Research article

Divergent agonist selectivity in activating β1- and β2-adrenoceptors for G-protein and arrestin coupling

Ida Casella, Caterina Ambrosio, Maria Cristina Grò, Paola Molinari, Tommaso Costa


The functional selectivity of adrenergic ligands for activation of β1- and β2-AR (adrenoceptor) subtypes has been extensively studied in cAMP signalling. Much less is known about ligand selectivity for arrestin-mediated signalling pathways. In the present study we used resonance energy transfer methods to compare the ability of β1- and β2-ARs to form a complex with the G-protein β-subunit or β-arrestin-2 in response to a variety of agonists with various degrees of efficacy. The profiles of β1-/β2-AR selectivity of the ligands for the two receptor–transducer interactions were sharply different. For G-protein coupling, the majority of ligands were more effective in activating the β2-AR, whereas for arrestin coupling the relationship was reversed. These data indicate that the β1-AR interacts more efficiently than β2-AR with arrestin, but less efficiently than β2-AR with G-protein. A group of ligands exhibited β1-AR-selective efficacy in driving the coupling to arrestin. Dobutamine, a member of this group, had 70% of the adrenaline (epinephrine) effect on arrestin via β1-AR, but acted as a competitive antagonist of adrenaline via β2-AR. Thus the structure of such ligands appears to induce an arrestin-interacting form of the receptor only when bound to the β1-AR subtype.

  • β-adrenoceptor
  • β-arrestin-2
  • G-protein-coupled receptor (GPCR)
  • ligand efficacy
  • protein–protein interaction
  • resonance energy transfer


Although β1- and β2-AR (adrenoceptor) subtypes are structurally alike [1,2] and interact similarly with the G-protein Gs, which controls adenylate cyclase activity and cAMP-mediated signalling, many studies show that there are marked differences among subtypes in activating downstream signalling cascades [3] and eliciting functional responses [46].

One element underlying the functional diversity between β1- and β2-AR subtypes is the different topology of membrane location and cAMP signalling patterns that these two proteins display when co-existing on the surface of the same adult myocardiocyte [7]. Selective compartmentalization into specialized membrane microdomains [8,9] and the differential interaction of the C-termini of the receptors with PDZ-domain-containing proteins [1012] may be key factors that determine such location-dependent differences in signalling properties of the receptors.

However, there is also evidence that the interactions of β-AR subtypes with distinct signal transducers, such as Gs, Gi and β-arrestins, may differ, thus generating signalling diversity. Striking differences among the three β-AR subtypes were reported for the interaction with arrestins. It has been observed that the β3-AR subtype does not interact with arrestin [13], nor undergoes GRK [GPCR (G-protein coupled receptor) kinase]-mediated phosphorylation in response to agonist occupation [14], whereas the β1-AR was found to be less efficient than the β2-AR in interacting with arrestin [15].

Arrestins were originally considered to be molecular devices specifically designed to ‘arrest’ G-protein-receptor signalling. It is now clear, however, that these proteins, particularly β-arrestin-1 and -2, are fully fledged signal transducers [16,17], and the broadness of the signalling network they can regulate has been recently exposed by comprehensive phosphoproteomic analysis [18,19].

Several studies have focused on agonists that, acting on the same receptor subtype, can display differential efficacy for the interaction with the alternative tranducers G-protein and arrestin [17]. This transducer-dependent difference in agonist efficacy, commonly called biased agonism or ligand-directed signalling [3,20,21], has been investigated for the β2-AR subtype, where a number of agonists with a slight preference for arrestin interactions were reported [22,23]. However, an evaluation of how ligands differentially activate β1-AR and β2-AR in promoting G-protein or arrestin interactions is not available.

In the present study we used RET (resonance energy transfer) to compare the differential ability of the two AR subtypes to form a complex with each transducer in response to occupation by 45 distinct adrenergic structures. The results of the present study show that the β1-AR/β2-AR selectivity of adrenergics is strikingly different between arrestin and G-protein interactions.


Reagents and drugs

Cell culture media, reagents and FBS (fetal bovine serum) were from Invitrogen. Restriction enzymes were from New England Biolabs. Pertussis toxin was from List Biologicals. Coelenterazine and bDOC (bis-desoxycoelenterazine, sold as coelenterazine 400a) was from Biotium. All other biochemicals were purchased from Sigma–Aldrich. A clonal murine cell line (2B2 cells) previously isolated from embryos carrying the targeted ablation of Gnas exon 2 [24] was made available to us by Dr M. Bastepe (Harvard Medical School, Boston, MA, U.S.A.) and Dr O.H. Onaran (Ankara University, Ankara, Turkey). Adrenergic ligands were purchased from Bachem or Tocris, or were kindly donated by Dr Ad Ijzermann (University of Leiden, Leiden, The Netherlands). Details on source, structures and abbreviations for the ligands used in this paper are listed in Supplementary Table S1 (at

Plasmid constructs

β1-AR and β2-AR fused to Rluc (Renilla luciferase) were obtained by inserting the PCR fragments encoding each receptor cDNA into the Renilla Luciferase Vector (Packard), upstream of the Rluc-coding sequence. The resulting β1-AR–Rluc and β2-AR–Rluc chimaeras were then transferred into neomycin-resistance retroviral expression vectors (pQC series, Clontech).

RGFP (Renilla green fluorescent protein)-tagged transducers were made by linking the amplified RGFP-coding sequence (from Prolume) to the second codon of either the bovine β1 subunit of the heterotrimeric G-protein (Gβ1), through a 21-mer linker peptide (EEQKLISEEDLGILDGGSGSG), or to the β-arrestin-2 cDNA sequence, through a 13-mer peptide corresponding to the c-Myc epitope (EEQKLISEEDLGRT). Both constructs were transferred into hygromicin-resistant retroviral vectors (pQH), which were used to generate the viral supernatants used for cell transduction. Retroviral vectors expressing the long form of the human Gα protein GαsL were obtained by transferring the GαsL-coding sequence from the pcDNA3.1 plasmid (Missouri S&T cDNA Resource Centre) to the retroviral vector pQP bearing the puromycin-resistant gene coding sequence. We also prepared a luminescent β-arrestin-2 chimaera using a similar procedure. Rluc cDNA was fused to the β-arrestin-2-coding sequence using a 21-mer linker peptide (GDLGELSREEQKLISEEDLRT), and subcloned into a neomycin-resistant retroviral vector. This construct was used to prepare cells co-expressing luminescent arrestin and a membrane-targeted variant of RGFP, which carries at the C-terminus the farnesylation–palmitoylation consensus sequence of hRAS [25].

Cell lines

HEK (human embryonic kidney)-293 cell lines were cultured in DMEM (Dulbecco's modified Eagle's medium) and 2B2 cell lines in a 50% mixture of DMEM and F12, both media containing 10% FBS. Cell lines stably co-expressing each luminescent AR in association with either RGFP-tagged Gβ1 or RGFP–β-arrestin-2 transducers were obtained by infecting cells sequentially with retroviruses encoding the single fusion proteins followed by selection with G418 (500 μg/ml) in combination with hygromicin B (100 μg/ml). 2B2 cell lines additionally expressing exogenous GαsL were also obtained by viral transduction and puromycin selection. Clones expressing different ratios of chimaeric proteins were isolated after low-density plating of virally transduced cells, and 2–3 weeks of culture in the presence of the proper antibiotics.

Expression levels of luminescent and fluorescent chimaeric proteins

To measure the receptor density of chimaeric receptors we used (−)-[125I]pindolol (Amersham). Cell monolayers were detached using Ca2+/Mg2+-free PBS (PBS) containing 1 mM EDTA, and pelleted at 600 g. After resuspension in PBS, duplicate aliquots of ~100 000 cells were incubated with the radioligand (10 pM) with or without 12 log-spaced concentrations of unlabelled pindolol in a total volume of 1 ml. The reaction was carried out for 90 min at 20 °C, and terminated by rapid filtration on to GF/B glass fibre microplates (Filtermate 196, Packard). Radioactivity was counted and binding constants computed as described previously [26]. The Bmax (fmol/μg of proteins±S.E.M., n=3) measured in the main cell lines used in the present study were as follows. In 2B2 cells co-expressing RGFP–β-arrestin-2 the Bmax for β2-AR was 1.43 (±0.2) and for β1-AR was 1.22 (±0.12). In HEK-293 cells co-expressing RGFP–Gβ1 the Bmax for β2-AR was 2.6 (±0.73) and for β1-AR was 5.3 (±0.85). The Kd values (in pM±S.E.M., n=6) for pindolol were 386 (±77) at β2-AR and 918 (±138) at β1-AR.

Further characterization of the level of fusion proteins and their ratios in the cell lines and additional clones generated in the present study was performed by measuring the intrinsic luminescence and fluorescence of cell-membrane preparations or whole-cell extracts, according to transducer localization, as reported previously [25]. Enriched plasma membranes were prepared as described in [26].

Cholera-toxin-catalysed ADP-ribosylation

Cholera toxin labelling of α-subunits was performed using 100 μg of membrane proteins prepared from 2B2-KO (knockout)- and 2B2-Gαs-reconstituted cells as described previously [26]. Proteins were then separated by SDS/PAGE. The radioactivity of the bands corresponding to Gα subunits was quantified on a vacuum-dried gel with a microchannel array detector counter (Packard Instant Imager). To ADP-ribosylate endogenous subunits, cell monolayers were exposed to pertussis toxin (50 ng/ml) for 18 h prior to harvesting.

BRET (bioluminescence RET) recording of receptor–transducer interactions

The use of Renilla photoproteins as reporters of protein–protein interactions has been described previously [27]. Luminescence was recorded using 96-well white plates (Packard View Plate) using a plate luminometer (VICTOR light, PerkinElmer) equipped with two independent automatic injectors.

G-protein-coupling assays were performed on membranes prepared from HEK-293 or 2B2 cells expressing each luminescent receptor plus RGFP–Gβ1; membranes were incubated in PBS containing 2.5 μM coelenterazine for 10 min. Aliquots (5 μg/75 μl) of suspension were rapidly distributed into a white 96-well OptiPlate (PerkinElmer) already preloaded with 25 μl of PBS containing or not adrenergic ligands at 4-fold the final desired concentration. The plate was counted in the luminometer after 3 min of further incubation. Receptor–arrestin interactions were measured on monolayers of intact 2B2 cells. After overnight growth in 96-well OptiPlates, cell culture medium was replaced with 90 μl of PBS containing 2.5 μM bDOC, and incubated for 2 min prior to the addition of 10 μl of serial dilutions of the ligands (10-fold the final concentration in PBS), which had been prepared in a companion plate. Luminescence was recorded after an additional 10 min of incubation. Maximal effects for all ligands were determined using a single saturating concentration (10–100 μM). In addition, agonists displaying a significant effect in at least one of the four assays were further characterized by construction of concentration–response curves, using seven or 11 half-log-spaced concentrations of each ligand in duplicate wells. In both types of assays, adrenaline (epinephrine) was used as the reference ligand and included in every test plate to account for inter-assay variability of the estimated parameters.

Kinetic experiments were performed as described previously [25].

The direct effect of adrenergics on the luciferase activity of Rluc was measured as described previously [25]. Since several compounds had significant inhibitory activity (20–40%) on the enzyme at 1 mM, concentrations greater than 100 μM were avoided in the determination of BRET ratios. Consequently, for a few ligands (e.g. dopamine, its N-methyl analogue MAPE, and sulfonterol) the concentration–response curves in some assays did not reach a true experimentally determined plateau level. In such cases, the Emax value was extrapolated through the fitting routine (see below), by constraining the a parameter of the curve to be equal to or less than that for adrenaline. Because such parameter estimates are not obtained with the same degree of experimental confidence as the others, they were not included in the final comparison of intrinsic activities (Figures 6B and 6C) and are marked by asterisks in the final Tables.

Data analysis

RET ratios were determined as the ratios of high energy (donor) and low energy (acceptor) emissions, sequentially recorded with different filters [25]. Using coelenterazine, light was recorded through 450/20 nm and 510/20 nm windows, and RET ratios, corrected for spectral overlap, are calculated as: Embedded Image where T is the maximal transmittance in the two filter sets. With bDOC light was recorded through short-pass (450 nm cut-off) and long-pass (490 nm cut-off) filters and the RET ratio(bDOC)=cpsLP/cpsSP (LP is long-pass and SP is short-pass). The relative Emax of all ligands (intrinsic activities) were computed as the fraction of the Emax for adrenaline, after subtraction of the RET signal recorded in the absence of ligand. Concentration–response curves were analysed by non-linear curve fitting to the general logistic function: Embedded Image where y and x are the RET ratio and ligand concentration, a and d are upper and lower asymptotes, c is the ligand concentration yielding half-maximal RET change, and b is the slope factor at c. From best-fitting parameters, the EC50 (c) and Emax (a−d) of each ligand were obtained. The significance of the difference of fitted parameters among ligands were assessed according to the extra sum of squares principle [28].

Agonist-induced changes of RET kinetics were fitted to an exponential change function: Embedded Image Embedded Image where, Δt is the time difference (s) from ligand injection, Y0 is the baseline RET signal, and Yi and τi are amplitude and time constant (s) of n exponential components.


RET analysis of receptor–Gβ subunit interaction in membranes and the role of Gα subunits

Receptor–G-protein efficacy was measured using a previously described cell-free BRET assay [25], to eliminate the inhibitory effect of endogenous arrestins on the receptor–G-protein interaction. We used membranes from HEK-293 cells co-expressing either β2-AR—Rluc or β1-AR–Rluc and RGFP–Gβ1. The agonist-induced interaction between receptor and the β1 subunit of heterotrimeric G-proteins (Gβ1) in membranes results in enhancement of the RET signal, denoting a diminution of distance between the C-terminus of the receptor and the Gβ1 N-terminus. As discussed previously [25], it is not possible to discriminate whether such a change is due to the rearrangement of a pre-associated supramolecular complex or to micro-association reactions among proteins that diffuse along the plane of the membrane (and may perhaps be confined within specialized microdomains).

The interaction kinetics of β-ARs was similar to that previously observed for opioid receptors (t1/2, 10–15 s), reaching a plateau within 1 min from agonist addition, with no apparent differences between β2- and β1-AR subtypes (results not shown). The rank order of the potency of catecholamine for RET enhancement (isoproterenol≥adrenaline≥noradrenaline in β2-AR compared with isoproterenol>noradrenaline≥adrenaline in β1-AR) follows the typical pattern established in physiological studies (Figures 1A and 1B). Both agonist effect and block by antagonist displayed the distinctive enantioselectivity expected for β2-AR (Figures 1C and 1D). Thus the receptor–G-protein interaction described by the resonance signal appears to maintain the characteristics of native ARs, despite the presence of fused reporter tags on both partners.

Figure 1 Catecholamine preference and stereoselectivity for AR–G-protein coupling

Concentration–response curves for ligand-induced enhancement of the RET ratio in membranes prepared from cells co-expressing fluorescent Gβ1 and luminescent β2-AR (A, C and D) or β1-AR (B). All data were fitted using a general logistic function (solid lines) as described in the Experimental section. (A and B) Data are means±S.E.M. for three experiments. See Supplementary Table S1 (at for the means of EC50 estimates. (C) Data show means from two experiments with ranges given by error bars, and were replicated in an additional experiment using different enantiomer concentrations. Log EC50 values (mean±S.E.M., n=3) are −8±0.22 and −6.8±0.21 for the (−) and (+) isomer respectively. (D) Concentration-dependent inhibition by propranolol (propr) stereoisomers of RET enhancement induced by adrenaline (EPI). A representative experiment performed in the presence of 20 nM of agonist is shown [IC50, 1.4 nM for the (−) and 96 nM for the (+) isomer]. The experiment was repeated using 100 nM EPI or 10 nM isoproterenol (ISO) with similar results. The average (+)/(−) IC50 ratio from the three experiments is 85±18. NE, noradrenaline.

As previously observed for opioids [25], this signal results from a trimolecular interaction that requires Gα subunits. To investigate which Gα subunits are involved, we used a MEF (mouse embryonic fibroblast) cell line (2B2) established from transgenic mouse embryos that carry a targeted deletion of the Gαs gene [24]. Several lines co-expressing β1-AR–Rluc or β2-AR–Rluc and RGFP–Gβ1 were engineered. In addition, we reintroduced a functional ‘dark’ Gαs subunit in this host, by transducing β2-AR/Gβ1-2B2 cells with a retroviral vector coding for the ‘long’ spliced-variant form of the GNAS gene (GαsL).

Activation of the β2-AR was compared in membranes from 2B2 and 2B2-Gαs cells, both of which were treated or not with pertussis toxin overnight. Agonist-induced enhancement of RET was reduced by 80% in membranes of cells lacking the Gαs subunits, but a smaller signal still persisted. Pertussis toxin treatment abolished the agonist response in cells lacking the Gαs subunit, but produced only slight inhibition (~20%) in the presence of Gαs (Figures 2A and 2B). Similar pertussis-toxin-sensitive signals were observed in 2B2 cells expressing β1-AR (results not shown). These results indicate that although Gαs accounts for the majority of the interaction reported by the resonance signal, a smaller interaction supported by pertussis-sensitive Gαi protein exists in both receptor subtypes.

Figure 2 Interaction of β2-AR–Rluc with endogenous Gα subunits

Membranes of 2B2 cell lines expressing β2-AR–Rluc and RGFP–Gβ1 with (A) or without (B) GαsL were assayed for RET enhancements in the absence (basal) or presence of the agonist isoproterenol (ISO, 1 μM) and the antagonist ICI 118551 (ICI, 10 μM). The same membranes were ADP-ribosylated in vitro with cholera toxin (see the Experimental section) in the absence or presence of the ligands as indicated (see lane numbering in D), and separated by SDS/PAGE. (C) Autoradiogram of the dried gel. (D) Radioactivity counted in the 41 kDa bands (Gαi, left-hand histogram) and in the 46/48 kDa doublet (Gαs, right-hand histogram) using a Packard Instant Imager.

To confirm this suggestion, we measured direct interactions of β2-AR–Rluc with endogenous α-subunits using agonist-induced enhancement of cholera-toxin-catalysed ADP-ribosylation in the same 2B2 and 2B2-Gαs membranes. In the absence of Gαs, isoproterenol enhanced [32P]ADP-ribose incorporation into the 40 kDa band corresponding to Gαi, and the effect was inhibited by the β-blocker ICI-118551. In membranes reconstituted with Gαs, the agonist primarily increased the labelling of a 46/48 kDa doublet corresponding to GαsL, but no clear effect on Gαi labelling was detected in this case, suggesting that in the presence of Gαs the interaction with Gαi may be negligible (Figure 2C). The agonist-induced enhancement of cholera toxin labelling of the 40 kDa band in 2B2 membranes was abolished following treatment of the cells with pertussis toxin (50 ng/ml, 18 h), suggesting that the labelled protein is a member of the Gi/o family of G-proteins (results not shown). Such results also indicate that the ability of the luciferase-fused receptor to interact with endogenous α-subunits in the membrane is preserved.

As observed in the opioid receptor system, the adrenergic RET signal was rapidly suppressed by guanine nucleotides. We used this allosteric signal-quenching reaction to compare the relative potency of GDP to inhibit adrenaline-induced interactions at the two receptor subtypes. The IC50 for GDP was 10-fold lower at the β1 receptor than at the β2 receptor (Figure 3), suggesting that the β2-AR subtype can form a much more stable complex with Gαβγ than the β1-AR, despite the similarity in interaction kinetics.

Figure 3 Different potency of GDP in inhibiting G-protein coupling at β1-AR and β2-ARs

(A) Membranes from cells expressing β1-AR-Gβ1 (●) or β2-AR–Gβ1 (○) were assayed for RET in the presence of 10 μM adrenaline (EPI) and increasing concentrations of GDP (x axis). Data are averaged from two separate experiments and are expressed as a fraction of the effect measured in the absence of nucleotide. (B) Membranes were prepared from two cell clones expressing different levels of β2-AR–Rluc and RGFP–Gβ1 and the effect of GDP was measured as described in (A). Data are means of triplicate determinations in a single experiment. To test for a difference of IC50 in the two clones, data were first fitted with no constraints, and then refitted by forcing a common parameter c. The second fit was not significantly worse (P=0.37) as determined by extra-sum-of-square statistics [28], indicating that the IC50 of GDP is similar in the two curves. Thus GDP potency does not depend on the magnitude of RET signal measured in the two membranes.

Receptor–arrestin interaction: kinetics and role of Gα subunits

Our objective was to determine the intrinsic activity of agonists for each AR–arrestin coupling without the influence of the concurrent receptor–G-protein interaction. Thus 2B2 Gαs-KO cells co-expressing each luminescent receptor type and fluorescent β-arrestin-2 were prepared. In addition, β2-AR–arrestin co-expressing cells were also transduced with dark GαsL, to evaluate if, and to what extent, the presence of Gαs might modify the interactions.

Both β1-AR and β2-AR interact rapidly with arrestin, approaching steady-state within 5 min following agonist injection to the cell monolayer. However, the initial rate of the β2-AR subtype is somewhat delayed by a slower transient of RET increase in the first 30 s (Figure 4). Modelled by mono-exponential functions, the t1/2 of the adrenaline-induced β1-AR–arrestin interaction was smaller (35±7 s) than that of β2-AR–arrestin (76±9 s). Partial agonists displayed reduced maximal effects with slower rates in both β1-AR and β2-AR (Figures 4A and 4B), whereas lowering the molar concentration of agonist primarily reduced amplitudes, and not rates, for concentrations >30 nM (Figure 4C).

Figure 4 Kinetics of the enhancement of RET ratios in 2B2 cell lines expressing luminescent β1-AR or β2-AR, and fluorescent β-arrestin-2

Microwells containing cells expressing β2-AR (A) or β1-AR (B) were injected at time zero with the agonists indicated (all 10 μM), and the progress of luminescence was recorded continuously at 0.5 s intervals. Values of t1/2 (s) from mono-exponential fits (solid lines) are: in (A), 76±7 for adrenaline and 161±21 for terbutaline; in (B), 32±3 for adrenaline, 72±5 for terbutaline and 143±15 for clenbuterol. (C) Different concentrations of adrenaline (as indicated) were added at time zero to cells expressing β2-AR; t1/2 (±S.E.M. from the fit) are: 54 (±9) at 10 μM, 85 (±11) at 0.1 μM, 57 (±7) at 100 nM and 173 (±21) at 30 nM. (D) Microwells containing cells expressing β2-AR (▲, Δ) or β1-AR (■, □), pre-treated (Δ, □) or not (▲, ■) with 100 ng/ml pertussis toxin (PTX) for 18 h, were injected at time zero with 10 μM adrenaline. Computed t1/2 (s) are: β2-AR, 87 (±9) control and 76 (±7) PTX; β1-AR, 39 (±4) control and 37 (±5) PTX. Data in (A) are means (±S.E.M., grey bars) for four experiments. All other plots are representative of experiments repeated at least three times (although using different time sampling or type of partial agonists), which yielded comparable results.

We also found that for β1-AR RET kinetics, a two-component exponential model (with t1/2 ranging from 3 to 7 s and from 30 to 100 s) often afforded significant reductions of the fit standard error. This is similar to the biphasic kinetics described previously for the β2-AR–β-arrestin-2 interaction in single-cell imaging FRET studies [29]. However, we failed to find a consistent relationship between agonist efficacy and t1/2 or amplitudes of the two components. For β2-AR kinetics, the s-shaped inflection in the initial time course prevented significant improvements by fitting a sum of exponentials, unless two components with amplitudes of opposite sign were allowed in the model. Although improving fitting statistics, the physical interpretation of such a model is unclear. For this reason and because estimating both amplitudes and rates from multi-exponential fits of experimental data is a notorious ill-posed problem [30], all comparisons of kinetic parameters in the present study are based on mono-exponential approximations of the data.

Regardless of the complexity, the data in Figure 4 indicate that the maximal enhancement of RET at steady-state is a good descriptor of ligand efficacy for both β1- and β2-AR subtypes. Thus this parameter was chosen to assess ligand intrinsic activities for receptor–arrestin coupling in subsequent studies.

As observed for the G-proteins, concentration–response curves of catecholamines for arrestin coupling displayed the typical signature of β1-/β2-AR pharmacology (catecholamine potencies and agonist stereoselectivity are in Figure 5). Interestingly, the EC50 of the three catecholamines for β1-AR were smaller at arrestin than at G-protein, whereas this trend was reversed for β2-AR (see Table 2 and Supplementary Table S1).

Figure 5 Catecholamine preference and stereo selectivity for AR–arrestin coupling

Concentration–response curve for catecholamine-induced enhancement of RET ratio in 2B2 cell lines co-expressing fluorescent β-arrestin-2 and luminescent β1-AR (top panel) or β2-AR (bottom panel). Results are from a representative experiment where the four catecholamines were compared in the same microplate. The log EC50 values for isoproterenol enatiomers are: at β1-AR, −8.1 (−) and −6.8 (+); at β2-AR, −7.4 (−) and −6.1 (+). Table 2 and Supplementary Table S1 (at provide means of catecholamine EC50 values. EPI, adrenaline; ISO, isoproterenol; NE, noradrenaline.

Figure 6 β1-/β2-AR selectivity profiles of ligand intrinsic activity for G-protein and arrestin

(A) The shift of intrinsic activity (y axis) was computed as the net difference between β1-AR and β2-AR intrinsic activities measured for G-protein coupling (grey) and arrestin coupling (black) using the cumulative data shown in Table 1. Error bars were calculated from the sum-of-variances of the data. Positive numbers indicate the net β1-AR preference and negative numbers indicate the β2-AR preference. Data are shown in histogram form, sorted according to decreasing net difference. Asterisks mark ligands for which plateau Emax values could not be reached with the highest concentration, in at least one of the four assays (see the Experimental section). Ligand abbreviations are as listed in Supplementary Table S1 (at (B and C) β1-AR intrinsic activities plotted as a function of β2-AR intrinsic activities for arrestin- (B) and G-protein- (C) coupling assays. Data are ordered according to increasing β2-AR effects. Broken lines are the line of identity, whereas the solid lines were traced using a polynomial function to approximate the general trend of the data points. Ligands marked with arrows (grey triangles) showing most evident variations from the general trend are discussed in the text. (D). The β1-AR and β2-AR–arrestin intrinsic activities of the group of ligands exhibiting highest β1-AR selectivity (arrows in B) are replotted with those of ligands having similar effects on both receptors. Note the reversal in rank ordering of efficacy. With respect to, for example, albuterol, the relative Emax is progressively increasing at β1-AR, while concomitantly fading to undetectable levels at β2-AR.

To verify that the C-terminal Rluc extension of ARs does not modify the intrinsic ability of the receptor to interact with arrestin, we also developed an indirect BRET system capable of detecting arrestin binding to intact wild-type receptors for the β2-AR subtype. We took advantage of an mtRGFP (membrane-targeted RGFP) variant previously used to monitor receptor internalization [25]. This protein is localized in close proximity to luminescent receptors thus producing a high RET signal, which fades away as agonist binding triggers receptor endocytosis. Exploiting the same principle, we could detect the docking of luminescent arrestin to ‘dark’ wild-type β2-AR through the increase of proximity-induced RET between the arrestin-tethered Rluc donor and the membrane-anchored RGFP acceptor. In HEK-293 cells co-expressing mtRGFP plus Rluc–β-arrestin-2, and further transduced with wild-type β2-AR, the intrinsic activity of agonists was in good agreement with that measured using the conventional assay in 2B2 cells (Supplementary Figure S1 at, indicating that Rluc tagging does not change receptor efficacy for arrestin. The agreement between the two assays also suggests that the bulk of BRET signal that we measure in the present study is generated by an arrestin–receptor complex present on the cell surface rather than segregated into endocytic vesicles. In fact, unlike the direct interaction between luminescent receptor and fluorescent arrestin, the indirect RET between receptor-bound luminescent arrestin and membrane-anchored RGFP is interrupted as the receptor–arrestin complex undergoes endocytosis [25].

The role of Gαs on the receptor–arrestin interaction was evaluated by comparing the relative Emax of a series of adrenergic agonists in β2-AR–β-arrestin-2 cells expressing or not the co-transfected GαsL gene. The correlation between the intrinsic activities of ligands measured in the two cell lines was tight (Supplementary Figure S2 at, indicating that the coupling with Gα subunits does not manifestly change the efficiency of ligands in inducing receptor–arrestin interaction. Similarly, there was no influence of functional Gαi subunits when agonist intrinsic activities for β2-AR or β1-AR were compared in cells that had been exposed or not to pertussis toxin (results not shown), nor did pertussis toxin significantly alter the kinetics of the RET signal triggered by full agonists in β2-AR–β-arrestin-2 or in β1-AR–β-arrestin-2 cells (Figure 4D).

Thus, taken together, these results show that the receptor–G-protein interaction does not appear to modify the binding of receptor to arrestin reported by RET, neither directly via protein–protein associations, nor indirectly via activation of signalling pathways.

Effect of apparent receptor–transducer stoichiometry on intrinsic activity

To investigate how the relative Emax of ligands may depend on the level of expression of receptor–Gβ-interacting partners for each receptor subtype, we selected clones with a different stoichiometry of the two chimaeric proteins, as determined by the luminescence and fluorescence of the tethered reporters. The stoichiometry of expression had no detectable effect on the relative intrinsic activity of agonists for both receptor subtypes (Supplementary Figure S3 at Interestingly, however, the overexpression of ‘dark’ Gαs subunit significantly increased the intrinsic activity of the partial agonist clenbuterol for the β2-AR (Supplementary Figure S3D). The cell lines used in the present study had comparable levels of Gαs expression in immunoblots (results not shown). Nonetheless the final results were collected using at least two different clones for each receptor subtype to minimize the chance that unknown differences in endogenous G-protein subunits could bias the final results.

For the study of arrestin coupling we could not obtain lines showing a significant variation in the expression of fluorescent arrestin. Thus cell lines expressing equal levels of luminescent β2-ARs or β1-ARs and fluorescently labelled arrestin were selected and used for pharmacological analysis. However, to evaluate how a change of expression stoichiometry could affect the results, we generated a cell line expressing a 2.5-fold higher level of β2-AR luminescence, with no difference in fluorescence expression. The raised receptor/arrestin ratio in cells expressing a higher level of receptor slightly increased the intrinsic activity of a number of partial agonists (Supplementary Figure S4 at This indicates that in the β2-AR and β1-AR cell lines used for the comparison, the receptor/arrestin ratio is safely below the critical 1:1 proportion, where even small differences (easily masked by experimental noise in the determination of intrinsic luminescence and fluorescence) might affect the comparison of relative Emax of ligands across receptors. (More discussion on this point is in the Supplementary Material section entitled “Analysis of ligand-induced receptor activation using RET methods” at

β1-AR and β2-AR ligand intrinsic activities for G-protein and arrestin interactions

Using the BRET interaction assays described above, we compared the intrinsic activity of 45 adrenergic ligands, including antagonists and agonists endowed with various degrees of efficacy. The entire data set is available in Supplementary Table S1. Mean intrinsic activity data pooled from both concentration–response curves and determinations at a single saturating concentration are also reported in Table 1, while the EC50 values are in Table 2.

View this table:
Table 1 Ligand intrinsic activities (Emax minus basal as fraction of adrenaline values)

Weighted means from both single saturating concentration experiments and concentration–response curves (the latter were given double weight in the final mean calculation).

View this table:
Table 2 Comparison of the potency of adrenergic ligands (expressed as a log10 molar value of the EC50) for inducing β2-AR or β1-AR coupling to G-protein and arrestin

Only ligands for which a comparison can be made are presented here. The complete data set is given in Supplementary Table S1 (at Err, the range of values for n=2 or S.E.M. for n>2; No crc, the ligand effect was not sufficient to obtain a reliable concentration–response curve. Data extrapolated from computer fittings are labelled with asterisks as in Table 1.

To mark the ligands exhibiting the largest β1-/β2-AR difference in maximal complex formed with each transducer we plotted the net difference between relative intrinsic activities (β-AR1 minus β2-AR) of all ligands. This index arbitrarily sets β1-AR selectivity as positive and β2-AR selectivity as negative numbers (Figure 6A).

For receptor–G-protein interactions, cimeterol is the ligand with the highest level of β2-AR selectivity, followed by MAPE and N-methyl-dopamine (although Emax values of the latter two ligands may be underestimated, given the very low potency for the β1-AR–G-protein interaction). Other structures with significant β2-AR preferential effects include well-known β2-AR agonists, such as albuterol, clenbuterol and terbutaline. A direct plot of G-protein intrinsic activity for the two receptor subtypes (Figure 6C) shows an overall β2-AR preference in promoting G-protein coupling for the majority of ligands, but no marked reversal of the ordering of efficacy. Only CGP-12177 and xamoterol (β2-AR antagonist and partial agonist respectively) show significant β1-AR preference (Figure 6C and Table 1). This preserved ranking of efficacy across β1 and β2 receptor subtypes for G-protein coupling largely agrees with a recent study based on cAMP signalling measurements [31].

For the receptor–arrestin interaction, the trend is reversed. The direct comparison of intrinsic activity data shows an inverse pattern compared with G-protein (Figure 6B). Many ligands with β2-AR preference on G-protein show β1-AR preference on arrestin. Even two notorious β2-AR agonists, such as cimeterol or clenbuterol, gain a slightly greater β1-AR intrinsic activity in promoting receptor–arrestin interaction (Figure 6A and Table 1).

Such a switch towards β1-AR preference as the transducer shifts from G-protein to arrestin is also apparent on examination of the EC50 values of ligands (Table 2). For G-protein coupling most ligands exhibit significantly smaller EC50 values at the β2-AR subtype, but the ratio of potency in favour of β2-AR is drastically reduced for arrestin coupling (Table 2 and Supplementary Table S1). Most impressive is the change for clenbuterol, with a 200-fold shift towards the β1-AR as we move from G-protein to arrestin. Also interesting is the potency of noradrenaline: the EC50 ratio shows only a 2-fold β1-AR selectivity for G-protein coupling, but rises to 160-fold for the arrestin interaction (Table 2).

Note that although carvedilol was reported to induce translocation of arrestin to the membrane via β2-AR [32,33], in the present study we only see a minor effect of carvedilol at β1-AR and none at β2-AR for arrestin coupling (Table 1). It would be interesting to investigate whether this discrepancy only reflects a technical difference, or underlies an important functional separation between the biological events that are monitored by the two types of assay.

Agonists with restricted β1-AR–arrestin efficacy

In addition to the global shift towards β1-AR observed for most structures, a number of agonists (particularly dobutamine, synephrine analogues, dopamine and ritodrine) exhibit a clear reversal in the relative order of β2-/β1-AR efficacy for arrestin. This is best illustrated in Figure 6(D), where such ligands are compared with other agonists that maintain similar relative effects at both subtypes. In contrast with albuterol (which shows a similar Emax in both subtypes), the maximal receptor–arrestin complex induced by such ligands decreases to negligible levels at β2-AR while concomitantly increasing at β1-AR (Figure 6D). The loss of β2-AR efficacy with high β1-AR efficacy for arrestin in such ligands suggests that they can act as competitive antagonists of β2-AR–arrestin interactions. This was further investigated using dobutamine.

The concentration–response curves shown in Figure 7 indicate that dobutamine is a strong partial agonist [IA (intrinsic activity)=0.5–0.6 relative to adrenaline] for both the β1-AR– and β2-AR–G-protein interactions, and even slightly stronger (IA=0.75) for the β1-AR–arrestin interaction. In contrast, dobutamine produces no effect on the β2-AR–arrestin interaction. However, the concentration–response curves of adrenaline in the presence of increasing concentrations of dobutamine show a typical rightward shift of EC50 indicating competitive antagonism (Figure 7). The pA2 constant computed from Schild plot analysis of such experiments (6.52±0.3) is in good agreement with the EC50 of dobutamine determined for the β2-AR–G-protein interaction (−6.85±0.2) (see Table 2). Although dobutamine is a racemic mix of two enantiomers [34], which were not tested in resolved form in the present study, the inability of racemic dobutamine to form a detectable β2-AR–arrestin complex suggests that both enantiomers have no efficacy on this interaction.

Figure 7 Competitive antagonism of dobutamine for the β2-AR–arrestin interaction

(AC) Concentration–response curves of dobutamine (DOB) and adrenaline (EPI) for the interactions indicated in the plots. Note that dobutamine is a relatively effective partial agonist in the three assays. (D) Concentration–response curves of adrenaline in the absence or presence of increasing concentrations of dobutamine (as indicated in the inset). The inset is a Schild plot of the data (CR, concentration ratio; i.e. ratio of adrenaline EC50 in the presence compared with the absence of dobutamine). The pA2 value computed from the plot is 6.5±0.32. Data were averaged from two independent experiments using the indicated concentrations of dobutamine, and were replicated in one additional experiment with different concentrations of antagonist.


We have compared the ability of the structurally similar β1-ARs and β2-ARs to bind the major transduction proteins that mediate their biological effects, G-proteins and β-arrestin-2, in response to occupation by 45 different ligand structures. The interaction was quantified as maximal enhancement of RET ratio induced by each ligand (relative to the reference agonist adrenaline).

This enhancement of RET ratio reflects two equiprobable mechanisms: (i) intramolecular rearrangement of a pre-existing complex that results in tighter receptor–transducer association or (ii) the formation of a new receptor–transducer complex. Any of the two mechanisms might be involved in G-protein coupling, whereas the second is most likely to be involved in arrestin coupling. In both reactions the ‘molecular efficacy’ of the ligand is given by the free-energy change that couples the binding sites of ligand and transducer on the receptor molecule [35]. This free-energy change and the maximal ratio transducer-bound compared with unbound receptor (i.e. the Emax in RET) are predicted to be linearly related according to the first mechanism, and are also virtually linear according to the second, unless the receptor exceeds the transducer concentration. However, theoretical analysis (given in-depth in the Supplementary Material section entitled “Analysis of ligand-induced receptor activation using RET methods”) also shows that any deviation from 100% RET efficiency of the reporter system introduces significant non-linearity between the optical signal and the molar fraction of transducer-bound receptor. Consequently, the differences of intrinsic activities between two receptor subtypes not only reflect divergences in ligand molecular efficacy, but also in the stoichiometry and the affinity/stability of transducer–receptor complexes.

For this reason, experiments were designed to assess how the stoichiometry between receptor and transducer can influence the observed differences. In the G-protein system, the intrinsic activity of the ligand was not affected by a wide range of receptor/Gβ1 expression ratios, except when the abundance of Gαs was enhanced by overexpression. Such a result is consistent with the idea that in this system the change of RET reflects internal rearrangements of a preformed receptor–Gαβγ complex [25], although it cannot rule out alternative mechanisms. The effect of Gαs overexpression may suggest that this subunit is the limiting factor in the stability of functional receptor–transducer complexes that can be preassembled in the membrane. In the arrestin system, we used cell lines with identical expression levels of luminescent receptors and fluorescent arrestin, to ensure that the results were independent of differences in relative expression. Doubling receptor expression of the β2 subtype at equal arrestin levels shifted the Emax of partial agonists only slightly, indicating that small differences of expression in the cell lines used for the analysis cannot affect the comparison.

Even if the role of expression stoichiometry can be discounted, changes of ligand Emax between the two receptors reflect both a difference in ligand molecular efficacy and a difference in receptor–transducer affinity. The first involves particular ligands individually, the second affects all ligands, but to a variable extent according to their efficacy, because of the hyperbolic relationship between the fraction of transducer–receptor complex and the optical RET signal (Supplementary Material section entitled “Analysis of ligand-induced receptor activation using RET methods”). The exact distinction of ‘individual’ and ‘global’ differences is obviously impossible in the presence of experimental noise. However, the large panel of ligands examined in the present study helps to identify general trends resulting from a difference in transducer–receptor affinity, as such a difference alters the linearity, but not the ordering, of ligand efficacies. In contrast, individual ligand divergences that significantly change the ranking of intrinsic activities between receptor subtypes more probably reflect a true diversity in the conformational perturbations that the structure of the ligand can transmit on the two receptor molecules.

According to this interpretation algorithm, most of the β1-/β2-ARs in ligand intrinsic activities for G-protein coupling primarily reflect a molecular difference of G-protein interaction between the two receptors, rather than individual divergences of efficacy among ligands. When compared on the same plot (Figure 6C), the bulk of Emax data deviates from the identity line and bends towards the β2-AR axis, with the largest differences occurring in the mid range (0.4–0.6) of intrinsic activity. This is the pattern expected if we assume that β2-AR can establish a more stable interaction than the β1-AR subtype with G-protein subunits. This conclusion agrees with our finding that the potency of GDP to quench adrenaline-induced coupling is 10-fold greater at β1-AR than at β2-AR. It also agrees with previous work where β1-/β2-AR subtype differences were assessed in conventional signalling assays [36]. Only a few ligands that appear to deviate significantly from this general trend (e.g. CGP12177 and xamoterol) are structures possibly endowed with the ability to induce a better G-protein-interacting conformation when occupying the β1-AR-, rather than the β2-AR-binding site.

In contrast, the mirror-like trend in the differences of ligand intrinsic activities observed for arrestin coupling (Figure 6B) indicates that the default β2-AR preference existing for G-proteins is abolished or reversed for arrestin. This suggests that the β1 subtype interacts with the arrestin/GRK system more efficiently than the β2-AR. Note that this inversion cannot be attributed to the stronger coupling of β2-AR to Gαs (which thus may ‘compete’ against arrestin), because in the present study arrestin interactions were recorded under a Gs-null background. A greater arrestin affinity and/or phosphorylation efficiency of the β1 subtype may be responsible for this global shift. Although for arrestins there is no allosteric inhibitor of transducer–receptor interaction (like GDP) to support this finding, the faster kinetics observed for the β1-AR–arrestin interaction would be in-line with such a conclusion.

Our data seem to contrast with other studies that reported a reduced ability of β1-AR to undergo internalization and down-regulation compared with β2-AR [15,3739]. However, while in previous work the comparison was based on events downstream of the arrestin–receptor complex, in the present study we directly measured the formation of that complex. Moreover, the comparison of two alternative ways to measure the β2-AR–arrestin interactions (Supplementary Figure S1) suggests that the arrestin–receptor complex responsible for the bulk of RET signal is probably on the cell surface. Thus it is possible that despite a stronger interaction with arrestin, the β1-AR–arrestin complex might be less efficient in progressing along the subsequent steps of endocytosis and recycling than the β2-AR. More experiments are necessary to clarify this surprising paradox.

Some agonists display levels of β1-AR efficacy for arrestin coupling which are greater than expected from the general trend. Besides xamoterol and CGP12177 (already displaying β1 preference in G-protein coupling), the most obvious is a cluster of agonists (e.g. dobutamine, synephrine analogues, ritodrine and dopamine) that show reversal of β2-/β1-AR efficacy profiles, thus producing considerable levels of receptor–arrestin complex via β1-AR, but undetectable levels via β2-AR. As demonstrated for dobutamine, these ligands are competitive antagonists of the β2-AR–arrestin interaction induced by a full agonist. The dobutamine Ki value for blocking adrenaline on β2-AR–arrestin is very close to its EC50 as an agonist on the β2-AR–G-protein interaction. Thus the loss of β2-AR efficacy on arrestin is not due to reduced binding affinity for the β2-AR–arrestin complex. The most logical conclusion is that these ligands are capable of inducing a proper ‘arrestin-fitting’ conformation of the receptor when binding to the β1-AR subtype, but fail to do so when occupying the site of the β2-AR subtype.

Although data derived from the direct measurement of protein–protein association are the best to gauge the conformational change that each ligand structure can transfer to the receptor–transducer interface, they cannot predict the relative effect of such a change on downstream signalling. In fact, a complex signalling network can generate strong non-linearity between receptor–transducer complex formation and resulting biological responses. Thus the profile of β1-/β2-AR selectivity that we have seen in the present study at the molecular level might be considerably altered at the stage of arrestin-mediated functional responses.

With such a caveat in mind it is, however, interesting to note that dobutamine and dopamine are clinically relevant sympathomimetics, considered to exert β1-AR-mediated inotropic effects via cAMP signalling. Yet we find that the transducer where such drugs show β1-AR selectivity of efficacy is arrestin, not Gs. This suggests a role for arrestin in mediating adrenergic control of the myocardial contractile response. Indeed, a recent study has demonstrated that arrestins can mediate enhancement of cardiomyocyte contractility via angiotensin 1 receptors [40], and rapid changes of local free Ca2+, apparently mediated by β-arrestin-2/ERK (extracellular-signal-regulated kinase) via β2-AR, were described in hippocampal neurons [32]. The efficacy profile of ‘dobutamine-like’ agonists identified in the present study predicts a peculiar pattern of receptor–transducer output in cells expressing both β1-ARs and β2-ARs; G-protein responses are activated via both subtypes, but arrestin signalling can only occur through the β1-AR, as these ligands block the influence of endogenous catecholamines on the β2-AR–arrestin interaction. One obvious question is whether such an imbalance in β1-/β2-AR-mediated signalling might be related to the adverse effects that dobutamine has shown in clinical trials of heart failure patients [41]. Interestingly, a functional distortion in the relative balance of β1-/β2-AR-mediated signalling, resulting from the loss of compartmentation, was found in myocardiocytes from failing hearts [7].

In conclusion, we have shown that the β1-/β2-AR efficacy profiles of adrenergics are diametrically different for G-protein and arrestin. This primarily reflects an inversion in the strength of receptor–transducer interactions (β2-AR>β1-AR for G-proteins and β1-AR≥β2-AR for arrestins). We have also identified a group of β1-selective arrestin agonists, which can induce formation of the receptor–arrestin complex only when bound to the β1 subtype.


Ida Casella, Caterina Ambrosio and Maria Cristina Grò conducted the experiments. Paola Molinari, Ida Casella and Maria Cristina Grò prepared the chimaeric constructs and transfected cell lines used in the present study. All authors contributed to the experimental design, data analysis and preparation of the paper, which was written by Ida Casella and Tommaso Costa.


This work was supported, in part, by the Basic Research Investment Fund ‘Internationalization’ programme [grant number RBIN04CKYN_001].


We thank Dr M. Bastepe (Harvard Medical School, Boston, MA, U.S.A.) and Dr O.H. Onaran (Ankara University, Ankara, Turkey) for the gift of 2B2 cells. Dr Ad Ijzerman (University of Leiden, Leiden, The Netherlands) generously provided some of the β-agonists used in the present study.

Abbreviations: AR, adrenoceptor; bDOC, bis-desoxycoelenterazine; BRET, bioluminescence resonance energy transfer; DMEM, Dulbecco's modified Eagle's medium; FBS, fetal bovine serum; GPCR, G-protein-coupled receptor; GRK, GPCR kinase; HEK, human embryonic kidney; IA, intrinsic activity; KO, knockout; RET, resonance energy transfer; RGFP, Renilla green fluorescent protein; mtRGFP, membrane-targeted RGFP; Rluc, Renilla luciferase


View Abstract