β-Arrestin Binding to the β2-Adrenergic Receptor Requires Both Receptor Phosphorylation and Receptor Activation*

Homologous desensitization of β2-adrenergic receptors has been shown to be mediated by phosphorylation of the agonist-stimulated receptor by G-protein-coupled receptor kinase 2 (GRK2) followed by binding of β-arrestins to the phosphorylated receptor. Binding of β-arrestin to the receptor is a prerequisite for subsequent receptor desensitization, internalization via clathrin-coated pits, and the initiation of alternative signaling pathways. In this study we have investigated the interactions between receptors and β-arrestin2 in living cells using fluorescence resonance energy transfer. We show that (a) the initial kinetics of β-arrestin2 binding to the receptor is limited by the kinetics of GRK2-mediated receptor phosphorylation; (b) repeated stimulation leads to the accumulation of GRK2-phosphorylated receptor, which can bind β-arrestin2 very rapidly; and (c) the interaction of β-arrestin2 with the receptor depends on the activation of the receptor by agonist because agonist withdrawal leads to swift dissociation of the receptor-β-arrestin2 complex. This fast agonist-controlled association and dissociation of β-arrestins from prephosphorylated receptors should permit rapid control of receptor sensitivity in repeatedly stimulated cells such as neurons.

A major mechanism for switching off many G-protein-coupled receptors is homologous desensitization. Homologous desensitization is a two-step process (1). First, agonist-activated receptors become substrates for G-protein-coupled receptor kinases (GRKs), 1 which phosphorylate serine and threonine residues in the C terminus or the intracellular loops of the receptors. The phosphorylated receptors then bind ␤-arrestins, and this interaction uncouples the receptors from their G-proteins and thereby blocks signaling to G-proteins. In addition to blocking "classical" receptor-G-protein signaling, ␤-arrestin binding to receptors can also mediate the activation of "novel" signaling pathways (often involving tyrosine kinases) and the targeting of the receptors to clathrin-coated pits. This targeting is then followed by internalization to endosomes and (after dephosphorylation (2)) either recycling to the plasma membrane or lysosomal degradation (reviewed in Ref. 3). Based on studies with knock-out mice, homologous desensitization seems to be involved in various physiological processes as diverse as heart development (4), nociception (5), thermoregulation (6), leukocyte chemotaxis (7), and regulation of locomotor activity (8).
A detailed investigation of homologous desensitization has been difficult to achieve because of the lack of good methods to monitor and quantify its extent. In particular, the interaction between receptors and ␤-arrestins has been difficult to quantify in intact cells. Initial experiments utilized coimmunoprecipitation of receptors and ␤-arrestins (9). Barak et al. (10) were the first to use fusion proteins of ␤-arrestins with green fluorescent protein (GFP) to visualize the translocation of ␤-arrestins to the plasma membrane upon receptor stimulation by confocal microscopy. More recently, resonance energy transfer methods have been developed to quantify receptor-arrestin interaction (reviewed in Ref. 11). It has been shown that both bioluminescence resonance energy transfer (BRET) and fluorescence resonance energy transfer (FRET) are suitable methods to quantify the interaction between receptors and ␤-arrestins (12)(13)(14)(15)(16). In general, FRET is most frequently applied between a CFP-tagged donor (e.g. the ␤-arrestin) and a YFP-tagged acceptor fluorophor (e.g. the receptor). In unstimulated cells, the receptor resides in the plasma membrane, whereas ␤-arrestin is located in the cytosol (10). During agonist-induced desensitization ␤-arrestin translocates to the plasma membrane and forms a complex with the receptor. The spatial proximity of the two proteins brings CFP and YFP close together, which, upon excitation of the CFP, leads to FRET.
Studies published so far have measured bioluminescence energy transfer (12, 14 -16) or FRET (13) between G-proteincoupled receptors and arrestins in populations consisting of at least thousands of cells. In this article, we show that FRET between the ␤ 2 -adrenergic receptor and ␤-arrestin can be detected with high temporal resolution in individual intact cells. Using the ␤ 2 AR as a model system, we find that (a) the ratelimiting step of the formation of the ␤-arrestin-receptor complex is receptor phosphorylation by G-protein-coupled receptor kinases (GRKs), (b) washout of agonist results in dissociation of the complex, and (c) repeated short term stimulation of a cell leads to accumulation of phosphorylated receptors that can interact very rapidly with ␤-arrestin when agonist is added again. Therefore, the desensitization kinetics of "used" receptors is quite different from the desensitization kinetics of "naive" receptors. We speculate that this fast mode of receptor desensitization may be important in settings where receptors are stimulated in a pulsed manner, e.g. in synapses. * This work was supported by Sonderforschungsbereich 487 "Regulatory Membrane Proteins," by a Leibniz grant from the Deutsche Forschungsgemeinschaft, and by the Fonds der Chemischen Industrie. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

EXPERIMENTAL PROCEDURES
Cell Lines and DNA Constructs-HEK293 cells stably expressing a conformational sensor of the murine ␣ 2A -adrenergic receptor (␣ 2A -cam) (17) were kindly supplied by Jean-Pierre Vilardaga. The cDNAs for bovine ␤-arrestin2 and dominant-negative bovine GRK2 (the K220R mutant) were kindly provided by Susanna Cotecchia (Lausanne, Switzerland), and the cDNA for the R170E mutant of bovine ␤-arrestin2 (18) was a kind gift by Vsevolod Gurevich (Nashville, TN). The cDNAs for phosphorylation-deficient ␤ 2 AR (19), the FLAG-tagged ␤ 2 AR (20) and the expression plasmid for human GRK2 (21) have been described. eCFP or eYFP (BD Biosciences Clontech, Heidelberg, Germany) were fused to the C terminus of bovine ␤-arrestin2, the human ␤ 2 -adrenergic receptor, or the murine ␣ 2A -adrenergic receptor as described recently for a ␤Arr2-GFP fusion protein (22), a ␤ 2 AR-GFP fusion protein (23), and the ␣ 2A -cam (17), respectively. Briefly, the stop codon was replaced with an XbaI site (TCTAGA), and the coding sequences of eCFP and eYFP except for the N-terminal methionine were appended in-frame. In additional constructs, eCFP and eYFP were also fused in front of the ␤-arrestin2 open reading frame, yielding CFP-␤Arr2 and YFP-␤Arr2, respectively. All constructs were cloned into pcDNA3 (Invitrogen) and verified by sequencing.
FRET Measurements-The ␣ 2A -cam was stably expressed in HEK293 cells as described previously (17). For all other measurements, HEK293 cells on 65-mm plates were transfected with Effectene (Qiagen, Hilden, Germany). The ratio between receptor DNA and ␤Arr2-CFP DNA was varied between 2:1 and 1:1. If GRK2 DNA was present, the ratio between receptor and GRK2 DNA was 3:1. 6 h after transfection, cells were split on round polylysine-coated coverslips (24 mm diameter), and experiments were performed after a further 18 h. The coverslips were mounted on an Axiovert 135 inverted microscope (Zeiss, Jena, Germany) using an "Attofluor" holder (Molecular Probes, Leiden, The Netherlands), and the cells were continuously superfused with HBS (150 mM NaCl, 10 mM HEPES, 10 mM glucose, 2.5 mM KCl, 4 mM CaCl 2 , 2 mM MgCl 2 , pH 7.2). The superfusion system has been described previously (17). Cells were observed using an oil immersion 63ϫ lens, a polychrome IV (Till Photonics, Grä felfing, Germany) for excitation, and a dual emission photometric system (Till Photonics). To minimize photobleaching, the illumination time was set to 10 -40 ms applied with a frequency of 10 Hz if not stated otherwise. Fluorescence was measured at 535 Ϯ 15 nm (F 535 ) and 480 Ϯ 20 nm (F 480 ) (beam splitter DCLP 505 nm, Chroma Technology, Rockingham, VT) upon excitation at 436 Ϯ 10 nm (beam splitter DCLP 460 nm, Chroma Technology). Signals detected by avalanche photodiodes were digitized using an analog/digital converter (Digidata 1322A, Axon Instruments, Union City, CA) and stored on PC using Clampex 8.1 software (Axon Instruments). The emission intensities were corrected by the respective spillover of CFP into the YFP channel and direct YFP excitation (spillover of YFP into the CFP channel was negligible) to give corrected intensities F CFP and F YFP . FRET was calculated as the ratio F YFP /F CFP . Ligands were applied using the superfusion system. Kinetics were fitted with monoexponential functions using the Levenberg-Marquardt algorithm as implemented in Kaleidagraph 3.6 (Synergy Software).
Confocal Microscopy-Confocal microscopy was performed on a Leica TCS SP2 system. HEK cells were transfected as above and split on round polylysine-coated coverslips that were mounted on a custom-built holder. YFP was excited with the 514 nm line of an argon laser and images were taken with a 63ϫ lens using the factory settings for YFP. Fluorescence loss in the cytoplasm over time was quantified using the Leica confocal software or ImageJ (NIH Image software) as described by Oakley et al. (24).
Biochemical Assays-Receptor phosphorylation was measured in HEK293 cells transfected with FLAG-tagged ␤ 2 AR and human GRK2 as described previously (21,25). Briefly, the transfected cells were incubated in phosphate-free DMEM and loaded with [ 32 P]orthophosphate. Cells were stimulated or not with 1 M isoproterenol (Sigma-Aldrich) for 1 min. Subsequently, alprenolol (Sigma-Aldrich) was added to a final concentration of 10 M, and the cells were incubated for a further 2 min at 37°C. Incorporation of phosphate into the receptor was detected by immunoprecipitation with anti-FLAG M2 antibody followed by SDS-polyacrylamide gel electrophoresis and autoradiography. ␤ 2 AR cell surface expression was measured using [ 3 H]CGP12177 (Amersham Biosciences) binding to intact cells (25,26). Briefly, cells were incubated for various times with 1 M isoproterenol. The cells were washed three times with ice-cold HBS and incubated with the radioactive ligand for 2 h at 4°C to determine the amount of cell surface receptors. Background binding was determined in the presence of 10 M alprenolol.

RESULTS
It was shown previously that HEK293 cells transiently transfected with ␤Arr2-GFP fusion proteins and ␤ 2 ARs will translocate the fluorescent protein to the plasma membrane upon stimulation of the receptor (10). Time-lapse confocal microscopy with HEK293 cells transiently transfected with ␤ 2 AR and ␤Arr2-YFP showed that stimulation of the receptors with an agonist caused a translocation of ␤-arrestin2 to the cell surface ( Fig. 1A and supplemental movie S1). This translocation was essentially complete after 1 min, which is in agreement with earlier data (27). Quantification of the translocation showed a time constant of 2.16 Ϯ 0.06 min Ϫ1 (Fig. 1A). Agonist washout resulted in redistribution of ␤Arr2-YFP back to the cytosol which was always partial and very variable between cells. An example trace is shown in Fig. 1B.
Can the interaction between a suitably tagged receptor and ␤-arrestin be detected by FRET between YFP-and CFP-tagged proteins? To investigate this question, the ␤ 2 AR was tagged with CFP or YFP at the C terminus to yield ␤ 2 AR-CFP and ␤ 2 AR-YFP, respectively. The biological properties of ␤ 2 AR-GFP fusion proteins in comparison with the wild-type receptor have been investigated extensively (10,28). Although these proteins are unable to recycle back to the plasma membrane due to the absence of a PDZ binding motif at the extreme C terminus (29), their binding, signaling, and internalization properties are very similar to those of the wild-type receptor. ␤Arr2 was FIG. 1. A, translocation of ␤Arr2-YFP to agonist-stimulated ␤ 2 AR. HEK293 cells were transiently transfected with ␤ 2 AR, ␤Arr2-YFP, and GRK2. Confocal images were taken every 10 s. Immediately after acquisition of the first image, 10 M isoproterenol was added. The translocation was quantified as loss of fluorescence in the cytoplasm, which was normalized to the initial value. Shown are means and S.E. of eight cells. Representative examples are shown at the top and in supplemental movie S1. The curve corresponds to a monoexponential fit with a rate constant of 2.16 min Ϫ1 . B, movement of ␤Arr2-YFP upon agonist washout. HEK293 cells were transfected as described, and a single cell was perfused with either 10 M isoproterenol (bar) or HBS. Confocal images were taken every 15 s, and fluorescence in the cytosol was quantified as described under "Experimental Procedures." Shown is a trace from a representative experiment. Four other experiments yielded similar results. tagged with CFP or YFP at either the N or the C terminus, which yielded CFP-␤Arr2 and YFP-␤Arr2 (N-terminally tagged) as well as ␤Arr2-CFP and ␤Arr2-YFP (C-terminally tagged). We have shown earlier that all four fluorescent ␤-arrestins undergo agonist-induced translocation to the parathyroid hormone receptor (22), suggesting that ␤Arr2 can be tagged at either the N or the C terminus without loss of receptor binding. To further demonstrate that the tagged ␤-arrestins are functional, we showed that ␤Arr2-CFP was able to enhance the internalization of ␤ 2 AR in a manner similar to what has been reported for untagged ␤Arr2 (Fig. 2).
In all cells expressing only a single fluorescent protein and in cells expressing identically colored receptor and ␤-arrestin2, excitation at 436 nm caused very little emission at 535 nm (YFP), and agonists did not elicit any changes in F YFP /F CFP . In contrast, when cells expressing any of the four combinations consisting of receptor and ␤-arrestin2 tagged with different colors were excited at 436 nm, a significant emission of yellow light was observed in the presence of agonist (data not shown), suggesting that agonist-induced FRET between the two fluorophores had occurred. The most robust agonist-induced changes were measured for the combination ␤ 2 AR-YFP ϩ ␤Arr2-CFP, which was therefore used in all subsequent experiments. To confirm that the change in F YFP /F CFP was indeed due to FRET between ␤ 2 AR-YFP and ␤Arr2-CFP, we photobleached the acceptor (YFP) and monitored F CFP . Bleaching the YFP by illumination with 500 nm light in the continuous presence of isoproterenol led to an increase in CFP fluorescence, as expected for FRET (Fig. 3).
Continuous monitoring of the fluorescence intensities allowed us to investigate the kinetics of the receptor/␤-arrestin2 interaction. A representative trace of an experiment derived from observation of a single cell expressing ␤ 2 AR-YFP and ␤Arr2-CFP is shown in Fig. 4A. Upon short term agonist stimulation (Ͻ120 s), a slow increase in the ratio F YFP /F CFP was observed, which was due to a simultaneous increase in F YFP and a decrease in F CFP , as would be expected for FRET. The rate constant of this increase was 2.10 Ϯ 0.60 min Ϫ1 , corresponding to a t 0.5 of 19.6 s (n ϭ 6). This is virtually identical to the rate constant of ␤-arrestin2 translocation to the membrane as observed by confocal microscopy (Fig. 1). When the agonist was washed out, the ratio decreased to approximately its initial value. This suggests that the existence of the complex between the ␤ 2 AR and ␤Arr2 in living cells requires the continuous presence of agonist. Complex dissociation could also be achieved by superfusing cells with an antagonist instead of buffer (data not shown). Fig. 4A was obtained with co-transfected human GRK2. When dominant-negative GRK2 (the K220R mutant) was cotransfected instead, no agonist-induced FRET was observed (Fig. 4B). When no GRK2 was co-transfected, changes in FRET were visible but were much less robust than with co-transfected GRK2 (data not shown). Furthermore, a phosphorylation-deficient mutant of the ␤ 2 AR (19) was unable to elicit any agonist-induced FRET (Fig. 4C). These results demonstrate that GRK-mediated phosphorylation is required for a functional interaction between the ␤ 2 -adrenergic receptor and ␤-arrestin2.
In the experiments shown thus far, the agonist was applied for a short time, never exceeding 2 min. If cells were superfused with the agonist for extended times, we observed that after several minutes, FRET started to decline again, reaching its basal value after ϳ30 min (Fig. 5A). However, in translocation assays, the association of ␤Arr2-YFP with the membrane remained stable over this time range (Fig. 5B), consistent with observations by others (see for example Ref. 30 for a 60-min stimulation). The difference between the two experiments is that the tagging of ␤ 2 AR with YFP (which is required for the FRET experiment) destroys a PDZ-binding domain at the extreme C terminus of the receptor, which is present in the wild-type receptor used for the translocation experiment. This PDZ-binding domain is important for recycling of the receptor to the plasma membrane (29). Thus, in the FRET experiment the decrease of receptor levels at the plasma membrane should be more dramatic, which would lead to a pronounced loss of ␤-arrestin2 binding sites.
When a single cell was repeatedly superfused with isoproterenol, we observed that the first stimulation resulted in a fairly slow FRET signal (see above), whereas subsequent stimulations caused much faster signals (k obs 19.5 Ϯ 2.4 min Ϫ1 , corresponding to a t 0.5 of 2.1 s (n ϭ 18)) (Fig. 6A, upper panel). The kinetics of the subsequent stimulations appeared identical. This suggests that the speed of complex formation between ␤ 2 AR-YFP and ␤Arr2-CFP was greatly enhanced by prior stimulation(s) of the receptor. Using a mutant of ␤Arr2 that can bind to agonist-activated ␤ 2 ARs in a phosphorylation-independent manner (R170E (18)), the kinetic difference between first and subsequent stimulations was abolished; we always observed a rapid increase in agonist-induced FRET (Fig. 6A,  lower panel), the speed of which was comparable with the second and subsequent binding of wild-type ␤Arr2-CFP to the receptors (k obs 26.6 Ϯ 5.9 min Ϫ1 ; n ϭ 6) (Fig. 6B). In contrast to the association rates, the dissociation rates after the first and subsequent stimulations were indistinguishable (k obs of first dissociation 2.88 Ϯ 0.54 min Ϫ1 (n ϭ 7), k obs of subsequent dissociations 3.96 Ϯ 0.54 min Ϫ1 (n ϭ 13), p Ͼ 0.2 (Student's t test; when all 20 experiments were averaged, k obs was 3.60 Ϯ 0.42 min Ϫ1 , corresponding to a t 0.5 of 11.6 s). These experi-ments show that the time-limiting step for arrestin binding is GRK2-mediated phosphorylation of the receptor. This would explain why the phosphorylation-independent R170E mutant of ␤-arrestin2 showed much faster binding to the receptor upon addition of agonist. Furthermore, after short term agonist stimulation and subsequent washout, the receptors appeared to remain phosphorylated at the cell surface. This phosphorylation alone was not sufficient for ␤-arrestin2 binding to the receptor, and the complex dissociated when the agonist was washed out. Upon subsequent agonist application, the receptor was already phosphorylated, which permitted much faster formation of the receptor-␤-arrestin2 complex.
Do prephosphorylated receptors indeed accumulate in cells after short agonist stimulation? We sought to answer this question by measuring ␤ 2 -adrenergic receptor phosphorylation in intact cells. HEK293 cells transfected with FLAG-tagged ␤ 2 AR (20) and GRK2 were stimulated for 1 min with 1 M isoproterenol. To mimic the washout of agonist in the superfusion system, cells were then treated with the neutral antagonist alprenolol (10 M) for another 2 min. Subsequently, receptor phosphorylation in these cells was determined by immunoprecipitation and compared with receptor phosphorylation in cells that had been stimulated but not undergone the agonist washout procedure. Fig. 7 shows a robust increase in receptor phosphorylation after a 1-min stimulation of cells with agonist that was somewhat reduced after 2 min of alprenolol addition. These results indicate that short stimulation of cells may in- HEK293 cells were transfected with ␤ 2 AR-YFP, ␤Arr2-CFP, and GRK2, and 10 M isoproterenol was superfused for 30 min (indicated by the black bar). The normalized ratio F YFP /F CFP is shown. Four other experiments yielded similar results. B, ␤-arrestin translocation remains stable during long-term agonist stimulation. HEK293 cells were transfected with ␤ 2 AR, ␤Arr2-YFP, and GRK2. Images were taken by confocal microscopy. Immediately after the first image had been taken, 10 M isoproterenol was added, and image acquisition continued for 30 min. The quantification of translocation was done as described in the legend for Fig. 1. Shown is a representative experiment; three other experiments yielded similar results. a.u., arbitrary units. deed lead to the accumulation of phosphorylated receptors, the majority of which remains phosphorylated for at least 2 min.
As illustrated in Figs. 4 and 6, removal of agonist leads to dissociation of the receptor-␤-arrestin2 complex. To clarify whether the observed dissociation kinetics is an intrinsic property of the receptor-␤-arrestin2 complex, we compared isoproterenol with two other agonists at the ␤ 2 -adrenergic receptor, epinephrine and norepinephrine. The affinities of isoproterenol, epinephrine, and norepinephrine for the human ␤ 2 AR are 0.5, 0.7, and 26 M, respectively, and all three ligands are full agonists (31,32). Concentrations of 10-fold K D (i.e. 5 M isoproterenol, 7.5 M epinephrine, and 250 M norepinephrine) induced an increase in FRET between ␤ 2 AR-YFP and ␤Arr2-CFP, although this increase was always reduced in the case of norepinephrine (Fig. 8A). The kinetics of loss of FRET after agonist washout was dependent on the identity of the agonist (Fig. 8B); norepinephrine washout resulted in the fastest dissociation of the receptor-␤-arrestin2 complex (k obs 19.2 Ϯ 3.0 min Ϫ1 , n ϭ 7), whereas isoproterenol washout resulted in the slowest dissociation (k obs 6.6 Ϯ 0.6 min Ϫ1 , n ϭ 14). Epinephrine showed an intermediate rate (k obs 10.8 Ϯ 1.2 min Ϫ1 , n ϭ 10). This observation can be explained in two ways; either the various agonist-receptor complexes differ in their affinities for ␤-arrestin2 (and therefore in their conformation), or it is the dissociation of the agonist from the receptor, rather than that of ␤-arrestin2, that is the limiting step of dissociation of the receptor-␤-arrestin2 complex.
Measurement of receptor-arrestin interaction with FRET is not limited to the ␤ 2 -adrenergic receptor. We have previously shown similar experiments for the parathyroid hormone receptor (17). In Fig. 9A we demonstrate the interaction of ␤Arr2-CFP with a YFP-tagged murine ␣ 2A -adrenergic receptor. Sim- ilar to the ␤ 2 AR, ␤-arrestin2 binding to the receptor could be stimulated with three different ligands: norepinephrine, epinephrine, and phenylephrine. Norepinephrine and epinephrine are full agonists at the ␣ 2A -AR, whereas phenylephrine is a partial agonist. All three ligands were able to increase FRET between the YFP-tagged ␣ 2A -AR and ␤Arr2-CFP (Fig. 9A). As expected, the magnitude of the FRET signal was similar for norepinephrine and epinephrine but much lower for phenylephrine. The washout kinetics were measured as well. In an attempt to get information on ligand off-rates we employed the ␣ 2A -cam mutant, which responds to the binding of agonists with a change in FRET (17). All three ligands were able to change the FRET signal of the ␣ 2A -cam mutant, which enabled us to determine the rate of the ligand dissociation-induced conformational change in the ␣ 2A -AR. These data are summarized in Fig. 9B. It is obvious that the ␤-arrestin2 dissociation is markedly slower than the relaxation of the active state of the receptor (which probably reflects the velocity of agonist dissociation). Nevertheless, Fig. 9C shows that there is a remarkable correlation between the ␣ 2A -cam relaxation rates and the ␤-arrestin2 off-rates; the conformational change triggered by ligand dissociation is approximately 4 -5-fold faster than the ␤-arrestin2 dissociation.

DISCUSSION
In this article we show how the kinetics of complex formation between the ␤ 2 -adrenergic receptor and ␤-arrestin2 can be measured in single living cells using FRET between fluorescently labeled proteins. A number of control experiments demonstrated that we indeed had observed FRET between the two proteins. Most importantly, photobleaching of the acceptor YFP resulted in increased fluorescence of the donor CFP. The increase in FRET was absolutely dependent on GRK2 activity, showing that ␤-arrestin2 binding to the ␤ 2 AR requires GRKmediated receptor phosphorylation. This is consistent with a large body of evidence from the literature.
The requirements for interactions between G-protein-coupled receptors and ␤-arrestins are still not fully understood. For most G-protein-coupled receptors, including the light receptor rhodopsin and the ␤ 2 -adrenergic receptor, phosphorylation by G-protein-coupled receptor kinases has been shown to be a mandatory prerequisite. An exception is the luteinizing hormone/choriogonadotropin receptor (33). It is believed that interaction of ␤-arrestin with phosphate groups on the intracellular domains of the receptor triggers a conformational change in the ␤-arrestin molecule, which enables it to bind to additional parts of the receptor (34). Although it has been shown that binding of ␤-arrestin to ␤ 2 -adrenergic receptors increases the affinity of the receptors for the agonist isoproterenol (35), in vitro studies suggest that ␤ 2 AR phosphorylation is essential for ␤-arrestin binding, whereas additional treatment with agonist leads only to a 2-3-fold (36) or even lower (37) increase in ␤-arrestin binding. Our experiments showed that the majority of ␤ 2 AR ceased to interact with ␤Arr2 after agonist washout, although the process was not always fully reversible (e.g. Fig. 4A). In contrast, translocation of ␤-arrestin2 to the plasma membrane seemed much less reversible upon agonist washout (Fig. 1B). This suggests that once ␤-arrestin has reached the plasma membrane, it may be attached there by additional mechanisms, for example, by a conformational change within the receptor and/or ␤-arrestin molecule, which stabilizes the receptor-␤-arrestin complex independently of agonist, by the ability of ␤-arrestins to bind phosphoinositides (38), by ␤-arrestin dephosphorylation (39 -41), or by ␤-arrestin ubiquitinylation (27). Alternatively, agonist-receptor-␤-arres-tin2 complexes might be internalized into endosomes where they would not be affected by agonist washout. It appears, however, that the ␤ 2 AR is unable to cause internalization of bound ␤-arrestins (30) (see also Figs. 1A and 5B).
Upon prolonged stimulation we observed a decrease of FRET between the ␤ 2 -adrenergic receptor and ␤-arrestin2 (Fig. 5A) but no translocation of membrane-bound ␤-arrestin2 back into the cytoplasm (Fig. 5B). A major difference between the two experiments is that in the case of the FRET measurement, we used a ␤ 2 AR that is YFP-tagged at the C terminus. Previous studies suggest that, in contrast to wild-type receptors, such receptors are unable to recycle to the plasma membrane (29). Thus, in the FRET experiment, the amount of receptor at the plasma membrane should decline steadily, eventually reaching very low levels, whereas in the translocation experiment, a steady state balancing receptor internalization and recycling should be reached, resulting only in a moderate reduction of receptor at the plasma membrane. An alternative explanation for our results is that the ␤-arrestin that is released during receptor internalization (30) remains "trapped" at the plasma membrane, for example by its ability to bind phosphoinositides (38).
The kinetics of the disruption of the receptor-␤-arrestin2 complex is dependent on the identity of the agonist used for generation of the complex. This observation can be explained in two ways. First, each ligand could induce a slightly different conformation of the receptor with different affinities of ␤-arres-tin2 to each of these conformations. The three agonists used in this study are all full agonists with respect to G s activation (31,32). However, we observed that norepinephrine elicited a smaller increase in FRET between the ␤ 2 AR and ␤Arr2 than epinephrine or isoproterenol (Fig. 8A), suggesting that the conformation of the norepinephrine-activated receptor differs in some respect from that of the epinephrine-or isoproterenolactivated receptor. It has been shown before that the conformational requirements for receptor-␤-arrestin2 interaction are similar but not identical to those for receptor-G-protein interaction (e.g. see Ref. 22). Second, the rate of agonist dissociation could actually be the limiting factor for disintegration of the receptor-␤-arrestin2 complex. Because it is virtually impossible to determine off-rates for ␤-adrenergic receptor agonists at physiological temperature by radioligand binding (42), off-rates of some ␤-adrenergic agonists have been calculated previously from measurements of the activation of signaling pathways (43,44). This has yielded off-rates for isoproterenol of Ն4 min Ϫ1 (43) and for epinephrine of Ͼ100 min Ϫ1 (calculated by the method described in Ref. 44). Very recently, the interaction of ligands with purified ␤ 2 -adrenergic receptors has been measured by plasmon waveguide resonance spectroscopy (45). These authors found time constants for isoproterenol displacement by alprenolol of 4.7 min Ϫ1 and for epinephrine displacement by alprenolol of 7.2 min Ϫ1 (45). These values, as well as the value determined by Mueller et al. (43), agree quite well with our dissociation rates of the receptor-␤-arrestin2 complex after agonist washout (6.6 min Ϫ1 for isoproterenol and 10.8 min Ϫ1 for epinephrine; Fig. 8B), suggesting that the dissociation rate of the agonist may be limiting for ␤-arrestin2 dissociation from the receptor.
We further explored this hypothesis with the ␣ 2A -adrenergic receptor. In this system, we were able to follow the conformational changes induced by agonists in real time (17) and thus compare the kinetics of relaxation of the ligand-induced receptor conformation with the kinetics of ␤-arrestin2 dissociation. Unfortunately, this approach could not be applied to the ␤ 2adrenergic receptor, because we have so far been unsuccessful in creating a ␤ 2 -cam mutant that is properly targeted to the plasma membrane. Although the dissociation of ␤-arrestin2 from the receptor was about 4 -5-fold slower than the confor-mational change caused by agonist dissociation, the correlation between the two velocities was remarkable (Ͼ99%). Obviously the conformational change within the receptor must occur before any receptor-associated proteins can be affected by it. For example, the agonist-induced conformational change of the ␣ 2A -adrenergic receptor has been determined to occur with a minimal half-time of approx. 25 ms (17), whereas the receptorinduced G-protein activation showed a half-time of 0.5-1 s (46), which is about 20 -40-fold slower. This may be the basis for the good correlation between the off-rate of a receptor agonist and the corresponding ␤-arrestin2 off-rate.
There are several differences between the ligand dissociation experiment employing the ␣ 2A -cam receptor and the ␤-arres-tin2 dissociation experiment using the ␣ 2A AR-YFP, which may be responsible for the difference between the kinetics of the conformational change observed in the ␣ 2A -cam and the dissociation kinetics of ␤-arrestin2 from the ␣ 2A AR-YFP receptor. In the ␣ 2A -cam, most of the third intracellular loop has been replaced with YFP (17), and therefore the receptor lacks GRK phosphorylation sites. To address whether the ␣ 2A -cam mutant can still bind arrestins, we constructed an analogous receptor in which the C-terminal CFP was also replaced by YFP. Thus, it carries two YFPs, one in the third intracellular loop and one at the C terminus. This receptor mutant was unable to interact with wild-type ␤-arrestin2, as measured by FRET (data not shown). However, it interacted robustly with the phosphorylation-independent ␤-arrestin2 R170E mutant (data not shown). We determined off-rates of ␤-arrestin2 R170E for epinephrine and norepinephrine, and although they were slightly faster than the ones presented in Fig. 9B for the interaction of ␣ 2A AR-YFP with ␤Arr2-CFP, the difference was not statistically significant (data not shown). Furthermore, arrestins may induce the high affinity state of the receptor (35), and therefore coexpression of arrestins with the ␣ 2A -cam might affect the agonist off-rates. Finally, it is possible that the conformational change measured in the ␣ 2A -cam mutant is only one change in a series of conformational changes that ultimately leads to dissociation of the receptor-arrestin complex. It has previously been suggested that the ␤ 2 -adrenergic receptor undergoes a series of conformational changes upon agonist binding (47,48). It is unclear whether this finding holds true only for the ␤ 2 AR or is also applicable to other G-protein-coupled receptors. Although the conformational change required for G-protein activation is probably the one that is visible in the ␣ 2A -cam receptor, the conformation required for ␤-arrestin binding may be different, as suggested previously (47).
Probably the most physiologically relevant aspect of our study is the finding that short term agonist stimulation leads to accumulation of GRK-phosphorylated receptors at the plasma membrane. There was little difference between the extent of ␤ 2 AR phosphorylation in the presence of agonist for 1 min and at 2 min after its removal. This is consistent with previous reports suggesting that dephosphorylation of ␤ 2 -adrenergic receptors requires receptor internalization (49), which is not observed after a short stimulation. In contrast, ␤Arr2 dissociation from the (apparently still phosphorylated) receptors was essentially complete after less than 1 min. Because phosphorylation of ␤ 2 ARs by GRKs does not affect receptor signaling in the absence of arrestins (1,50), receptors thus can "remember" a previous stimulus without being impaired in their signaling ability. Subsequent stimulations can then induce very rapid ␤-arrestin binding, and this may lead not only to very rapid homologous desensitization but also to rapid switching of signaling mechanisms from G-proteins to ␤-arrestin-mediated "non-classical" signaling pathways. Such mechanisms are probably not very prominent for hormonal receptors, where agonists appear and disappear rather slowly. However, they should be operative in settings in which repeated and rapid stimulation of receptors occurs, i.e. in synapses.
In summary, our investigation of the ␤ 2 AR-␤Arr2 interactions in intact cells using FRET reveals that the interaction depends on both GRK2 phosphorylation of the receptor and on the presence of agonist. Short term stimulation leads to accumulation of GRK2-phosphorylated receptors at the cell surface. These GRK2-phosphorylated receptors retain the information of the previous stimulus, and they can thereby interact very rapidly with ␤-arrestins upon subsequent agonist binding. This represents a simple mechanism of "memory" in the receptor protein and may have important implications for the desensitization and signaling of other G-protein-coupled receptors at synaptic contacts.