Atomic Structure of GRK5 Reveals Distinct Structural Features Novel for G Protein-coupled Receptor Kinases*♦

Background: GRK5 is implicated in several human pathologies, but relatively little is known about its structure and function. Results: High resolution crystal structures of human GRK5 with AMP-PNP and sangivamycin were determined. Conclusion: GRK5 is found in a partially closed state with its kinase domain C-tail forming novel interactions with nucleotide and the N-lobe. Significance: The GRK5 structure provides important insight into its function. G protein-coupled receptor kinases (GRKs) are members of the protein kinase A, G, and C families (AGC) and play a central role in mediating G protein-coupled receptor phosphorylation and desensitization. One member of the family, GRK5, has been implicated in several human pathologies, including heart failure, hypertension, cancer, diabetes, and Alzheimer disease. To gain mechanistic insight into GRK5 function, we determined a crystal structure of full-length human GRK5 at 1.8 Å resolution. GRK5 in complex with the ATP analog 5′-adenylyl β,γ-imidodiphosphate or the nucleoside sangivamycin crystallized as a monomer. The C-terminal tail (C-tail) of AGC kinase domains is a highly conserved feature that is divided into three segments as follows: the C-lobe tether, the active-site tether (AST), and the N-lobe tether (NLT). This domain is fully resolved in GRK5 and reveals novel interactions with the nucleotide and N-lobe. Similar to other AGC kinases, the GRK5 AST is an integral part of the nucleotide-binding pocket, a feature not observed in other GRKs. The AST also mediates contact between the kinase N- and C-lobes facilitating closure of the kinase domain. The GRK5 NLT is largely displaced from its previously observed position in other GRKs. Moreover, although the autophosphorylation sites in the NLT are >20 Å away from the catalytic cleft, they are capable of rapid cis-autophosphorylation suggesting high mobility of this region. In summary, we provide a snapshot of GRK5 in a partially closed state, where structural elements of the kinase domain C-tail are aligned to form novel interactions to the nucleotide and N-lobe not previously observed in other GRKs.

The G protein-coupled receptor kinase (GRK) 3 family dates back to the discovery of rhodopsin kinase (GRK1) and now includes seven members that fall into the following three subfamilies: GRK1 (GRK1 and -7), GRK2 (GRK2 and -3), and GRK4 (GRK4 -6) (1-3). GRKs are best known for their ability to phosphorylate activated G protein-coupled receptors (GPCRs) and promote arrestin binding, a process that drives receptor desensitization and trafficking as well as arrestin-mediated signaling (1)(2)(3). GRKs also phosphorylate many other proteins and can mediate kinase-independent regulation of receptor and effector function (2). Although most GRKs are localized in the cytoplasm or at the plasma membrane, GRK5 and GRK6 are also found in the nucleus with GRK5 localization being regulated by G q signaling (4,5) and GRK6 localization regulated by palmitoylation (6). These findings suggest many novel ways whereby GRKs may mediate their biological effects.
GRKs have been implicated in many pathological conditions, including cardiovascular disease, cancer, and various neurological and metabolic disorders (2,3). This has been best studied in cardiovascular disease where increased GRK expression is associated with congestive heart failure, hypertension, and myocardial ischemia. GRK2 appears to play a particularly important role in this process and is overexpressed in human heart failure and hypertension as well as in experimentally induced heart failure models (7)(8)(9). GRK5 is also linked with cardiovascular disease and contributes to cardiac hypertrophy (10) and hypertension (11). Moreover, a Q41L polymorphism within the regulator of G protein signaling homology (RH) domain in GRK5 is prevalent in African Americans and has an enhanced ability to desensitize the ␤ 2 -adrenergic receptor (␤ 2 AR) (12) and protect against the development of congestive heart failure (13). GRK5 has also been linked with type 2 diabetes (14), prostate tumor growth (15) and metastasis (16), Parkinson disease with dementia (17), synuclein phosphorylation and aggregation in sporadic Parkinson disease (18), and Alzheimer pathology in mice and humans (19), possibly due to its role in neurite outgrowth, learning, and memory (20). Taken together, these studies reveal that GRK5 is implicated in many different human diseases and may provide an important therapeutic target in the treatment of cardiovascular disease, neurological and metabolic disorders, and cancer.
Although it is important to understand the in vivo function of GRKs, it is also important to understand the mechanisms that mediate GRK function. Significant insight into GRK function has come from detailed biochemical and biophysical analysis. This has included defining important functional domains, including detailed mapping of interfaces such as GRK2 interaction with G␣ q (21) and G␤␥ (22) and GRK5 interaction with Ca 2ϩ /calmodulin (23). Such studies have been greatly complemented by x-ray crystallography, and structures for GRK1 (24), GRK2 (25), GRK2/G␤ 1 ␥ 2 (25,26), GRK2/G␤ 1 ␥ 2 /G␣ q (27), and GRK6 (28) have provided important insight. These structures reveal that the RH, catalytic, and C-terminal domains are in extensive contact and help to hold the kinase domain in an inactive open conformation. Several studies suggest that an N-terminal ␣-helical domain plays an important role in regulating catalytic domain closure via a process that is likely regulated by receptor binding (29 -31). Indeed, this domain has been observed in one crystal form of GRK1 (24,31) as well as in GRK6 in complex with either sangivamycin or AMP (30) and appears to stabilize catalytic domain closure. Although crystallography has provided significant insight on GRK1, -2, and -6, there have been no studies to date on the structural analysis of GRK5.
Here, we use x-ray crystallography to better understand the structure and function of GRK5. We find that human GRK5 shares significant structural similarity to other GRKs, albeit with some unique aspects. In contrast to crystal structures of GRK1 and GRK6, GRK5 is a monomer in the crystal and contains a well ordered kinase domain C-tail that includes interactions between the active-site tether and bound nucleotide. Such studies have high significance for understanding GRKs and how structure/function changes might mediate pathological consequences. Moreover, these studies should set the stage for developing strategies to specifically regulate GRK5 function in the treatment of disease.

Experimental Procedures
Materials-Sangivamycin, ADP, AMP-PNP, and phosphatidylcholine (soybean type II-S) were purchased from Sigma, and [␥-32 P]ATP was from PerkinElmer Life Sciences. Crystallization screening reagents were purchased from Hampton Research (Aliso Viejo, CA).
Protein Expression and Purification-The site-specific mutants GRK5-K215R and GRK5-R470A were generated using the QuikChange site-directed mutagenesis kit (Agilent) and verified by DNA sequencing. The Bac-to-Bac Baculovirus Expression System (Invitrogen) was used to generate baculoviruses to express human full-length wild-type GRK5, GRK5-R470A, and kinase-dead GRK5-K215R in Sf9 insect cells. The recombinant proteins were purified as described previously with some modifications (32). Briefly, cells were harvested by low speed centrifugation 48 h after infection, and the cell pellet was resuspended in lysis buffer (20 mM Hepes, pH 7.2, 250 mM NaCl, 5 mM EDTA, 1 mM phenylmethylsulfonyl fluoride, 10 M leupeptin, 3 mM benzamidine, and 0.02% Triton X-100). Cells were lysed using a Brinkman Polytron (two times for 30 s at 25,000 rpm) followed by high speed centrifugation to remove cell debris. The supernatant was diluted 4-fold with buffer A (20 mM Hepes, pH 7.2, 2 mM EDTA, 0.02% Triton X-100, 1 mM dithiothreitol (DTT)) and applied to an SP-Sepharose cationexchange column. The column was washed and eluted with a 50 -750 mM linear NaCl gradient in buffer A. GRK5 peak fractions were pooled, diluted with buffer A, loaded onto a heparin-Sepharose 6 FF affinity column, and eluted with a 200 -800 mM linear NaCl gradient in buffer A. GRK5-containing fractions were pooled, diluted with buffer B (20 mM Hepes, pH 7.2, 1 mM DTT), and injected onto a 1-ml Mono S cation-exchange FPLC column. Proteins were eluted with a 200 -700 mM NaCl linear gradient in buffer B, and fractions containing GRK5 were combined, diluted to 200 mM NaCl with buffer B, concentrated in a 50-kDa molecular mass cutoff filter (Millipore Corp.) to ϳ20 mg/ml, aliquoted, and stored at Ϫ80°C. The yield of pure GRK5 was 2-5 mg/liter of Sf9 cells with protein purity Ͼ95% as analyzed by SDS-PAGE.
Crystallization-Prior to crystallization, 8 mg/ml GRK5 (ϳ120 M) was incubated for 2 h with 4 mM AMP-PNP and 2 mM MgCl 2 (GRK5⅐AMP-PNP complex) or 0.4 mM sangivamycin and 0.2 mM MgCl 2 (GRK5⅐sangivamycin complex). Initial robotic screening for crystallization conditions was performed in a 96-well format using crystallization kits from Hampton Research. Protein droplets were prepared by mixing 0.4 l of protein sample with 0.4 l of reservoir solution and equilibrating against 60 l of reservoir solution as a sitting drop at 4°C. Several conditions under which crystals appeared were further optimized using the hanging-drop vapor-diffusion method and increasing the droplet size and reservoir volume by 5-8-fold, varying the concentration of precipitants and salts at different pH values. The complexes crystallized under similar conditions (AMP-PNP: 0.2 M NaCl, 0.1 M BisTris, pH 6.5, 20% w/v PEG 3350; sangivamycin: 0.15 M NaCl, 0.1 M BisTris, pH 5.5, 21% w/v PEG 3350). The best single diffracting crystals were obtained by micro-seeding. GRK5 crystals appeared within 3 days and grew over the course of 2-3 weeks with tetragonal bipyramidal morphology. SDS-PAGE analysis of a dissolved crystal revealed fulllength GRK5. Crystals were cryoprotected using the reservoir solution supplemented to 35% PEG 3350 and flash-frozen in liquid nitrogen.
Data Collection, Structure Determination, and Refinement-Diffraction data for crystals were collected at beamline X6A at the Brookhaven National Synchrotron Light Source (NSLS) on an Area Detector System Corp. Quantum-270 detector using an x-ray wavelength of ϳ1.0 Å and 0.2°oscillations under a constant stream of liquid nitrogen maintained at 100 K. The best crystals for GRK5⅐AMP-PNP and GRK5⅐sangivamycin complexes diffracted to 1.8 and 2.1 Å, respectively. Diffraction data for both datasets were processed and scaled to primitive tetragonal space group P4 3 2 1 2 using HKL2000 suite (Table 1) (33). The ligand-bound structures of GRK5 were solved by molecular replacement with Phaser (34) using two fragments of the GRK6 crystal structure (PDB code 2ACX) as the search model, corresponding to residues 30 -179 and 180 -507 (kinase domain). One copy of the GRK5 molecule was located in the asymmetric unit, which results in an estimated solvent content of ϳ43%. The model was subjected to iterative cycles of positional refinement and isotropic B-factor refinement using five TLS groups in Phenix.refine (35) as well as manual model building using Coot (36). To ascertain N-terminal residues 15-23 and a few loop regions of the model, omit-maps and featureenhanced maps were calculated using PHENIX. Mg 2ϩ ion, AMP-PNP, and sangivamycin were refined into their corresponding electron density as identified by difference maps. The final refined models for GRK5⅐AMP-PNP and GRK5⅐ sangivamycin have R work /R free of 17.2/20.4 and 16.8/23.1%, respectively ( Table 1). The R free set was calculated using randomly selected reflections over 14 thin resolution shells. Mol-Probity (37) evaluation of main chain geometry suggests that Ͼ98.3% of residues occupy favored regions of the Ramachandran plot, with no outliers. The validity of refined structure models to the unmerged diffraction data was assessed using the correlation coefficient of half-datasets, CC*, using PHENIX. Also, CC*, CC work , and CC free were calculated for the final deposited structures in PHENIX and suggest that the structures were not over-fitted (Table 1) (38). Coordinates were deposited in RCSB structure database with PDB accession codes 4TND and 4TNB for GRK5⅐AMP-PNP and GRK5⅐sangivamycin complexes, respectively.
Structure Analysis-Structural superpositions were performed using the secondary structure matching algorithm in Coot (36,39) and PDBeFold server (39) along with r.m.s.d. calculations and distance deviation analysis between regions. Domain hinge analysis was carried out using program DynDoM (40). All ribbon diagrams and surface representations in this work were generated using PyMOL (41). Nonlinear Poisson-Boltzmann electrostatic calculations were performed using APBS Tools plugin in PyMOL (42). Angles among superimposed regions were calculated using phenix.angle script of PHENIX 1.9 -1692 (43). GRK5 intramolecular interactions and interactions with ligand were mapped using Ligplot plus (44) and PDBePISA server (80).
Analytical Ultracentrifugation-GRK5 was diluted in 20 mM Hepes, pH 7.2, 200 mM NaCl, 1 mM DTT to a concentration of 15 M and subjected to analytical ultracentrifugation analysis in a Beckman ProteomeLab XL-I analytical ultracentrifuge in sedimentation velocity mode. GRK5 (400 l) and reference buffer (420 l) were loaded into a two-sector 1.2-cm path length Epon centerpiece sample cell. The runs were performed at fixed speed of 50,000 rpm Ϫ1 at a constant temperature of 4°C, and absorbance values were collected at a fixed wavelength of 276 nm. The partial specific volume of GRK5 (v ), solvent density, and relative viscosity values in buffer solution along with theoretical molecular mass were calculated using SEDNTERP Version 1.09 (45). The resulting data were fitted to a continuous c(s) distribution model in program SEDFIT (46) and an estimated molecular mass was obtained. The fitted data were visualized and presented using program GUSSI (University of Texas Southwestern).
Circular Dichroism-Thermal unfolding of GRK5 and ligand-bound complexes was monitored by recording variations in ellipticity at 223 nm as a function of temperature in 1.0°C increments from 20 to 70°C using a Jasco J-810 spectropolarimeter equipped with a Peltier temperature control system. GRK5 at a final concentration of 5 M was measured using a 0.1 cm quartz cuvette (Starna Cells, Inc.) in phosphate-buffered saline solution (3.2 mM Na 2 HPO 4 , 0.5 mM KH 2 PO 4 , pH 7.4, 135 mM NaCl), 0.4 mM ligand, 2 mM MgCl 2 , and 0.4% dimethyl sulfoxide.
Michaelis-Menten Kinetic Analysis-K m , V max , and k cat values for ATP were determined by varying ATP concentrations from 2 to 200 M in reactions containing 10 M rhodopsin (bleached for 5 min) and either 50 nM WT GRK5 or GRK5-R470A in 20 mM Tris-HCl, pH 7.4, 5 mM MgCl 2 , 30 mM NaCl, 0.5 mM EDTA, and 4000 cpm/pmol [␥-32 P]ATP. Reactions proceeded for 2 min at 30°C in room light and were stopped with SDS sample buffer. Rhodopsin phosphorylation was quantified by measurement of 32 P incorporation into rhodopsin after SDS-PAGE as described previously (32). Reaction velocities at the various ATP concentrations were fit to the Michaelis-Menten equation using GraphPad Prism.

Results and Discussion
Crystal Structure of GRK5-Previous studies suggest that the crystallization of ligand-free GRKs is very difficult because only an ϳ8 Å crystal structure of apo-GRK1 has been reported (24). Thus, we initially focused on using nucleotides to stabilize GRK5 and facilitate crystallization. Circular dichroism was used to investigate the structural stability of apo-and nucleotide-bound GRK5. Thermal unfolding of ligand-free GRK5 yielded an apparent melting temperature (T m ) of 31.5°C, although the addition of nucleotides such as ADP, AMP-PNP, and sangivamycin resulted in an increase in T m , consistent with stabilization ( Fig. 1A). Sangivamycin, a potent nucleoside inhibitor of protein kinases (47), increased the T m by 5.5°C (T m ϭ 37°C), ADP by 2.5°C (T m ϭ 34°C), AMP-PNP by 1.5°C (T m ϭ 33°C), and AMP had no effect (T m ϭ 31.5°C). Thus, sangivamycin bind-ing induces significant stabilization of GRK5, whereas the effect of other nucleotides was less prominent.
Screening full-length human GRK5⅐AMP-PNP and GRK5⅐ sangivamycin complexes led to diffracting crystals of tetragonal bipyramidal morphology (Fig. 1B). The data collection and refinement statistics are listed in Table 1. Structures of AMP-PNP and sangivamycin-bound GRK5 were solved to 1.8 and 2.1 Å resolution, respectively, and refined to R work /R free values of 17.2/20.4% (GRK5⅐AMP-PNP) and 16.8/23.1% (GRK5⅐sangiva-mycin) at given resolution. Both ligand-bound complexes were crystallized with nearly identical unit cell dimensions, crystal packing arrangement, and space group. The domain organization of GRK5 includes N-terminal ␣-helical, calmodulin/PIP 2 binding and RH domains, a central kinase domain, and a C-terminal region that includes a portion of the RH domain and a second calmodulin/PIP 2 binding domain (Fig. 1C). The crystal structures revealed a refined model lacking 14 N-terminal and 47 C-terminal residues with residues 15-543 clearly resolved, FIGURE 1. Atomic structure of GRK5. A, thermostabilization of GRK5 induced by different ligands. Stability of GRK5 against thermal denaturation was monitored by circular dichroism measurement of changes in the ellipticity intensity at 223 nm as a function of temperature. The measurements were performed at 5 M GRK5 in the absence (apo-GRK5) or presence of 0.4 mM ligand. B, GRK5⅐AMP-PNP⅐Mg 2ϩ crystals were grown at 4°C to dimensions 0.04 ϫ 0.04 ϫ 0.12 mm. The crystals show bipyramidal morphology in contrast to rod-like thin plate crystals of its close homolog GRK6 (28). C, domain organization and important functional regions of GRK5. The RH domain is shown in blue, the catalytic domain in yellow (N-lobe) and orange (C-lobe), and the C-tail of kinase domain in purple. The two clusters of positively charged residues at the N and C terminus encompass GRK5 CaM-and PIP 2 -binding sites (magenta). N-terminal amphipathic ␣-helix is in green. Autophosphorylation sites on the C-tail of the kinase domain are indicated by yellow circles. D, ribbon representation of GRK5⅐AMP-PNP crystal structure. Full-length GRK5(1-590) was crystallized, and residues 15-543 are clearly resolved (the first and last residues are labeled). The color coding is the same as in C. AMP-PNP ligand and Mg 2ϩ are shown as sticks and black sphere, respectively.

Atomic Structure of GRK5
including the kinase domain C-tail, which is not visible in most GRK structures (Fig. 1D). Mass spectroscopic analysis reveals that the GRK5 used in crystallography is intact and lacks any post-translational modifications except for some very minor phosphorylation at the extreme C terminus (data not shown). This reveals that the N-and C-terminal regions missing from the structure are disordered and suggests high flexibility of these regions, a property common to most GRK crystal structures.
Overall, the crystal structures of AMP-PNP and sangivamycin-bound GRK5 were nearly identical (r.m.s.d. 0.35 Å), with minor changes in the structural arrangement of the kinase-active site (discussed below). This is in strict contrast to GRK6 (70.4% sequence identity to GRK5), which adopts distinct conformations in the presence of AMP-PNP (28) and sangivamycin (r.m.s.d. 1.94 Å) (30). The observed difference between GRK5 and GRK6 is likely dictated by structural restraints imposed by crystal contacts, which appeared to be ligand-independent in

PDB code 4TND 4TNB
a Statistics for the highest resolution shell are shown in parentheses. b The simple merging R factor for the multiple observations is shown (75,76). c Redundancy-independent merging R factor is shown (77). d Precision-indicating merging R factor is shown (78). e The CC1 ⁄ 2 is the correlation coefficient between two randomly selected half-datasets; CC* is a statistic metric for assessing the effective resolution limits of data and quality of unmerged data in the context of a refined model; CC work and CC free are the standard and cross-validated correlations of the observed intensities to the refined modelbased intensities, for the work and test sets, respectively (38). f R free value is calculated using the small subset of randomly selected reflections (test-set) that are set aside prior to refinement and not used in the refinement of the structural model (79). g Data are from Chen et al. (37).

JOURNAL OF BIOLOGICAL CHEMISTRY 20633
the GRK5 complexes with AMP-PNP and sangivamycin, although the GRK6 complexes crystallized in two distinct packing arrangements and space groups. However, the minor differences between the two crystal structures of GRK5 may not reflect the propensity of the kinase to adopt distinct conformations in solution, because crystal structures often trap the lowest energy and most stable protein conformation. The absence of structural restraints imposed by crystal lattice contacts provides the necessary dynamics in the GRK5 molecule in solution resulting in significant differences in GRK5 stabilization induced by different ligands (Fig. 1A). However, because both structures show a high degree of structural identity and likely trap the most stable conformation of GRK5, we focused our analysis and discussion on the higher resolution GRK5 structure co-crystallized with the nonhydrolysable ATP analog AMP-PNP.
The GRK5 crystal structure contains features typical for protein kinases in general and GRKs in particular. The kinase domain possesses a bilobal-fold with a smaller N-terminal lobe (N-lobe, yellow) and a larger C-terminal lobe (C-lobe, orange) (Fig. 1D). The N-lobe consists of a canonical five-strand antiparallel ␤-sheet (denoted ␤1-␤5) forming characteristic groove and flanking helices ␣B and ␣C. The GRK5 C-lobe is composed of a two-strand antiparallel ␤-sheet (␤6-␤7), located under the interlobe hinge region, and six ␣-helices (␣D-␣I). Fully ordered AMP-PNP is coordinated within the kinase interlobe cleft, representing the kinase-active site. Regulatory C-terminal extension of the kinase domain (Fig. 1D, purple) is fully ordered in GRK5 and is found in a position near the RH domain that has not been observed previously in other GRKs.
The bilobal architecture is also a characteristic feature of GRK RH domains that are structurally related with the regulator of G protein-signaling proteins (48). One lobe is composed of helices ␣4 -␣7 packed in an antiparallel manner to assemble the GRK5 RH bundle subdomain (Fig. 1D). The GRK5 RH terminal subdomain (second lobe) is by N-terminal helices ␣0 -␣3, ␣8 -␣9, and C-terminal ␣10 -␣11. The RH bundle and terminal subdomains generate a characteristic V-shaped notch with two ends forming a contact surface with adjacent kinase domain N-and C-lobes. The ␣10-helix of the RH terminal subdomain has a large interface with the kinase N-lobe, whereas hydrophilic residues in the N-terminal ends of ␣4 (backbone oxygen of Val 92 ) and ␣7 (Lys 139 and Glu 140 ) interact with the kinase domain C-lobe via polar contacts with Asn 452 , Lys 454 , and Arg 455 of the ␣J helix (at the beginning of the kinase domain C-tail). Thus, the RH and kinase domains of GRK5 are in intimate contact and establish a close relationship between the two domains that provides a structural basis for allosteric effects (e.g. receptor binding and kinase domain closure).
No domain-swapped interface, as is observed in prior structures of GRK1 and GRK6, was observed in GRK5, and both ligand complexes crystallized as a monomer in the asymmetric unit. The C-terminal calmodulin-and PIP 2 -binding domain (residues 546 -565) (23,49,50) is located within the disordered C-terminal segment, immediately adjacent to the last residue observed in the structure (Pro 543 ). Thus, this domain appears to be located near the ␣1-helix of the RH domain and the N-terminal calmodulin/PIP 2 -binding region (residues 20 -39, col-ored magenta in Fig. 1, C and D) (23,51). This suggests close proximity of these two important domains that mediate GRK5 interaction with phospholipids and calmodulin.
A noticeable divergence in the crystal structures of GRK5 and other GRKs is also observed in the relative position of the RH and kinase domains. The long axis of the RH and kinase domain of GRK5 is twisted compared with GRK1, GRK2, and GRK6 (Fig. 2). DynDoM analysis reveals that the hinge for this motion is located within the interdomain loop of GRK5 that connects the ␣9-helix of the RH domain and the ␤1-sheet of the kinase domain (Fig. 1D). When the kinase domains are superimposed, the RH bundle subdomain demonstrates larger displacement than the RH terminal subdomain, consistent with the position of the RH bundle subdomain far distant of the hinge. It moves by ϳ10 Å on average (␣6 helix) relative to its position in other GRKs (the kinase domains are superimposed in Fig. 2). The significance of this observed movement of the two major domains of GRK5 is not clear, because it mainly affects the relative arrangement of the RH bundle and kinase C-lobe domains, regions that are remote from the predicted membrane-and receptor-binding surfaces. Moreover, no significant rearrangement of the contact interface between the RH and kinase domains of GRK5 was identified, and the individual structures of the RH subdomains were also changed very little (r.m.s.d. 0.4 -0.9 and 1.4 -1.6 Å for terminal and bundle, respectively, compared with GRK1 and GRK6). This movement is also unlikely to be related to the crystallization of GRK5 as a monomer, because a GRK1-L166K mutant that crystallized in a monomeric form lacks this structural feature (52). Interestingly, the 2.4 Å crystal structure of bovine GRK5 in complex with the GRK-selective inhibitor CCCG215022 (see accompanying paper (81)) also does not have the RH and catalytic domain rotation observed in our GRK5⅐AMP-PNP and GRK5⅐sangivamycin structures. This suggests some degree of structural plasticity in the relative orientation of these two major domains of GRK5.
GRK5 Is a Monomer Both in the Crystal Structure and in Solution-GRK1 and GRK6 crystallized as dimers in the asymmetric unit with the C termini of each molecule mediating a domain-swapped dimer interface (24,28,30). In contrast, GRK5 is a monomer in the crystal structure. Sequence alignment identified a unique Pro 529 in the C-terminal region of GRK5 that is not present in other GRKs (Fig. 3A). Pro 529 causes a kink in the polypeptide chain of the C terminus and bends it back upon itself. Thus, the direction of the visible segment of the C terminus (the fragment between Asn 530 and the last visible residue Pro 543 ) is changed preventing antiparallel alignment of the C termini, which were found to stabilize the dimer interface in other GRKs. Arg 169 is at a key position to support this architecture and packs against the C-terminal portion of GRK5 with its guanidinium group forming hydrogen bonds with the side chain of Asp 536 , Asn 525 , and the backbone oxygen of Val 526 . In addition, residues Met 165 , Phe 166 , and Arg 169 form a hydrophobic pocket to sandwich Phe 527 instead of mediating a dimer interface as observed in GRK6 (28). A similar packing arrangement was found in GRK1-L166K, where dimerization was disrupted by mutation of Leu 166 in the hydrophobic dimer interface resulting in GRK1 crystallization as a monomer (52).
This confirms that this topology of the C terminus promotes crystallization of a monomeric form of GRKs.
To determine whether GRK5 also exists as a monomer in solution, we performed size-exclusion chromatography (Fig.  3B) and analytical ultracentrifugation (AUC) (Fig. 3C). These experiments revealed molecular masses of ϳ70 kDa by sizeexclusion chromatography and ϳ62.4 kDa by AUC under sedimentation velocity mode, consistent with a monomeric state, and no dimer or aggregate forms were detected. There is also little evidence that other GRKs form dimers in solution, although some contain a domain-swapped dimer interface in the crystal structures (28). Although the crystal structure and in-solution analysis of GRK5 does not exclude dynamic selfassociation of GRK5 under some conditions, it demonstrates that the C-terminal domain-swapped dimer interface typical in other GRKs is unlikely to be formed in GRK5.
Although our studies suggest that GRK5 is primarily a monomer in solution, there are a few recent reports that GRK5 can also dimerize. For example, a recent study on the role of GRK5 dimerization in plasma membrane localization demonstrated that a M165E/F166E mutation disrupts GRK5 dimerization and results in GRK5 redistribution from the plasma membrane to the cytoplasm (53). Although Met 165 and Phe 166 mediate a hydrophobic dimer interface in GRK6 (28), they are involved in intramolecular stabilization with Phe 527 in GRK5 (Fig. 3A). Although destabilization of this intramolecular interface might result in a change of membrane-binding properties of GRK5, the Xu et al. (53) study provides several additional lines of evi-dence supporting a role for GRK5 dimerization in plasma membrane binding. In addition, a role for GRK5 dimerization was also suggested in actin nucleation (20). Moreover, we find that GRK5 can undergo intermolecular autophosphorylation in the presence of phospholipids (see Fig. 6B), also supporting its ability to dimerize under certain conditions. Taken together, these studies suggest that GRK5 dimerization may be linked to phospholipid binding.
Kinase Domain of GRK5 Adopts a Partially Closed Conformation and Binds Ligands in the Canonical ATP-binding Pocket-In AGC kinases, the small and large lobes of the kinase domain are observed to "close" upon binding of nucleotides and substrate peptides, phosphorylation of the activation loop or, in PKA, complex formation between catalytic and regulatory subunits (54,55). It is believed that the fully closed conformation of the kinase domain allows proper alignment of substrate and nucleotide in the active site needed for effective catalysis (56).
Superposition of GRK5 with its close homolog GRK6 and the prototypical AGC kinase PKA reveals an intermediate partially closed conformation of the GRK5 kinase domain (structural coordinates of kinase domain C-lobe were superimposed to observe differences in conformation of N-lobe, indicative of kinase domain closure) (Fig. 4A). GRK5⅐AMP-PNP adopts a more closed conformation compared with the open state of apo-PKA (PDB code 1J3H) and GRK6⅐AMP-PNP (PDB code 2ACX), yet it is somewhat less closed than the fully closed conformation of PKA (catalytic subunit of the holo tetrameric complex; PDB code 3TNP). The catalytic cleft of GRK5 is nota-bly narrower than in apo-PKA and GRK6⅐AMP-PNP, consistent with a more catalytically competent conformation of the kinase domain. Overall, the conformation of the GRK5 kinase domain closely resembles the relatively closed state observed in the GRK6⅐sangivamycin structure (PDB code 3NYN), the most closed structure of a GRK determined to date.
AMP-PNP and sangivamycin both have strong electron density and low B-factors in the refined GRK5 structure. They reside in the canonical ATP-binding pocket in a large cleft formed in the interlobe space of the kinase domain. The purine ring of the ligands is buried within a hydrophobic pocket in the active site. Two additional conserved hydrogen bonds with Thr 264 and Met 266 of the hinge region fulfill binding for the adenine base (Fig. 4B). The triphosphate tail is more solventaccessible than in the fully closed PKA structure (57) due to partial opening of the cleft and rising of the P-loop. The ribose portion of the ligand is partially shielded by a direct contact with the AST of the C-terminal extension of the kinase domain. The sangivamycin-binding site largely overlaps the adenineand ribose-binding sites of AMP-PNP. AMP-PNP retains only one magnesium ion in a tridentate chelation by oxygen atoms of ␣-, ␤-, and ␥-phosphates as well as Asp 329 of the "DLG" motif ("DFG" in PKA/PKB/PKC) and Asn 316 (Fig. 4B), similar to PKA⅐AMP-PNP (57) and PKB⅐AMP-PNP (58) structures, although magnesium ions are coordinated in a bidentate fashion by phosphate groups in this case. Magnesium also binds a catalytic water molecule to fulfill the octahedral coordination state (59). In the GRK6⅐AMP-PNP structure, Asp 329 and Asn 316 are displaced due to the more open conformation of the kinase domain and do not coordinate magnesium, although in the GRK5⅐sangivamycin complex, the side chain of Asp 329 forms hydrogen bonds with the amide group at the C7 position of the diazaindole ring of sangivamycin instead of magnesium coordination.
One characteristic of the protein kinase active state is interaction between a conserved glutamate residue from the ␣C-helix and lysine residue from the ␤3-strand (60). Mutation of this catalytic lysine residue (Lys 215 in GRK5/6, Lys 216 in GRK1, and Lys 220 in GRK2) into arginine inactivates GRKs. Lys 215 is positioned by Glu 234 to bind ␣and ␤-phosphates of the ligand in GRK6⅐AMP-PNP and GRK1⅐ATP (Lys 216 and Glu 235 ). In GRK5⅐AMP-PNP, Lys 215 also forms a hydrogen bond with Glu 234 , but it lacks contact with the triphosphate tail of the ligand (Fig. 4C). The triphosphate tail of AMP-PNP in GRK5 adopts a conformation with the ␣and ␤-phosphates shifted 1.5 and 4 Å away from Lys 215 , respectively, as compared with GRK6⅐AMP-PNP (28) and thus is less buried in the catalytic cleft (Fig. 4C). The crystal structure of GRK5⅐AMP-PNP was obtained under low magnesium concentration and therefore shows electron density for only one magnesium ion. As noted above, this magnesium ion is coordinated in a tridentate arrangement by the triphosphate tail in GRK5, although normally magnesium is coordinated in a bidentate manner by two phosphate groups of ATP or AMP-PNP. It is possible that the binding of a second Mg 2ϩ (at high magnesium concentration) would change the triphosphate tail conformation and help to position it deeper in the active site cleft of GRK5, where it can interact with Lys 215 .
The phosphate-binding glycine-rich loop (P-loop) of the N-lobe is formed by the ␤1-␤2 turn and is directly engaged in binding of the triphosphate tail of ATP when the kinase is in its active state (full closure of kinase domain) and helps to stabilize the phosphorylation transition state (61). Because of incomplete closure of the kinase domain in GRK5, the P-loop is raised a bit above the triphosphate tail and is not fully engaged in phosphate group binding as observed in the fully closed conformation of PKA⅐AMP-PNP (57). P-loop is engaged in coordination of ␣and ␤-phosphates, although it is still misaligned with the ␥-phosphate. However, the P-loop of GRK5 is observed at a lower position when compared with the open conformations of apo-PKA (62) and GRK6⅐AMP-PNP (28). Interestingly, the P-loop of GRK5 seems to be in more intimate contact with the adjacent ␣B-helix of the N-lobe than in other GRK structures. Side chains of Glu 218 , Arg 221 , and Arg 225 residues are situated on the side of ␣B-helix facing toward the P-loop and form a network of hydrogen bonds with the last two being directly bound to the backbone oxygen of Gly 196 from P-loop (Fig. 4D). The adjacent residue Phe 197 of the P-loop packs against the guanidinium group of Arg 225 and is engaged in weak cationinteraction. Overall, these contacts should stabilize the position of the P-loop. Compared with other GRKs, the interface between ␣B-helix and P-loop in GRK5 is stronger due to involvement of both arginines from the ␣B-helix, although in other GRKs the interaction is provided only by one Arg residue. The tight contact between the P-loop and the adjacent ␣B-helix of the N-lobe provides a basis for synergistic changes within these two sides of the kinase catalytic groove.  Fig. 1D, and its C terminus is shown in magenta, and the GRK6 C terminus is in red. The GRK6 C terminus is domain-swapped; its position is fixed by symmetry-related GRK6 molecule from the dimer complex (data not shown), and the GRK5 C terminus is aligned against its own RH domain. Key intramolecular contacts between the GRK5 C terminus and RH domain that support the specific architecture of the C terminus are highlighted in the left box, and the hydrophobic pocket consisting of Met 165 , Phe 166 , and Arg 169 (RH domain) that retains Phe 527 (C terminus) is highlighted in the right box. Sequence alignment of C-terminal regions of human GRKs indicates unique Pro 529 (framed in magenta) that forms a kink, thereby hindering a domain-swapped interface in GRK5. Identical residues are boxed in red, and residues showing similarity are in red and grouped in a blue frame. B, size-exclusion chromatographic analysis of purified GRK5. A Superose 12 16/60 gel filtration column was calibrated using molecular mass markers, whose elution volumes and relative molecular masses were used to build a calibration curve (inset). GRK5 eluted as a single species consistent with an ϳ70-kDa monomeric state of the protein. C, sedimentation-velocity profile of GRK5. The top panel shows the raw absorbance (solid circles) collected at 276 nm with the interval of 40 min together with corresponding fitted data (solid line) plotted as a function of radial position. The low values (Ͻ 5% of signal) for residuals confirm fidelity of fit between raw and fitted data. The monophasic sedimentation boundaries suggest that GRK5 exists as a monodisperse single species in solution. The bottom panel shows the fitted distribution of the apparent sedimentation coefficient calculated for GRK5 is 2.6 S, which corresponds to an estimated molecular mass of ϳ62.4 kDa, consistent with a monomer of 67.8 kDa (theoretical molecular mass). spans the kinase domain C-lobe (C-lobe tether; residues 451-467 in GRK5), active site (AST; residues 468 -473), and N-lobe (N-lobe tether (NLT); residues 474 -502) (Fig. 5A). The C-tail plays a regulatory role in AGC kinases, acting either as a cisinteracting regulatory element with direct influence on kinase activity or as a scaffold for trans-interacting elements that dock on the C-tail and modulate AGC kinase activity (63).

Active-site Tether Is an Integral Part of the Nucleotide-binding Pocket of GRK5-The C-terminal extension of the kinase domain (C-tail) is a conserved feature of all AGC kinases (63). It
The AST segment of the C-tail regulates the catalytic cycle of substrate phosphorylation in AGC kinases and helps to control entry and exit of nucleotide in the kinase-active site (so-called "nucleotide gate") (54). It is typically disordered in the absence of nucleotide but becomes an integral part of the ATP-binding site when nucleotide is bound and the kinase domain is closed (64). A conserved Phe-Asp-Asp-Tyr (FDDY) motif (residues 327 to 330 in PKA) in the AST, particularly Phe 327 and Tyr 330 , is part of the hydrophobic pocket that anchors the adenine and ribose rings of ATP in PKA and many other AGC kinases (Fig.  5B) (63). Replacement of these residues with Ala reduces affinity for ATP and ultimately inactivates the enzyme (65,66). In GRKs, the FDDY motif is not conserved, and the AST fragment does not coordinate nucleotide even when fully ordered as in the GRK6/sangivamycin structure (30). However, we observed a direct contact of the AST segment to the ribose ring of AMP-PNP and sangivamycin in GRK5, which resembles the ASTnucleotide interaction observed in PKA and PKB (Fig. 5B).
The AST spans the active site with the side chain of Arg 470 intruding into the nucleotide-binding pocket of GRK5 to hydrogen bond with the ribose ring of AMP-PNP. Guanidinium group of Arg 470 directly interacts with 3Ј-OH of the AMP-PNP ribose ring, although analogous interaction of Tyr 330 of PKA requires an active-site conserved water molecule to mediate AST interaction with 2Ј-OH of the ribose ring (Fig.  5B). Importantly, Arg 470 also mediates a hydrogen bond network and plays a central role in bridging the C-tail with the N-lobe (Arg 470 backbone oxygen and guanidinium group interact with Arg 190 and Leu 192 of the ␤1-strand, respectively) and the hinge region (Arg 470 backbone amide binds to Asn 267 ) (Fig.  5C). Although GRK5 adopts an intermediate state with the structural elements of the kinase domain not fully aligned for phosphotransfer, the contact interface between Arg 470 and other parts of the kinase domain facilitate its closure. Interestingly, in the GRK6⅐sangivamycin structure, the ordered ␣N-helix and Arg 190 of ␤1-strand were at the center of the interface stabilizing the closed state of the kinase domain (30). We speculate that Arg 470 has a similar role in GRK5, which also adopts a partially closed conformation even in the absence of ␣N-helix ordering. Furthermore, Arg 470 seems to have an even more prominent role in the GRK5⅐sangivamycin complex compared with GRK5⅐AMP-PNP, because its side chain is bent toward the C-lobe, and the guanidinium group also hydrogen bonds with Asp 270 of C-lobe (Fig. 5C), in addition to the observed interaction to the ribose ring in the AMP-PNP complex. Thus, tighter contact between N-and C-lobes provided by AST in GRK5⅐sangivamycin might better stabilize the closed state of the kinase domain, which is in agreement with increased thermostability demonstrated in our biophysical studies (Fig. 1A). It might also explain the slight lowering of the P-loop in comparison with GRK5⅐AMP-PNP (Fig. 4A, inset), although the lack of phosphate groups in sangivamycin as compared with AMP-PNP is a more compelling argument in this case. Similar lowering of P-loop was observed in PKA complex with adenosine (similar to sangivamycin it lacks a triphosphate tail; PDB code 1FMO) as compared with PKA⅐AMP-PNP crystal structure (PDB code 4DH3).
Interestingly, despite a large sequence difference between the AST segments of PKA and GRK5, it maintains similar spatial arrangement of intramolecular contacts with nucleotide and Nand C-lobes (Fig. 5D). It appears that the guanidinium group of Arg 470 in GRK5 occupies a similar locus as conserved catalytic water in PKA (red sphere in Fig. 5D), which was shown to form a network of interactions with nucleotide (links Tyr 330 to 2Ј-OH of ribose ring), N-lobe (through the carbonyl oxygen of Leu 49 of P-loop), and C-lobe (through the carboxylate of Glu 127 ) in a fully closed state of catalytic subunit in PKA. Similar interactions are observed in GRK5 but without assistance of the catalytic water molecule. Also, backbone carbonyl and amide of Arg 470 in GRK5 form similar contacts with the hinge and the N-lobe as observed for Asn 326 and Asp 329 of the AST fragment in PKA (Fig. 5D). Thus, Arg 470 in the GRK5 AST maintains a similar set of contacts as distinct AST residues in PKA. Structural conservation of these contacts in GRK5 suggests an essentially similar role of AST for kinase domain closure that was well characterized in PKA (59,63).
To further assess a role for C-tail interaction with nucleotide in GRK5, we replaced Arg 470 with alanine and compared catalytic parameters of mutant and wild-type GRK5. GRK5-R470A did not demonstrate a dramatic effect on Michaelis-Menten kinetics for ATP, albeit catalytic efficiency (k cat /K m ) was reduced by 22% compared with WT GRK5 (Fig. 5E). Interestingly, the effect was primarily due to a change in V max (K cat ), although the K m value was unchanged. This suggests that the contribution of Arg 470 to nucleotide binding is minimal (no effect on K m ), although it facilitates the GRK5 catalytic reaction (effect on k cat and V max ). In agreement with this, a lower rate of substrate phosphorylation by GRK5-R470A was revealed in time courses of rhodopsin and tubulin phosphorylation by GRK5 (data not shown). Although a mechanistic understanding of this at the molecular level requires additional studies, Arg 470 interaction with the ribose ring might help to stabilize the orientation of nucleotide within the active site needed for effective catalysis. Another possible interpretation can be related to a potential role of Arg 470 in facilitating ADP release, the rate-limiting step in product turnover by protein kinases (67). The AST is engaged in nucleotide coordination and is found in close proximity to the kinase-active site in fully closed (active) conformation of PKA, although it is further away from the active site in the intermediate and open states (68). Thus, the AST might provide an additional driving force for ADP release when the catalytic cycle is complete and the enzyme adopts a more open conformation with the AST moving away from the active site and pulling out the nucleotide from the kinase-binding pocket. If true, AST interaction with nucleotide should increase nucleotide off-rate, which is in agreement with observed higher catalytic turnover rate for ATP of WT GRK5 as compared with GRK5-R470A where AST is disengaged.
Thus, the kinase domain C-tail of GRK5 is fully ordered, and its AST fragment directly interacts with nucleotide in the catalytic cleft thereby playing a similar nucleotide gate role as in many other AGC kinases. In contrast, the AST in GRK1 (24) and GRK6 (28) is disordered and not part of the nucleotidebinding pocket, despite the active site being occupied with ADP or ATP (AMP-PNP). AST in GRK6⅐sangivamycin is ordered, mainly due to interactions with ␣N-helix, but it is not involved in coordination of the ligand. Available crystal structures of GRK2 lack electron density for nucleotide in the nucleotidebinding pocket, although GRK2-G␤␥ was crystallized in the presence of ATP, and therefore, it is not clear whether the AST is involved in nucleotide coordination. However, the AST is partially ordered in a structure of GRK2 bound to the serotonin reuptake inhibitor paroxetine, mainly due to a few weak van der Waals contacts with the piperidine B ring of paroxetine (69). Thus, a role of AST in other GRKs is not clear. Interestingly, Arg 470 is not conserved even within the GRK4 subfamily (His in GRK4 and Gln in GRK6, see GRK sequence alignment in Fig.  5A); therefore, the contribution of the AST in nucleotide coordination could be unique to GRK5.
Structural Mobility of the Kinase Domain C-tail-The kinase domain C-tail (Fig. 5A) is generally disordered to a different degree in GRK crystal structures and has only been completely modeled in partially closed GRK1⅐ATP (24) and GRK6⅐ sangivamycin (30). Although the GRK5 crystal structure also adopts a partially closed state, there are marked differences in the position of the C-tail and its intramolecular contacts. The entire kinase domain C-tail of GRK5 is well ordered with consistently low B-factor values (32.0 Å 2 for the C-tail versus 35.6 Å 2 for an entire ligand-bound model, see Table 1) compared with the structures of GRK1 and GRK6 (data not shown). Extensive intramolecular and intermolecular (crystal packing) contacts account for the low B-factors of the C-tail in GRK5. Fragment 470 -475 and ␣K-helix are stabilized through crystal contacts to the adjacent GRK5 molecule in the crystal lattice. The AST is also stabilized via hydrogen bonding to nucleotide, whereas the NLT is anchored in the binding pocket formed by ␤1-␤5 strands of the N-lobe. Comparing its position relative to the N-lobe in GRK1 and GRK6, the GRK5 NLT, including the autophosphorylation sites, moves more than 20 Å away from the close proximity of the catalytic cleft and occupies a new position close to the RH domain (Fig. 6A). In GRK6⅐sangivamycin, the AST forms a typical "tail loop" (Fig. 6A, inset) interacting with ␣N-helix (which is disordered in GRK5). The tail loop ("AST loop") is not formed in GRK5, and instead of binding to the ␣N-helix, the AST region participates in nucleotide coordination with Arg 470 . Moreover, unwinding of the tail loop results in displacement of AST residues upward from the kinase-active site (e.g. 12 Å shift of Tyr 473 in GRK5 relative to its the N-lobe in GRK5 (29 amino acids, residues 474 -502) as compared with GRK6 (23 amino acids, residues 480 -502) and GRK1 (26 amino acids, residues 481-506). This enables the NLT to follow a novel path and be positioned inside the N-lobe ␤-sheet groove, a site that is not occupied in GRK1 and GRK6. This new position of the NLT resembles the position of the NLT in PKA/PKB/PKC, which also span the N-lobe groove, although the NLT is not as deeply buried as in GRK5.
The NLT in PKA/PKB/PKC serves as a docking site for an upstreamkinasePDK1thatmediatesPKA/PKB/PKCphosphorylation and positions the ␣C-helix within the ATP-binding pocket (63). In the GRK1 and GRK4 subfamilies, the NLT contains adjacent autophosphorylation sites that appear to regulate GPCR phosphorylation (Fig. 5A) (2). We examined the mobility of the NLT segment by analyzing intramolecular autophosphorylation of GRK5 (Fig. 6B). In addition, we also tested the ability of wild-type GRK5 (WT) to phosphorylate a kinase-dead mutant GRK5-K215R (KD) to assess intermolecular autophosphorylation. In the presence of ATP and phospholipids, GRK5 undergoes rapid autophosphorylation at Ser 484 and Thr 485 within the NLT (70). KD and WT GRK5 were incubated separately or together for 2 or 60 min, and incorporation of 32 P into GRK5 was evaluated by SDS-PAGE. Autophosphorylated forms of GRK5 migrate slower, which leads to the appearance of two shifted bands of partially and fully autophosphorylated WT GRK5 after 2 min and one band of fully autophosphorylated WT GRK5 after 60 min. Only autophosphorylation of the WT enzyme was detected upon 2 min of reaction despite a 5-fold molar excess of KD GRK5, providing evidence for an FIGURE 6. Structural mobility of the kinase domain C-tail in GRKs. A, comparison of kinase domain C-tail orientation in GRK5 relative to its position in GRK6 (PDB code 3NYN, upper left), GRK1 (PDB code 3C4W, lower left), and PKA (PDB code 3TNP, lower right) reveals a large displacement of the GRK5 NLT segment compared with GRK1 and GRK6. The GRK5 NLT moves away from the catalytic cleft and occupies a position close to the RH domain with autophosphorylation sites (red sticks) largely shifted from their position in GRK1 and GRK6 (23 Å and 26 Å, respectively). The structural coordinates of the GRK5 NLT more closely resemble the conformation of the NLT in PKA (12 Å distance between corresponding positions of autophosphorylation sites, Ser 338 in PKA and Ser 484 in GRK5). Inset, upper right, Tyr 473 , which is conserved in the GRK1 and GRK4 subfamilies, forms a hydrogen bond with the backbone amide of Asp 468 and stabilizes the tail loop (AST loop) in GRK6. In GRK5, Tyr 473 shifts 12 Å away from its position in GRK6 and does not stabilize formation of the AST loop. B, GRK5 is autophosphorylated in an intra-as well as intermolecular manner. GRK5 (WT) and kinase-dead GRK5-K215R (KD) were autophosphorylated alone or in a WT/KD mixture at a 1:5 molar ratio. The reaction was done in the presence of [␥-32 P]ATP and phospholipid vesicles at 30°C for 2 or 60 min as described under "Experimental Procedures." Only intramolecular autophosphorylation of WT GRK5 was observed after 2 min, which demonstrates the mobility of the GRK5 kinase domain C-tail, which can access the active site of GRK5 situated ϳ25 Å away from the catalytic cleft in the crystal structure. Phosphorylation of the KD mutant was only detected at 60 min when WT GRK5 was present, demonstrating the ability of GRK5 to autophosphorylate in an intermolecular manner. GRK5 that is fully autophosphorylated (F), intermediately autophosphorylated (I), or nonphosphorylated (N) is denoted on the right. intramolecular reaction (Fig. 6B). The KD mutant is not capable of any autophosphorylation, and therefore its phosphorylation is indicative of intermolecular phosphorylation by WT GRK5. Phosphorylation of KD mutant was detected after 60 min suggesting that intermolecular phosphorylation of GRK5 also occurs, albeit with much slower kinetics than intramolecular autophosphorylation (Fig. 6B). Thus, autophosphorylation of GRK5 can occur in an intra-as well as an intermolecular manner (cis-and trans-autophosphorylation). Because autophosphorylation sites in the NLT are situated relatively far from the catalytic cleft in the crystal structure of GRK5, intramolecular autophosphorylation suggests high mobility of this region in solution. In the crystal structure, mobility of this region is restricted by crystal packing contacts, although conformational variability of the C-tail observed in different ligand complexes of GRKs (either fully or more often partially visible C-tail) also implies "malleability" of the C-tail.
Molecular Basis of GRK5 Association with the Plasma Membrane-Several mechanisms of membrane localization are utilized by GRKs to be in close proximity to their substrates, the integral membrane G protein-coupled receptors. These include C-terminal prenylation (GRK1 and GRK7), palmitoylation (GRK4 and GRK6), pleckstrin homology domain binding to acidic phospholipids and membrane-associated G␤␥-subunits (GRK2 and GRK3), and electrostatic interaction with phospholipids (GRK4 -6) (1). Two clusters of basic/hydrophobic residues within the N-and C-terminal regions appear to play a primary role in GRK5 association with phospholipids (Fig. 1C). When oriented as in Fig. 1D, an electrostatic contour reveals a bipartite distribution of the surface charge with the top (including N-and C-terminal regions) and the right side (N-lobe and catalytic cleft of kinase domain) of GRK5 abundant in basic residues, although the RH domain and C-lobe (except for its active site surface area) are predominantly negatively charged (Fig. 7A, left center). The N-terminal membrane-binding determinant (residues 22-35) is part of the positively charged area on the top of GRK5 and, together with several other basic residues in the N-lobe and the C-tail of kinase domain that lie in the same plane, constitutes a large membrane-binding surface (Fig.   7A). Based on the position of the last visible residue in the structure (Pro 543 ), the C-terminal membrane-binding determinant (residues 546 -562) should be located in close proximity to the N-terminal membrane-binding interface and therefore extend the area of positive charge on the membrane-binding region of GRK5. Simultaneous anchoring of the N-and C-terminal basic regions might result in formation of a single broad surface that promotes stable association of GRK5 with phospholipids (Fig.  7B). Two-dimensional diffusion within the membrane plane would enable GRK5 to rapidly encounter and dock on an activated GPCR (Fig. 7B). This would be followed by kinase domain closure and proper alignment of the receptor phosphoacceptor sites (C terminus) onto the C-lobe of the catalytic cleft in a manner analogous to how substrate peptide docks into the catalytic cleft of PKA, which was used to model substrate binding for GRK5 in Fig. 7A (C-lobes of PKA and GRK5 were aligned). Once receptor phosphorylation is complete, GRK5 would likely dissociate from the receptor and possibly the membrane via a process that may be regulated by GRK5 autophosphorylation (Fig. 7B). This would be similar to GRK1, where autophosphorylation facilitates dissociation of GRK1 from phosphorylated rhodopsin (71).
Another aspect of GRK5 regulation to consider is its ability to interact with calmodulin (CaM), a ubiquitous Ca 2ϩ sensor that links elevation of intracellular calcium levels with GRK inhibition. Although the exact mechanism of CaM binding to GRK5 is unknown, previous studies demonstrated that CaM binds to N-and C-terminal regions of GRK5 (residues 20 -39 and 546 -562) (23,72), which largely overlap with the membrane-binding determinants. This leads to the hypothesis that CaM binding would block GRK5 association with membranes in response to elevation of intracellular calcium levels and thereby attenuate GPCR phosphorylation (Fig. 7B). Moreover, CaM enhances the redistribution of GRK5 from the plasma membrane to the cytoplasm and/or nucleus, where GRK5 targets a different set of substrates (5). CaM wrapping around an amphipathic ␣-helix of a target protein represents the canonical binding mode of CaM (73). Moreover, CaM can induce helical formation when bound to regions that were previously extended or disordered FIGURE 7. GRK5 electrostatic surface and a model for GRK5 binding to phospholipid membranes. A, electrostatic surface potential of GRK5. The surface potential of GRK5 is depicted from Ϫ5 (red, acidic) to ϩ5 (blue, basic) kT/e Ϫ , and neutral regions are white. The stretch of basic residues is clustered on the top of GRK5 and maps the presumed membrane-binding surface. The N-terminal membrane-binding region (residues 22-35) is abundant in Lys residues (labeled) and is at the core of the interface providing electrostatic interaction of GRK5 with negatively charged lipids in the plasma membrane. A few positively charged residues of the N-lobe and the C-tail of kinase domain might also contribute to electrostatic anchoring of GRK5 on membranes. It should also be emphasized that although the C-terminal membrane-binding region (residues 546 -562) is not resolved in the GRK5 crystal structure, it is near the last residue that was observed (Pro 543 , labeled) and thus should be in the same plane with the N-terminal membrane-binding region likely constituting a common electrostatic surface for membrane attachment. The electrostatic contour of the kinase-active site also has a basic character. Substrate peptide (yellow stick model) from a PKA⅐AMP-PNP⅐SP20 complex (PDB code 4DG0) was mapped onto GRK5 C-lobe to model substrate binding into the catalytic cleft of GRK5 (coordinates of C-lobes of GRK5 and PKA were aligned). The zoomed-in view demonstrates the peptide orientation relative to AMP-PNP in GRK5. The site of phosphorylation on the substrate peptide is labeled and is positioned against the ␥-phosphate of AMP-PNP. This illustrates the initial step of the kinase catalytic cycle (substrate and nucleotide alignment into the kinase-active site). B, structural model depicting binding and CaM-driven dissociation of GRK5 from phospholipid membrane and GPCRs. The model is based on the GRK5 crystal structure, which is oriented relative to the phospholipid surface in a manner that would bring the N-terminal membrane domain (residues [22][23][24][25][26][27][28][29][30][31][32][33][34][35] in close contact with PIP 2 in the plasma membrane. Color coding is the same as in Fig. 1D. The GRK5 C terminus (residues 544 -590) is incomplete in the crystal structure and therefore unresolved regions are depicted as a gray dashed line, and the C-terminal membranebinding region (residues 546 -562) is depicted as a gray ␣-helix. The crystal structure of GRK5 suggests close location of both N-and C-terminal membranebinding determinants implying that they can form a single interface and work synergistically to target GRK5 to the membrane. Binding of agonist to inactive GPCR (light green, from the crystal structure of ␤ 2 AR in complex with the inverse agonist ICI 118551, PDB code 3NY8) promotes GPCR activation. Docking of GRK5 on an activated GPCR (green, from the ␤ 2 AR⅐G s complex structure, PDB code 3SN6) promotes full closure of the kinase domain and proper alignment of phosphoacceptor sites on the receptor (green dashed line). CaM and GRK5 autophosphorylation regulate the stability of GRK5 association to the membrane and GPCR. CaM (shown as a brown model) drives GRK5 redistribution from membrane compartment to cytoplasm or nucleus, where GRK5 targets a different set of substrates. GRK5 autophosphorylation sites (Ser 484 and Thr 485 ) are located at the presumed GRK5/lipid bilayer interface in the crystal structure, and hence, autophosphorylation might change the electrostatic potential of GRK5 and destabilize GRK5 anchoring on the membrane. (74). Although both CaM-binding regions on GRK5 have a propensity to form amphipathic ␣-helices as predicted by secondary structure simulation (data not shown), the N-terminal CaM-targeting region (residues 20 -39) is largely unstructured (Fig. 1D). Moreover, it is also involved in extensive intramolecular contacts with RH domain residues. This includes Trp 30 , a residue predicted to form an interface with CaM (72) that is buried within a hydrophobic pocket of GRK5 and is thus inaccessible to CaM. Thus, it is reasonable to speculate that binding of CaM to the N-terminal CaM-targeting region would involve a structural rearrangement within the N-terminal CaM-binding region that would include dissociation from the RH domain and subsequent helical formation induced by CaM binding.
Author Contributions-K. E. K. purified proteins, designed, performed, and analyzed all enzymatic, CD, and SEC experiments, prepared figures, and wrote the paper. A. B. and K. E. K. crystallized the complexes and carried out x-ray data collection and structural analysis. A. B. solved the crystal structures and performed AUC experiment and analysis and wrote the paper. J. L. B. conceived and coordinated the study and wrote the paper. All authors reviewed the results and approved the final version of the manuscript.