Ligand recognition and G-protein coupling selectivity of cholecystokinin A receptor

Cholecystokinin A receptor (CCKAR) belongs to family A G-protein-coupled receptors and regulates nutrient homeostasis upon stimulation by cholecystokinin (CCK). It is an attractive drug target for gastrointestinal and metabolic diseases. One distinguishing feature of CCKAR is its ability to interact with a sulfated ligand and to couple with divergent G-protein subtypes, including Gs, Gi and Gq. However, the basis for G-protein coupling promiscuity and ligand recognition by CCKAR remains unknown. Here, we present three cryo-electron microscopy structures of sulfated CCK-8-activated CCKAR in complex with Gs, Gi and Gq heterotrimers, respectively. CCKAR presents a similar conformation in the three structures, whereas conformational differences in the ‘wavy hook’ of the Gα subunits and ICL3 of the receptor serve as determinants in G-protein coupling selectivity. Our findings provide a framework for understanding G-protein coupling promiscuity by CCKAR and uncover the mechanism of receptor recognition by sulfated CCK-8.

complexes. Three G-protein subtypes were engineered to stabilize the CCK A R-G protein complexes ( Supplementary Fig. 4). Gα q is chimerized by replacing its αN-helix with the equivalent region of Gα i1 to facilitate scFv16 binding 30 . Gα s was modified based on the mini-Gα s that was used in the crystal structure determination of the G s -coupled adenosine A 2A receptor (A 2A R) 31 . Two dominant-negative (DN) mutations (G203A and A326S 32 ) were introduced to Gα i1 , and corresponding DN mutations at equivalent sites of Gα s and Gα q were also introduced ( Supplementary Fig. 4b). Unless otherwise specified, G q , G s and G i refer to the respective engineered G proteins that are used in the CCK A R structure determination.
Globally, CCK A R adopts similar overall conformations in all three structures, with an all-atom root-mean-square deviation (r.m.s.d.) of 0.84 for G q /G s -coupled receptors and 1.03 for G q /G i -coupled receptors. The structure of the CCK-8-CCK A R-G q complex, which has the highest resolution at 2.9 Å, was used for detailed analysis and mechanistic evaluation of ligand recognition and receptor activation. The inactive and active structures of the closed homolog receptors (inactive: ghrelin receptor, PDB 6KO5 34 ; active: neurotensin receptor 1 (NTSR1), PDB 6OS9 18 ), all belonging to the β-branch of the rhodopsin family, are applied for structural comparison. CCK A R presents a fully active conformation, resembling the G i -coupled NTSR1, displaying an ~9-Å outward movement of TM6 (measured at the Cα of the residue at position 6.27 in CCK A R and the ghrelin receptor) and an ~4-Å inward shift of TM7 (Cα carbons of Y7.53) compared with the inactive ghrelin receptor (Extended Data Fig. 1a,b). Similar to the active NTSR1 complex, the conserved residues in the 'micro-switches' (PIF, ERY, CWxP and NPxxY) of CCK A R display the conserved conformations observed in active GPCRs (Extended Data Fig. 1c).
It is of interest that the octapeptide CCK-8 almost completely occupies the polypeptide-binding pocket, structurally supporting the fact that it is the smallest active form of CCK isoforms.  Fig. 1 | Cryo-EM structures of CCK A R-G protein complexes. a, Schematic of G-protein coupling promiscuity of CCK A R. b-g, Three-dimensional maps and models of the CCK-8-CCK A R-G q -scFv16 (b,c), CCK-8-CCK A R-G s (d,e) and CCK-8-CCK A R-G i -scFv16 (f,g) complexes. CCK-8, magenta; G q -coupled CCK A R, green (b,c); G s -coupled CCK A R, pink (d,e); G i -coupled CCK A R, dark yellow (f,g); Gα q , orange; Gα s , blue; Gα i , cyan; Gβ, light blue; Gγ, yellow; scFv16, light purple.
in G q /G s -coupled complexes, and 1.18 for CCK-8 in G q /G i -coupled complexes), supported by clear EM density maps ( Fig. 2a and Supplementary Fig. 5). The region of ligand recognition by CCK A R can be divided into three major parts: (1) the extracellular loops, (2) hydrophobic cavities beneath ECLs and (3) the bottom of the TMD pocket (Fig. 2a).
At the extracellular side, three ECLs are folded to embrace the N-terminal amino acids of CCK-8 (Fig. 2a). The sulfate group of ionic Y 2P interacts with the side chain of R197 ECL2 . This polar interaction prompts the aromatic ring of Y 2P to form hydrophobic contacts with F185 ECL2 , M195 ECL2 and the main chain of K105 ECL1 , thus connecting CCK-8 to ECL1 and ECL2 ( Fig. 2b and Extended Data Fig. 3). These structural observations are consistent with the previous finding that the R197 ECL2 M mutation was 1,470-fold less potent than the wild-type (WT) CCK A R 11 . The alanine mutation of R197 ECL2 completely abolishes the binding of CCK-8, thus strongly supporting the contention that R197 ECL2 serves as a determinant to discriminate between sulfated and non-sulfated CCK (Fig. 2c and  Supplementary Table 2). Likewise, poor ligand selectivity of CCK B R may be attributed to a substitution of arginine for valine at the corresponding position (Extended Data Fig. 4). Meanwhile, M 3P , G 4P and W 5P clamp the interior surface of ECL3 (Fig. 2b).
Two hydrophobic cavities exist below the ECLs to accommodate W 5P and M 6P (Fig. 2d,e). The side chain of W 5P is sandwiched by the side chains of I352 7.35 and R336 6.58 and buries in a deep hydrophobic pocket composed of TM6, ECL3 and TM7 (Fig. 2d). The backbone CO group of W 5P forms a hydrogen bond with R336 6.58 , and its indole nitrogen atom makes another hydrogen bond with N333 6.55 ( Fig. 2d and Extended Data Fig. 3), which is reported to be critical to CCK A R activation 38 . Alanine mutations in residues N333 6.55 , R336 6.58 , A343 ECL3 , E344 ECL3 , L347 ECL3 and S348 ECL3 completely abolish the binding of CCK-8, suggesting the key roles of these residues in CCK-8 recognition (Fig. 2c and Supplementary  Table 2). In contrast to the W 5P -occupied hydrophobic pocket, M 6P sits in a relatively shallow hydrophobic cavity in the opposite direction, constituted by F107 ECL1 , C196 ECL2 , T118 3.29 and M121 3.32 ( Fig. 2e and Extended Data Fig. 3). Mutating F107 ECL1 and residues in ECL2 and ECL3 to alanine eliminated the binding ability of CCK-8 entirely, highlighting an essential function of the three ECLs in peptide recognition ( Fig. 2c and Supplementary Table 2).
At the bottom of the binding pocket, D 7P and main chain CO group of CCK-8 form a stabilizing polar interaction network with TM5 (H210 5.39 ), TM6 (N333 6.55 and R336 6.58 ) and TM7 (Y360 7.43 ) ( Fig. 2f and Extended Data Fig. 3). The phenyl ring of F 8P makes a polar hydrogen-π interaction with Y176 4.60 , and inserts into a large hydrophobic crevice composed of residues from TM3, TM4, TM5 and TM6 ( Fig. 2f and Extended Data Fig. 3). Besides N333 6.55 and R336 6.58 , which also interact in a polar    manner with W 5P , I329 6.51 is closely related to CCK-8 binding ( Fig. 2c and Supplementary Table 2). Elucidation of the recognition mechanism of CCK-8 provides clues for therapeutic development against CCK A R. GW-5823, CE-326597 and Glaxo-11p (Extended Data Fig. 5a) are small-molecule agonists for CCK A R with moderate activities 10,39,40 . Docking of these agonists to the CCK A R shows that they only occupy the bottom half of the TMD binding pocket, thus lacking essential interactions with ECLs 1-3 of CCK A R (Extended Data Fig. 5b). This structural feature may lead to a weaker activity of these small-molecule agonists relative to CCK-8. Together, our data provide a framework for understanding the mechanism of small-molecule agonist recognition and offer a template for guiding drug design targeting CCK A R.

Overall coupling mode of CCK A R-G protein complexes
Although all four G-protein subtypes were reported to interact with CCK A R 24 , only three of the CCK A R-G protein samples (CCK A R-G q , CCK A R-G s and CCK A R-G i protein complexes) were obtained for high-resolution cryo-EM structure determination (Fig. 1). Structural comparison indicated that TM6 and ICL2 in CCK A R adopt nearly identical conformations in G q -, G i -and G s -coupled structures ( Fig. 3a and Extended Data Fig. 6a-c). However, slightly different tilts of the Gα α5-helix were seen among the three heterotrimeric G proteins (4° for Gα q /Gα s and 8° for Gα q /Gα i ; Fig. 3a). Meanwhile, the distal end of the Gα s α5-helix moves 7 Å outward, away from the TMD core relative to the equivalent Gα q residue (measured at the Cα atom of L H5. 25 , where superscripts refer to the common Gα numbering (CGN) system 41 ; Fig. 3a). G q presents the largest solvent-accessible surface area (SASA, 1,492 Å 2 ) with the receptor, compared to a G s value of 1,293 Å 2 and G i value of 1,167 Å 2 , consistent with a 6.6-to 20.3-fold increased potency of G q coupling to CCK A R in comparison to coupling with G s and G i (Supplementary Table 3). This finding supports the hypothesis that the size of the G-protein coupling interface may correlate with the ability of a receptor to link with different G proteins 21,22 . In addition, coupling of different G-protein subtypes exhibits distinct effects on CCK-8 binding. Compared to G s or G i proteins, G q coupling increases the binding affinity of CCK-8 (Supplementary Table 3), consistent with the increased binding activity of isoproterenol against β 2A R in the presence of G s protein 16 . This finding indicates an allosteric modulation effect of G q protein on CCK-8 binding, supporting the positive cooperativity between agonists and G proteins 42 .
In addition, comparisons of these three complex structures to previously reported G-protein-coupled class A GPCRs reveal the different extent of TM6 displacement and the concomitant shift of the Gα α5-helix ( Fig. 3b-d). TM6 of CCK A R in all three G-protein complexes displays an 11-12-Å (measured at the Cα atom of the residue at position 6.27) smaller outward displacement compared to G s -coupled GPCRs, which translates into a notable swing of the Gα α5-helix in the same direction (9-11° relative to G s -coupled β 2A R and A 2A R as measured at the Cα atom of Y H5. 23 ). This smaller displacement of TM6 is contrary to the previous assumption that TM6 of G s -coupled GPCRs undergoes a notable outward movement, thus opening a larger cytoplasmic pocket to accommodate bulkier residues at the distal end of the Gα s α5-helix relative to G i/o -coupled receptors 21,43 . To avoid a potential clash with TM6, the distal end of the Gα s α5-helix in the CCK A R-G s complex stretches away from the TMD core and inserts into the crevice between the TM6 and TM7-helix 8 joint. This featured conformation of the Gα s α5-helix in the CCK A R-G s complex is unique compared to that in structures of the G s -coupled β 2A R and A 2A R, supporting the complexity of the GPCR-G protein coupling mechanism (Fig. 3b).
TM6 and the Gα α5-helix of CCK A R-G protein complexes display similar conformational changes to other G i -and G q -coupled GPCRs, such as the G i -coupled NTSR1 and the G q -coupled 5-HT 2A R (Fig. 3c,d). TM6 of the CCKA R -G i protein complex is highly overlaid with that of G i -coupled NTSR1, while the cytoplasmic end of TM6 shows a 4-Å smaller outward displacement compared to that of G o -coupled M 2 R (Fig. 3c). On the G-protein side, the α5-helix of Gα i in the CCK A R-G i complex shows a nearly overlapped conformation with that of the NTSR1-G i complex. In contrast, it exhibits a 3-Å (measured at the Cα atom of Y H5.23 ) shift away from TM6 relative to that of G o -coupled M 2 R (Fig. 3c). Structural comparison of G q -coupled CCK A R with G q /G 11 -coupled GPCRs demonstrates a 2-Å (measured at the Cα atom of Y H5.23 ) upward shift toward the cytoplasmic cavity in comparison to the G q -coupled 5-HT 2A R and a 28° rotation away from TM6 relative to G 11 -coupled M 1 R (Fig. 3d).

Interactions of the 'wavy hook' of CCK A R-G protein complexes
The 'wavy hook' at the extreme C terminus of the Gα α5-helix is thought to be one of the coupling specificity determinants for G proteins 44,45 , and undergoes distinct conformational rearrangements among the three CCK A R-G protein complexes (Fig. 3a).
A structural comparison of the interaction interface between the receptor cytoplasmic cavity and the Gα 'wavy hook' reveals distinct features of CCK A R-G protein coupling. Well-defined densities of Gα-protein 'wavy hook' residues allow for detailed structural analyses except for residues at the -1 position. The L(-2) H5. 25 in the α5-helix is highly conserved across the G-protein families and plays a pivotal role in G-protein coupling. Both L358 H5. 25 in Gα q and L353 H5. 25 in Gα i hydrophobically interact with residues in TM3 and TM6 (R139 3.50 , I143 3.54 , V311 6.33 and L315 6.37 ; Fig. 4a,b). Owing to the notable displacement of the Gα s C terminus, L393 H5. 25 in Gα s moves 7 Å outward away from the TMD core relative to the equivalent Gα q residue (Fig. 3a), repositioning it in a hydrophobic subpocket formed by M314 6.36 and M373 7.56 (Fig. 4c). In contrast to L(-2) H5. 25 24 in Gα q makes a hydrogen bond with the backbone CO group of Y370 7.53 (Fig. 4a). As a result of the replacement of Gα i G352(-3) H5.24 and the repositioning of Gα s E392(-3) H5. 24 , the corresponding hydrogen bond is absent in CCK A R-G i and CCK A R-G s complex structures. Additionally, Y356(-4) H5. 23 in Gα q forms extensive interactions with the receptor cytoplasmic cavity by making hydrogen bonds with R139 3.50 and Q153 ICL2 (Fig. 4d). By contrast, C351(-4) H5. 23 in Gα i only forms a weak hydrogen bond with R139 3.50 via its backbone CO group (Fig. 4e). Y391(-4) H5. 23 in Gα s exhibits limited hydrophobic and van der Waals interactions with residues in TM2 and TM3 (T76 2.39 , R139 3.50 and A142 3.53 ; Fig. 4f). Furthermore, both E355(-5) H5. 22 in Gα q and D350(-5) H5. 22 in Gα i form salt bridges with R376 8.49 in CCK A R, while Q390(-5) H5. 22 in Gα s disfavors the formation of the corresponding electrostatic interaction (Fig. 4d-f). To understand the 'wavy hook'-mediated G-protein selectivity, we displaced the amino acids (H5. 22-H5.25) in the Gα q subunit with the corresponding ones in the Gα s and Gα i subunits. Bioluminescence resonance energy transfer (BRET) assay results show that the Gα i displacement has no impact on CCK A R-G protein coupling compared to the WT Gα q subunit. However, partial (E355Q or N357E) or complete Gα s substitution remarkably decreased the G-protein coupling activity of CCK A R (Fig. 4g). These results indicate that the 'wavy hook' may play a crucial role in the coupling selectivity of CCK A R with G q over the G s protein.

Contribution of CCK A R ICL3 to G q coupling selectivity
In the CCK A R-G q protein complex structure, CCK A R displays a comparable length of TM5 relative to the M 1 R-G 11 complex 19 . However, the cytoplasmic end of the CCK A R TM5 exhibits an 8-Å outward bend (measured at the Cα atoms of A 5.73 ), which prevents it from interacting with the Gα q subunit (Fig. 5a). Instead, the ICL3 inserts into the cleft between TM5 of CCK A R and the α5-helix of the Gα q subunit (Fig. 5a). Compared to L225 5.75 in M 1 R, I296 ICL3 in CCK A R interacts with the same hydrophobic patch formed by the side chains of Y325 S6.02 , F339 H5.06 and A342 H5.09 in the Gα q subunit, but is buried deeper to create more closely packed hydrophobic contacts (Fig. 5a,b). These hydrophobic interactions are critical to CCK A R-G q coupling, as evidenced by our BRET analysis   showing that the I296 ICL3 G mutation significantly weakens G q coupling to CCK A R but has no impact on G s and G i coupling (Fig. 5c and Supplementary Table 4). This hydrophobic patch, which lies on the outer surface, may be unique for the G q/11 subunit. The equivalent residues in the Gα s and Gα i subunits are polar or charged residues, for which it would be energetically unfavorable to form hydrophobic interactions (Extended Data Fig. 7a-d). Indeed, this unconventional ICL3-G q interaction is not seen in the structures of G s -and G i/o -coupled CCK A Rs (Fig. 3b,c). Together, our findings offer structural evidence on the possible role of ICL3 in CCK A R-G q coupling preference. Hydrophobic residues on the inner surface of the ICL3 loop of CCK A R or the extended TM5 of M 1 R may represent a common feature of G q/11 -coupled GPCRs.

Discussion
As the largest family of cell surface receptors, GPCRs have more than 800 members but only couple to four G-protein subtypes. Specific GPCR signaling requires the receptor to couple with either a single or multiple G-protein subtypes [45][46][47] . Thus, one of the main questions is how does a given GPCR select a G-protein subtype for downstream signal transduction. The critical G-protein determinants of selectivity vary widely for different receptors that couple to specific G proteins. It is thought that G s -or G q -coupled receptors are relatively promiscuous and to some extent couple to G i1 (ref. 22 ). However, G i -coupled receptors are more selective 22 . The minor outward movement of TM6 contributes to such a superior G i coupling selection in comparison to that of G s (refs. 17,23,44,48,49 ). Although proven to be promiscuous, G q -coupled receptors tend to adopt an active conformation similar to that of G i -coupled GPCRs, reflecting the complexity of the GPCR-G protein coupling mechanism 19,20 . Because CCK A R has the ability to couple with different G-protein subtypes, it stands out as a suitable model for studying the promiscuity of G-protein coupling. In this article, we show that TM6 of CCK A R undergoes a similar outward displacement relative to G i/o -coupled (NTSR1 and M 2 R) and G q/11 -coupled GPCRs (5-HT 2A R and M 1 R), but has a smaller shift relative to G s -coupled GPCRs (β 2A R and A 2A R). CCK A Rs share almost identical conformations, whereas G q , G s and G i proteins vary in distinct orientations, producing different sizes of receptor-G protein interface. The predominant coupling to G q by CCK A R can be explained by its largest interface of the three CCK A R-G protein complexes. Structural comparison of the three CCK A R-G protein complexes reveals that 'wavy hook' residues of the Gα α5-helix and ICL3 of the receptor are important for the coupling promiscuity. In addition, detailed inspections disclose structural clues relative to the recognition mechanism of sulfated CCK-8 by CCK A R, in which R197 ECL2 is a major determinant. Together, our structures provide a framework for better understanding ligand recognition as well as G-protein coupling selectivity and promiscuity by CCK A R.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41589-021-00841-3.

Expression and purification of CCK A R-G protein complexes.
The WT CCK A R (residues 1-428) was used for cryo-EM studies. Full-length CCK A R complementary DNA was cloned into a modified pFastBac vector (Invitrogen) containing a hemagglutinin (HA) signal sequence followed by an 8× histidine tag, a double-maltose binding protein tag and a tobacco etch virus (TEV) protease site before the receptor sequence using homologous recombination (using a CloneExpress One Step Cloning Kit, Vazyme; Supplementary Fig. 4a). The N-terminal 1-29 amino acids of Gα q were replaced by the equivalent residues of Gα i1 to facilitate scFv16 binding 19 . An engineered Gα s construct was generated based on mini-Gα s 31 . The N-terminal 1-18 amino acids and the α-helical domain of Gα s were replaced by human Gα i1 , thus providing binding sites for scFv16 and Fab-G50, respectively 17,19 . Additionally, human Gα i1 with two dominant-negative mutations (G203A and A326S 32 ) was used to assemble a stable GPCR-G i protein complex. These two cognate mutations also exist in engineered Gα q and Gα s (Supplementary Fig. 4b). Receptor, rat H6-Gβ, bovine Gγ and the specific Gα subunit were co-expressed in Spodoptera frugiperda (sf9) insect cells (Invitrogen) as previously described 50 . In addition, GST-Ric-8A (a gift from B. Kobilka) was applied to improve the expression of Gα q .
ScFv16 was applied to improve the protein stability of CCK A R-G q and CCK A R-G i complex samples. The monomeric scFv16 was prepared as previously reported 51 . Cell pellets of the co-expression culture were thawed and lysed in 20 mM HEPES, pH 7.4, 100 mM NaCl, 10% glycerol, 5 mM MgCl 2 and 10 mM CaCl 2 supplemented with EDTA-free protease inhibitor cocktail (TargetMol). CCK A R-G protein complexes were assembled at room temperature (r.t.) for 1 h by the addition of 10 μM CCK-8 (GenScript) and 25 mU ml −1 apyrase. The lysate was then solubilized in 0.5% lauryl maltose neopentyl glycol (LMNG), 0.1% cholesteryl hemisuccinate TRIS salt (CHS), and the soluble fraction was purified by nickel affinity chromatography (Ni Smart Beads 6FF, SMART Lifesciences). In the case of CCK A R-G i and CCK A R-G q complexes, a three-molar excess of scFv16 was added to the protein elute. The mixture was incubated with amylose resin for 2 h at 4 °C. The excess G protein and scFv16 were washed with 20 column volumes of 20 mM HEPES, pH 7.4, 100 mM NaCl, 10% glycerol, 0.01% LMNG, 0.002% CHS and 2 μM CCK-8. TEV protease was then included to remove the N-terminal fusion tags of CCK A R. After 1 h of incubation at r.t., the flow-through was collected, concentrated and injected onto a Superdex 200 10/300 column equilibrated in buffer containing 20 mM HEPES, pH 7.4, 100 mM NaCl, 0.00075% LMNG, 0.00025% glycol-diosgenin (GDN), 0.0002% CHS and 10 μM CCK-8. The monomeric complex peak was collected and concentrated to ~5 mg ml −1 for cryo-EM studies.

Cryo-electron microscopy grid preparation and image collection.
For preparation of cryo-EM grids, 2.5 µl of each purified CCK A R-G protein complex was applied individually onto glow-discharged holey carbon grids (Quantifoil, Au300 R1.2/1.3) in a Vitrobot chamber (FEI Vitrobot Mark IV). The chamber was set to 100% humidity at 4 °C. Extra samples were blotted for 2 s and vitrified by plunging into liquid ethane. Grids were stored in liquid nitrogen for condition screening and data collection usage.
Automatic data collection of CCK-8-CCK A R-G protein complexes was performed on an FEI Titan Krios system at 300 kV. The microscope was operated with a nominal magnification of ×81,000 in counting mode, corresponding to a pixel size of 1.045 Å for the micrographs. A total of 5,415 videos for the dataset of the CCK-8-CCK A R-G q -scFv16 complex, 5,008 for the dataset of the CCK-8-CCK A R-G s complex and 4,811 for the dataset of the CCK-8-CCK A R-G i -scFv16 complex were collected, respectively, by a Gatan K3 Summit direct electron detector with a Gatan energy filter (operated with a slit width of 20 eV; GIF) using SerialEM software. The images were recorded at a dose rate of ~26.7 e Å −2 s −1 with a defocus ranging from −0.5 to −3.0 μm. The total exposure time was 3 s and intermediate frames were recorded in intervals of 0.083 s, resulting in a total of 36 frames per micrograph.
Image processing and map reconstruction. Image stacks were subjected to beam-induced motion correction and aligned using MotionCor 2.1. Contrast transfer function (CTF) parameters were estimated by Ctffind4. Data processing was performed using RELION-3.0 52 . Micrographs with measured resolution worse than 4.0 Å and micrographs imaged within the carbon area were discarded, generating 3,806 micrographs for the CCK-8-CCK A R-G q -scFv16 dataset, 4,963 for the CCK-8-CCK A R-G s dataset and 4,543 for the CCK-8-CCK A R-G i -scFv16 dataset for further data processing. For particle selection, two-dimensional (2D) and 3D classifications were performed on a binned dataset with a pixel size of 2.09 Å. About 2,000 particles were manually selected and subjected to 2D classification. Representative averages were chosen as a template for particle autopicking. The autopicking process produced 3,405,355 particles for the CCK-8-CCK A R-G q -scFv16 complex, 4,680,972 for the CCK-8-CCK A R-G s complex and 4,270,010 for the CCK-8-CCK A R-G i -scFv16 complex, which were subjected to reference-free 2D classifications to discard bad particles. Initial reference map models for 3D classification were generated by Relion using representative 2D averages. For the CCK-8-CCK A R-G q -scFv16 complex, the particles selected from 2D classification were subjected to six rounds of 3D classification, resulting in a single well-defined subset with 555,628 particles. For the CCK-8-CCK A R-G s complex, particles resulting from 2D classification were subjected to five rounds of 3D classification, resulting in two well-defined subsets with 499,924 particles. For the CCK-8-CCK A R-G i -scFv16 complex, particles selected from 2D classification were subjected to seven rounds of 3D classifications, resulting in two well-defined subsets with 140,602 particles. Further 3D refinement, CTF refinement, Bayesian polishing and DeepEnhancer processing generated density maps with an indicated global resolution of 2.9 Å for the CCK-8-CCK A R-G q -scFv16 complex, 3.1 Å for the CCK-8-CCK A R-G s complex and 3.2 Å for the CCK-8-CCK A R-G i -scFv16 complex, respectively, at a Fourier shell correlation of 0.143.
Model building and refinement. For the CCK A R-G q complex, the initial G q protein and scFv16 model were adopted from the cryo-EM structure of the M 1 R-G 11 protein complex (PDB 6OIJ) 19 . The initial CCK A R model was generated by an online homology model building tool 53 . All models were docked into the EM density map using Chimera 54 , followed by iterative manual adjustment and rebuilding in COOT 55 and ISOLDE 56 , and real-space refinement using Phenix programs 57 . The model statistics were validated using Phenix comprehensive validation. A model of the refined CCK A R from the CCK A R-G q complex was used for the other two complexes. Models from PTH1R-G s (PDB 6NBF) and FPR2-G i (PDB 6OMM) were used as templates for the model building of G s in the CCK A R-G s complex and G i1 -scFv16 in the CCK A R-G i complex, respectively. The fitted models were then built in the same way as the CCK A R-G q complex. The final refinement statistics are provided in Supplementary Table 1. All figures were prepared using PyMol and Chimera software. G-protein dissociation assay. G-protein dissociation was monitored by BRET experiments performed as previously reported 58 . Briefly, a C-terminal fragment of the G-protein-coupled receptor kinase 3 (GRK3ct) fused to a luciferase serves as a BRET donor. Gβγ dimer is labeled with the fluorescent protein Venus, a BRET acceptor. Upon G-protein heterotrimer activation, free Gβγ-Venus is released and binds to membrane-associated GRK3ct-luciferase, leading to an increased signal detectable by BRET.
NanoBiT G-protein recruitment assay. The recruitment of CCK A R to G i protein was detected in sf9 cells using the NanoBiT method as previously reported 59 . Briefly, the LgBiT fragment of NanoBiT luciferase was fused to the C terminus of CCK A R. SmBiT was fused to the C terminus of the Gβ subunit with a 15-amino-acid flexible linker. CCK A R-LgBiT, Gα i1 , SmBiT-fused human Gβ1 and human Gγ2 were co-expressed in sf9 insect cells. Cell pellets were collected by centrifugation after infection for 48 h. The cell suspension was dispensed in a 96-well plate (64,000 cells per well) at a volume of 80 μl diluted in the assay buffer (Hanks' balanced salt solution buffer supplemented with 10 mM HEPES, pH 7.4) and incubated for 30 min at 37 °C. The cells were then reacted with 10 μl of 50 mM coelenterazine H (Yeasen) for 2 h at r.t. The luminescence signal was measured using an EnVision plate reader (PerkinElmer) at 30-s intervals (25 °C). The baseline was measured before CCK-8 addition for eight intervals, and the measurements continued for 20 intervals following ligand addition. Data were corrected to baseline measurements and the results were analyzed using GraphPad Prism 8.0 (Graphpad Software).
NanoBiT G-protein dissociation assay. G s activation was measured by a NanoBiT dissociation assay. G-protein NanoBiT split luciferase constructs were generated by fusing the LgBiT in Gα s and the SmBiT to Gγ (a gift from A. Inoue, Tohoku University) as previously reported 60 . In brief, HEK 293T/17 cells were plated in 10-cm plates at a density of 3 × 10 6 cells per plate. After 24 h, cells were transfected with 1.62 μg receptor plasmids, 0.81 μg Gα s -LgBiT, 4.1 μg Gβ and 4.1 μg SmBiT-Gγ using Lipofectamine LTX reagent (Invitrogen). The transiently transfected cells were then seeded into poly-d-lysine-coated 96-well plates (50,000 cells per well) and grown overnight before incubation in assay buffer. Measurement of the luminescence signal was identical to the steps described above.
Surface expression assay. HEK 293T/17 cells were seeded into a six-well plate and incubated overnight. After transient transfection with WT or mutant plasmids for 24 h, the cells were collected and blocked with 5% BSA in PBS at r.t. for 15 min and incubated with primary anti-Flag antibody (1:300, Sigma-Aldrich) at r.t. for 1 h. The cells were then washed three times with PBS containing 1% BSA followed by 1 h of incubation with donkey anti-mouse Alexa Fluor 488-conjugated secondary antibody (1:1,000, Thermo Fisher) at 4 °C in the dark. After three washes, the cells were resuspended in 200 µl of PBS containing 1% BSA for detection in a NovoCyte flow cytometer (ACEA Biosciences) utilizing laser excitation and emission wavelengths of 488 nm and 519 nm, respectively. For each assay point, ~15,000 cellular events were collected, and the total fluorescence intensity of the positive expression cell population was calculated. The gating strategy and the method for the calculation of expression are shown in Supplementary Fig. 6. Molecular docking. Before docking, hydrogens were added to CCK A R and the whole system coordinates were optimized with a pH of 7.0. A grid file was then generated on the peptide pocket in our G q -coupled CCK A R structure. Small-molecule ligands Glaxo-11p, GW-5823 and CE-326597 were prepared in the OPLS3 force field with a pH of 7.0 to generate 3D structures. Finally, glide docking with standard precision was applied to all ligands and the structures with the best docking score were picked as outputs.
Statistics. All functional study data were analyzed using Prism 8 (GraphPad) and are shown as the mean ± s.e.m. from at least three independent experiments. Concentration-response curves were evaluated with a three-parameter logistic equation. Significance was determined with either a two-tailed Student's t-test or one-way analysis of variance Dunnett multiple comparisons test, and P < 0.05 was considered statistically significant.