic. Crystals grew in 1 – 3 days, and were harvested and frozen in liquid nitrogen for data collection.

Crystallographic data collection and refinement

X-ray diffraction data were collected at Advanced Photon Source GM/CA beamlines 23ID-B and 23ID-D. The best diffracting crystals were identified by rastering, and wedges of 1–10° were collected using a 10 μm beam with typically 2 seconds exposure, 0.6° oscillation, and no beam attenuation. Data collection statistics are summarized in Supplementary table 1. Diffraction data were processed in HKL2000²⁴, and the structure was solved using molecular replacement with the structures of active β₂AR, Nb80 (PDB ID: 3P0G), and T4 lysozyme (PDB ID: 2RH1) used as search models in Phaser²⁵. The resulting structure was iteratively refined in Phenix²⁶ and manually rebuilt in Coot²⁷. Final refinement statistics are summarized in Supplementary table 1. Figures were prepared in PyMol.