In most GPCRs, GRK phosphorylation sites come in clusters (3–4 serines or threonines within 5–6 residues), and it has been established in several models that multi-phosphor-ylation of the receptor is necessary for arrestin binding (Gurevich & Benovic, 1993; Gurevich et al., 1995; Mendez et al., 2000). Thus, even before the arrestin crystal structure was solved, it was pretty clear that the arrestin phosphate sensor is likely to contain a cluster of 3 or more positive charges within a relatively short sequence. One would also expect the relevant charges to be conserved in the arrestin family and localized in the N-terminal half of the molecule (Gurevich & Benovic, 1993). This identifies the stretch of residues 166–176 as the “prime suspect” (residue numbers here and below are given for bovine visual arrestin unless otherwise specified) (Fig. 4A). Neutralization of several positive charges in this element (K167S, R171Q, and K176S) yields mutants with reduced binding to P-Rh* and inactive P-Rh, and normal binding to Rh*, consistent with direct participation of these residues in phosphate interaction (Fig. 4B and C) (Gurevich & Benovic, 1995). However, the mutant that turned out to be the most interesting in this series, R175N, binds inactive P-Rh and P-Rh* better than WT arrestin. Most importantly, this mutant binds Rh* essentially as well as WT arrestin binds P-Rh* (Fig. 4B and C). In the context of the model, this phosphorylation-independent binding suggests that Arg175 is the phosphate sensor, and that the neutralization of its charge by mutagenesis turns it “on”. There seemed to be one inconsistency, though: the model posits that the sensor directly binds phosphates, hence one would expect a decrease (rather than the observed increase) of the mutant binding to dark P-Rh and P-Rh*. However, in the context of full-length arrestin, the interpretation of the effect of a mutation of this type is far from straightforward: the expected reduction of the direct binding of this site to the receptor-attached phosphates can be counterbalanced by its “activating” effect. Luckily, the impact of any mutation on phosphate binding per se can be determined directly in the context of arrestin(1–191), which does not have any sophisticated regulatory mechanisms. In this context, the R175N mutation behaves just like the neutralization of the other phosphate-binding positive charges; it decreases P-Rh* binding without affecting Rh* interaction (Fig. 4D and E).

Thus, Arg175 fits the description of the phosphate sensor: it binds phosphates, and its neutralization by mutation makes arrestin phosphorylation-independent. Actually, charge reversal in this position (R175E) is even more effective than neutralization in promoting arrestin binding to Rh*, whereas the conservative R175K mutation that preserves the charge also preserves arrestin selectivity for P-Rh* (Gurevich & Benovic, 1997). Based on these data, we proposed the simplest conceivable mechanism of the function of the phosphate sensor (Gurevich & Benovic, 1995). In the basal (inactive) state, Arg175 interacts with a negatively charged residue within the arrestin molecule itself. The binding of receptor-attached phosphate neutralizes its charge (like the mutation), breaking this intra-molecular interaction. The disruption of this salt bridge lets the rest of the arrestin molecule “know” that the phosphate is in place. When the salt bridge is pre-disrupted by an appropriate mutation, the phosphates’ “job” is already done, hence the mutant does not need receptor-attached phosphates for high-affinity binding. Obviously, this hypothesis calls for the identification of the intra-molecular interaction partner of Arg175. The arrestin crystal structure revealed that Arg175 is right in the middle of the “polar core”, an unusual (for a soluble protein) arrangement of 5 shielded, essentially solvent-excluded, interacting charged residues in the center of the arrestin molecule (Hirsch et al., 1999) (Fig. 5). Exhaustive mutagenesis of all 5 residues corroborated the salt bridge idea and identified Asp296 as the most important negatively charged partner of Arg175 (Vishnivetskiy et al., 1999). Breaking this salt bridge by charge reversal mutations on either side (R175E or D296R) yields phosphorylation-independent arrestin species, whereas its reconstruction by the combination of these mutations (R175E+D296R) restores arrestin selectivity for P-Rh* (Fig. 5). These data clearly demonstrate that this salt bridge is the phosphate sensor, and that receptor-attached phosphates simply need to break it to make possible arrestin transition into the high-affinity-binding state.

Thus, structurally both “hot spots” in the arrestin molecule where mutations facilitate the binding to the preferred and non-preferred forms of the receptor are the intra-molecular interactions that serve as “clasps” holding it in its basal conformation. Notably, both support the relative orientation of the 2 domains in free (inactive) arrestin (Fig. 7). This strongly suggests that in the process of its binding to the receptor arrestin undergoes a global conformational rearrangement, likely involving the movement of the 2 domains relative to each other. As will be discussed in the following sections, this conclusion has profound implications for various aspects of arrestin function, yet so far it is supported only by indirect evidence. The idea that a major conformational change in arrestin accompanies its binding to the receptor was originally proposed on the basis of the unusually high energy barrier of this process (Schleicher et al., 1989). The finding that arrestin sites that do not participate in its binding to inactive P-Rh or Rh* are mobilized for the interaction with P-Rh* lent further support to this model (Gurevich & Benovic, 1993). The data that the C-tail of visual arrestin is released upon receptor binding (Palczewski et al., 1991b; Vishnivetskiy et al., 2002; Raman et al., 2003) and that the overall pattern of arrestin3 proteolysis changes in the presence of heparin or phosphopep-tide mimicking the phosphorylated vasopressin type II receptor C-terminus (Xiao et al., 2004) are also consistent with this notion. However, the absence of the C-tail in visual arrestin splice variant p44 (Smith et al., 1994) reduces the activation energy by 50% but does not completely eliminate the barrier (Pulvermuller et al., 1997), suggesting that there is more to this conformational rearrangement than just the release of the C-tail. Structurally, any global rearrangement of the arrestin molecule must involve domain movement, which would be limited by the length of the inter-domain connector termed the “hinge region” (Fig. 7) (Vishnivetskiy et al., 2002). The hinge has substantial “slack” in all 3 arrestin types with known crystal structure: it is 12 residues long, whereas just 5 residues in a fully extended conformation can cover the distance between the domains (Granzin et al., 1998; Hirsch et al., 1999; Han et al., 2001; Milano et al., 2002; Sutton et al., 2005). In its sequence, the only conserved residues are prolines, glycines, and alanines, indicating the importance of its conformational flexibility rather than anything else. If arrestin–receptor interaction requires domain movement, and if the main function of the extended hinge is to make this movement possible, one can predict that an increase in hinge length or the scrambling of its sequence should not be detrimental for receptor binding, whereas increasing deletions in the hinge region should progressively reduce arrestin’s ability to bind the receptor. The binding should be completely abolished in mutants where the hinge is just long enough to cover the distance between the 2 domains. These predictions were rigorously tested and found to be correct (Vishnivetskiy et al., 2002), adding the most convincing piece of indirect evidence supporting the idea that domain movement is necessary for arrestin binding to the receptor.

Direct studies of the binding of different arrestin proteins to the 4 functional forms of these receptors reveal striking differences in their selectivity (Gurevich et al., 1993b, 1995; Kovoor et al., 1999; Celver et al., 2002; Sutton et al., 2005) (Fig. 10). First, non-visual arrestins do not demonstrate the remarkable preference for active phosphorylated receptor (P-R*) over inactive phosphoreceptor (P-R). Arrestins 2 and 3 show only a 2-fold difference in binding levels at best, as compared to about a 10-fold difference, which is the hallmark of visual arrestin. This appears to be to a large extent an intrinsic characteristic of these arrestins because their binding to inactive and active phosphorhodopsin is also similar (Fig. 10A). Second, the difference between non-visual arrestins binding to active phosphorylated (P-R*) and unphosphorylated (ligand-activated receptor; R*) forms is also less dramatic, no more than 5-fold (Fig. 10B and C).

Thus, other arrestins show qualitatively the same binding pattern, but their receptor functional form selectivity is nowhere near that of rod arrestin. This raises the question whether all arrestins use the same activation mechanism, and if they do, what is the structural basis of their remarkably different behavior. Thus far, crystal structures of 3 out 4 types of vertebrate arrestin proteins have been solved: rod (Granzin et al., 1998; Hirsch et al., 1999), arrestin2 (Han et al., 2001; Milano et al., 2002), and cone (Sutton et al., 2005). All arrestins show high conservation of the overall fold. In fact, the differences between arrestins from bovine rod and arrestin2 to salamander cone do not exceed the differences between alternative crystal forms of the same protein (Han et al., 2001; Sutton et al., 2005). Moreover, the polar core and the 3-element interaction are invariably present and virtually identically configured, suggesting that the phosphate-sensing mechanism must be shared by all arrestins. Indeed, the introduction of various mutations homologous to known “activating” mutations in visual arrestin into arrestin2, arrestin3, and salamander or human cone arrestin has very similar functional consequences (Gurevich et al., 1997; Kovoor et al., 1999; Smith et al., 2000; Celver et al., 2002; Pan et al., 2003; Sutton et al., 2005) (Fig. 11). First, the binding of these mutants to the unphosphorylated active forms of their cognate receptors dramatically increases, often to the levels that corresponding WT arrestins show with P-R* (Fig. 11B). Second, these mutants become more promiscuous: their binding to the active phosphorylated non-cognate receptors also increases. Interestingly, the binding of these mutants to the active unphosphorylated forms of non-cognate receptors does not increase (Fig. 11), indicating that when the remaining “push” for the transition into the active receptor-binding state must be provided by the receptor via the arrestin “activation sensor”, proper fit is still required. The idea that “activating” mutations in the polar core and 3-element interaction promotes arrestin binding to non-preferred forms of their cognate receptor and to phosphorylated active non-cognate receptors by enhancing the conformational flexibility of the molecule was first proposed based on the studies of visual arrestin–rhodopsin model (Gurevich & Benovic, 1993). The positions of the residues affected by “activating” mutations in subsequently solved crystal structures of visual arrestin and arrestin2 are consistent with this idea. The results of a recent hydrogen/deuterium exchange study of arrestin2 yielded the first direct evidence supporting this hypothesis (Carter et al., 2005). Essentially the same increase in accessibility of several regions implicated in receptor binding was detected in 2 structurally distinct phosphorylation-independent mutant forms of arrestin2, R169E (polar core mutation homologous to R175E in visual) and 3A (triple alanine substitution of the bulky hydrophobic residues that anchor arrestin C-tail) (Carter et al., 2005).

Collectively, these data suggest that the phosphate sensing mechanism in all arrestin proteins works similarly, if not identically. Thus, for the structural basis of their functional differences one has to look elsewhere. Several more subtle structural characteristics that distinguish visual and non-visual arrestins have been described. Visual arrestin has an arginine (Arg18) in the loop following β-strand I, which in all arrestins contains 2 phosphate-binding lysines discussed above. Neutralization and reversal of its charge progressively reduce visual arrestin binding to P-Rh*. Most importantly, these mutations dramatically reduce the binding to inactive P-Rh, which is mediated solely by arrestin interactions with receptor-attached phosphates (Fig. 12). This identifies it as a phosphate-binding residue (Sutton et al., 2005). None of the other arrestins has any positive charge in this loop. Interestingly, both non-visual and cone arrestin in various species have at least one proline or glycine residue in this loop that rod arrestins do not have (Fig. 12). Considering that the 2 phosphate-bound lysines in β-strand I need to move in the direction of the polar core to “deliver” the phosphates, extra flexibility of the adjacent loop would likely facilitate arrestin activation, making it less “picky”, whereas the rigidity of this loop in rod arrestin would make it more selective. Thus, an extra phosphate-binding residue in the loop following β-strand I together with the reduced flexibility of this loop may account for the better discrimination by visual arrestin between the functional forms of its receptor target (Sutton et al., 2005).

Each arrestin domain is a β-strand “sandwich”, in which the two β-strand sheets are “glued” together via hydrophobic interactions between the side chains pointing inside the “sandwich”. In visual arrestin, Val90 is one of these residues, participating in multiple interactions (with Val45, Val57, Val59, and Phe118) (Hirsch et al., 1999). In both non-visual arrestins, this valine is conspicuously absent, being replaced with serine (arrestin2) or alanine (arrestin3). Even though all its potential “partners” are conserved in arrestin2 (Val41, Val53, Val55, and Phe115), the absence of this valine makes the N-domain more flexible. Visual arrestin shows very little binding to the phosphorylated active m2 muscarinic cholinergic receptor (P-m2 mAChR*), whereas arrestin2 binds this receptor well (Fig. 13). Importantly, the V90S mutation in visual arrestin, which “loosens up” its N-domain, dramatically reduces its receptor specificity, enhancing its binding to P-m2 mAChR* virtually to the level of arrestin2, while somewhat reducing binding to its cognate P-Rh* (Han et al., 2001) (Fig. 13). The magnitude of the effect of the mutation of a single residue (the side chain of which is not even exposed) strongly suggests that a relatively rigid N-domain stabilized by the interactions of Val90 with its partners is an important contributor to the high specificity of visual arrestin. The absence of these stabilizing interactions in non-visual arrestins makes their N-domains more flexible, thereby enhancing their ability to “mold” themselves to fit a wide variety of structurally diverse GPCRs. However, the introduction of the valine in the homologous position in arrestin2 (S86V) does not appreciably change its receptor preference (Fig. 13), suggesting that even with this valine its N-domain is still “loose” enough to bind its cognate receptor. Thus, there are likely other structural elements that make rod arrestin more rigid that its non-visual homologues.

The arrestin phosphate sensor is specifically designed to respond to at least 2 phosphates (Gurevich & Benovic, 1993; Gurevich et al., 1995; Mendez et al., 2000; Huttenrauch et al., 2002) in close proximity to each other largely independently of the sequence context in which the phosphorylated residues are found (Vishnivetskiy et al., 1999). This type of mechanism explains how just 2 non-visual arrestins in mammals are “primed” for binding to hundreds of different GPCRs. The requirement for a spatially concentrated charge explains why relevant serines and threonines are usually found in clusters, such as STSVSKTETS in rhodopsin (Wilden & Kuhn, 1982), STSVS and TVSTS in the m2 muscarinic cholinergic receptor (Lee et al., 2000), TTIST and TPSS in the substance P receptor, SSS and TSS in the vasopressin-2 receptor, STS in the neurotensin-1 receptor, two SSS clusters in the oxytocin receptor (Oakley et al., 2001), STQTS and TATNST in the N-formyl peptide receptor (FPR; Key et al., 2003), SS and TRTS in the D2 dopamine receptor (Namkung & Sibley, 2004), SSTS in the prostaglandin E2 receptor EP4 (Neuschafer-Rube et al., 2004), or STTT in all splice variants of the 5-HT4 serotonin receptor (Barthet et al., 2005). This mechanism also explains why arrestins do not care which particular kinase phosphorylates the receptor (as long as it adds 2 or more phosphates in close proximity); although in most cases relevant phosphates are added by GRKs (Carman & Benovic, 1998), phosphates introduced by PKC on the D2 dopamine receptor (Namkung & Sibley, 2004) or by casein kinase II on the thyrotropin-releasing hormone receptor (TRHR; Hanyaloglu et al., 2001) work for arrestins just as well.

This design of the phosphate sensor also suggests that the activating element does not necessarily have to contain receptor-attached phosphates: any spatially concentrated negative charge can disrupt the polar core. Indeed, poly-anions such as heparin or phosphopeptides were found to induce the release of the arrestin C-tail (Palczewski et al., 1991b; Gurevich et al., 1994; McDowell et al., 1999; Vishnivetskiy et al., 2002; Xiao et al., 2004), and even stimulate the binding of the most selective member of the family, visual arrestin, to unphosphorylated light-activated rhodopsin (Gurevich et al., 1994; Puig et al., 1995).

To activate the phosphate sensor of arrestins, receptors have to be phosphorylated (or contain negatively charged phosphate mimics). The first identified sites of receptor phosphorylation relevant for arrestin binding were localized in the C-terminus of rhodopsin (Sitaramayya & Liebman, 1983). Phosphorylation sites were also found to be localized in the C-terminus of another prototypical GPCR, the b2AR (Dohlman et al., 1987; Bouvier et al., 1988). Moreover, PKA phosphorylation of the 3rd cytoplasmic loop (i3) of the b2AR did not promote arrestin binding (Lohse et al., 1992). These data led to the belief that the GRCR C-terminus is the place of phosphorylation necessary for arrestin interaction, and that only GRKs can do it right. However, as soon as the studies moved beyond these two classic models, it became clear that relevant phosphorylation sites in different receptors could be localized on any cytoplasmic element (Table 1). In many receptors they are in the C-terminus, but a large group of GPCRs carry relevant phosphorylation sites in the 3rd cytoplasmic loop, and in some cases the phosphates promoting arrestin binding are found in the first (i1) or the second (i2) cytoplasmic loops (Table 1).

Thus, the phosphates (or negatively charged mimics) that activate the arrestin phosphate sensor could be localized almost anywhere on the intracellular surface of the receptor. Apparently, this is also true for other non-phosphorylated arrestin-binding receptor elements. Several experimental paradigms were used by a number of groups to identify arrestin-binding elements of >40 different GPCRs. The data are often fragmentary and both existing relatively comprehensive data sets were generated by systematic receptor mutagenesis. Dr. Weiss’ group performed alanine-scanning mutagenesis of rhodopsin cytoplasmic loops and identified L72 and N73 in i1 and P142+M143 in i2 as residues crucial for arrestin interaction (Raman et al., 1999). Simultaneous mutation of any 2 of these 3 residues virtually eliminates arrestin binding. These effects were not caused by the phosphorylation deficit of mutant rhodopsin, suggesting direct involvement of these residues in the interaction. One important caveat of this approach is that if a mutation in a particular element reduces rhodopsin phosphorylation (which was observed with several promising i3 mutants), the defects in arrestin binding cannot be unambiguously interpreted (Raman et al., 1999). Using the phosphorylation-independent arrestin, mutant R175E (Gray-Keller et al., 1997) and unphosphorylated rhodopsin circumvented this problem and confirmed the central role of these residues in arrestin binding (Raman et al., 2003). These i1 and i2 residues also proved to be important in inducing arrestin transition into its active state (Raman et al., 2003). Dr. Ascoli’s group took advantage of the fact that rat and human lutropin receptors are highly homologous, yet the human version internalizes 7 times faster than the rat receptor via an arrestin-dependent pathway (Nakamura et al., 2000). This allowed the authors to zero in on the very few amino acids that differ in the 2 receptors and elegantly identify 7 non-contiguous residues that determine arrestin interactions. The substitutions that dramatically improved arrestin binding to the rat lutropin receptor turned out to be VQ→IH in i2; R→K, Q→R, T→M, and P→T in the i3; and L→F in i4 (proximal C-terminus preceding the palmitoylation site). The relatively mild (often conservative) nature of these substitutions suggests that they are unlikely to change the global conformation of the respective receptor domains, increasing the probability that these are indeed the residues that directly interact with arrestin.

In contrast to rhodopsin, many GPCRs have very large cytoplasmic loops and/or C-tails (Probst et al., 1992), making comprehensive alanine-scanning mutagenesis impractical. Convenient pairs of highly homologous receptors with dramatically different internalization and arrestin binding are also an exception rather than the rule. Therefore, a variety of alternative approaches have been used to identify arrestin-binding receptor elements. Several groups used synthetic or, more often, overexpressed and purified (e.g., as GST fusions) peptides representing intracellular receptor domains. Rhodopsin (Krupnick et al., 1994) and lutropin receptor (Mukherjee et al., 1999b) loop peptides were tested for their ability to compete with the native receptor for arrestin. Rhodopsin i1 and i3 competed with P-Rh* with 30–1000 μM affinities, whereas lutropin receptor i3 had an IC50 of 10 μM (Mukherjee et al., 1999b). The studies of arrestin interactions with receptor elements using surface plasmon resonance also yielded micromolar affinities (Cen et al., 2001a; Liu et al., 2004). Wild-type arrestins bound to the immobilized phosphorylated C-terminal peptide of the N-formyl peptide receptor with micromolar affinities, although several constitutively active mutants demonstrated sub-micromolar dissociation constants (Potter et al., 2002). Interestingly, in a few studies where the affinity of arrestin proteins for the native phosphoreceptor has been measured, it was found to be nanomolar for rhodopsin (Schleicher et al., 1989; Pulvermuller et al., 1997; Osawa et al., 2000) and sub-nanomolar for the b2AR or m2 muscarinic cholinergic receptor (Gurevich et al., 1993a, 1995). As the binding affinity directly reflects the energy of the interaction (ΔG°=−RT lnK_A), one can calculate that cooperative 3- to 4-point binding, where each individual interaction has micromolar affinity, would yield a sub-nanomolar K_D. Thus, all reported affinity measurements are in good agreement with the rest of the data, clearly indicating the multi-site nature of the arrestin–receptor interaction. Qualitative studies of direct arrestin binding to various elements of different receptors have implicated i3 of the m2 and m3 muscarinic, the α2A-adrenergic receptors (Wu et al., 1997), and the 5-HT2A receptor (Gelber et al., 1999), i3 and the C-terminus of delta-opioid and the C-terminus of kappa-opioid receptors (Cen et al., 2001b), as well as i2, i3, and the C-terminus of dopamine D2 (Macey et al., 2004) and D1 receptors (Macey et al., 2005). Where the binding of a combination of elements was tested (e.g., i3+C-terminus of the delta-opioid receptor (Cen et al., 2001b), it was found to be additive, suggesting that these elements bind to different sites on arrestin.

The most popular (although less mechanistically rigorous) approach used to delineate arrestin-binding receptor elements is the use of changes in (presumably) arrestin-dependent trafficking of mutant and chimerical receptors as the readout. These experiments also identified the same “usual suspects”, singly or in combinations: various cytoplasmic loops in many GPCRs, as well as the C-termini of most receptors known to man (Table 2).

However, a compelling set of studies of the m2 muscarinic receptor (Pals-Rylaarsdam et al., 1997; Lee et al., 2000) suggested that there is more to receptor phosphorylation than previously thought. This receptor is primarily phosphorylated on the 2 clusters in i3. The phosphorylation of the C cluster (TVSTS) is absolutely required for arrestin binding in vivo and in vitro (Pals-Rylaarsdam et al., 1997). Mutation of this sequence to AVAAA virtually eliminates the receptor’s ability to bind wild-type arrestins, similar to mutations or deletions of the phosphorylation sites in the b2AR or delta opioid receptor (Kovoor et al., 1999). Phosphorylation-independent arrestin2 mutants with the phosphate sensor “pre-activated” bind phosphorylation-deficient b2AR and delta opioid receptors (Kovoor et al., 1999; Celver et al., 2002), but the same “super-arrestins” still fail to bind the C-cluster mutant of m2 receptor (Lee et al., 2000). Unexpectedly, the deletion of 15 residues (including the C-cluster) yields an m2 receptor that binds even wild-type arrestins 2 and 3 normally. Thus, it appears that the C-cluster and surrounding sequence is an inhibitory element suppressing m2 interactions with arrestins. The deletion or, in the wild-type receptor, the phosphorylation of this element relieves the inhibition, likely moving it out of the way to allow arrestin binding to other m2 receptor elements (Pals-Rylaarsdam et al., 1997; Lee et al., 2000). This was the first description in a mammalian GPCR of a receptor element serving as a “brake” for arrestin binding that is released by receptor phosphorylation (although the term was coined a bit later; Whistler et al., 2001). Soon it became clear that similar mechanisms operate in other receptors, as well. Alanine substitution of serines and threonines in the distal C-terminus of the delta opioid receptor yielded a mutant that did not bind arrestins and failed to internalize, whereas the deletion of this element containing the regulatory phosphorylation sites was found to enable phosphorylation-independent recruitment of arrestin3 and normal receptor internalization (Whistler et al., 2001). Interestingly, the substitution of serines and threonines with aspartates also enables arrestin binding and receptor internalization, suggesting that the brake is removed from its inhibitory position by electrostatic interactions. Another example of a similar brake was described for the rat follitropin receptor (Kishi et al., 2002). Agonist activation leads to the phosphorylation of this receptor on i1 and i3 rather than on the C-tail. The deletion of 10 internal residues in the C-tail of this receptor that do not appear to be phosphorylated facilitated arrestin binding and arrestin-mediated internalization severalfold (Kishi et al., 2002). Although this issue has not been explored, it is tempting to speculate that in this case the phosphorylation of other receptor elements, rather than that of the brake itself, relieves the inhibition of arrestin binding. In yet another example, in i3 of the D1 dopamine receptor, alanine substitutions of all serines and threonines, or of the 3-serine cluster (256, 258, 259), yielded receptors with reduced phosphorylation and impaired arrestin binding and desensitization (Kim et al., 2004). C-terminal truncation of the D1 receptor also reduced phosphorylation, but even the most severely truncated mutant bound arrestin and desensitized normally. These data suggest a similar scenario; activation-induced phosphorylation of the D1 receptor moves its C-tail out of the way, enabling direct phosphorylation independent association of arrestin with the elements of the 3rd cytoplasmic loop (Kim et al., 2004).

It is likely that the position of phosphorylated elements in the 3-dimensional structure of different GPCRs varies as much as it does in the linear sequence of these receptors. Then bound arrestin simultaneously anchored via an unknown element at the inter-helical cavity and via its phosphate-binding residues at the phosphorylated cluster may well be oriented quite differently relative to the “long” horizontal axis (helix I–helix V) of the receptor (Fig. 15).

One might ask whether these theoretical considerations have any relevance in real life. Several unexpected findings in the last 2 years suggest that they have. The first indication that the complex of the same arrestin with the same receptor can come in different “flavors” came from studies of the N-formyl peptide receptor (FPR), which has 2 main serine/threonine clusters in its C-terminus: residues 328–332 (STQTS) and 334–339 (TATNST) (Key et al., 2003). It was found that although phosphorylation-independent arrestin2(1–382) binds the phosphorylation-deficient receptor, the complex does not demonstrate high agonist affinity, in sharp contrast to the wild-type arrestin2-phospho-FPR complex (Key et al., 2001). The restoration of the STQTS cluster increased mutant arrestin affinity for the receptor but did not restore high agonist affinity of the complex. In contrast, the restoration of the TATNST cluster did not significantly affect arrestin affinity for the receptor, but it restored high agonist affinity of the complex (Key et al., 2003). This was the first demonstration that the functional properties of the arrestin complex with the receptor phosphorylated at alternative clusters are different. The simplest structural interpretation of these data is that depending on the position of the phospho-cluster that is in contact with the phosphate binding “patch” of positively charged arrestin residues (Fig. 15) an appropriate arrestin element can or cannot reach the inter-helical cavity (or some other part of the receptor) and therefore does or does not stabilize the high agonist affinity state of the receptor.

The results of two recent studies revealed unexpectedly distinct functional consequences of the phosphorylation of the angiotensin II receptor (Kim et al., 2005a) and V2 vasopres-sin receptor (Ren et al., 2005) by different GRKs. In both cases, the bulk of receptor phosphorylation, arrestin recruitment, desensitization of G-protein-mediated signaling, and receptor internalization was shown to be mediated by GRKs 2 and 3. In contrast, most of the arrestin-mediated ERK1/2 activation depends on receptor phosphorylation by GRKs 5 and 6. The inhibition of GRK5/6 expression abolishes arrestin-mediated ERK signaling, whereas the inhibition of GRK2/3 expression enhances it. Accordingly, arrestin-mediated ERK activation is enhanced by GRK5/6 overexpression and inhibited by the overexpression of GRK2/3. These data clearly establish different functional capabilities of the arrestin complex with receptor phosphorylated by different GRKs. If one accepts the only logical assumption that GRK2/3 and GRK5/6 phosphorylate different receptor elements, which subsequently “compete” for the phosphate binding “patch” on arrestin, these apparently mystifying results lend themselves to the same mechanistic explanation as the FPR data. Different positions of phosphorylated receptor elements differentially orient arrestin on the receptor, yielding structurally and functionally distinct arrestin–receptor complexes (which, however, include the same arrestin and the same receptor).

Another recent study suggests that the same arrestin bound to different receptors assumes different conformations (Shenoy & Lefkowitz, 2005). The authors demonstrate that in complex with the angiotensin II type 1A (AT1A) receptor, lysines 11 and 12 in arrestin3 become ubiquitinated. This ubiquitination is required for stable arrestin3 association with the receptor, their co-trafficking, robust ERK activation, and the retention of ERK in the cytoplasm. The arrestin3(K11,12R) mutant cannot perform these functions, whereas the fusion of a ubiquitin moiety at the C-terminus of this double arginine mutant restores them. In contrast, the binding of the same arrest-in3(K11,12R) mutant to vasopressin V2 and neurokinin 1 receptors is indistinguishable from wild-type arrestin3, and it mediates vasopressin-induced ERK activation normally. However, quintuple mutant arrestin3(K18,107,108,207,296R) is impaired in endosomal trafficking with V2 vasopressin, but not with the AT1A receptor (Shenoy & Lefkowitz, 2005). Because ubiquitin ligases are not particularly “picky” and often modify the most exposed lysines, these data suggest that these arrestin3 residues are differentially exposed depending on the receptor type it binds.

Nonetheless the arrestin aspect of the “dimer problem” actually has a scientifically relevant dimension. Even when one arrestin binds one receptor, we need to know whether arrestin prefers its target in the form of monomer or as a part of the dimer (or whether it cares one way or the other). This issue has never been directly addressed experimentally. However, a few recent studies suggest that at least non-visual arrestins can bind receptors within a dimer. One study took advantage of the difference between arrestin preference of the 2 types of thyrotropin-releasing hormone receptor (TRHR): TRHR1 interacts with both arrestin2 and arrestin3, whereas TRHR2 interacts only with arrestin3, as evidenced by the enhanced internalization of TRHR1 by overexpression of either non-visual arrestin; TRHR2 endocytosis was enhanced only by arrestin3 (Hanyaloglu et al., 2002). Accordingly, the trafficking of TRHR2 was impaired in mouse embryonic fibroblasts lacking either arrestin3 or both arrestins 2 and 3, whereas the trafficking of TRHR1 was only impaired in fibroblasts lacking both non-visual arrestins. The formation of TRHR1/2 heterodimers increased the interaction of TRHR2 with arrestin2. Most importantly, even heterodimerization between TRHR2 and truncated TRHR1(C335Stop), which by itself does not interact with non-visual arrestins, enhanced TRHR2–arrestin2 interaction (Hanyaloglu et al., 2002). The simplest mechanistic interpretation of these findings is that in a heterodimer with TRHR1 (but not in the TRHR2 homo-dimer) the conformation of the TRHR2 cytoplasmic elements changes in such a way as to permit arrestin2 binding tight enough to promote arrestin2-dependent internalization. This interpretation implies that arrestin2 binds a receptor molecule that is part of a dimer.

Another study used two GPCRs with distinct modes of arrestin interaction and different trafficking patterns: the mu-opioid receptor (MOR) and neurokinin1 receptor (NK1) (Pfeiffer et al., 2003). Arrestin binds MOR expressed alone and brings it to the coated pit but does not internalize with the receptor, whereas arrestin internalizes in complex with NK1. Co-expressed MOR and NK1 form heterodimers, as evidenced by BRET and co-immunoprecipitation. Although heterodimerization does not appreciably change the ligand binding and signaling properties of these receptors, it gives rise to several interesting phenomena (Pfeiffer et al., 2003). First, exposure of the heterodimer to either MOR- or NK1-selective agonists promotes cross-phosphorylation and co-internalization of the other receptor. Second, regardless of which receptor is stimulated, the heterodimer co-internalizes with arrestin to endosomes. Consequently, resensitization of MOR is severely delayed in cells co-expressing both receptors, as compared to cells expressing MOR alone (Pfeiffer et al., 2003). Cross-phosphorylation is easily explained by the fact that GRK is activated by the active receptor itself, whereupon it phosphorylates any substrate that happens to be around (Palczewski et al., 1991a), in this case the other receptor in a dimer. Cross-internalization and especially co-trafficking of the whole dimer along with arrestin suggest not only that arrestin interacts with the receptor in a dimer, but also that this interaction is tight enough to bring the whole dimer inside. Moreover, it appears that in the pair the receptor with higher affinity for arrestin (in this case NK1) “wins”, dragging its partner (MOR) through the long recycling pathway, thereby slowing down its resensitization.

The stability of different arrestin–receptor complexes varies. Extensive use of arrestin-GFP fusions revealed that the activation of most GPCRs induces rapid arrestin translocation to the plasma membrane, indicative of its binding to the receptor (Barak et al., 1997). After that, the scenario can be quite different and largely depends on the receptor studied. Some GPCRs, such as the b2AR, D1 dopamine, and endothelin type A receptors release bound arrestin near the plasma membrane, apparently soon after internalization, whereas others, including angiotensin II type 1A (AT1A) or neurotensin receptors, travel with bound arrestins inside the cell (Zhang et al., 1999). The exchange of the C-termini between the b2AR and AT1A reverses this pattern, suggesting that this receptor element largely determines the stability of the complex (Zhang et al., 1999). The difference in complex stability determines the rate of receptor resensitization: the exchange of the C-tail between the b2AR that recycles and resensitizes rapidly and the vasopressin V2 receptor that keeps bound arrestin longer and recycles and resensitizes slowly, switches the pattern of their trafficking and the rate of their dephosphorylation, recycling, and resensitization (Oakley et al., 1999). Similarly, the exchange of the C-termini between 2 vasopressin receptors, V1a that recycles rapidly via peripheral endosomes, and V2 that moves slowly via the perinuclear recycling compartment reversed their trafficking pattern and the rate of their recycling (Innamorati et al., 2001).

Based on the relative affinity for different arrestins and the stability of the arrestin–receptor complex, GPCRs were divided into 2 classes (Oakley et al., 2000). Class A receptors (b2AR, mu-opioid, endothelin type A, dopamine D1, and α1B-adrenergic) bind arrestin3 with higher affinity than arrestin2 and do not interact with visual arrestin. These receptors also tend to release arrestin soon after internalization and recycle rapidly. Class B receptors (AT1A, neurotensin receptor 1, vasopressin V2, TRHR1, and NK1) bind both non-visual arrestins equally well, interact with visual arrestin, and tend to internalize with bound arrestin and recycle more slowly (Oakley et al., 1999; Zhang et al., 1999; Oakley et al., 2000). Obviously, a number of GPCRs do not fit into this classification. For example, the metabotropic glutamate receptor 1a interacts selectively with arrestin2, but not with arrestin3 (Dale et al., 2001), somatostatin receptor 2A internalizes along with arrestin, yet is recycled and resensitized rapidly, whereas somatostatin receptor 3 releases arrestin near the plasma membrane, yet does not recycle rapidly and is largely degraded after internalization (Tulipano et al., 2004). Nonetheless it is clear that the stability of the arrestin–receptor complex significantly affects the trafficking pattern of different GPCRs.

These experiments do not explain how the choice between the degradation and recycling is made for the same receptor in the same cell. Even the b2AR that is rapidly internalized and recycled even in the presence of agonist (Morrison et al., 1996) becomes progressively degraded upon stimulation for many hours (Pan et al., 2003). Several studies using various arrestin mutants suggest that the stability of the complex may play a role in this choice, as well as in arrestin-dependent receptor trafficking beyond internalization. Arrestin release must precede receptor dephosphorylation, leaving the loss of active receptor conformation due to agonist dissociation in acidic endosomes as the only logical signal for arrestin to get off. Then the receptor has to undergo several cycles of dephosphorylation (because receptor multi-phosphorylation is required for arrestin binding) before it can finally emerge in a recycling-competent fully dephosphorylated form (Hsieh et al., 1999). However, both non-visual arrestins poorly discriminate between phosphorylated active and inactive forms of the receptor (Fig. 10), suggesting that arrestin release may be rate limiting in this process. On the other hand, the binding of phosphorylation-independent mutants to the unphosphorylated receptor is strictly activation dependent (Fig. 16A), so that in this case receptor deactivation would be immediately followed by arrestin release. Moreover, because both phosphorylation-independent arrestins and GRKs compete for the same functional form of the receptor, R*, these mutants directly inhibit receptor phosphorylation (Fig. 16B) (Pan et al., 2003). Thus, if the expression of the mutant arrestin is high enough to out-compete endogenous wild-type arrestins and GRKs, the receptor primarily internalizes in an unphosphorylated form in complex with the constitutively active arrestin. As could be expected, the b2AR internalizes in these complexes at a normal rate but recycles much faster than the same receptor internalized in “normal” phosphoreceptor–arrestin complexes (Pan et al., 2003). It turned out that this accelerated cycling prevents receptor down-regulation even after very long (up to 24 h) agonist exposure (Fig. 16C). Phosphorylation-independent arrestin mutants bind phosphorylated receptor at least as well as wild-type arrestin, and in this situation their ability to discriminate between active and inactive phosphoreceptor is, if anything, even worse than that of wild-type arrestin (Fig. 16A). Thus, one can predict that overexpression of a GRK along with the mutant would ensure that most receptors get phosphorylated before they have a chance to encounter the “super-arrestin” mutant. Indeed, overexpression of GRK2 along with mutant arrestin returns the situation back to normal, “rescuing” receptor down-regulation (Fig. 16C) (Pan et al., 2003). Thus, the formation of a less stable complex with exactly the same receptor tips the balance towards recycling, whereas the stabilization of the complex favors receptor degradation. This conclusion is in remarkable agreement with the correlation observed for a variety of different GPCRs that belong to classes A or B (Oakley et al., 1999; Zhang et al., 1999).

This conclusion is also supported by studies where a different type of arrestin mutant was used to make the arrestin–receptor complex unusually stable. Rapid internalization of the b2AR requires transient ubiquitination of receptor-bound arrestin3, which is accomplished by E3 ubiquitin ligase Mdm2, that directly interacts with arrestin (Shenoy et al., 2001). The time course of arrestin3 ubiquitination and de-ubiquitination is consistent with the idea that arrestin carries a ubiquitin moiety as long as it is bound to the receptor (Shenoy et al., 2001). Indeed, the activation of the V2 vasopressin receptor, which stays in complex with arrestin longer than b2AR, results in more prolonged arrestin ubiquitination (Shenoy & Lefkowitz, 2003). Switching the b2AR and V2 C-termini, which to a large extent determine the stability of the complex (Oakley et al., 1999), also reverses the kinetics of arrestin ubiquitination induced by the activation of these receptors (Shenoy & Lefkowitz, 2003). An arrestin3–ubiquitin fusion protein that cannot be de-ubiquitinated in the cell stays in complex with the b2AR just as long as wild-type arrestin remains bound to class B receptors. Most importantly, the expression of the arrestin–ubiquitin fusion enhances the degradation of the b2AR (Shenoy & Lefkowitz, 2003). Thus, arrestin mutations that make b2AR–arrestin complex more transient prevent receptor degradation (Pan et al., 2003), whereas mutants forming unusually stable complexes shift the equilibrium towards degradation (Shenoy & Lefkowitz, 2003), suggesting that the stability of the complex plays a role in the determination of the fate of internalized receptor. One of the mechanisms whereby this can be accomplished appears to be arrestin-mediated recruitment of enzymes that ubiquitinate the receptor itself, as has been demonstrated for the b2AR (Shenoy et al., 2001) and the V2 vasopressin receptor (Martin et al., 2003).

The properties of the arrestin–receptor complex also affect post-endocytic receptor trafficking in other ways. An example of this is the N-formyl peptide receptor, which internalizes in an arrestin-independent fashion but requires arrestins for recycling (Vines et al., 2003). In normal cells this receptor recycles relatively fast, whereas in arrestin-deficient cells the receptor accumulates in the perinuclear recycling compartment and fails to recycle. The recycling can be rescued by the overexpression of arrestin2 or 3 (Vines et al., 2003). The recycling of this receptor is inhibited by the expression of a constitutively active arrestin2(I386A, V387A, F388A) mutant that binds the phosphorylated receptor with higher affinity than wild-type arrestin (Key et al., 2005). Importantly, the recycling was restored when receptor was only partially phosphorylated (Key et al., 2005), which reduces arrestin affinity for the receptor. Thus, excessively stable arrestin association stops the receptor in its tracks, inhibiting recycling of the N-formyl peptide receptor.

Thus, the stability of the arrestin–receptor complex is an important factor, which determines the receptor trafficking pattern and the ultimate fate of the internalized receptor. However, most of the data leading to this conclusion were generated using various receptor and arrestin mutants and chimeras, that is, the tools that the cell does not have at its disposal, raising the question whether the cell can actually regulate the properties of the arrestin–receptor complex by some normal, natural means. Biochemical studies of the mechanisms of the arrestin–receptor interaction suggest at least one relatively simple method the cell can employ to change the stability of the complex. The properties of the complex and the biological consequences of its formation depend to a large extent on the nature of the phosphorylatable receptor elements (Oakley et al., 1999; Zhang et al., 1999; Innamorati et al., 2001) and the actual level of their phosphorylation (Oakley et al., 2001; Pan et al., 2003; Key et al., 2005). Virtually all receptors have many more GRK phosphorylation sites than the minimum of 2 required for the activation of the arrestin phosphate sensor to induce arrestin transition into its high-affinity receptor-binding state. When arrestin binds, it physically covers the phosphorylation sites, whether they are “hit” by GRK or not, preventing further receptor phosphorylation (Fig. 16B) (Pan et al., 2003). Because active receptor encounters with GRK and arrestin are stochastic events governed by the law of mass action, in a cell that has relatively little GRK activity and high levels of arrestins most receptors will meet and bind arrestin as soon as they have the minimum requisite number of phosphates attached. Conversely, in cells expressing a lot of GRKs and relatively little arrestin, most receptors will be phosphorylated many times before they encounter and bind an arrestin molecule. The first situation would mostly generate arrestin–receptor complexes with the lowest possible stability, whereas in the latter case more stable complexes would emerge in abundance. The functionally relevant differences would persist even beyond the eventual dissociation of arrestin because the receptor that emerges would have varying numbers of attached phosphates requiring different time for full dephosphorylation.

Sometimes the reports that free or receptor-bound arrestin bind a particular partner are perceived as conflicting, as is the case with ARF6 and ARNO (Mukherjee et al., 2000; Claing et al., 2001; Hunzicker-Dunn et al., 2002), whereas it is likely that these partners simply bind both functional forms of arrestin. Another example is JNK3. Arrestin3 was reported to serve as a scaffold for receptor activation-dependent JNK3 phosphoryla-tion by the ASK1-MKK4-JNK3 activation cascade (McDonald et al., 2000). Yet arrestin3 that has a nuclear export signal (NES) in its C-terminus also redistributes JNK3 from the nucleus to the cytoplasm (Scott et al., 2002). Because membrane-bound GPCRs obviously cannot move between the nucleus and the cytoplasm via aqueous nuclear pore where NES-dependent transport operates, this must be the function of free arrestin, suggesting that it also binds JNK3. Similarly, bound non-visual arrestins mobilize ubiquitin ligase Mdm2 to GPCRs where it ubiquitinates the receptor (Shenoy et al., 2001), whereas free arrestins also redistribute Mdm2 from the nucleus to the cytoplasm (Wang et al., 2003).

Mechanistically, the change in arrestin conformation upon receptor binding likely explains an enhanced affinity of receptor-bound arrestin for many non-receptor-binding partners (Gurevich & Gurevich, 2003). However, free arrestin (like any normal protein) is not as rigid as the crystal structure may lead one to believe: it “breathes”, likely “sampling” a number of conformational variations, although for the sake of simplicity we call this whole group of conformational states “basal”. In the process, it may occasionally expose the same sites for its binding partners that become fully exposed in the receptor-bound state (which again is likely a whole family of conformations that we classify as “active”). The destabilization of the basal conformation of free arrestin by mutagenesis (Kim & Benovic, 2002), by the phosphopeptide mimicking the phosphorylated receptor C-terminus, or even by heparin (Xiao et al., 2004), enhances clathrin binding apparently because “loose” arrestin assumes the conformation somewhere in between the basal and active-like more often. It is likely that this would also be true for other binding partners that prefer receptor-bound arrestin, but so far this has never been tested experimentally.

Unless a single arrestin molecule can simultaneously scaffold the c-Raf1-MEK1-ERK1/2 and ASK1-MKK4-JNK3 cascades (which, given the relative sizes of arrestin and these kinases, seems virtually impossible; Fig. 17), there also must be inter-dependence of the binding of certain partners. For example, if arrestin with bound JNK3 or ERK1/2 were equally likely to bind c-Raf1 and ASK1, a lot of resulting complexes would be unproductive. Therefore, it would make sense biologically if the binding of JNK3 made the binding of ASK1 more likely (and/or vice versa), whereas the binding of ERK1/2 should facilitate the binding of c-Raf1 at the expense of ASK1. The mechanistic information regarding these interactions is sadly missing.