β-Arrestin-biased Agonism at the β2-Adrenergic Receptor*

Classically, the β2-adrenergic receptor (β2AR) and other members of the seven-transmembrane receptor (7TMR) superfamily activate G protein-dependent signaling pathways in response to ligand stimulus. It has recently been discovered, however, that a number of 7TMRs, including β2AR, can signal via β-arrestin-dependent pathways independent of G protein activation. It is currently unclear if among β2AR agonists there exist ligands that disproportionately signal via G proteins or β-arrestins and are hence “biased.” Using a variety of approaches that include highly sensitive fluorescence resonance energy transfer-based methodologies, including a novel assay for receptor internalization, we show that the majority of known β2AR agonists exhibit relative efficacies for β-arrestin-associated activities (β-arrestin membrane translocation and β2AR internalization) identical to the irrelative efficacies for G protein-dependent signaling (cyclic AMP generation). However, for three βAR ligands there is a marked bias toward β-arrestin signaling; these ligands stimulate β-arrestin-dependent receptor activities to a much greater extent than would be expected given their efficacy for G protein-dependent activity. Structural comparison of these biased ligands reveals that all three are catecholamines containing an ethyl substitution on the α-carbon, a motif absent on all of the other, unbiased ligands tested. Thus, these studies demonstrate the potential for developing a novel class of 7TMR ligands with a distinct bias for β-arrestin-mediated signaling.

7TMRs, 5 also known as G protein-coupled receptors, are the largest family of proteins involved in the transduction of signals from the extracellular milieu to intracellular effectors (1) and account for nearly 2% of all human genes (2). Clinically, 7TMR ligands are the single largest class of pharmacologic agents (3). According to a well established and evolutionarily conserved paradigm, 7TMRs signal through direct activation of heterotrimeric G proteins to promote the activation or inhibition of second messenger-generating enzymes and changes in second messenger-dependent effector activities (4). Thus, for example, agonist stimulation of ␤ 2 AR promotes G s -mediated activation of adenylyl cyclase, with subsequent cAMP production and cAMP-associated signaling events. Signal termination and 7TMR desensitization result from recruitment of ␤-arrestin proteins to the cytoplasmic surface of 7TMR following agoniststimulated receptor phosphorylation by the G protein-coupled receptor kinase (GRK) family of proteins (5,6).
When originally identified, the ␤-arrestin proteins were believed to play a role only in limiting 7TMR signaling by physically interceding between the receptor and G protein. Recent work has demonstrated, however, that for a variety of 7TMRs, ␤-arrestin proteins can mediate G protein-independent 7TMR signaling by selectively scaffolding signaling cascade components, including small GTP-binding proteins and members of the MAPK cascade (5,7). These findings have had a profound impact on our understanding of 7TMR ligand pharmacology.
Historically, 7TMR ligands have been classified according to their ability to promote receptor activation (agonists) or inhibit agonist-mediated receptor activation (antagonists). This binary categorization underestimates, however, the true complexity of 7TMR ligand behavior. The identification of inverse agonists, which inhibit ligand-independent (i.e. constitutive) activity of 7TMRs for G protein activation (8), broadened the spectrum of 7TMR ligands to at least three fundamentally distinct classes. In this pharmacologic paradigm, 7TMR ligands were thought to exhibit "correlated efficacies," stimulating or inhibiting all functions of a receptor to the same extent. Consistent with this were studies performed 2 decades ago in which purified ␤ 2 AR was reconstituted in a phospholipid vesicle system either with GRK2 or with both G s and adenylyl cyclase (9). Analysis of the intrinsic abilities of a series of ␤AR ligands defined as partial agonists for G protein activation demonstrated a near-perfect correlation (coefficient of 0.996) in the ability of each ligand to stimulate G protein-mediated (adenylyl cyclase activation) or ␤-arrestin-associated (␤AR phosphorylation) activities (9). More recent results, however, suggest that some ␤ 2 AR antagonists are actually agonists for some signaling pathways, thus, for example, stimulating MAPK activation while blocking G protein (10 -12). We refer to such selective signal activation as "ligand bias" (13), but the same phenomenon has also been described as "ligand-directed trafficking" (14), "protean agonism" (15), "pleuridimensional efficacy" (12), and "collateral efficacy" (16).
Recent evidence demonstrates that biased ligands can selectively activate ␤-arrestin signaling without activating G protein signaling. For example, angiotensin II-induced signaling via angiotensin II receptor type 1a (AT 1a R) leads to G␣ q/11 -mediated activation of phospholipase C-and ␤-arrestin-dependent functions, whereas AT 1a R activation by a peptide analogue of angiotensin II ([Sar 1 ,Ile 4 ,Ile 4 ]angiotensin II, denoted SII) leads to ␤-arrestin-dependent MAPK activation alone (17)(18)(19)(20). The physiologic relevance of this AT 1a R/G protein-independent signaling has recently been demonstrated in mice with cardiacspecific overexpression of a mutant AT 1a R capable of activating only G protein-independent pathways in response to angiotensin II (21), as well as in isolated cardiomyocytes derived from wild-type mice treated with SII (19). Additional studies of ␤ 2 AR showed that a mutant ␤ 2 AR incapable of agonist-induced signaling via G s promoted agonist-induced activation of the MAPK cascade via ␤-arrestins (22). Other studies have also demonstrated ligand-induced, G protein-independent, ␤arrestin-dependent signaling by 7TMRs believed previously to signal only via conventional G protein-dependent mechanisms, including the V2 vasopressin receptor (23) and parathyroid hormone receptor (24), among others (see Ref. 25).
Because the standard approach to 7TMR ligand discovery and biochemical characterization is based solely on G proteindependent activities such as second messenger accumulation, we set out to develop a systematic approach that would allow us to unequivocally distinguish the ability of a ligand to modulate classic G protein-dependent activities from distinct signals. Accordingly, we chose to use the prototypical 7TMR ␤ 2 AR as a model system both because we have shown previously that a mutant ␤ 2 AR was capable of G protein-independent/␤-arrestindependent signaling and because a wealth of well characterized ligands that have been demonstrated previously to modulate G protein-dependent activities at ␤ 2 AR are readily available commercially. Using a set of FRET-based live-cell biosensors, we interrogated a series of ligands in search of ␤-arrestin-biased agonists in hopes of establishing structure-activity relationships for ␤-arrestin ligand bias at ␤ 2 AR.

EXPERIMENTAL PROCEDURES
Materials-Deoxyepinephrine, epinephrine, formoterol, ICI-118,551, isoproterenol, norepinephrine, propranol, salbutamol, and salmeterol were obtained from Sigma. N-Cyclopentylbutanephrine (CPB) was supplied previously by Sterling-Winthrop. Additional CPB was obtained by the Duke University Small Molecule Synthesis Facility and determined to be equivalent to that provided by Sterling-Winthrop for all assays tested. Anti-phospho-␤ 2 AR (Ser 355 / Ser 356 ) was obtained from Santa Cruz Biotechnology. Ligands were used at receptor-saturating concentrations, calculated as 100 ϫ k d as follows: isoproterenol ( Plasmids-pcDNA3.1 expressing ␤ 2 AR-mCFP and ␤-arres-tin2-mYFP, as well as cell lines stably expressing these plasmids, are described elsewhere (26). pcDNA3 expressing the cAMP biosensor ICUE2 (indicator of cAMP using Epac) was obtained from Jin Zhang (Johns Hopkins Medical Institutions) and is described elsewhere (27,28). MyrPalm-mYFP was a gift of Roger Tsien (University of California, San Diego) (29).
Cell Culture-HEK-293 cells were maintained in minimal essential Eagle's medium (M2279; Sigma) plus 10% fetal bovine serum and penicillin/streptomycin. For stable cell line generation, G418 was used at 400 g/ml for selection and 100 g/ml for maintenance, and hygromycin was used at 250 g/ml for selection and 150 g/ml for maintenance. Stably transfected cell lines were generated as described elsewhere, with ␤ 2 AR cell surface expression determined by 125 I-cyanopindolol binding of 1.0 pmol/mg of protein (26).
Imaging-For all imaging experiments, cells were seeded onto fibronectin-coated imaging dishes 24 h prior to assay. Cells were washed once with 1ϫ phosphate-buffered saline, placed in imaging buffer (125 mM NaCl, 5 mM KCl, 1.5 mM MgCl 2 , 1.5 mM CaCl 2 , 10 mM glucose, 0.2% bovine serum albumin, 10 mM HEPES, pH 7.4), and imaged in the dark on a stage heated to 37°C. Images were acquired on a Zeiss Axiovert 200M microscope (Carl Zeiss MicroImaging Inc.) with a Roper Micromax cooled charge-coupled device camera (Photometrics) controlled by SlideBook 4.0 (Intelligent Imaging Innovations). CFP and FRET images were obtained through a 436/20 excitation filter (20-nm bandpass centered at 436 nm), a 455DCLP dichroic longpass mirror, and separate emission filters (480/30 for CFP and 535/30 for FRET). YFP intensity was imaged from a 500/20 excitation filter, a 515LP dichroic mirror, and a 535/30 emission filter. All optical filters were obtained from Chroma Technologies. Excitation and emission filters were switched in filter wheels (Lambda 10-2; Sutter Instruments). Integration times were varied between 100 and 300 ms to optimize signal and minimize photobleaching.
Cyclic AMP FRET Assays-G protein efficacy was measured by detection of the second messenger cAMP. cAMP was measured using the biosensor ICUE2 (28), an improved version of ICUE (27). ICUE2 is an Epac1 (exchange protein directly activated by cAMP)-based sensor fused to both CFP and YFP; cAMP binds to Epac, causing conformational changes that induce a change in intramolecular FRET. G protein efficacy was quantified as the integrated change in the FRET ratio (CFP intensity over YFP intensity) over the 5 min following ligand addition. This measure captures both the kinetics and amplitude of G protein efficacy.
Endocytosis FRET Assays-␤ 2 AR internalization was measured by co-expressing ␤ 2 AR-mCFP and a membrane-targeted mYFP bearing a sequence coding for myristoylation and palmitoylation, MyrPalm-mYFP (29). In the basal state, these two proteins colocalize in the plasma membrane, where their effective concentration is sufficiently high to result in intermolecular FRET, even in the absence of any intermolecular affinity. This "packing" effect can lead to artifactual results when studying protein-protein interactions by FRET (30), but is used here simply to generate a marker of membrane colocalization. When ␤ 2 AR-mCFP internalizes from the cell surface after stimulation, MyrPalm-mYFP remains behind; this loss of colocalization leads to a reduction in basal FRET in proportion to the amount of ␤ 2 AR-mCFP that is sequestered. Thus, this assay provides a real-time kinetic measurement of receptor internalization and can in principle be extended to the study of any protein that traffics to or from the plasma membrane.
pERK Assays-HEK-293 cells in 6-or 12-well plates were starved for 4 h in serum-free medium prior to stimulation. After stimulation with the appropriate ligand for 5 min, cells were solubilized directly by adding 2ϫ SDS sample buffer, followed by boiling at 100°C for 5 min. For each transfection, an equal aliquot of cells was set aside for protein determination (Bradford).

RESULTS
Given the recent recognition that 7TMRs are capable of G protein-independent/␤-arrestin-dependent signaling, we tested a broad panel of previously described full and partial agonists of G protein activity at ␤ 2 AR for ␤-arrestin-biased activities using HEK-293 cells as the basis for our ␤ 2 AR model systems.
Accordingly, the ability of each compound to induce cAMP production via the endogenously expressed ␤ 2 AR was assessed in HEK-293 cells stably expressing ICUE2, an improved version of the FRET-based cAMP reporter ICUE (27). This highly sensitive technique allows for the quantitative comparison of both the relative amount of cAMP generated in response to ligand  (27,28) were incubated with the appropriate vehicle or with one of seven ligands at concentrations 100-fold above the reported EC 50 . Time course data are representative of three to five independent experiments. The FRET emission ratio, corresponding to cAMP concentration, was sampled every 2 s. NE, norepinephrine; Salb, salbutamol; Salm, salmeterol; Prop, propranol; ICI, ICI-118,551. B, changes in ␤ 2 AR and ␤-arrestin2 association determined by FRET. HEK-293 cells with stable co-expression of human ␤ 2 AR-mCFP (1.0 pmol/mg) and rat ␤-arrestin2-mYFP were incubated with the same ligands described in A. FRET is quantified as a percentage of whole-cell total CFP-excited fluorescence (%F). Data shown are representative of three to five independent experiments. ␤ 2 -Adrenergic Receptor ␤-Arrestin-biased Agonism FEBRUARY 29, 2008 • VOLUME 283 • NUMBER 9 stimulation and the rigorous quantification of the rate at which cAMP production occurs. As shown in Fig. 1A, the classic ␤AR agonist isoproterenol (Iso) resulted in a rapid but transient cAMP response, as did a series of ␤AR partial agonists. The ICUE2 response was completely abrogated by pretreatment of cells with the ␤AR antagonist propranolol or ICI-118,551 and exhibited a standard dose-response relationship for varying concentrations of isoproterenol, with an EC 50 similar to previous findings (28). The ICUE2 dose-response relationship was the same regardless of whether the ICUE2 response was measured as the initial linear rate, the maximum response, or the integrated response over time (data not shown). Because the integrated response captures both the kinetics and amplitude of cAMP dynamics, we chose that measure as the simplest representation of G protein efficacy.
The proximal event in ␤-arrestin-dependent signaling is thought to be ligand-stimulated 7TMR phosphorylation by GRKs, followed by ␤-arrestin translocation from the cytosol to the phosphorylated 7TMR (5). We thus used a FRET-based measurement of receptor/␤-arrestin interaction as an indicator of ␤-arrestin-dependent signaling. We have found previously that the rate of ␤-arrestin recruitment, and not the amount of ␤-arrestin recruited, is sensitive to changes in GRK concentrations (26); thus, differences in GRK/␤-arrestin efficacy will manifest most acutely as changes in the kinetics of ␤-arrestin recruitment. Accordingly, we used the rate of ␤-arrestin recruitment to measure GRK/␤-arrestin efficacy. HEK-293 cells with stable co-expression of ␤ 2 AR-mCFP and ␤-arrestin2-mYFP were incubated with the same expanded panel of ␤AR ligands utilized in the initial screen described for Fig. 1A. As seen in Fig. 1B, these ligands demonstrated a range of efficacies. Interestingly, we found one ligand, CPB, that exhibits a marked bias: robust ␤-arrestin efficacy despite only moderate G protein efficacy.
We investigated this bias more fully by characterizing a larger set of ␤AR ligands, including a number that are used clinically. The addition of each ligand at a receptor-saturating concentration (100-fold higher than published k d values) revealed a range of efficacies for G protein activation and ␤-arrestin recruitment. As seen in Fig. 2A, a correlation plot clearly shows that although most ligands were essentially Iso-efficacious for G protein-dependent and ␤-arrestin-dependent activities, several ligands were more efficacious for ␤-arrestin translocation to ␤ 2 AR than for cAMP production. CPB exhibits G protein efficacy nearly identical to that of the partial agonist norepinephrine (Fig. 1A), but ␤-arrestin efficacy greater than that of the full agonist isoproterenol. Indeed, CPB engendered ␤-arrestin recruitment 3.2-fold more rapidly than did norepinephrine. Furthermore, relative to the full G protein agonist Iso, which did not demonstrate any bias in its ability to promote cAMP production or ␤-arrestin translocation, CPB was only 58% as efficient for G protein activation, but 1.3fold more efficient for ␤-arrestin recruitment. Several additional ligands ( Fig. 2A, shown in red) were also found to have significant ␤-arrestin bias. These were ethylnorepinephrine and isoetharine, structural derivatives of norepinephrine and Iso, respectively. Specificity of these ligands' activities was demonstrated by blocking all effects with the ␤ 2 AR-spe-cific antagonist ICI-118,551 and by the absence of any effect of blocking endogenous ␣-adrenergic receptors with the antagonist yohimbine (data not shown).
We also measured the bias present among the tested ligands by calculating a "bias factor," the ratio of ␤-arrestin efficacy to G protein (cAMP) efficacy (Fig. 2B). This analysis allowed us to rapidly discern which ligands appear biased and are of interest for further study. As a benchmark, we chose a guideline of twice the S.D. of isoproterenol's bias factor: although isoetharine and ethylnorepinephrine exhibit only modest bias factors, CPB exhibits a bias factor appreciably larger than this benchmark. Thus, we hypothesize that isoetharine, ethylnorepinephrine, and  Fig. 1 to promote ␤-arrestin recruitment (ordinate) and cAMP formation (abscissa) shows that the majority of ligands do not demonstrate any bias. Three ligands (shown in red) demonstrate enhanced ability to promote ␤-arrestin recruitment to ␤ 2 AR relative to their ability to induce cAMP production. Efficacy for G protein activation is reported as the integrated FRET ratio change of ICUE2, indicative of cumulative cAMP production. Efficacy for ␤-arrestin is reported as the rate of recruitment (min Ϫ1 ), which corresponds to the amount of elicited GRK activity (26). B, the ratio of ␤-arrestin efficacy to G protein efficacy is expressed as the bias factor for each ligand. A bias factor of 1.0 indicates no bias. As a benchmark for inferring bias, a line is shown demarcating the mean ϩ 2 S.D. of the bias factor of isoproterenol. CPB, shown in red, exhibits a bias factor unlikely to be randomly detected by assay variability and leads to the hypothesis that CPB is significantly biased. Epi, epinephrine; Norepi, norepinephrine; EtNE, ethylnorepinephrine; MeNE, methylnorepinephrine; deoxyepi, deoxyepinephrine.
especially CPB are ␤-arrestin-biased ligands and as such will differ in downstream effects when compared with unbiased agonists.
␤-Arrestin translocation to 7TMRs in response to ligand stimulation follows receptor phosphorylation at cytoplasmic serines and threonines by the GRK family of kinases (6). As another measure of efficacy for GRK/␤-arrestin functions, we assayed the rate of agonist-stimulated, GRK-mediated phosphorylation of ␤ 2 AR. HEK-293 cells stably expressing ␤ 2 AR (22) were stimulated at times ranging from 30 s to 20 min and were analyzed for receptor phosphorylation with an antibody specific for ␤ 2 AR phosphorylated on serines 355 and 356, sites shown previously to undergo phosphorylation by GRKs (33,34). The initial rate of CPB-stimulated ␤ 2 AR phosphorylation closely mirrored that stimulated by Iso, whereas ligands with partial agonist activity for both G protein and ␤-arrestin activities (norepinephrine, salbutamol, and salmeterol) stimulated rates much slower than that stimulated by Iso (Fig. 3A). The initial rates of GRK-mediated phosphorylation of ␤ 2 AR are shown graphically in Fig. 3B. These results show that, consistent with its effects on ␤-arrestin recruitment, CPB is a full agonist for receptor phosphorylation by GRKs despite its modest partial agonism for G protein activity.
As CPB leads to both rapid ␤-arrestin recruitment (Fig. 2) and GRK-mediated ␤ 2 AR phosphorylation (Fig. 3), we hypothesized that CPB may also promote more efficient ␤ 2 AR internalization. Indeed, an enhanced rate or amount of ␤ 2 AR inter-nalization would be a functional readout for CPB bias. To examine this directly, we developed a FRET-based receptor internalization assay in which ␤ 2 AR-CFP was co-expressed in HEK-293 cells with YFP modified by the covalent linkage of both myristoyl and palmitoyl moieties (29). With these lipid modifications, YFP is localized nearly exclusively to the plasma membrane, where it exhibits density-dependent FRET with ␤ 2 AR-CFP (data not shown), arising from the high effective concentration of fluorophores confined to a two-dimensional plane, even in the absence of intermolecular affinity (29). This constitutive colocalization is disrupted by internalization of the receptor, but not the lipid-modified YFP, into endocytic vesicles after agonist stimulation (Fig. 4A). This corresponds to a loss of FRET after agonist stimulation, with a time course matching the change in spatial separation (Fig. 4B). Importantly, preincubation with 450 mM sucrose, a well established technique for the inhibition of internalization (35), completely abrogated ␤ 2 AR endocytosis in response to isoproterenol (Fig.  4B). As anticipated, ligands demonstrated to be less efficacious for ␤-arrestin recruitment (Fig. 1B), such as salbutamol and norepinephrine, resulted in both less rapid and less robust ␤ 2 AR internalization than the reference compound, Iso (Fig.  4C). Importantly and concordant with its high efficacy for ␤-arrestin recruitment (Fig. 1B), stimulation with CPB resulted in rapid ␤ 2 AR internalization (Fig. 4, C and D).
To highlight the extent to which ␤ 2 AR internalization reflects the ability of a ligand to promote either G protein-or ␤-arrestin-dependent actions, we compared the rate of ␤ 2 AR internalization for several ligands (data from Fig. 4C) with cAMP formation and with the rate of ␤-arrestin recruitment (from Fig. 1). As shown in Fig. 5A, the correlation of the ␤ 2 AR internalization rate and ␤-arrestin recruitment rate for a range of responses matched a hyperbolic model of ligand efficacy, consistent with a single "activated" receptor state (33). This model assumes that every ligand stabilizes "active" and "inactive" receptor conformations and that efficacy results from the proportion of receptors shifted to the active state. Different assays of efficacy measure different saturable cellular responses, resulting in a hyperbolic efficacy function. However, correlation of the ␤ 2 AR internalization rate and cAMP response is inconsistent with this hyperbolic model: the biased ligand CPB is disproportionately efficacious for ␤ 2 AR internalization relative to its efficacy for cAMP generation (Fig. 5B). This finding is inconsistent with a single active state of the receptor and implies that multiple functionally distinct receptor conformations can be stabilized by agonists.
One potential concern regarding our measurement of ligand bias is that we used both kinetic and integrated responses to characterize ligand efficacy. Signal kinetics and signal magnitude are difficult to cleanly differentiate, so we sought a single downstream measurement to verify that CPB is indeed biased toward ␤-arrestin. In our previous studies of ␤-arrestin-biased signaling, we elucidated both G protein-dependent and ␤arrestin-dependent contributions to ␤ 2 AR-stimulated MAPK activation (22). Thus, we used activation of the MAPK ERK1/2 to verify the ␤-arrestin bias of CPB. Serum-starved HEK-293 cells were treated with Iso and CPB in the presence of either control siRNA or siRNA directed against ␤-arrestin2. We . ␤-Arrestin-biased agonists induce more rapid ␤ 2 AR phosphorylation. A, HEK-293 cells with stable overexpression of FLAG-␤ 2 AR/pcDNA3 as described previously (22) were stimulated for the indicated times (30 s or 1, 2, 4, 8, or 20 min) with the ligand and concentrations as described in the Fig. 1 legend. Following SDS-PAGE and transfer to nitrocellulose, ␤ 2 AR phosphorylated by GRKs on serines 355 and 356 (p355/p356) in the C-terminal tail of the receptor was detected with a phospho-specific antibody (33). Each data point represents the mean Ϯ S.E. from three independent experiments. NE, norepinephrine; Salb, salbutamol; Salm, salmeterol. B, for visual comparison, the initial rates of phosphorylation derived from A are shown graphically; the rate of phosphorylation stimulated by CPB is indistinguishable from that elicited by Iso.
observed that in the presence of siRNA targeting ␤-arres-tin2, Iso-stimulated pERK was reduced by 43.8 Ϯ 1.9%, whereas for CPB, pERK stimulation was reduced by 60.3 Ϯ 5.3% (p Ͻ 0.05; n ϭ 4) (Fig. 6). Accordingly, these results demonstrate that although Iso and CPB induce similar levels of pERK stimulation, their utilization of G proteinversus ␤-arrestin-dependent signaling pathways to activate ERK differs significantly (i.e. there is a relatively greater contribution of ␤-arrestin signaling to the CPB ERK response). As such, these results further verify the ␤-arrestin bias of CPB relative to Iso and demonstrate that the observed bias leads to alterations in a downstream signaling kinase cascade. Furthermore, these results validate the use of proximal signal transduction signals, in this case the rate of ␤-arrestin recruitment and the integrated cAMP response, to discern ␤-arrestin bias among a set of ligands.

DISCUSSION
Although the ability of 7TMRs to induce cellular responses through G protein activation in response to ligand has long been recognized, an accumulation of evidence now clearly demonstrates that 7TMRs can also impart G protein-dependent or -independent ␤-arrestin-dependent signals. The present study is consistent with these more recent observations (18, 20, . Exposure to isoproterenol results in ␤ 2 AR translocation from the plasma membrane into punctate endocytic vesicles. Images were deconvolved by the no-neighbors algorithm in SlideBook software to sharpen the definition of membrane and vesicle fluorescence. B, FRET time lapse reveals that after the addition of 1 M Iso, FRET is lost between ␤ 2 AR-mCFP and MyrPalm-mYFP as ␤ 2 AR is endocytosed. This signal was completely abrogated by the addition of 450 mM sucrose (Suc) to the cell imaging medium. C, the rate and extent of ␤ 2 AR internalization vary greatly among different ␤ 2 AR agonists. CPB stimulates ␤ 2 AR internalization with a rate and extent indistinguishable from that caused by isoproterenol. Internalization is not found after treatment with either vehicle (MeOH) or the ␤ 2 AR antagonist ICI-118,551 (ICI). Salm, salmeterol; NE, norepinephrine; Salb, salbutamol. D, a comparison of the initial ␤ 2 AR internalization rates determined from Fig. 3C is shown; internalization stimulated by CPB is indistinguishable from that elicited by Iso. FIGURE 5. A comparison of G protein and ␤-arrestin efficacies for ␤ 2 AR internalization. A, the hyperbolic correlation of the ␤ 2 AR internalization rate (ordinate) and ␤-arrestin recruitment rate (abscissa) for a range of responses is consistent with a single activated receptor state. Salm, salmeterol; Salb, salbutamol; NE, norepinephrine; Form, formoterol. B, the correlation of the ␤ 2 AR internalization rate (ordinate) and cAMP response (abscissa) is inconsistent with this model: the biased ligand CPB is disproportionately efficacious for ␤ 2 AR internalization relative to its efficacy for cAMP generation. Data are means Ϯ S.E. from Figs. 2 and 4. [22][23][24]31) and further provides a robust approach that can be readily adapted for the identification of 7TMR ligands that promote ␤-arrestin-selective signaling.
As postulated by Kenakin (16), all 7TMRs are likely capable of adopting a range of distinct conformations, each of which can lead to the activation of distinct intracellular signaling pathways. The adoption of these distinct conformations can, in turn, be modulated by the presence of ligand, such that distinct receptor configurations are stabilized or induced in response to a particular ligand. Our evaluation of a broad range of ligands, characterized previously by their ability to stimulate G proteindependent activities at ␤ 2 AR, is in agreement with the hypothesis that biased 7TMR signaling can be induced by ligand. Thus, although the majority of ligands assayed demonstrated comparable efficacies for G protein-and ␤-arrestin-associated activities, at least three ligands showed a clear bias for ␤-arrestinassociated activity. Previous biophysical studies of purified ␤ 2 ARs exposed to an array of ligands also support this concept (36,37). In addition, recent cellular studies examining the capacity of a variety of ␤AR ligands to activate adenylyl cyclase and MAPK signaling pathways at ␤ 2 AR showed clear ligand-dependent activation profile differences, although the extent to which MAPK signaling was independent of G protein activation and the relationship of this response to ␤-arrestin function were not determined (12).
A corollary to these findings is that to determine the true efficacy of a ligand for a specific 7TMR, one must define the entire range of effector signaling pathways modulated in response to the ligand. 7TMR ligands have been described classically solely according to their relative abilities to affect G protein-dependent pathways (i.e. as agonists, partial agonists, antagonists, or inverse agonists). However, our data, as well as the work of several other groups, suggest that the efficacy of each ligand for multiple G protein-mediated as well as non-G protein-mediated pathways must also be considered during classification. It is unknown whether important therapeutic differences that have been described among members of a 7TMR ligand class relate to these non-G protein effects, but this provides an attractive and testable hypothesis.
Interestingly, although we observed ligands that demonstrated a bias toward ␤-arrestin-dependent cellular processes, we did not detect any ligands that exhibited a G protein bias relative to ␤-arrestin activity. However, there is no theoretical reason to dismiss the possibility of ligands that promote 7TMR conformations with a selective bias toward G protein activation. Our studies included only ligands demonstrated previously to have ␤AR biological activity (i.e. agonists); careful characterization of a more comprehensive collection of ligands may in fact identify some with a G protein bias relative to ␤-arrestin. It is interesting to note, however, that in a recent study in which a collection of ␤AR ligands was used to assess biased signaling at both the ␤ 1 -and ␤ 2 ARs, no compounds were found that promoted more adenylyl cyclase activity relative to MAPK activation (12), a result that is consistent with our own findings.
One of the most intriguing findings revealed by our study is that the three ␤-arrestin-biased ligands identified share a common structural feature absent from all of the nonbiased ligands assayed. CPB, ethylnorepinephrine, and isoetharine all contain an ethyl substituent on the catecholamine ␣-carbon. Structurally, ethylnorepinephrine is identical to norepinephrine except for the presence of the ␣-carbon ethyl substituent, whereas isoetharine is structurally identical to isoproterenol except for the addition of an ␣-carbon ethyl moiety. As seen in Fig. 2A, the addition of an ␣-carbon ethyl group to isoproterenol (creating isoetharine) resulted in a substantial decrease in ␤ 2 AR-mediated G protein activation, with no change in the ␤-arrestin translocation rate. Moreover, addition of an ethyl substituent to the ␣-carbon of norepinephrine (to generate ethylnorepinephrine) substantially improved the rate of ␤-arrestin recruitment while having no significant effect on adenylyl cyclase activity. Thus, although the inclusion of the same structural motif on two ligands resulted in ␤-arrestin bias at ␤ 2 AR relative to the nonbiased "parent" compounds, the chemical basis for this bias is unclear. Accordingly, the importance of ␣-carbon alkyl substitutions in ␤-arrestin-biased ␤ 2 AR ligands, although suggestive, remains to be determined. Further investigations into the structure-activity relationship of ␤-arrestin-biased ligands for ␤ 2 AR, using newly synthesized derivatives of the ligands studied here, should greatly aid in understanding the possibilities for even more highly biased ligands for this receptor.
␤AR antagonists currently used in clinical practice for the management of hypertension and heart failure and following acute myocardial infarction were developed based upon their ability to block catecholamine-induced ␤AR activation of G proteins. It is reasonable to hypothesize, however, that ␤AR antagonists with a similar ability to inhibit ␤AR G protein activation, but with the simultaneous capacity to stimulate ␤-arrestin-dependent signaling pathways, may have additional salutary effects to those already recognized for beta-blockers. Indeed, recent data from our laboratory demonstrate that carvedilol, a nonselective ␤AR antagonist with particular efficacy in the management of heart failure, appears to be unique among beta-blockers in current clinical use in possessing the ability to promote ␤-arrestinmediated signaling while concomitantly antagonizing G protein activity (38).
Whether 7TMR ligands that promote biased signaling activities, as demonstrated for carvedilol, have superior efficacy for the management and treatment of other pathologic human conditions is at present unknown. The potential for ␤-arrestinbiased ligands to have physiologic properties that differ from ligands without bias is suggested, however, by studies of the ␤-arrestin-biased angiotensin receptor ligand, SII. This modified angiotensin peptide promotes positive inotropic and lusitropic responses in adult mouse cardiomyocytes (19) in the complete absence of G protein signaling. Whether other ␤-arrestin-biased ligands, such as the [D-Trp 12 ,Tyr 34 ]parathyroid hormone-(7-34) analogue (24) or any of the three ligands identified here, have unique physiologic properties in vivo remains to be determined. Accordingly, similar arguments may be proffered for nearly any current 7TMR that is a target of modern pharmacotherapy.
The recent recognition that 7TMRs are capable of G proteinindependent/␤-arrestin-dependent signaling has altered longheld dogmas in our understanding of 7TMR signal transduction mechanisms. Thus, it is perhaps not surprising that with the development of careful screening methods for the detection of both ␤-arrestinand G protein-dependent activities, we were able to identify ligands that induce ␤-arrestin-biased activities at ␤ 2 AR. Although our previous classification of 7TMR ligands was based solely on the efficacy of the ligand for modulation of G protein activity, our new appreciation for the growing complexity of 7TMR signaling suggests that future ligand characterizations will necessitate the inclusion of other signaling pathway(s) or effector systems modulated in response to ligand.