Crystal structure of rhodopsin in complex with a mini-Go sheds light on the principles of G protein selectivity

The structure of the rhodopsin/mini-Go complex reveals new insights on G protein selectivity.


INTRODUCTION
G protein (heterotrimeric guanine nucleotide-binding protein)-coupled receptors (GPCRs) are a large family of membrane receptors that interact with heterotrimeric G proteins to transform extracellular signals into cellular responses. Selective coupling of GPCRs to specific G-protein subtypes is a critical step that determines their physiology and their response to natural ligands and clinical drugs. While previous studies have revealed regions on GPCRs and G proteins involved in selectivity (1,2), the molecular details remained elusive. These details can be elucidated by examining and comparing the structure of GPCR-G protein complexes, which lie at the center of this signal transduction event.
Crystal structures of GPCR signaling complexes have been determined only for the G s protein and arrestin. Several techniques are being used to stabilize these GPCR-G protein complexes for structural determination, such as fusion of T4 lysozyme on the receptor and the use of a nanobody (3) or mini-G protein (4), or the creation of a fusion chimera in the rhodopsin-arrestin complex (5). Mini-G proteins comprise the guanosine triphosphatase (GTPase) domain of the  subunit of heterotrimeric G proteins and are fully capable of stabilizing GPCRs in their active state, generating characteristic changes in agonist affinity and recapitulating G protein specificity in vivo (6)(7)(8).
The light-sensitive visual receptor rhodopsin is the prototypical class A (rhodopsin-like) GPCR. It is activated by light-induced isomerization of the covalently bound ligand retinal. Upon activation, rhodopsin couples to and activates G t (transducin), a member of the G i/o subfamily (2), thus triggering the cyclic guanosine monophosphate cascade. At the center of this signal transduction mechanism lies the formation of a complex between active rhodopsin and the G protein.
Rhodopsin is the first GPCR that had structures determined in the inactive and active states with and without agonist (9) and in an arrestin-bound state (5). In addition, rhodopsin is the only GPCR for which structures of human disease mutants have been solved (10,11). Comparisons of these structures revealed that rhodopsin activation results in the opening of a cleft on the cytoplasmic side of the receptor and the formation of a binding cavity for the C-terminal domain (5) of G t . The rhodopsin system has significantly contributed to our understanding of class A GPCR activation (12)(13)(14)(15)(16).
The organization of rhodopsin and the G proteins G i and G t in the signaling complex has been recently analyzed by combining sitedirected spin labeling and electron paramagnetic resonance spectroscopy data with molecular dynamics simulations (17,18). In addition, a mutagenesis analysis performed with native rhodopsin and G i allowed for the location of residues involved in the stabilization of the complex (19). While these results have provided valuable information on the overall arrangement of the complex, a high-resolution, threedimensional (3D) map of the interaction interface is still missing. Obtaining additional structures with this G protein subtype will allow us to identify the structural elements defining G protein selectivity.

RESULTS
Here, we report the crystal structure of the stabilized constitutively active mutant of bovine rhodopsin N2C/M257Y/D282C in complex with mini-G o . These three mutations do not affect negatively the structure or functional properties of active rhodopsin (20). Attempts at creating a mini-G protein from G t were unsuccessful, and therefore, we used the related G o (62% sequence identity) (6). Similar to the full heterotrimeric G protein (21), mini-G o couples effectively to light-activated rhodopsin solubilized in detergent, creating a stable complex that shows the characteristic spectral properties of the fully active state of rhodopsin bound to all-trans retinal ( fig. S1A). The complex was purified in octyl glucose neopentyl glycol (OGNG) and crystallized by vapor diffusion at 4°C. Merging diffraction data from five crystals allowed structure determination by molecular replacement and refinement up to a resolution of 3.1 Å (Material and Methods,  table S1, and fig. S2). Crystals grew with p6 1 symmetry, in which the crystal packing is mainly mediated by hydrophilic contacts between mini-G o molecules. The orthosteric ligand-binding pocket contains only partial density for the agonist all-trans retinal ( fig. S3), but the electron density is well defined at the mini-G o binding interface in the cytoplasmic cleft (Fig. 1).

Mini-G o binding interface to rhodopsin
The crystal structure of the engineered G protein mini-G o bound to rhodopsin displays a high overall similarity to mini-G s bound to the adenosine A 2A receptor [A 2A R; root mean square deviation (RMSD), 1.11 Å] (4), as well as to the guanosine triphosphate (GTP)-binding domains of heterotrimeric G s bound to the  2 -adrenergic receptor ( 2 AR; RMSD, 1.17 Å) ( fig. S4A) (3). The protein-protein interfaces of mini-G o and rhodopsin in the complex are composed of 17 residues and 23 amino acid residues (Fig. 2), with a surface area of 1104 and 1375 Å 2 , respectively. An evolutionary analysis (22) classifies this interface as biological (that is, not as merely a crystallographic contact). The regions on mini-G o that make contact with the receptor comprise the 4-6 loop, the 6 sheet, and the 5 helix (Figs. 1 and 2 and fig. S5). In the receptor, the major contact regions to mini-G o are located across the cytoplasmic ends of transmembrane helices TM2, TM3, TM5, TM6, and the TM7/H8 joint, with additional contacts mediated by the ICL3. These contact interfaces agree with a previous cross-linking study using rhodopsin mutants and native transducin (23) and with a mutagenesis analysis performed with native rhodopsin and G i mutants (19). These data, combined with the structural similarities to existing GPCR-G protein complexes, strongly suggest that the complex between rhodopsin and mini-G o recapitulates the main features (orientation and contact interface) of the native interaction with G t .
The C-terminal portion of the 5 helix in mini-G o contributes most of the interactions to the receptor, as previously observed in G s coupled to other GPCRs (Figs. 2C, 3, and 4A). In contrast, the contacts that we observe between the 6 sheet and the receptor are not present between either A 2A R (4) or  2 AR (3) and G s ; this is probably a consequence of the different modes of binding of G o to rhodopsin (see below) (18). In addition, the contacts between the helical ICL2 of A 2A R,  2 AR, and the serotonin 5-HT 1B receptor (5-HT 1B R) (24) and the 2-3 loop/3 sheet in the  subunit of G s are absent in the rhodopsin/mini-G o structure, in which ICL2 is unstructured (Fig. 3A). Thus, these differences may reflect a certain degree of plasticity in the interface between GPCRs and G proteins outside the 5 helix-binding crevice.
The residues in rhodopsin involved in making contacts with mini-G o are also very similar to those between A 2A R or  2 AR and G s , although there are some subtle differences (Fig. 4A). Only in rhodopsin and class B GPCRs are there contacts mediated by TM2. Moreover, the contacts with TM7/H8 are more prominent in G o -than in G s -coupled complexes. There are also differences in the number of residues in each structural element making contact to the  subunit. In particular, the smaller outward movement of TM6 in the rhodopsin/mini-G o structure, coupled with the different tilt of the 5 helix in the G protein, results in most of the rhodopsin/mini-G o contacts being in TM6 and fewer in TM5 (Figs. 1C, 3, and 4A). In contrast, in the  2 AR/G s structure and the A 2A R/mini-G s structure, most of the contacts are observed in TM5 and the end of TM3 running into ICL2 (Figs. 3 and 4A).
In summary, the contact residues in the complex interfaces spread over the cytoplasmic region of the receptors, and we cannot observe concrete interaction patterns. Thus, the selectivity between receptors and G proteins is not just encoded in the 2D protein sequence of the effector-binding site but rather in its 3D shape.     The ionic interaction between E134 3.49 and R135 3.50 at the D/ERY motif is broken; opening of this interhelical ionic lock is one of the hallmarks of GPCR activation. The constitutively activating M257 6.40 Y mutation assists in stabilizing the extended active conformation of R135 3.50 , as observed in the structure of this mutant bound to a G t -derived peptide (20). (C) Contact interface between the C terminus of mini-G o and TM2 and TM6 in the receptor.
(D) Contact interface between the C terminus of mini-G o and TM3 and TM7 in the receptor. The last four residues of the G protein (C-terminal hook) are located between R135 3.50 and the joint between TM7 and helix 8 (H8). (E) Contact interface between the 6 sheet in mini-G o and intracellular loop 3 (ICL3) in the receptor.

Selectivity between receptor and G protein
The major difference between the G o -and G s -GPCR complexes lies in the pose of the C-terminal 5 domain within the effector-binding site, with a difference in tilt angle of about 30° ( Fig. 3 and fig. S4B).
Remarkably, the last four amino acid residues of the 5 helix in G i/o and G s fold in an identical "hook-like" structure upon binding the receptor (3,4,20). However, their different side chains mold these elements into dissimilar shapes (Fig. 4B). This short C-terminal hook and the adjacent residues in 5 all make contacts to the receptor and are the most important structural elements in defining the subtype specificity of G proteins (2,(25)(26)(27).
The degree of opening in the cytoplasmic binding cavity partially determines the overall placement of the C-terminal hook, as its last two residues interact with the "displaced" cytoplasmic domains of TM5/6 (Figs. 3C and 4C). This hook also latches to TM3 [a structural hub of the GPCR fold located in the opposite side of the binding cavity (28)] by a subtype-specific interaction [G i/o , Cys351 H5.23 ; G s , Tyr391 H5. 23 ; the residue numbering refers to the human wildtype G o1  subunit (UniProtKB P09471), and the superscripts refer to the common G numbering system (25)] with Arg 3.50 [conserved in class A GPCRs; superscripts in receptor residue numbers refer to the Ballesteros-Weinstein general numbering scheme (29)]. As a result, the hook locates at the interface between TM5/6 and the TM7/H8 joint in the binding cavity of the receptor. We observed that the conserved Gly352 H5.24 of mini-G o forms a backbone-side chain interaction with Asn310 8.47 in the receptor, placing it very   close to the TM7/H8 joint (Fig. 3C). In contrast, G s has a much bulkier Glu392 H5.24 , which rests between the TM7/H8 turn and the TM5/6, interacting with Lys270 6.32 (Figs. 3C and 4C). As a result, the relative orientation between the C-terminal hook and TM7/H8 of the receptor is the main factor that differentiates the binding of G i/o and G s . Thus, GPCRs could modulate their selectivity for G proteins by shaping the cytoplasmic binding cavity to fit the Cterminal hook in a specific pose that results in a pulling motion of the 5 helix in the G protein, leading to its activation (30,31). In G i , this pose requires that the C-terminal hook packs its backbone tightly to the TM7/H8 turn. In contrast, recognition of the G s C-terminal hook involves a structurally looser interaction with the TM7/H8 joint and TM5/6.

Two conformationally distinct states of rhodopsin
The reaction pathway of rhodopsin activation has been delineated in great detail using a variety of spectroscopy methods, revealing the existence of several identifiable intermediate states (15,16). Furthermore, the structure of some of these intermediates has been determined by x-ray and electron crystallography (9). The early agonist-bound intermediates batho, lumi, and meta I form in microseconds and show only minor structural changes limited to the ligand-binding pocket. Similar small and local changes have also been observed in other agonist-bound inactive-state GPCR structures (32,33). In contrast, transition to later stages in rhodopsin activation (meta IIa, meta IIb, and meta IIbH + ) involves the opening of a cytoplasmic effector-binding site within a few milliseconds through the relocation of the cytoplasmic end of TM6. The equilibrium between these late active states is partially regulated by subtle changes, such as in the protonation state at the cytoplasmic D/ERY motif (ionic lock) (13) or in the electronic properties of retinal (34). All identified spectroscopic states of rhodopsin fall into two main structural categories: inactive (including the inverse agonist-bound state and the agonistbound early active intermediates) and active (apoprotein at low pH, agonist-bound late active states and arrestin-bound rhodopsin). The active conformation of rhodopsin displays a relatively small opening in the cytoplasmic effector-binding site. To date, it has been unclear whether this was a true feature of rhodopsin bound to G t and, by extension, of other GPCRs bound to G i/o . Surprisingly, the structure of rhodopsin in complex with mini-G o is virtually identical to all the previous structures of active rhodopsin (average RMSD, 0.51 ± 0.07 Å; Figs. 5 and 6A and table S2) (5,20,(35)(36)(37)(38). Thus, our results show that a GPCR can expose an identical site for the recognition of the 5 helix of G and the finger loop of arrestin (Fig. 5). Rhodopsin thus effectively acts like a structural switch upon activation, adopting predominantly two thermodynamically stable conformational states, an inactive state and an active state. In contrast, spectroscopic and structural studies suggest that GPCRs that bind diffusible ligands exist as an ensemble of conformations that include a number of inactive, intermediate, and active conformations (1,39). Accordingly, their structures show a higher degree of variability (Fig. 6A).

DISCUSSION
The crystal structure of rhodopsin bound to an engineered mini-G o protein provides new insights into the molecular basis of GPCR-G protein recognition and selectivity. Comparison with existing complexes suggests that the sequence of the binding interface alone does not determine selectivity (Fig. 4A). On the other hand, we notice that the interactions between ICL2 and ICL3 of the GPCRs and the GTPase domain of G proteins show a certain degree of plasticity, which could account for selective protein-protein interactions. In addition, the different extent in the movement of TM6 of GPCRs upon activation has been proposed to play a fundamental role in G protein selectivity (30,31). While all these factors probably contribute to pulling the 5 helix to selectively activate specific  G proteins, we suggest that another significant determinant in this mechanism is the precise placement of the C-terminal hook-like element of the G protein (its last four residues) relative to the TM7/ H8 joint. Activation of GPCRs involves a precise coupling between the physicochemical properties of their ligands and the dynamic properties of the receptor that results in the engagement and activation of intracellular signaling proteins to promote downstream signaling. GPCRs that bind diffusible ligands have been shown to exist in an ensemble of conformations (Fig. 6B) (40). We show here that the visual receptor rhodopsin, on the other hand, functions like a "conformational switch" upon activation. This behavior remains even when the covalently bound ligand retinal is replaced by diffusible, pharmacologically designed compounds ( fig. S6) (41), suggesting that this is an inherent property of the receptor. This remarkable property of rhodopsin is likely related to the control of visual perception. First, a stable inactive conformation (in combination with other adaptations such as the presence of high local concentrations of both rhodopsin and G proteins in the rod photoreceptor membranes) allows for a tight control of basal activity, which is critical to reduce dark noise in visual systems. In addition, by having a unique active conformation, binding to G proteins and the turnover time for allowing arrestin binding after receptor phosphorylation are probably optimized, allowing for the activation of about 600 G t molecules per second (42).
The structure of rhodopsin bound to mini-G o presented here adds a key piece to the gallery of states along its activation pathway, making rhodopsin the best-described GPCR in terms of structure. We can now sequentially track the conformations of a GPCR from its inactive state to agonist-induced activation, G protein coupling, and binding to arrestin. Furthermore, the structure provides new insights into the binding interface between activated GPCRs and G i/o proteins. This information advances our understanding of GPCR-G protein selectivity and how GPCR conformations promote certain signaling pathways.

Protein preparation
Mini-G o was prepared as described (6,43). The N2C/M257Y/D282C mutant of bovine rhodopsin was expressed in human embryonic kidney (HEK) 293 GnTI − cells as described (20,44). Buffers for every purification step were cooled to 4°C before use, and the steps after the addition of 9-cis retinal were performed under dim red light before light activation of rhodopsin. HEK293 cells were homogenized in phosphate-buffered saline (PBS) buffer and then solubilized by supplementing dodecyl maltoside (DDM) (Sol-Grade, Anatrace) to 1.25% (w/v). After gentle mixing for 1 hour, the cell lysate was centrifuged at 200,000g for 1 hour to remove unsolubilized residual. The rhodopsin apoprotein in the solubilisate was captured using immunoaffinity agarose resin coupled to 1D4 antibody. The resin was collected and washed with 10 column volume (CV) of PBS buffer containing 0.04% DDM, and then 2 CV of PBS buffer containing 0.04% DDM and 50 mM 9-cis retinal (Sigma-Aldrich) or F and Endo F1 were tested, and only Endo F1 rendered a homogeneous deglycosylated product, which was confirmed by SDS-PAGE ( fig. S1C).

Crystallization and synchrotron data collection
Crystallization of the gel-filtrated rhodopsin/mini-G o complex was set up using the vapor diffusion method. Crystallization drops were dispensed using a mosquito crystallization robot by mixing 200 nl of protein and 200 nl of crystallization buffer containing 0.1 M MES (pH 5.5) and 10 to 20% polyethylene glycol 4000 (PEG 4000) in an MRC 2-well crystallization plate (Swissci) at 4°C. Sword-or rodshaped crystals appeared in 1 to 3 days, and crystals wider than 10 m mainly grew from drops using the crystallization buffer containing 13 to 17% PEG 4000. Before crystals were harvested on crystal loops (MiTeGen), crystals were cryoprotected by adding 400 nl of mother liquor supplemented with 10% glycerol or 10% trehalose to the crystallization drop 1 day before harvesting. Harvested crystals were flashfrozen in liquid nitrogen ( fig. S1B). Frozen crystals were evaluated at the PXI beamline at the Swiss Light Source (SLS) using raster scanning to identify the best diffracting locations on a frozen crystal. Diffraction datasets were collected by exposing crystals to a 10-or 20-m-sized beam with 0.05° oscillation angle per frame.

Refinement and model building
Individual diffraction datasets were analyzed using XDS (45) for integrating Bragg peaks. All datasets showed the same unit cell dimensions with negligible deviation (less than 0.5%) and space group p6 1 . The best five datasets were chosen for merging together to improve the signal-to-noise ratio. The datasets were combined using XSCALE without scaling and merging, and the pooled reflection list was further analyzed using the STARANISO server (Global Phasing Ltd.). The STARANISO server first analyzed the anisotropy for each dataset, giving a resolution of 3.69 Å in the hk plane and 3.11 Å in the l direction using the criterion of CC 1 / 2 = 0.3. The scaling factor was determined and applied after applying the anisotropic mask. After scaling, the datasets were merged, and the final resolution was improved to 3.02 Å and anisotropically truncated to 3.12 Å in the best direction (table S1). Structures of rhodopsin (PDB ID: 4A4M) and mini-G s (PDB ID: 5G53) were used as search models to perform molecular replacement with the program Phaser. The sequence of mini-G o was modeled using the Swiss-Prot server (46). Structure refinement was performed using phenix.refine (47) and Rosetta refinement (48) from the Phenix suite. Manual model-building and adjustment of the coordinates were performed using the visualization program Coot. Polder maps (49) to verify the presence of retinal were calculated with Phenix using a low-resolution cutoff of 20 Å. For the retinal-only polder omit map, three cross-correlations were calculated between the following maps: (i) calculated F obs with and without ligand (0.70), (ii) calculated and real F obs with ligand (0.85), and (iii) calculated and real F obs without ligand (0.77). These results verified the presence of retinal ( fig. S3). In the F obs -F calc (observed and calculated structure factors) map, there is a strong peak density above the 5 cutoff, and a water molecule was accordingly modeled into this position. The outlier in the Ramachandran plot (R206 in the 3-2 loop of mini-G o ) was likely caused by a crystal contact with ICL3 of rhodopsin in a symmetry-related complex molecule.

Comparison and analysis of structures
Structural models were downloaded from the Protein Data Bank. RMSDs (table S2) were obtained by superimposing the 3D struc-tures with the super command in PyMOL (Schrödinger). The contacts between receptors and G proteins (Figs. 2 and 4A) were identified by selecting atoms within <4 Å cutoff. Figures were prepared using PyMOL.

SUPPLEMENTARY MATERIALS
Supplementary material for this article is available at http://advances.sciencemag.org/cgi/ content/full/4/9/eaat7052/DC1 Fig. S1. Sample preparation and crystals of the rhodopsin/mini-Go complex. Fig. S2. Data completeness versus resolution. Fig. S3. Electron density of the all-trans retinal. Fig. S4. Structural comparison of the G proteins. Fig. S5. Sequence alignment of Gi/o proteins, mini-Go, and Gt peptide. Fig. S6. Comparison of active rhodopsin structures. Table S1. Statistics of the crystallographic data and the structural refinement. Table S2. Quantitative measure of similarity between active rhodopsin structures.