Functional map of arrestin-1 at single amino acid resolution

Scanning mutagenesis is a powerful protein-engineering technique that has been particularly successful in the stabilization of GPCRs (14) or in the study of specific protein–protein interactions (for example, ref. 15). To facilitate large-scale mutagenesis projects, we recently developed software tools for primer design and sequence analysis (16). Here we have used these tools to scan the complete arrestin-1 sequence. We then compared the binding of all 403 mutants to the phosphorylated GPCR rhodopsin. To do this, we changed each arrestin-1 residue, except the first methionine, to alanine, or in case of a native alanine at a particular position, to glycine. These mutations remove all side-chain atoms beyond the β-carbon, which allows us to probe the functional role of side chains throughout the arrestin molecule with a direct binding assay to phosphorylated light-activated rhodopsin (Fig. S1). To facilitate the large-scale screening of mutants we labeled arrestin-1 by fusion to the fluorescent protein mCherry, which allowed binding experiments directly from Escherichia coli cell lysates. We obtained the highest level of binding to light-activated (*) phosphorylated rhodopsin in rod outer segment (P-ROS*) membranes, whereas dark phosphorylated (P-ROS) and light-activated unphosphorylated (ROS*) membranes did not bind arrestin-1 efficiently (Fig. 1A). These results are in agreement with the requirement of rhodopsin to be both activated and phosphorylated for high-affinity binding of arrestin (17). To compare the relative binding affinity of mutants, we developed a high-throughput assay that takes advantage of the dependency of arrestin–rhodopsin interactions on low ionic strength (17, 18). For each mutant, the amount of arrestin-1 bound to P-ROS* was determined at eight different salt concentrations to fit the data to sigmoidal dose–response curves (Fig. 1B). Importantly the half-maximal inhibitory concentration (IC₅₀) for NaCl determined with this assay is largely independent from variations in the arrestin-1 expression level, as we chose a rhodopsin concentration of 1.25 µM, far in excess of the 5–50 nM apparent binding affinity of arrestin-1 (19, 20). As a further precaution we expressed wild-type arrestin in parallel with each set of mutants. The variation in expression level between these 59 independent experiments was ±61%, yet the determined dose–response curves yielded a very reproducible IC₅₀ of 0.41 ± 0.04 M for wild-type arrestin-1 (a frequency distribution analysis is provided in Fig. S2). The obtained value is in good agreement with previous reports using radioactively (17) or fluorescently labeled arrestin-1 (18), suggesting that the mCherry fusion did not inhibit arrestin binding and can be used for the relative comparison of arrestin mutations. Besides the wild type, we measured dose–response curves and determined IC₅₀ values for 403 arrestin mutations. (A selection of data is available as Table 1, and all mutations are listed in Table S1). From this dataset, we excluded 16 mutants from further analysis because they were binding at a level less than 10% of wild-type arrestin-1 expressed in parallel. Further analysis by in-gel fluorescence of mCherry showed that 12 of these mutants did express very poorly and the remaining 4 did not express at all within the detection limit. Once these low-expressing mutants had been excluded, the remaining dataset did not show a correlation between the expression level of a mutant and the determined IC₅₀ value. In total our functional data thus covers 96% of the arrestin-1 sequence, providing an extensive dataset for single amino acid contributions to the binding of arrestin to phosphorylated rhodopsin.

About 50% of mutations leave the IC₅₀ values basically unchanged, whereas the remaining values varied between 1.28 ± 0.17 M and 0.24 ± 0.01 M for the strongest and weakest binders (Fig. 1C), respectively. Increased ionic strength predominantly reduces the strength of hydrophilic interactions. Nevertheless out of 25 of the best-binding mutations, 13 affected polar residues including 10 residues that are charged under physiological conditions. Similarly, 10 of the worst 25 binding mutations affect polar residues including 4 charged residues. This even distribution between polar and hydrophobic residues throughout the dataset is in agreement with the idea that arrestin binding involves a multitude of hydrophilic and hydrophobic interactions, as well as specific conformational changes and is not dominated by a few charged interactions. However, weakening of polar interactions by increasing salt concentrations is sufficient to inhibit complex formation, an effect that is altered when additional hydrophobic interactions are weakened by mutagenesis. Principally, salt acts as a classical multisite inhibitor in a competition-binding experiment, allowing us to compare the relative effects of arrestin-1 mutations affecting both hydrophilic and hydrophobic interactions. Whereas many mutations affected specific arrestin activation sensors and receptor-binding sites (see Results and Discussion, Functional Map of Arrestin-1), several mutations affected the packing of the arrestin β-sheets. Mutation of all residues pointing from β-strand XIV into the C domain, for example, moderately increased binding, likely because they increase the structural flexibility of the protein. The identification of such stretches and structural regions within the protein is one of the strengths of our comprehensive approach, as it allows us to connect functional regions and overall increases the reliability of the individual measurements.

Plotted on the crystal structure of the arrestin basal state (Fig. 2), our data highlight the regions involved in binding to rhodopsin and allow a quantitative interpretation of the mechanisms of arrestin regulation. Arrestins are elongated molecules consisting of two cup-like domains predominately formed by the N- and C-terminal parts of the protein. Each of the two domains, named “N domain” (residues 8–180) and “C domain” (residues 188–362) is constructed from a seven-strand sandwich whose loops form much of the interface between the domains. Mutations affecting rhodopsin binding were sparser in the C domain and had weaker effects compared with those in the N domain (Fig. 2). The strongest effect was achieved by several mutations along the arrestin C tail (C terminus starting from residue 375), which folds back over the N domain. This region is known as a three-element interaction site because the C tail interacts with β-strand I and α-helix I. Disruption of these interactions by mutagenesis yields constitutively active arrestins with dramatically reduced selectivity: high binding to P-ROS* and significantly enhanced binding to the nonpreferred forms of rhodopsin in inactive P-ROS and unphosphorylated ROS* (21, 22). In agreement with these results we observed increased binding after mutation of Phe375, Val376, and Phe377 of the C tail; His10, Val11, and Phe13 of β-sheet I; and Leu111 in α-helix I. Interestingly, mutation of Phe375 and Phe377 are among the strongest binding mutations with IC₅₀ values of 1.28 ± 0.17 M and 1.08 ± 0.11 M respectively. In comparison the effect of mutations in β-strand I and α-helix I was less pronounced and increased salt tolerance to between 0.67 ± 0.06 M and 0.84 ± 0.06 M. The smaller effect of mutations in the two elements (β-strand I and α-helix I) that clamp the C tail in inactive arrestin-1, suggests that the main role of the three-element interaction is to restrain the C tail in the basal arrestin conformation. The critical role of the arrestin C tail is further confirmed by mutation of Phe380, whose phenolic side chain is buried deeply within arrestin where it interacts with Arg29 in the inactive conformation (Fig. 3). Two residues further along the polypeptide chain, Arg382 anchors the C tail within the polar core, one of the phosphate-sensing regions in arrestin (17, 21). The polar core is maintained by extensive charge–charge interactions between the N and C domains of the molecule and comprises charged residues from the very N terminus (Asp30), the body of the N domain (Arg175), the interfacial loop of the C domain (Asp303), and the C tail (Arg382). Disruption of either of these charged interactions by alanine mutations leads to increased IC₅₀ values with the strongest effects observed by mutation Arg382Ala in the C tail. Besides mutagenesis of charged residues in the polar core, mutating the uncharged residue Gly297 strongly increased binding to rhodopsin. Likely the methyl group introduced by the Gly-to-Ala substitution introduces a steric disturbance into the polar core by clashing with the position of Thr304, another residue facing the polar core whose mutation leads to significantly increased binding. Altogether it is striking to note that 9 of the 10 best binding mutations are located in either the polar core or the C tail of arrestin, highlighting the special importance of these regions.

The importance of the arrestin C tail had been recognized early on based on limited proteolysis studies (23). Release of the C tail is one of the major conformational changes occurring during arrestin activation as suggested by the sequential multisite binding model (17) and confirmed by EPR (24) and NMR (20) spectroscopic studies. The recent crystal structure of p44 arrestin-1, lacking the arrestin C tail, revealed marked conformational changes compared with a previous structure of p44 arrestin-1 (25), including a 20° rotation between the C and N domains (9) as had previously been suggested based on arrestin truncations (26) and molecular modeling (27). It is yet unclear to what extent this structure represents arrestin in the receptor-bound conformation. However, plotting of our functional data against these conformational changes (Fig. S3) shows how the critical interactions along the C tail (described above) are released and result in reorganization of the three-element interaction and polar core regulatory sites (Fig. 3). Release of the C tail from its basal position and reorganization of the protein increases solvent accessibility of several positively charged residues that reduce binding to phosphorylated rhodopsin upon mutation. Among those are Lys14, Lys15, and Lys110 close to the three-element interaction site. Another one is Lys300 in the gate loop (residues Asp296–Asn305) that reorganizes during breakage of the polar core (9). The weakest binder revealed by our screen, Arg29, is located in the center of this positively charged region. Homology modeling of arrestin-1 on the structure of arrestin-2 containing the vasopressin receptor phosphopeptide V2Rpp (10) confirms this region as a direct interaction site for phosphorylated rhodopsin residues along the rhodopsin C terminus (Fig. S4). In the inactive conformation, the charged Arg29 side chain is exposed to the solvent but also interacts with Phe380, the phenylalanine that anchors the C tail between the polar core and the three-element regulatory site. This special location, together with the strong effect of Arg29 mutation on P-ROS* binding, support a phosphosensing mechanism in which Arg29 controls a switch between the arrestin-1 C tail for the phosphorylated C terminus of rhodopsin.

After arrestin-1 recognizes the phosphorylated C terminus of the receptor, the sequential multisite binding model predicts that arrestin detects the active receptor state (17). Suggested recognition sites in arrestin include the middle loop (Gln133–Ser142) (24, 28, 29), the finger loop (Gly68–Leu77) (24, 30, 31), and the lariat loop (Asn246–Tyr254) (11), all located in the central region between the C and N domains. Within our scan, mutations in the middle loop had only a marginal effect, whereas five mutations in the finger loop (Gln69, Asp73, Val74, Met75, and Leu77) and five in the lariat loop (Leu249, Tyr250, Ser251, Ser252, and Tyr254) diminished binding to light-activated rhodopsin (Fig. 2). Interestingly, only the middle and finger loops undergo large conformational changes during activation, whereas those in the lariat loop are much less pronounced (Fig. S3). Release of the C tail however exposes the lariat loop so that both finger and lariat loop are available as the major interaction site with the activated receptor.

The binding stoichiometry between GPCRs and arrestins remains to be fully established. In the visual system, arrestin-1 saturates phosphorylated light-activated rhodopsin at a 1:1 molar ratio (32) and binds monomeric rhodopsin reconstituted into nanodiscs with high affinity (19, 33). Our mutagenesis data show that both arrestin grooves do not bind rhodopsin to the same extent (Fig. 1) with a clear preference for the N domain containing the phosphosensing region and the lariat and finger loops in the central region between the two domains. To study whether these interaction surfaces could be satisfied simultaneously in a 1:1 arrestin–receptor complex we modeled the C terminus of rhodopsin along the position of the V2Rpp peptide into arrestin-1 and performed molecular docking studies with the structure of light-activated rhodopsin (4) (Fig. 4). As initial constraints for the model of the complex, we placed H8 in a position where it could join the modeled rhodopsin C terminus and close to the arrestin-1 phosphosensor. It should be noted that the C terminus of the V2 receptor is four residues shorter compared with rhodopsin (Figs. S4 and S5) and differs in the relative position of phosphorylatable Thr and Ser residues. Rhodopsin, for example, lacks a phosphate that could interact with Gln69 in the arrestin finger loop (Fig. 3). Relative positions of phosphorylatable residues may thus be involved in determining receptor specificity.

As a second anchoring point, beside the phosphorylated C terminus, we placed the finger loop into the central crevice that opens upon activation of rhodopsin and mediates interaction with the C terminus of the Gα subunit (34). This initial pose was optimized by multiple runs of molecular docking using Rosetta (35). Among the best-scoring hits we chose the model in best agreement with the experimental binding map. This experimentally guided model of the arrestin–rhodopsin complex shows the arrestin finger loop (Fig. 4, Right Top Inset) near the cytoplasmic side of TM4 and in close contact with the intracellular loop 1 connecting TM1 and TM2. On the other side of the finger loop, the tip of the lariat loop interacts with the cytoplasmic ends of TM6 and TM7–H8 in rhodopsin (Fig. 4, Right Middle Inset), whose relative position has been implicated in arrestin-biased agonism in β-adrenergic receptors (36) and arrestin binding to rhodopsin (37). Also, the introduction of a salt-bridge between TM3 and TM7 by the night blindness causing rhodopsin mutant G90D reduces its ability to bind arrestin (38, 39). Together, the finger and lariat loop could thus create an activity sensor that specifically interacts with helices critical for GPCR activation.

Although this binding mode explains the major interaction sites in the N domain, it does not account for two stretches of residues at the edge of the C domain (Trp194–Ser199 and Gly337–Thr345 of the C loop), whose mutation strongly reduces binding. This region is part of the arrestin dimerization interface with mutation of Phe197 directly interfering with arrestin oligomerization (40, 41). However, arrestin binds rhodopsin as a monomer and the high self-association constant of arrestin (K_Dtet = 7.5µM) (40) suggests that arrestin oligomerization is marginal at the assayed conditions. Interference with oligomerization is thus an unlikely reason for the lower binding upon mutagenesis of this region. Interestingly preactivation of arrestin leads to reorganization of the C edge into a continuous interface (Fig. 4, Right Bottom Inset and Figs. S6 and S7). Principally, the reorganized interface could thus provide a secondary binding site for the accommodation of a GPCR dimer. Indeed it has been suggested that the arrestin C loop binds a second rhodopsin molecule in disc membranes containing high levels of activated rhodopsin (18, 42). Site-directed spin-labeling studies, on the other hand, revealed a high flexibility of this arrestin region upon binding to both highly activated P-ROS* and nanodiscs containing only monomeric P-rhodopsin* (28). In absence of a second receptor, the mostly hydrophobic C edge could thus interface with the phospholipid membrane in agreement with the strong lipid dependence of arrestin–rhodopsin complex formation (19, 33, 43). Lys232 and Gln195 close to the C edge may interact with acidic phospholipids known to facilitate rhodopsin binding (43). As a third possibility, beside a rhodopsin dimer or interaction of the C edge with the phospholipid membrane, both arrestin and rhodopsin will undergo conformational changes that cannot be anticipated from the currently available crystal structures. Such conformational changes are in agreement with the structure of the β2-adrenergic receptor–G protein complex (6) which, in comparison with metarhodopsin-II (4, 5) and an active nanobody stabilized β2-adrenergic receptor state (44), showed a large outward movement of TM6 upon G protein binding. In addition to rearrangements in the receptor, solution NMR suggested a global transition of arrestin-1 upon binding to rhodopsin, due to the adaptation of a dynamic molten globule-like structure (20). Conformational changes in both arrestin and rhodopsin upon complex formation may thus explain the large distance between the major interaction surfaces revealed by our scanning mutagenesis.

Using high-throughput mutagenesis and binding assays we have compared 403 mutants covering the complete arrestin-1 sequence. This substantial effort provides a single amino acid resolution map of the residues involved in rhodopsin binding that will be of particular importance once a crystal structure of the arrestin–rhodopsin complex becomes available. Systematic mutagenesis has been tremendously successful for the conformational thermostabilization and structure determination of GPCRs (14). The methods described here provide the means to use a similar strategy for the crystallographic structure determination of the arrestin–rhodopsin complex. The combination of mutants will facilitate engineering of arrestin-1 with specific functional characteristics (45, 46). Transfer of mutations to arrestin-2 and arrestin-3 sets the stage to develop diagnostic and therapeutic tools to study diseases caused by hyperactivity of other GPCRs.

ROS were prepared from bovine retinas under dim red light as described (47). Rhodopsin was phosphorylated, essentially as described (48), yielding a mixture of rhodopsin species containing any number of phosphates up to seven phosphate groups per molecule with three phosphates sufficient for high-affinity rhodopsin binding (49).

Bovine arrestin-1 was cloned into the EgWoMiPi vector for expression in bacterial and mammalian cells (Fig. S8). Mutations were introduced by PCR assisted by the AAscan program suite (16) available from our Web site, www.psi.ch/lbr/aascan. Scanning mutagenesis is further described in SI Materials and Methods.

Arrestin mutants were expressed in E. coli in sets of 12, each including wild type as reference for relative expression levels. Per day, ROS-P* binding of mutants in 3 sets was compared using a 96-well centrifugal, pull-down assay (Fig. S1). The data were fitted to sigmoidal dose–response curves with variable slope to extract IC₅₀ values. IC₅₀ and R² values, 95% confidence intervals, and the number of measurements are listed in Table S1.

Peptide-bound bovine arrestin-1 was modeled with Modeller (50) using the structure of rat arrestin-2 bound to a GPCR phosphopeptide (10) as a template. Next we docked the model of peptide-bound arrestin-1 to the structure of light-activated rhodopsin (Protein Data Bank ID code 4A4M) (4) guided by the scanning mutagenesis data. Energy minimization was used to optimize the geometry of the side chains and the residue–residue interactions in the protein interface. Additional details are described in SI Materials and Methods.