Introduction

The beta-adrenergic receptors (βARs) belong to the G protein-coupled receptor (GPCR) family. As part of the sympathetic nervous system, βARs mediate physiological responses to the catecholamines norepinephrine and epinephrine to regulate cardiovascular, respiratory and metabolic functions. Norepinephrine is a neurotransmitter released from sympathetic nerves, whereas epinephrine is a hormone primarily released from the adrenal medulla into the systemic circulation. β1AR and β2AR are highly homologous and both are expressed in the heart,1 but they play distinct roles in regulating cardiac function. While these receptors have the same affinity for epinephrine, the β1AR has a tenfold higher affinity for norepinephrine than the β2AR. While both β1AR and β2AR activate Gs leading to activation of adenylyl cyclase and an increase in cAMP, they reside in distinct signaling domains in cardiac myocytes,2,3,4,5 which may contribute to their distinct roles in cardiac physiology.6,7,8 Epinephrine-activated β2AR also couples to Gi in cardiac myocytes, which inhibits adenylyl cyclase. In addition to having a lower affinity for the β2AR, norepinephrine does not promote coupling to Gi in cardiac myocytes.9 As discussed below, the differences in norepinephrine affinity and G protein coupling specificity suggest that the β2AR plays a minor role in physiologic regulation of cardiac function, but is protective in pathologic stress.

The sympathetic nervous system regulates cardiovascular response to changes in body position (supine to erect) and activity (sitting to walking and running). Under normal physiologic conditions, cardiovascular function is regulated primarily by norepinephrine released from sympathetic nerve terminals. Consistent with this, the major circulating catecholamine is norepinephrine that is released from sympathetic nerves and escapes into the circulation before undergoing reuptake at the synapse.10 Under more severe stress such as fight or flight, hypotension from hemorrhage or sepsis, or loss of cardiac function from infarction or heart failure, the adrenal gland releases epinephrine into circulation where it becomes the major circulating hormone.10 Thus, under normal physiologic conditions cardiac β2ARs contribute relatively little toward cardiac function because of their lower affinity for norepinephrine, their location relative to the sympathetic synapse,2 and the fact that norepinephrine promotes coupling of β2AR only to Gs, and not Gi. Under these physiologic conditions, coupling to Gi would be expected to antagonize the effects of Gs activation in enhancing cardiac performance. However, under extreme and prolonged stress, stimulation of the β1AR by high levels of epinephrine and norepinephrine leads to cardiac myocyte apoptosis, necrosis and remodeling,11,12 while stimulation of the β2AR leads to activation of anti-apoptosis and anti-necrosis pathways to protect the heart.13,14,15,16 The cardioprotective effects of β2AR activation can be inhibited by disrupting β2AR-Gi signaling, converting β2AR activation from anti-apoptotic to pro-apoptotic.14

Thus, β2AR activation is beneficial under conditions of prolonged pathologic stress to protect the heart, but should be minimized under physiological conditions to avoid opposing β1AR-mediated cardiac regulation through β2AR coupling to Gi or contribute to cardiotoxicity through β2AR coupling to Gs. Therefore, the lower affinity of the β2AR for norepinephrine appears to play an important role in cardiac physiology. This difference is surprising given that norepinephrine is only slightly smaller than epinephrine, which has the same affinity for both subtypes, and the amino acids that form the orthosteric binding pocket for epinephrine (defined here as within 4 Å of epinephrine bound to the β2AR) are identical for the β1AR and β2AR.

The human β2AR17,18,19 and turkey β1AR20,21 were among the first hormone-activated GPCR structures to be determined. So far, there are over 50 structure entries of the human β2AR and turkey β1AR bound to different agonists and antagonists in the protein data bank, yet none of these structures captured a complex between the receptor and norepinephrine. Despite the high sequence homology between the human β1AR, the turkey β1AR, and the human β2AR, these βARs have different pharmacologic properties (Supplementary information, Fig. S1).22 Structural information can explain the selectivity mechanism for ligands like the β2AR-selective salmeterol, that extend out of the orthosteric binding pocket into the more divergent extracellular vestibule.23 However, existing structures cannot explain the tenfold selectivity of norepinephrine for the human β1AR over the human β2AR, where the binding pockets are expected to be identical. Understanding the mechanism of the β1AR selectivity of norepinephrine may aid in the development of subtype-selective drugs not only for βARs, which are the primary drug targets of cardiovascular and pulmonary diseases, but also for other GPCRs. Factors that contribute to the affinity, but not evident in the crystal structure, include the rates of ligand association and dissociation.24 We therefore sought to investigate the molecular mechanism of norepinephrine selectivity for the human β1AR using binding kinetics studies, structural biology and metadynamics simulations.

Results

Association rate determines selectivity of norepinephrine

The binding affinity is determined by both association rate (Kon) and dissociation rate (Koff). We measured the binding kinetics of norepinephrine to both human β1AR and β2AR using competition association binding assays. Norepinephrine displayed faster association rates (on-rates, ~22-fold) but comparable dissociation rates (off-rate, ~1.5-fold) for the β1AR over the β2AR. The combined effects on the association and dissociation rate result in a lower Kd for norepinephrine for the β1AR, and agree with Ki measured in competition assays at equilibrium. While binding affinity differences can usually be attributed to differences in the dissociation rate, our results suggest that differences in the affinity of β1AR and β2AR for norepinephrine are mainly determined by the differences in on-rates (Fig. 1a; Supplementary information, Table S1).

Fig. 1: Kinetics studies of norepinephrine binding to the β1AR, β2AR and β1ARin / β2ARout, β2ARin / β1ARout chimeras.
figure 1

a Comparison of the Kon, Koff and Kd of norepinephrine to the β1AR, β2AR and β1ARin / β2ARout, β2ARin / β1ARout chimeras. Data are given as means ± SEM of 3–8 independent experiments. b Design of the β1ARin / β2ARout, β2ARin / β1ARout chimeras.

Given that the residues forming the catecholamine-binding pockets are identical between the β1AR and β2AR, we explored the role of the differences in the amino acid composition and thus shape of the extracellular vestibules. Two chimeric constructs were generated; one with the transmembrane core of the β1AR and extracellular vestibule of the β2AR (hereby named β1ARin2ARout), the other contains the transmembrane core of the β2AR and extracellular vestibule of the β1AR (hereby named β2ARin1ARout) (Fig. 1b; Supplementary information, Fig. S2). Binding kinetics studies on these two chimeric receptors were performed (Fig. 1a). Accordingly, the β2ARin1ARout chimera showed a much faster on-rate than β2AR (~60-fold) and β1ARin2ARout chimera showed a slower on-rate than β1AR (~30-fold). This suggest that the extracellular vestibules of the β1AR and β2AR are the key determinants of the different association rates of norepinephrine to the receptors, and thus of the different affinities. Interestingly, the chimeric receptor β2ARin1ARout displays an even faster association rate and higher affinity for norepinephrine than the wild-type β1AR. This may in part be due to the higher basal activity of β2AR compared to β1AR.25 The transmembrane core of β2AR might affect the equilibrium between inactive (low agonist affinity) and active (high agonist affinity) states; thereby allosterically affect the shape of the β1AR extracellular vestibule. In the β-arrestin recruitment assay, the EC50 values of the ligands to the different receptor constructs are in agreement with the ligand affinities measured in the study (Supplementary information, Fig. S3), the results suggest that mutating the extracellular surface does not interfere with the signal transduction at the intracellular side, since the residues involved in the downstream singling remain unchanged.

Shape of the extracellular vestibule differs between β1AR and β2AR

Delineating structural information on the extracellular vestibules is the key to understand the molecular mechanism of the different norepinephrine association rates between the human β1AR and β2AR. High-resolution structures of the human β2AR in inactive and active states have been solved,18,26,27 including β2AR bound to epinephrine.28 We therefore determined the structures of the human β1AR in both inactive and active conformations, bound to an antagonist (carazolol) and agonists (BI-167107, epinephrine and norepinephrine), respectively (Fig. 2a; Supplementary information, Fig. S4a, b). The crystal structure resolutions range from 2.5 to 3.1 Å (Supplementary information, Table S2). Clear densities of the ligands are revealed by fo-fc simulated annealing omit maps (Supplementary information, Fig. S5).

Fig. 2: Comparison of the orthosteric pockets and extracellular vestibules between the human β1AR and β2AR.
figure 2

a Chemical structures of the following β1AR ligands: inverse agonist carazolol, high-affinity agonist BI-167107, endogenous catecholamines epinephrine and norepinephrine. b Comparison of the orthosteric pockets of β1AR–norepinephrine structure (yellow) and β1AR–epinephrine (magenta) structure. c Comparison of the orthosteric pockets of β1AR–norepinephrine structure (yellow) and β2AR–epinephrine (blue, PDB code: 4LDO) structure. d Comparison of the extracellular vestibules of inactive (carazolol-bound, green) and active (BI-167107-bound, cyan) β1AR with inactive (carazolol-bound, PDB code: 2RH1, orange) and active (BI-167107-bound, PDB code: 4LDE, pink) β2AR structures.

The inactive and active β1AR structures display remarkable overall similarity to those of the β2AR. The classical structural change associated with receptor activation, the outward movement of TM5 and TM6 and the inward displacement of TM3 and TM7 on the cytoplasmic side, are also observed in β1AR structures. The features for conformational change also include the rearrangement of P2365.50, I2463.40 and F3336.44, representing the conserved PIF motif (Supplementary information, Fig. S4c, d), as well as the water-mediated hydrogen bond between Y5.58 and Y7.53 in the NPxxY motif (Supplementary information, Fig. S4e, f).

As expected, the catecholamine-binding pockets are similar between the β1AR and β2AR. When comparing the β1AR-norepinephrine and β1AR–epinephrine structures, the orthosteric binding pockets are almost identical. The additional methyl group of epinephrine is accommodated without any notable side chain rearrangement (Fig. 2b). When comparing the β1AR-norepinephrine structure with the β2AR–epinephrine structure (PDB: 4LDO), all of the orthosteric binding pocket residues occupy similar positions, except for slightly different conformation of F45.52 in the ECL2 (Fig. 2c). However, we are not certain if the conformational difference of F45.52 observed in the crystal structure is due to real conformational differences or due to local energy minimum of receptors affected by different crystallization conditions. Unbiased Molecular dynamics (MD) simulations (3 × 4 μs, each) did not reveal significant differences of F45.52 rotamers (Supplementary information, Fig. S6). This observation is consistent with the fact that the affinity of epinephrine is the same for β1AR and β2AR.

MD simulations of norepinephrine bound to the β2AR reveal a similar pose as observed in the β1AR crystal structure (Supplementary information, Fig. S4g, h). Hence, the binding poses do not explain the higher affinity of norepinephrine towards β1AR. Comparison of the extracellular vestibules of β1AR and β2AR reveals considerable structural differences, suggesting that differences in the connection between solvent and the orthosteric pocket may account for their differing affinities for norepinephrine (Fig. 2d).

Norepinephrine takes different paths to bind to β1AR and β2AR

To investigate how the different structures of the extracellular vestibule affect norepinephrine binding, we sought to understand the mechanism of ligand recognition by β1AR and β2AR using simulations. As ligand binding events tend to occur on time scales typically not broadly accessible by unbiased MD simulations, metadynamics simulations were performed to calculate the ligand entrance pathway using an established protocol to investigate the interaction of small molecules and GPCRs.29 Here, the protonated and thus positively charged states of norepinephrine and epinephrine and the inactive-state conformation of the two receptors were applied, as the extracellular sites of inactive-state receptors are expected to better resemble the unoccupied, non-G protein-coupled form of the receptors.30

Interestingly, the simulations suggest that norepinephrine takes different pathways to the orthosteric binding pockets in the β1AR and β2AR (Fig. 3). For the β2AR, norepinephrine first contacts the receptor through two negatively charged residues E180ECL2 and D300ECL3 located on the TM6 side of the ECL2/ECL3 vestibule (Fig. 3a, b, green residues), then moves through a tunnel formed by F194ECL2 and H2966.58 (Fig. 3a, b, yellow residues) to the TM2 side of the ECL2/ECL3 vestibule. To enter the final orthosteric pocket, norepinephrine needs to pass through a gate formed by the two aromatic residues F19345.52 and Y3087.35 (Fig. 3a, b, purple residues). The results agree well with the previously simulated alprenolol-binding pathway.31 As with the β2AR, norepinephrine also initially interacts with the β1AR through two negatively charged residues, D21745.51 and D3567.32 (Fig. 3c, d, green residues) and also needs to pass through two aromatic residues F21845.52 and F3597.35 (Fig. 3c, d, purple residues) to enter the orthosteric pocket. However, D21745.51 and D3567.32 are close to the TM2 side of the ECL2/ECL3 vestibule and as a result, norepinephrine does not need to move from one side of the ECL2/ECL3 vestibule to the other side in the β1AR.

Fig. 3: Metadynamics simulations reveal different norepinephrine entrance pathway in the β1AR and β2AR.
figure 3

a Side view of the norepinephrine-binding pathway in the β2AR. Norepinephrine first interacts with the receptor through two negatively charged residues (green), then moves through a tunnel formed by F194ECL2 and H2966.58 (yellow), then enters the orthosteric pocket by passing through a gate formed by two aromatic residues (purple). b Top view of the norepinephrine-binding pathway in the β2AR. c Side view of the norepinephrine-binding pathway in the β1AR. Norepinephrine first interacts with the receptor through two negatively charged residues (green). Before entering the orthosteric pocket, norepinephrine needs to pass through a gate formed by two aromatic residues (purple). d Top view of the norepinephrine-binding pathway in the β1AR.

Subtype-specific differences in the binding pathways

To ascertain why norepinephrine appears to take different pathways in these two receptors, the residues that may affect the ligand-binding process were thus interrogated. The well-studied turkey β1AR, which has a sequence identity of 82% to the human β1AR for the residues W571.31–D2595.73 and E3196.30–C3928.59 (the entire receptor excluding the N- and C-termini, as well as a part of ICL3), was also added to the analysis.

A pair of negatively charged residues and a salt bridge between the ECL2 and ECL3 are found in all the three receptors (Fig. 4a–c). The negatively charged residue pair where norepinephrine first interacts with the receptor is conserved in the human β1AR (D21745.51 and D3567.32) and turkey β1AR (D20045.51 and D3227.32), and is in a different position in the human β2AR (E180ECL2 and D300ECL3). The salt bridges are in different positions in the three receptors; they are formed by E205ECL2 and R351ECL3 in the human β1AR, by D19245.51 and K3057.32 in the human β2AR, and by D184ECL2 and R317ECL3 in the turkey β1AR.

Fig. 4: Analysis of different residues on the extracellular domain of the receptors that contribute to the different norepinephrine-binding pathway.
figure 4

a The negatively charged residue pair (E180ECL2 and D300ECL3) and salt bridge (D19245.51 and K3057.32) between the ECL2 and ECL3 of the human β2AR (PDB code: 2RH1). b The negatively charged residue pair (D21745.51 and D3567.32) and salt bridge (E205ECL2 and R351ECL3) between the ECL2 and ECL3 of the human β1AR. c The negatively charged residue pair (D20045.51 and D322ECL3) and salt bridge (D184ECL2 and R317ECL3) between the ECL2 and ECL3 of the turkey β1AR (PDB code: 2YCW). d The conserved F45.52 (cyan) is surrounded by two aromatic residues Y174ECL2 and Y3087.35 in the human β2AR (PDB code: 2RH1). e The conserved F45.52 (cyan) is surrounded by two aromatic residues W199ECL2 and F3597.35 in the human β1AR. f The conserved F45.52 (cyan) is surrounded by two aromatic residues W182ECL2 and F3257.35 in the turkey β1AR (PDB code: 2YCW). g Sequence alignment of ECL2 and ECL3 of the three human βARs and three turkey βARs. The residues that account for the different ligand entrance pathway are highlighted. h Mutating the six residues aligning the ligand entrance pathway of the β1AR to their counterparts in the β2AR (β1AR_6mut) decreased norepinephrine affinity, while the reverse mutations of the β2AR (β2AR_6mut) increased norepinephrine affinity. Data are given as means ±  SEM from three independent experiments performed in duplicate. The concentrations of [3H]-DHA for competition binding are varied from 0.15 to 0.9 nM based on the Kd for the distinct constructs.

The F194ECL2 and H2966.58 pair that coordinates norepinephrine entrance in the human β2AR is not conserved in either human β1AR (V219ECL2 and K3476.58) or turkey β1AR (V202ECL2 and N3136.58). The final aromatic gate through which norepinephrine enters the orthosteric pocket in the human β2AR is formed by a Phe (F19345.52) and a Tyr (Y3087.35). In both the human and turkey β1AR, the gate is formed by two Phe residues (F21845.52 and F3597.35 and F20145.52 and F3257.35, respectively). Interestingly, there is another aromatic residue on ECL2 that is within 4 Å distance of F45.52 and may affect the conformation or dynamics of F45.52, which is a Tyr in the human β2AR (Y174ECL2) and a Trp in the human and turkey β1AR (W199ECL2 and W182ECL2, respectively) (Fig. 4d–f).

Sequence alignment suggests eight residues on the extracellular side of the human β1AR, the human β2AR and the turkey β1AR contribute toward the different shapes of the extracellular vestibule of the receptors (Fig. 4g). Comparing the sequences of all three human βARs and the three turkey βARs at these eight positions reveals that the turkey β3CAR is identical to the human β2AR in all eight positions, while the turkey β4CAR has seven residues in common with the turkey β1AR and has a similar salt bridge between ECL2 and ECL3 as the turkey β1AR. The human β1AR and β3AR are more different compared to the other four subtypes (Fig. 4g). Interestingly, a full pharmacology study of these six receptors suggests that turkey β3CAR has a pharmacological profile similar to the human β2AR, and the turkey β4CAR is more similar to the turkey β1AR. Of note, these six βARs have identical orthosteric pocket residues22 (Supplementary information, Fig. S7).

Functional characterization of binding pathway mutations

Out of the above-mentioned eight residues, six residues are different between the human β1AR and β2AR, which are W199ECL2, V219ECL2, K3476.58, R351ECL3, D3567.32 and F3597.35 in the human β1AR and Y174ECL2, F194ECL2, H2966.58, D300ECL3, K3057.32 and Y3087.35 in the human β2AR. To determine whether the six amino acids define the entry path to the orthosteric site, and are responsible for norepinephrine’s selectivity for β1AR over β2AR, we characterized the pharmacology of mutants where the six residues were exchanged between the receptor subtypes. Substituting all six residues in β2AR with those of β1AR (β2AR-6mut) led to an enhanced affinity for norepinephrine (Fig. 4h). Further mutagenesis studies indicate that all six substitutions were minimally required to observe norepinephrine’s enhanced affinity (Supplementary information, Fig. S8). In contrast, while substituting all six residues in β1AR with those of β2AR (β1AR-6mut) decreases norepinephrine affinity (Fig. 4h), different combinations of three mutations produce a near-maximal decrease in affinity (Supplementary information, Fig. S8). The β1AR-F359Y showed the largest decrease on norepinephrine affinity among the 6 single mutations of the β1AR (Supplementary information, Fig. S8a), but the reverse mutation in the β2AR (β2AR-Y308F) did not show any increase in norepinephrine affinity (Supplementary information, Fig. S8c) and a combination of mutations were required to increase norepinephrine affinity in the β2AR, including a mutant without the Y308F mutation (β2AR-5mut-1) (Supplementary information, Fig. S8d). Similar results were reported for mutagenesis studies in turkey β1AR and human β2AR.21 Furthermore, a triple mutant (β1AR-3mut-3) without the F359Y mutation showed a comparable effect in decreasing norepinephrine affinity as the β1AR-6mut (Supplementary information, Fig. S8b). The complex behaviors of the mutants suggest that norepinephrine selectivity is determined by a combination of residues that cooperate to form the ligand-binding pathway (Supplementary information, Fig. S8). It should be noted that the β1AR-6mut and β1AR have similar affinities for epinephrine, while the β2AR-6mut has a slightly (less than threefold) increased affinity for epinephrine compared to the β2AR (Supplementary information, Fig. S9). Similar results are obtained with the β-arrestin recruitment assay where both epinephrine and norepinephrine maintain full agonist activity for the β1AR, β2AR, β1AR_6mut and β2AR_6mut receptors, but with different EC50 values in agreement with the Ki values measured in the binding assay (Supplementary information, Fig. S10). The results suggest the six residue mutations do not affect the overall conformation of the orthosteric binding pocket or the downstream signaling. Binding kinetics studies confirmed that mutating the six residues mainly affect the association rate of norepinephrine, the Kon of norepinephrine is ~7-fold lower in β1AR-6mut compared to β1AR, and ~16-fold higher in β2AR-6mut compared to β2AR (Supplementary information, Table S1). Of note, the fold-change in the association rate (and affinity) of norepinephrine for β1AR-6mut, or the β2AR-6mut, is not as large nor as complete as with the β1ARin2ARout or β2ARin1ARout chimeric receptors, which were constructed by exchanging the N-terminus (from the first residue to W1.31) plus another 55 amino acids on the extracellular half of the receptor that differ between the β1AR and β2AR (Fig. 1b; Supplementary information, Fig. S2). We speculate that other regions of the extracellular face contained in the chimeric receptors, but more distant from the orthosteric pocket may influence the electrostatic potential of the receptor, alter the conformation equilibrium between active and inactive states, or influence the overall shape of the extracellular vestibule (Supplementary information, Fig. S11).

The results suggest the existence of allosteric effects where other regions of the receptor may affect the conformation or dynamics of the residues forming the ligand-binding pathway or orthosteric pocket and thus affect the ligand affinity. To examine such an allosteric effect, we chose F45.52 as an example to perform mutagenesis studies. F45.52 is an important part of the entrance pathway and orthosteric pocket for both β1AR and β2AR. The residue is conserved as Phe in both the β1AR and β2AR, but several of the residues surrounding F45.52 are different between the two receptors (Fig. 4a–f) and as a result, F45.52 may have different conformational and dynamics properties, and play different roles in ligand affinities in the β1AR and β2AR. Indeed, while the F45.52A mutation only slightly reduced norepinephrine affinity for the β2AR (approximately threefold decrease in Ki value), it reduced affinity for the β1AR by more than 250-fold (Supplementary information, Fig. S12). Molecular dynamics studies suggest that the F45.52A mutation has a greater impact on the binding pathway for the β1AR than for the β2AR (Supplementary information, Fig. S13). The F45.52A mutation results in a higher fluctuation of ECL2 in the β1AR and higher fluctuation of ECL3 in the β2AR (Supplementary information, Fig. S13). Interestingly, F45.52A mutation induces alternative conformations of D21745.51 only in the β1AR, but not in the β2AR. As mentioned earlier, D21745.51 is part of the negatively charged residue pair where norepinephrine first interacts with the receptor. The results may explain why F45.52A mutation has larger effects on norepinephrine affinity in the β1AR than the β2AR. Of note, F45.52A mutation also has larger effects on epinephrine affinity in the β1AR than the β2AR, suggesting some common mechanisms of epinephrine and norepinephrine binding to the receptor (Supplementary information, Fig. S12).

Electrostatics distinguish epinephrine and norepinephrine binding

Simulating epinephrine binding to human β1AR and β2AR yielded surprisingly similar binding pathways to those of norepinephrine, thus to say, different between the β1AR and β2AR (Supplementary information, Fig. S14a–d). The question is why norepinephrine has β1AR selectivity whereas epinephrine does not. The chemical structures of norepinephrine and epinephrine are very similar except that epinephrine contains an additional methyl group on the primary amine (Fig. 5a). Both ligands are significantly basic (pKa 8.5–8.732) and hence largely exist in a protonated form under physiological conditions. The addition of the methyl group with an electron-donating effect at the amine modifies the electron distribution. According to quantum chemical calculations, the protonated nitrogen of epinephrine (+0.559) is less positively charged than the protonated nitrogen of norepinephrine (+0.631) (Fig. 5a). Due to the charge difference, the molecular electrostatic potential (MEP) of epinephrine is 3.9 kcal/mol smaller compared to norepinephrine along the N-H axis at the distance observed between the N-H hydrogen of the ligands and the negatively charged oxygen of Asp3.32 in the crystal structures of β1AR and β2AR (1.8 Å), indicating a weaker interaction of epinephrine with negatively charged parts of the receptor (Supplementary information, Fig. S15).

Fig. 5: Epinephrine does not have preference for the β1AR or the β2AR as norepinephrine does due to difference in electrostatic properties.
figure 5

a Comparison of partial charge of the protonated nitrogen and charge distribution of norepinephrine and epinephrine. Illustrated are the electrostatic potential charges of the two nitrogens, based on the sum of partial charges of the nitrogen and its attached hydrogen. The MEP is mapped onto the quantum mechanical isodensity surface (0.001 au) of norepinephrine and epinephrine, respectively. b The ligand-binding pathway of the human β1AR is composed of a continuous negatively charged tunnel. c The ligand-binding pathway of the human β2AR is composed of two negatively charged areas connected by a neutral gap. d The norepinephrine-binding energy profiles are different between the β1AR and β2AR. The red arrow indicates a local energy minimum of norepinephrine–β2AR complex at a CV distance of 1.2 nm. e The epinephrine-binding energy profiles show similar patterns in the β1AR and β2AR. f Snapshot of metadynamics simulations illustrating norepinephrine–β2AR complex for energy minimum at CV distance of 1.2 nm. Norepinephrine is engaged in an extensive hydrogen network with D1133.32, N3127.39 and Y3167.43, and water-mediated interactions to C19145.50, D19245.51, K3057.32, Y3087.35, and W3137.40. Of these residues, K3057.32 and Y3087.35 are different between the β1AR and β2AR (labeled in red font). The plot shows the distance representing the collective variable applied during metadynamics simulations during unbiased MD of norepinephrine in complex with β2AR. Norepinephrine maintains the binding mode for 1275 ns in unbiased MD simulations.

The electrostatic surface map suggests the ligand entrance pathway of the β1AR contains a continuous negatively charged tunnel. In contrast, the ligand entrance pathway of the β2AR is composed of two negatively charged areas separated by a neutral gap (Fig. 5b, c). As the protonated nitrogen of norepinephrine is more positively charged than for epinephrine, it may prefer the β1AR entrance over that of the β2AR, while epinephrine does not have such a preference. Indeed, even though epinephrine and norepinephrine may take the same binding pathway, they have different binding energy profiles. The norepinephrine binding energy profiles are different between the β1AR and β2AR (Fig. 5d; Supplementary information, Fig. S14e), while the epinephrine binding energy profiles show similar patterns between the two receptors (Fig. 5e; Supplementary information, Fig. S14e). Interestingly, norepinephrine seems to be transiently trapped in a pocket located before the orthosteric pocket in the β2AR (at a collective variable (CV) distance of 1.2 nm) (Fig. 5d, red arrow), which may partially explain the slower association rate to the orthosteric pocket. The binding mode is well maintained for 1275 ns in unbiased MD simulations (Fig. 5f). In binding kinetics studies, epinephrine showed an approximately threefold slower association rate than norepinephrine in the β1AR, which may be due to both the slightly larger size and the charge distribution difference (Supplementary information, Table S1). While in the β2AR, epinephrine showed a ~14-fold faster association rate than norepinephrine. Hence, epinephrine appears to migrate faster through the less continuously charged pathway of β2AR. This agrees with the lower charge and MEP value, produced by the shielding effect of the N-methyl group on the positively charged nitrogen. The epinephrine association rate is only approximately twofold faster in the β2AR than in the β1AR, suggesting a similar preference for the extracellular vestibule structures of the β1AR and β2AR. Epinephrine binding studies with the β1ARin2ARout and the β2ARin1ARout chimera receptors indicated that the observed difference in epinephrine association may be contributed by the transmembrane core of the receptor (Supplementary information, Fig. S16, Table S1b). The twofold difference in association rate may be a reflection of subtle differences in the equilibrium between active and inactive states and could partly explain why β1ARin2ARout has a faster dissociation rate and lower affinity than the other three receptor variants.

Discussion

We sought to address an interesting and physiologically important difference in the binding affinity of norepinephrine for the human β1AR and β2AR. Our results suggest that the selectivity is largely dictated by the association rate of the catecholamine, which might be the key to sympathetic nervous system function. The relatively slow association rate of norepinephrine to the β2AR might also contribute to the inefficient coupling of β2AR to Gi and differences in β2AR trafficking in neonatal cardiac myocytes compared to receptor activated by epinephrine.9 Crystal structures revealed identical orthosteric catecholamine-binding pockets, but different shapes and electrostatics of the extracellular vestibules of the β1AR and β2AR. Metadynamics simulations indicated that the two receptor isoforms have different ligand entrance pathways that account for the association rate difference. Interrogation of the entrance pathway through site-directed mutagenesis suggests that residues within the extracellular vestibule may serve as “selectivity filter” for norepinephrine. Even though epinephrine takes the same binding pathway as norepinephrine, the additional methyl group at its nitrogen alters the electron distribution on the catecholamine. Consequently, epinephrine does not show a preference for the extracellular vestibule of the β1AR or β2AR, while norepinephrine is transiently trapped in an intermediate binding site in the extracellular vestibule of the β2AR due to its higher positive charge density.

The results have a broader implication for drug development where receptors with identical orthosteric pockets could have different selectivity filters that define the pharmacology. Here, the selectivity filter is defined by the extracellular vestibule, a solvent-accessible tunnel, located above the orthosteric binding site, which is found in many class A GPCRs. Identifying subtype-selective drugs is a major goal in the development of safer and more efficacious therapeutics that target GPCRs. Our data suggest that efforts to develop subtype-selective ligands using structure-based drug design should consider the extracellular vestibule not only in the context of allosteric modulators or bitopic ligands, which bridge both the orthosteric site and the vestibule, but also for smaller orthosteric ligands.

Materials and methods

Human β1AR construct design and expression

In the human T4L–β1AR construct, T4 lysozyme was connected to S541.28 of β1AR with two alanine residues and the receptor was truncated at position 399 as previously reported for T4L–β2AR construct.33 The flexible ICL3 of T4L–β1AR (C261–L314) was removed and FLAG epitope (DYKDDDA) was fused to the amino-terminus of T4L–β1AR. The receptor was expressed in Sf9 insect cells with recombinant baculovirus (Bac-to-Bac expression systems) for 48 h at 27 °C.

Protein purification

For inactive-state β1AR structure determination, T4L–β1AR was solubilized in solubilization buffer (20 mM HEPES, pH 7.5, 100 mM NaCl, 1% LMNG and 0.02% CHS) and purified by M1 Flag affinity chromatography (Sigma) followed with a size exclusion chromatography in a final buffer comprised of 20 mM HEPES, pH 7.5, 100 mM NaCl, 0.01% LMNG, 0.0002% CHS and 10 μM carazolol.

For active-state β1AR structure determination, Nb6B9 with carboxyl-terminus His tag was expressed in E. coli strain BL21 (DE3) and purified by nickel affinity chromatography as previously reported.28 The T4L–β1AR was purified by M1 Flag affinity chromatography in presence of the target agonists (100 nM BI-167107, 1 mM epinephrine or 1 mM norepinephrine) and mixed with a 1.2-fold molar excess of Nb6B9 overnight at 4 °C. A nickel affinity chromatography was performed to pull down functional receptor. A final size exclusion chromatography purification was performed to remove excess Nb6B9. The size exclusion chromatography buffer was comprised of 20 mM HEPES, pH 7.5, 100 mM NaCl, 0.01% LMNG and 0.0002% CHS plus target agonists (100 nM BI-167107, 1 mM epinephrine or 1 mM norepinephrine, respectively).

Crystallization of β1AR

Purified protein was reconstituted into lipidic cubic phase by mixing with monoolein (Sigma) containing 10% (w / w) cholesterol (Sigma) in a 2 : 3 protein to lipid ratio (w / w) using previously reported two-syringe method.34 The β1AR–carazolol crystals were grown in 100 mM sodium citrate, pH 5.0, 150 to 175 mM lithium sulfate, 38% to 42% PEG300 and 3% to 6% 1,3-butanediol (Sigma). The β1AR–BI-167107 crystals were grown in 100 mM HEPES, pH 7.5, 100 to 175 mM ammonium sulfate, 38% to 43% PEG400 and 0.15 to 0.2 M Glycine. β1AR–norepinephrine and β1AR–epinephrine crystals were grown in 100 mM HEPES, pH 7.5, 100 to 175 mM sodium sulfate, 38% to 43% PEG300, and 1 mM ligands. Crystals appeared after 1 d and reached full size within 4 d at 20 °C.

Diffraction data collection and processing

The diffraction data was collected with the automatic data collection system ZOO35 at beamline BL32XU, SPring-8, Japan. The micro-focused beam with 1 Å wavelength and 10 × 15 μm size was used for automatic data collection. For each crystal, 5° or 10° dataset was collected with 0.1° oscillation per frame.

The diffraction data was automatically processed by KAMO.36 For β1AR–carazolol dataset, 1038 crystals were merged together to generate the final 2.5 Å dataset. For β1AR–BI-167107, β1AR–norepinephrine and β1AR–epinephrine datasets, 101 crystals, 318 crystals and 94 crystals were used to generate the final datasets, respectively.

Structure determination and refinement

The β1AR–carazolol and β1AR–BI-167107 structures were solved by molecular replacement using phenix37 with β2AR–carazolol (PDB code: 2RH1) and β2AR–BI-167107 (PDB code: 4LDE) structures as searching models, respectively, while the β1AR–norepinephrine and β1AR–epinephrine structures were solved using β1AR–BI-167107 structure as searching model. Structure refinement was carried out with phenix.refine in combination with manual building in coot.38 The final structure validations were performed with Molprobity.39 Statistics for data collection and structure refinement are summarized in Supplementary information, Table S2. All structure figures were prepared using PyMOL (the PyMOL Molecular Graphics System, Schrödinger, LLC).

Radioligand binding on βARs and mutants

The wild-type human β1AR, β2AR, β1ARin / β2ARout and β2ARin / β1ARout chimeras and turkey β1AR constructs were synthesized (GENEWIZ) and subcloned into pFastbac vector. Mutations for the human β1AR and human β2AR were generated by site-directed mutagenesis. Receptors were expressed in Sf9 insect cells with Bac-to-Bac expression system. For membrane preparation, 50 mL cells were centrifuged and homogenized in 8 mL lysis buffer (20 mM Tris, pH 7.5, 1 mM EDTA). The lysate solution was centrifuged at 800 rpm for 10 min. The supernatant was then isolated and centrifuged at 18,000 rpm for 20 min. Finally, the membrane-containing pellet was resuspended with binding buffer (20 mM HEPES, pH 7.5, 100 mM NaCl).

For competition binding assay, 100 μL diluted membrane-suspension was incubated with varying concentrations of cold ligands and 0.15–2 nM [3H]DHA in a buffer containing 20 mM HEPES, pH 7.5, 100 mM NaCl and 0.5 mg/mL BSA, to a final volume of 500 μL and incubated for 1.5 h with shaking at 200 rpm. The membranes were collected by filtration using Brandel 48-well harvester and filter papers containing the membrane were incubated with 3 mL OptiPhase HiSafe 3 liquid scintillation cocktail. The radioactivity was counted by Microbeta2 scintillation counter. Binding curves were fitted by GraphPad Prism 6.0 (GraphPad LLC, CA).

The binding kinetics assay

Dissociation binding kinetics of [3H]-DHA was measured by incubating membranes from Sf9 insect cells expressing wild-type human β1AR, β2AR, β1ARin / β2ARout or β2ARin / β1ARout chimeras with 0.1–0.5 nM [3H]-DHA in assay buffer (20 mM HEPES, pH 7.4, 100 mM NaCl, and 1 mM ascorbic acid) for 1 h, followed by the addition of an equal volume of propranolol (final concentration of 50 μM to prevent re-association of [3H]-DHA) and then harvesting on GF/C Unifilter™ (Perkin-Elmer) plates by vacuum filtration at different time points. The amount of [3H]-DHA bound to the filters was measured by liquid scintillation counting in a Top Count™ (Perkin Elmer) and the dissociation constants were determined using one phase exponential decay fit with GraphPad Prism 6.0 (GraphPad LLC, CA). Association binding kinetics were measured by incubating membranes with 3 different concentrations of [3H]-DHA (0.1–0.5 nM) for different times and determining the association binding constants using GraphPad Prism 6.0 (GraphPad LLC, CA). To determine the binding kinetics of unlabeled agonists, competition association assays were performed as previously described.40 Membranes were incubated with at least 3 different concentrations of the unlabeled agonist and a single concentration of [3H]-DHA (0.1–0.5 nM) for different times. Binding kinetics of the agonists were determined using the Kinetics of Competitive Binding fit in GraphPad Prism 6.0 (GraphPad LLC, CA) using the measured kinetic parameters for [3H]-DHA in the analysis.

β-arrestin-2 recruitment assay

The β-arrestin-2 recruitment assay was performed using the PathHunter assay (DiscoverX, Birmingham, UK). In general, HEK293 cells stably expressing (EA)-β-arrestin-2 were transiently transfected with a receptor containing the PK fragment tag using Mirus TransIT-293 (peqlab, Erlangen, Germany). After 24 h incubation in DMEM/F12 medium (Life Technologies, Darmstadt, Germany) at 37 °C and 5% of CO2, the cells were detached and transferred into 384-well plates at around 5000 cells/well density. After incubation for another 24 h, the catecholamines were added to the cell and incubated at 37 °C for an optimized time for each receptor. The detection mix was added and incubated at room temperature for 1 h. The chemoluminescence measurement for β-arrestin-2 recruitment was performed with a Clariostar plate reader (BMG, Ortenberg, Germany).

Setup of MD simulations

The simulations of the active state are based on the here reported crystal structures of β1AR in complex with norepinephrine and epinephrine or based on the β2AR crystal structure in complex with epinephrine (PDB code: 4LDO).28 For β2AR in complex with norepinephrine, the β2AR crystal structure in complex with epinephrine was applied. The structure was aligned with the active state structure of β1AR in complex with norepinephrine and the coordinates of norepinephrine were transferred. For the simulations systems of the inactive state used for the metadynamics simulations, the crystal structures of β1AR in complex with carazolol (reported here) and β2AR in complex with carazolol (PDB code: 2RH1)18 were applied. If simulated in complex with norepinephrine, the respective structure was aligned with the active state structure of β1AR in complex with norepinephrine and the coordinates of norepinephrine were transferred. In the case of simulations with epinephrine, either the coordinates of epinephrine in the crystal structure of β1AR or β2AR were used.

The structures were modified that they cover the same part of the protein sequence. The structures used for β1AR simulations covered the residues from number 54 to 256 and 318 to 391 and the structures used for β2AR simulations covered the residues from number 29 to 231 and 367 to 340. Missing side chains were completed utilizing UCSF Chimera41 and a palmitoyl group was added to Cys392 or Cys341 according to the UniProt entry of human β1AR42 and β2AR,43 respectively. The open N- and C-termini, as well as the terminal residues at the intracellular ends of TM5 and 6, were capped with neutral acetyl and methylamide groups.

For simulations of active states, all titratable residues were left in their dominant protonation state at pH 7.0 except for Asp2.50, Glu3.41 and Asp3.49. Previous studies of the β2AR suggest that Asp2.50 and Asp3.49 are protonated in the active state,44,45 and residue Glu3.41 directly contacts the lipid interface and therefore will also exist predominantly in its protonated state.46,47 We thus protonated these three residues in our simulations of the active state. For the simulation of inactive states, all titratable residues were left in their dominant protonation state at pH 7.0 except for Glu3.41, which was protonated in these simulations as well.

The protein structures were then aligned to the Orientation of Proteins in Membranes (OPM)48 structure of either active β2AR (PDB code: 3SN6) for simulations of the active state, or aligned to the structure of inactive β2AR (PDB code: 2RH1) for simulations of the inactive state. Each complex was inserted into a solvated and pre-equilibrated membrane of dioleoyl-phosphatidylcholine (DOPC) lipids using the GROMACS tool g_membed.49 Subsequently, water molecules were replaced by sodium and chloride ions to give a neutral system with 0.15 M NaCl. For active state simulations, the final system dimensions were roughly 80 × 80 × 120  Å3, containing about 150 lipids, about 17.500 water molecules, 71 sodium ions and 88 or 78 chloride ions for β1AR and β2AR, respectively. In the case of simulations of the inactive state, the final system dimensions were roughly 80 × 80 × 100 Å3, containing about 150 lipids, about 13.000 water molecules, 59 sodium ions and 73 or 64 chloride ions for β1AR and β2AR, respectively.

Parameter topology and coordinate files were built up using the tleap module of AMBER1850 and subsequently converted into GROMACS input files. For all simulations, the general AMBER force field (GAFF)51 was used for the ligands, the lipid14 force field52 for the DOPC molecules and ff14SB53 for the protein residues. The SPC/E water model54 was applied. Parameters for norepinephrine and epinephrine were assigned using antechamber.50 The structure of norepinephrine and epinephrine were optimized using Gaussian 1655 at the B3LYP/6-31G(d) level of theory and charges calculated at HF/6-31G(d) level of theory. Subsequently, atom point charges were assigned according to the RESP procedure described in literature.54 A formal charge of +1 was defined for norepinephrine and epinephrine.

MD simulations

Unbiased simulations were performed using GROMACS 2019.3.56,57 Multiple simulations were started. Each simulation system was energy minimized and equilibrated in the NVT ensemble at 310 K for 1 ns followed by the NPT ensemble for 1 ns with harmonic restraints of 10.0 kcal·mol–1 on protein and ligands. In the NVT ensemble, the V-rescale thermostat was used. In the NPT ensemble the Berendsen barostat, a surface tension of 22 dyn·cm–1, and a compressibility of 4.5 × 10–5 bar–1 was applied. The system was further equilibrated for 25 ns with restraints on protein backbone and ligand atoms. Here, the restraints were reduced in a stepwise fashion to be 10.0, 5.0, 1.0, 0.5 and 0.1 kcal·mol–1, respectively. Productive simulations were performed using periodic boundary conditions and a time step of 2 fs with bonds involving hydrogen constrained using LINCS.58 Long-range electrostatic interactions were computed using the particle mesh Ewald (PME)59 method with interpolation of order 4 and fast Fourier transform (FFT) grid spacing of 1.6 Å. Non-bonded interactions were cut off at 12.0 Å.

Analysis of the trajectories was performed using Visual Molecular Dynamics (VMD),60 and the CPPTRAJ61 module of AMBER18. Visualization was performed using PyMOL (the PyMOL Molecular Graphics System, Schrödinger, LLC). Plots were created using Matplotlib 3.0.2.62

Metadynamics simulations

Metadynamics simulations were performed to obtain the free energy profile of norepinephrine and epinephrine at the β1AR and β2AR aiming to derive meta-stable intermediate states suggesting a potential binding pathway of the two ligands. The simulations were performed using an inactive state structure of the respective receptor. The simulations were performed using GROMACS 2018.456,57 patched with the open-source, community-developed PLUMED library,63 version 2.564 and follows a recently described protocol by Saleh and coworkers to determine ligand binding modes at GPCRs.29 In brief, we used a combination of the well-tempered metadynamics (WT)65,66 and funnel-shaped walls. Following the above-described equilibration protocol, a metadynamics-history-dependent bias was applied along with the z component of the distance between the center of mass of the Cα atoms of Val3.36 and Trp6.48, representing the center of the receptor, and center of mass of the ligand atoms. This distance was used as a single collective variable. A funnel restraint was applied to the relative position on the xy plane to ensure better sampling for the relevant region. Gaussian hills with an initial height of 0.48 kcal·mol–1 applied every 1 ps were used. The hill width was chosen to be 1 Å. The Gaussian functions were rescaled in the WT scheme using a bias factor of 20. For each simulation system, an initial metadynamics simulation was performed until the ligand was unbound from the receptor. Using 16 frames extracted from this initial simulation, a multiple-walker metadynamics simulation was performed. For all systems, these frames cover a spectrum of ligand conformations ranging from the ligand in the orthosteric pocket to the ligand at the extracellular side of the receptor. The free energies were calculated using the sum_hills function of the PLUMED plugin64 and plotted against the distance of the CV using Matplotlib 3.0.2.62 For each distance representing a minimum or maximum on the free energy landscape, coordinates were extracted and clustered into ten groups based on the ligand atoms only using the CPPTRAJ61 module of AMBER18. Representative structures of a cluster were considered as potential meta-stable intermediate states if the cluster included at least 10% of the analyzed frames.

Charge and MEP calculations

The structure of norepinephrine and epinephrine were optimized, and charges were calculated using Gaussian 1655 at the MP2/aug-cc-pVDZ67,68,69 level of theory. The partial charges of the protonated nitrogens present the electrostatic potential charges with charges of hydrogens summed into heavy atoms. The molecular electrostatic potentials were visualized using UCSF Chimera.41