Structure of the human galanin receptor 2 bound to galanin and Gq reveals the basis of ligand specificity and how binding affects the G-protein interface

Galanin is a neuropeptide expressed in the central and peripheral nervous systems, where it regulates various processes including neuroendocrine release, cognition, and nerve regeneration. Three G-protein coupled receptors (GPCRs) for galanin have been discovered, which is the focus of efforts to treat diseases including Alzheimer’s disease, anxiety, and addiction. To understand the basis of the ligand preferences of the receptors and to assist structure-based drug design, we used cryo-electron microscopy (cryo-EM) to solve the molecular structure of GALR2 bound to galanin and a cognate heterotrimeric G-protein, providing a molecular view of the neuropeptide binding site. Mutant proteins were assayed to help reveal the basis of ligand specificity, and structural comparison between the activated GALR2 and inactive hβ2AR was used to relate galanin binding to the movements of transmembrane (TM) helices and the G-protein interface.


Introduction
Human galanin is a neuropeptide composed of 30 amino acids whose highly conserved N-terminus is linked to biological activities [1]. Its GPCR family receptors (from GALR1 to GALR3) are associated with different disease states and have been used as markers for certain cancer types [2]. Fragments of galanin may have distinct roles from the integral neuropeptide; for example, galanin [2][3][4][5][6][7][8][9][10][11] binds to GALR2 roughly 500 times more tightly than GALR1 [3][4][5]. Nerve damage induces synthesis of galanin, which appears to show neuroprotective effects mediated by GALR2 [6]. Agonists for GALR2 are believed to have a potential as treatments for a1111111111 a1111111111 a1111111111 a1111111111 a1111111111 a variety of nervous disorders by acting as antidepressants or anticonvulsants as well as having analgesic properties, stimulating nerve growth, and reducing neuronal damage [7][8][9][10]. Human GALR2 and human GALR3 show roughly 55% sequence conservation with each other, while human GALR2 shows only 37% sequence identity with human GALR1 [11]. All of them signal via Gi/o and inhibit the adenylyl cyclase, but GALR2 can also activate phospholipase C (PLC) through Gq/11, leading to inositol triphosphate accumulation and a subsequent increase of the concentration of intracellular calcium [12]. In order to understand the signaling pathway of GALR2 and interactions between GALR2 and galanin, we determined the cryoelectron microscopy (cryo-EM) structure of human GALR2 with holo-galanin and associated G-proteins.

Results and discussion
To determine the molecular structure of activated GALR2, a complex was purified composed of GALR2, galanin, heterotrimeric mGαq iN /Gβ1γ2, and scFv16 (S1 Fig). The sample was plunge-frozen, and micrographs were collected using a 300-kV Titan Krios G4 with a Gatan K3 direct detector in the movie mode. The data were processed with cryoSPARC (v.3.3.1) and were refined with PHENIX (v. 1.19.2) (Table 1 and  The final model of GALR2 includes all residues from Glu24 1.31 to Arg300 8.51 , including the intracellular loops (ICLs) and extracellular loops (ECLs). The scFv16 stabilizes the complex by binding to the N-terminal helix of mGαq iN and a surface loop of Gβ1 (Fig 1A).
Activated GPCRs act as nucleotide exchange factors through direct interactions with heterotrimeric G-proteins. The C-terminal helix of mGαq iN fits into a hydrophobic pocket formed by TM 5-7 of GALR2 ( Fig 1B). Phe228, Ile235, Asn239, and Tyr243 of mGαq iN form the principal interface with GALR2, and a salt bridge forms between Glu135 ICL2 of GALR2 and Arg32 of mGαq iN (Fig 1B). The TM helices of GALR2 form a relatively shallow galaninbinding pocket at the extracellular face of the protein (Fig 1C), where the first 16 residues of galanin were modeled into the density map (Fig 2A). The modeled fragment adopts a compact shape with helix-like conformation from Leu4 P to Leu11 P (Fig 2A). Galanin touches each ECL, and it also forms hydrophobic contacts near the N-termini of TM2 and TM7 ( Fig 2B). Galanin has a similar sequence to galanin-like peptide (GALP) and spexin (S3A Fig), which can also act as endogenous ligands and activate GALR2 [13]. Trp2 P , Thr3 P , Tyr9 P , Leu10 P , and Gly12 P of galanin are common to these peptides (S3A Fig). Galanin is well conserved among various species, but the last 15 residues are more variable than the N-terminal region, which is known to interact with GALRs (S3B Fig). Replacing Trp2 P of galanin with alanine prevents binding to GALRs [4]. This tryptophan side chain makes a hydrophobic interaction with Leu266 ECL3 of GALR2 ( Fig 2C). The side chain of Tyr9 P reaches furthest into the pocket, where it makes interactions with Ile85 2.64×63 , His102 3.29 and Tyr164 4.64×65 of the receptor (Fig 2C). Leu10 P interacts with Phe264 ECL3 , while Pro13 P packs against His176 ECL2 and Pro177 ECL2 (Fig 2C). His14 P , Ala15 P , and Val16 P are completely exposed on the surface of the complex and do not make any interactions with the receptor (Fig 2C). While the majority of the observed galanin residues interact with the receptor to some extent, Trp2 P , Asn5 P , and Tyr9 P appear to be central to receptor binding, while Leu10 P also makes substantial hydrophobic interactions (Fig 2C).
Comparing the sequences of the human galanin receptors (GALR1-GALR3), the majority of the neuropeptide binding site is found to be preserved, but not perfectly. Cys98 3.25 and Cys175 ECL2 of GALR2 form a disulfide bridge, and an equivalent bond is presumably also found in the other 2 receptors (S3C Fig). This bond, connecting the N-terminus of TM3 and ECL2 will create a more rigid platform for galanin to pack against. Next to the disulfide bond sits His176 ECL2 , which contacts Asn5 P , Gly8 P , and Pro13 P . This histidine is unique to GALR2 (S3C Fig) and makes a modest contribution to galanin binding; the H176 ECL2 A mutant shows a 5-fold higher EC 50 value ( Table 2). This residue is replaced by tryptophan (Trp188 ECL2 ) in GALR1 and valine in GALR3 (S3C Fig), suggesting that the 3 receptors make interactions of different strengths at this position.
The GALR2 model was used to design a number of mutants that were assayed for galanin binding, considering the expression level of the mutants (Figs 3, S4A, and S4B). The Y164 4.64×65 F mutant shows only slightly weakened galanin binding, while the Y164 4.64×65 A mutant shows none at all, due to the loss of hydrophobic contact with galanin Tyr9 P (Figs 2C and 3). His102 3.29 is common to all 3 receptors, and H102 3.29 A of GALR2 showed strongly decreased binding affinity to galanin, again emphasizing the importance of Tyr9 P for binding (Fig 3). Phe264 and Leu266 in ECL3 contact Leu10 P and Trp2 P , respectively, and replacing either residue with alanine gave roughly 200-fold drops in binding affinity, presumably because ECL3 becomes highly flexible in these mutants, and the same interactions with galanin are lost ( Fig 2C and Table 2). Leu255 6.54 and Arg274 7.35×34 lie close together and contact Leu10 P , so that the mutants L255 6.54 A and R274 7.35×34 A both showed 12-to 15-fold drops in affinity for galanin ( Fig 2C and Table 2). Mutant R184 5.35×36 V also showed a significant drop in galanin binding, possibly due to its role in stabilizing ECL2 (Fig 3). Although Arg184 5.35×36 forms a contact with Pro13 P , the galanin residue is not required for interaction with GALR2. The truncated peptide galanin [2][3][4][5][6][7][8][9][10][11] has been reported to show specificity for GALR2 and GALR3 over GALR1 [3,14]. Duan and colleagues [15] reported that the N-terminal amino group of galanin forms a hydrogen bond with Glu32 1.31 of GALR1, but no such bond forms with GALR2. The experimental model of activated GALR2 shows notable differences from the inactive model of hβ 2 AR (PDB code 2RH1) [16]. The conformational changes associated with agonistinduced activation of class A GPCRs are well known [17,18], mainly involving a highly conserved "toggle" tryptophan residue (Trp 6.48 ) within the CWxP motif of TM6, close to the

PLOS BIOLOGY
NPxxY motif of TM7; agonists trigger pronounced movements of TM6 relative to the inactive state, and GALR2 follows the same pattern (Fig 4A and 4B). The DRY and PIF motifs are other motifs common to class A GPCRs, and their conformational changes on activation are shown ( Fig 4B). The isoleucine residue of the PIF motif is replaced by serine in GALR2 and GALR3 (S3C Fig). This residue, Ser113 3.40 of GALR2, shows no significant movement, while Phe245 6.44 of PI(S)F motif slides against it as TM6 extends toward the cytoplasmic side and rotates, as in the case of other reported GPCR models [17] (Fig 4B). The movement of TM6 is illustrated by the predicted shift of 8 Å in the position of Lys231 6.30 of the DRY motif ( Fig 4B). In common with other class A GPCRs, the shift in TM6 of GALR2 creates a binding site for the cognate G-proteins [17], but unlike the dopamine receptors, the agonist itself (galanin) is considerably distant (>12 Å) from Trp249 6.48 .
Recently, Duan and colleagues published cryo-EM models of GALR1 coupled to G i and GALR2 coupled to G q , showing the role of ICL2 in partner selectivity [15]. This model of GALR2 (PDB code 7WQ4) includes the antibody Nb35 used to stabilize the complex, instead of scFv16, and 2 cholesterol molecules are modeled lying against TM6. Therefore, our structure provides an independent view of the complex, showing that the antibodies have little effect on the GPCR itself. Several residues (from 219 to 222) of ICL3 that lie close to G q are not modeled in the structure (PDB code 7WQ4). The RMS deviations between the 273 C α atoms shared by the GALR2 models is 1.16 Å; the largest difference is found around ICL3, which is slightly shifted by indirect effects of Nb35. No notable bonds are formed between ICL3 of the GPCR and G q in our model; Val221 5.72 comes within 4 Å of Tyr212 of mGαq iN . As concluded by the Duan group, the selectivity of GALR2 for G-protein partners is mainly controlled by ICL2 [15]. The protein-protein interface is essentially the same between the model (PDB code 7WQ4) and our model, except that the latter places Glu135 ICL2 of GALR2 close enough to Arg32 of mGαq iN to form a salt bridge. Although our model includes residues from 1 to 16 of galanin rather than residues from 1 to 13, the ligand overlays closely and makes the same interactions with the protein when compared to each other. Minor changes of rotamer indicate the flexibility of the binding pocket, but there are no significant differences between the ligand interfaces in the 2 GALR2 models.
In conclusion, GALR2 shares the activation mechanism common to class A GPCRs. Our model reveals details of the interactions between GALR2 and galanin, whose N-terminal half adopts a compact form, and this model will assist structure-based drug design selectively targeting one or more galanin receptors to address various human diseases.

Construct design
WT full-length human GALR2 carrying an N-terminal hemagglutinin (HA) signal sequence, FLAG tag, followed by an 8× His tag and HRV 3C protease cleavage site was cloned into the  (M7W, H102I, R106L) was added after the 8× His tag at the N-terminus [19]. The heterotrimeric mGαq iN /Gβ1γ2 was designed as previously described in mini-Gαq/Gβ1γ2 [20]. Single chain antibody scFv16 containing GP67 secretion signal sequence was also inserted into pFastBac1 vector [21].

Expression of GALR2, mGαq iN /Gβ1γ2 heterotrimer, and scFv16
GALR2 and mGαq iN /Gβ1γ2 were expressed respectively using Bac-to-Bac Baculovirus Expression system (Invitrogen) in Spodoptera frugiperda (Sf9) cells using ESF media (Expression Systems, Davis, CA, USA). Using the high titer virus at a multiplicity of infection (MOI) of 3, Sf9 cells density of 2 × 10 6 cells/mL in 400 mL biomass were infected. The cells were incubated with shaking at 27˚C for 72 h and harvested, followed by washing with phosphate-buffered saline (PBS), flash-freezing in liquid nitrogen, and storage at −80˚C until further use. The single chain antibody scFv16 was expressed using Bac-to-Bac Expression system in Sf9 cells, and high titer virus was generated. The cells were incubated with shaking at 27˚C for 72 h, and secreted scFv16 in the supernatant was separated from the cells by centrifugation.

Cryo-EM grid preparation and data collection
A volume of 3 μl of the sample was applied to a glow-discharged holey carbon grid (Quantifoil R1.2/1.3, Cu, 300 mesh). The grid was blotted with blotting force of 10 for 5 s at 4˚C, 100% humidity and flash-frozen into liquid ethane using Vitrobot Mark IV instrument (Thermo Fisher Scientific, Waltham, MA, USA). After the grid was stored in liquid nitrogen, cryo-EM images of the sample were collected on 300 kV Titan Krios G4 (Thermo Fisher Scientific, Waltham, MA, USA) in Hokkaido, Japan. The images were collected by K3-summit camera with 20 eV slit. A total of 7,216 movies were collected by standard mode for 50 frames with total dose of 53.17 e/Å 2 , exposure time of 1.5 s, and dose on camera of 15.967 e −1 /px/s. Magnification of micrographs were ×130,000, and pixel size was 0.67 Å/pixel. Defocus range was from −0.8 to −2.0 μm. Data were automatically collected using EPU software.

Cryo-EM data processing
The processing of the collected data were carried out by cryoSPARC (v.3.3.1) [22]. Motion correction was done by Patch motion correction. CTF estimation for micrographs was done by Patch CTF estimation. Micrographs under 5 Å CTF resolution were cut off by Curate Exposures, and 7,040 micrographs were selected. Particles were auto picked by blob picker using only 10 micrographs, and the particles were extracted using binning state (3.35 Å/pixel).
Extracting from 10 micrographs, 1,972 particles were picked, and 2D models from these were made. Particles in all micrographs were auto picked by Template picker referenced from 2D models. Approximately 2,666,499 particles were picked from micrographs and extracted using binning state (3.35 Å/pixel). After extracting, all suitable particles were selected and classified by Ab-Initio Reconstruction into 6 classes. Class 2 had the clearest model, and 858,756 particles from the class were selected. After reextracting using the particles in 1.12 Å/pixel (Nyquist Resolution 2.24 Å), 703,468 particles were selected by 2D classification. These particles were classified by Hetero Refinement into 6 classes using Ab-initio model as reference model. After 2 times Hetero Refinement, 3D reconstruction was performed by NU-refinement, and 3.11 Å map was obtained. Finally, the map using DeepEMhancer was used [23].

Model building and refinement
The first model of GALR2 was used from AlphaFold Protein Structure Database (UniProt: O43603). Model of G-proteins and scFv16 complexes was used from the structure of Cholecystokinin A receptor (CCKAR)-Gq complex (PDB:7EZM) [24]. Models were roughly fitted to the cryo-EM map by COOT at first, and Real-Space Refine in PHENIX was used for further refinement [25]. The model was built manually based on C α and side chain maps using COOT and then refined using Real-Space Refine in PHENIX.

NanoBiT-based Gq activation assay
Galanin/GALR2-induced Gq activation was measured by the NanoBiT-based Gq-PLCβ association assay [26,27], in which interaction between an activated, GTP-bound Gαq subunit and its effector PLCβ2 is measured by the NanoBiT enzyme complementation system [28]. Plasmid transfection for HEK293A cells (Thermo Fisher Scientific, Waltham, MA, USA) was performed by combining 5 μl (volume is per well in a 6-well culture plate) of polyethylenimine solution (1 mg/mL) and a mixture of plasmids each encoding the Gαq subunit fused with the large fragment (LgBiT) at the amino acid position 123-124 (500 ng), the small fragment (SmBiT)-fused PLCβ2 subunit (500 ng), the untagged Gβ1 and Gγ2 subunits (500 ng each), and the RIC8A chaperone (100 ng), along with a test GALR2 construct (500 ng; containing Nterminal HA-derived signal sequence followed by the FLAG-epitope tag). After an incubation for 1 day, the transfected cells were harvested, pelleted with centrifugation, and suspended in 2 mL of Hank's balanced saline solution containing 0.01% bovine serum albumin (BSA fatty acid-free grade, SERVA) and 5 mM HEPES (pH 7.4) (assay buffer). The cell suspension was dispensed in a white 96-well plate at a volume of 80 μl per well and mixed with 20 μl of 50 μM coelenterazine (custom-synthesized by Amadis Chemical) diluted in the assay buffer. After 2-h incubation at room temperature, the plate was measured for baseline luminescence (Spectra-Max L, Molecular Devices). Thereafter, 20 μl of titrated concentrations of galanin diluted in the assay were manually added, and the plate was positioned for luminescent measurement. Kinetics data points from 5 min to 10 min were averaged and normalized to the initial count and used as a G-protein activation index. The G-protein activation signals were fitted to a 4-parameter sigmoidal concentration-response curve (GraphPad Prism8), and pEC 50 values (negative logarithmic values of EC 50 values) and Span values ("Top"-"Bottom") were obtained. For individual experiments, we calculated E max by normalizing Span to the WT GALR2 and ΔpEC 50 by subtracting pEC 50 of the WT GALR2 performed in parallel. We also calculated E max /EC 50 of GALR2 mutants relative to that of GALR2 WT, a dimensionless parameter known as relative intrinsic activity (RAi) [29] and used its log-transformed value (Log RAi) to denote receptor activity.

Flow cytometry analysis
Transfection was performed according to the same procedure as described in the "NanoBiTbased Gq activation assay" section. One day after transfection, the cells were collected by adding 200 μl of 0.53 mM EDTA-containing Dulbecco's PBS (D-PBS), followed by 200 μl of 5 mM HEPES (pH 7.4)-containing Hank's balanced salt solution. The cell suspension was transferred to a 96-well V-bottom plate in duplicate and fluorescently labeled with an Anti-DDDDK-tag mAb-Alexa Fluor 647 (Clone FLA-1, MBL Life Sciences; 5 μg/mL diluted in 2% goat serumand 2 mM EDTA-containing D-PBS (blocking buffer)). After washing with D-PBS, the cells were resuspended in 200 μl of 2 mM EDTA-containing-D-PBS and filtered through a 40-μm filter. The fluorescent intensity of single cells was quantified by an EC800 flow cytometer (Sony). The fluorescent signal derived from Alexa Fluor 647 was recorded in an FL3 channel, and the flow cytometry data were analyzed with the FlowJo software (FlowJo). Live cells were gated with a forward scatter (FS-Peak-Lin) cutoff at the 390 setting, with a gain value of 1.7. Values of mean fluorescence intensity (MFI) from approximately 20,000 cells per sample were used for analysis. For each replicate experiment, MFI counts of GALR2 mutant samples were normalized to those of GALR2 WT (100% level) and the mock-transfected samples (0% level), and the resulting values were used to denote surface expression levels of the mutants.

S1 Fig. Purification of recombinant GALR2-galanin-mGαq iN /Gβ1γ2-scFv16 complex.
After injecting the complex sample into an SEC column, 3 peaks appeared. The complex of GALR2, galanin, heterotrimeric mGαq iN /Gβ1γ2, and scFv16 was eluted at the first peak around 10.50 ml. The peak fraction was visualized using an SDS-PAGE. Uncropped gel of this figure can be found in S1 Raw Images.  (Fig 3), RAi of the mutant to that of WT was calculated from Emax and EC50 values, and its logarithm-transformed value (Log RAi) was used to represent Gq-activity parameter. The parameter of Y164A was not available due to the undetectable Gq activity of the mutant. Colors in the mutant bars indicate an expression level matching to that of titrated WT. Bars and error bars represent mean and SEM of 3 independent experiments (dots). NA, parameter not available because of lack of the ligand response. Statistical analyses were performed using the ordinary one-way ANOVA followed by the Dunnett's post hoc test with the expression-matched (colored) WT response. ns, p > 0.05; ��� , p < 0.001; ���� , p < 0.0001. The data underlying this figure can be found in S3 Data. MFI, mean fluorescence intensity; RAi, relative intrinsic activity; WT, wild-type.