A unique hormonal recognition feature of the human glucagon-like peptide-2 receptor

Glucagon-like peptides (GLP-1 and GLP-2) are two proglucagon-derived intestinal hormones that mediate distinct physiological functions through two related receptors (GLP-1R and GLP-2R) which are important drug targets for metabolic disorders and Crohn’s disease, respectively. Despite great progress in GLP-1R structure determination, our understanding on the differences of peptide binding and signal transduction between these two receptors remains elusive. Here we report the electron microscopy structure of the human GLP-2R in complex with GLP-2 and a Gs heterotrimer. To accommodate GLP-2 rather than GLP-1, GLP-2R fine-tunes the conformations of the extracellular parts of transmembrane helices (TMs) 1, 5, 7 and extracellular loop 1 (ECL1). In contrast to GLP-1, the N-terminal histidine of GLP-2 penetrates into the receptor core with a unique orientation. The middle region of GLP-2 engages with TM1 and TM7 more extensively than with ECL2, and the GLP-2 C-terminus closely attaches to ECL1, which is the most protruded among 9 class B G protein-coupled receptors (GPCRs). Functional studies revealed that the above three segments of GLP-2 are essential for GLP-2 recognition and receptor activation, especially the middle region. These results provide new insights into the molecular basis of ligand specificity in class B GPCRs and may facilitate the development of more specific therapeutics.

INTRODUCTION GLP-2R belongs to class B GPCR subfamily, which is mainly expressed in the gut, pancreas and brain. 1 Its endogenous ligand is GLP-2 (GLP-2(1-33)), a member of the glucagon-like peptide family that also includes GLP-1 and glucagon. GLP-2 is encoded by the proglucagon gene and becomes active after post-translational processing by prohormone convertases. As a gastrointestinal hormone, it mainly regulates intestinal epithelial cell growth and functions that are crucial for digestion and absorption of nutrients. 2 Similar to GLP-1, GLP-2 is rapidly inactivated by dipeptidyl peptidase 4 in vivo. 3 Clinically, GLP-2 analogue, teduglutide, is used to treat short bowel syndrome and Crohn's disease. Other therapeutic indications, including colitis, pediatric gastrointestinal disorders and inflammation of the intestinal mucosa, are currently under development. 4 Recent studies in animal models have shown that GLP-2 also exerts beneficial effects on glucose homeostasis, 5,6 spontaneous hypertension [7][8][9] and depression. [10][11][12] This prompts a renewed interest in examining the role of this peptide beyond the gastrointestinal tract.
The cryo-electron microscopy (cryo-EM) emerges as a primary methodology to determine the structure of GPCR-G protein complexes starting with the GLP-1R 13 and the calcitonin receptor, 14 followed by eight other class B GPCRs in complex with various G proteins. [15][16][17][18][19][20] These structures suggest that class B GPCRs have a common activation pattern which basically coincides with the previously proposed two-domain binding model. 21 Although class B receptors have roughly the same activation mode, the molecular details are receptor specific, especially in the ligand-binding region. Therefore, exploring the recognition mechanism between GLP-2 and GLP-2R is of interest for further understanding the mechanism of ligand recognition and receptor activation among class B GPCRs. In this study, we employed cryo-EM to determine the high-resolution structure of the human GLP-2R in complex with G s protein. Based on our prior experience, 22 NanoBiT strategy 23,24 was used in this study to stabilize the GLP-2-GLP-2R-G s complex and strengthen the interactions between GLP-2R and Gβ, resulting in a cryo-EM structure at a global resolution of 3.0 Å (Supplementary information, Table S1). Our data provide a rational template to facilitate the design of better GLP-2 analogues for therapeutic use and expand our knowledge on the biology of this receptor family.

Structure determination
We used NanoBiT strategy [22][23][24] to stabilize the complex, in which GLP-2R was directly attached to LgBiT and rat Gβ was linked to HiBiT with a 15-amino acid linker in between. Transiently transfected HEK 293T cells showed that the C-terminustruncated GLP-2R(1-490)-LgBiT had a 6-fold stronger interaction with Gβ-SmBiT than the wild-type (WT) receptor(1-553)-LgBiT (Supplementary information, Fig. S1a, b). When SmBiT was replaced by HiBiT, the luminescence signal detected between GLP-2R and Gβ was 10-fold stronger (Supplementary information, Fig. S1c). A haemagglutinin (HA) signal peptide was added to the N-terminus of GLP-2R to increase the expression level. In addition to linking with LgBiT, a TEV protease site and an OMBP (optimized maltose binding protein)-MBP tag were attached to the Cterminus (Supplementary information, Fig. S2a). The OMBP tag is codon-optimized on the basis of MBP to increase its expression in insect cells. At the same time, using OMBP-MBP tag can enhance affinity between the protein complex and amylose resin in order to obtain more complexes. Rat Gβ1 was attached to HiBiT with a 15-amino acid linker between them (Supplementary information, Fig. S2b). To form an active G protein-coupled complex, HA-GLP-2R(1-490)-LgBiT-TEV-OMBP-MBP was co-expressed with a dominant-negative bovine Gα s (DNGα s ), rat Gβ1-HiBiT and bovine Gγ2 in Sf9 insect cells. Compared with the protein expressed without NanoBiT strategy, the yield decreased but the stability increased (Supplementary information, Fig. S1d, e). Negativestaining EM visualization displayed more particles with similar size (Supplementary information, Fig. S1f).
A large-scale purification was subsequently performed using this approach. Cells were stimulated by 20 μM GLP-2 together with the camelid-derived nanobody Nb35 to stabilize the interface between Gα s and Gβ. 25 The complex was solubilized in lauryl maltose neopentyl glycol (LMNG) and cholesteryl hemi succinate (CHS), treated with TEV enzyme to remove tags from GLP-2R and then isolated by amylose resin. It was further purified by sizeexclusion chromatography for cryo-EM specimen preparation (Supplementary information, Fig. S3a, b). The production of cAMP and ligand-binding ability of our modified hGLP-2R construct were reduced presumably caused by its lower expression level (~73% of WT; Supplementary information, Fig. S3c, d and Table S2).
Frozen-hydrated GLP-2-GLP-2R-G s complexes were imaged using a Titan Krios microscope (Supplementary information, Fig. S4a). 2D classification showed clear secondary structure features and random distribution of particles in different directions enabling high-resolution cryo-EM map reconstruction (Supplementary information, Fig. S4b). A total of 284,669 particles were selected after consecutive 3D classifications, which yielded a cryo-EM density map with a global resolution of 3.0 Å (Supplementary information, Fig. S4c, d).
Overall architecture In the cryo-EM structure of the GLP-2-GLP-2R-G s complex, most of the receptor core, G protein subunits and the first 31 amino acids of GLP-2 were observed with high-resolution features, thereby allowing for accurate modeling of side-chain rotamers ( Fig. 1;  Supplementary information, Fig. S5). The receptor extracellular domain (ECD) was omitted from the final reported structure because of the relatively low resolution owning to its intrinsic flexibility, which is a general feature in most reported class B GPCR-G s complex structures.
The overall structure of active GLP-2R-G s complex is similar to other class B GPCR-G s complexes such as GLP-1-GLP-1R-G s , 13 ExP5-GLP-1R-G s , 17 glucagon-GCGR-G s , 18 LA-PTH-PTH1R-G s 19 and GHRH-GHRHR-G s , 26 with root mean square deviation (RMSD) values of 1.76, 1.05, 1.55, 1.09 and 0.95 Å for the whole complex, respectively. When pointing toward the glucagon receptor subfamily, notable conformational differences were observed in the extracellular half of GLP-2R where the peptide hormone binds, especially in TM1, ECL1 and TM7, indicating that GLP-2R utilizes a distinct peptide-recognition mode specific for GLP-2, but not for GLP-1, glucagon or GHRH.
The middle region of GLP-2 forms both a polar network with several charged and polar amino acids in TM7 (K414 7.35b and R417 7.38b ) and ECL2 (T332 ECL2 and N334 ECL2 ) via Gly 4P , Ser 5P , Asp 8P , Glu 9P and Asn 11P and a complementary nonpolar network with TM1 (R171 1.32b , Y172 1.33b , L175 1.36b , L178 1.39b and Q179 1.40b ) via Phe 6P , Met 10P , Ile 13P and Leu 14P (Fig. 2a). Intriguingly, we observed one water molecule bridging Ser 7P and D232 2.68b , which is absent in the GLP-1-GLP-1R structure. 13 When comparing the active GLP-2R structure with the GLP-1-bound GLP-1R structure, 13 the extracellular end of TM1 in GLP-2R is extended by six additional residues, thereby bending toward GLP-2 with the formation of additional hydrogen bond and hydrophobic contacts. These interactions further stabilize the peptide conformation in the bound state and may contribute to specific recognition of GLP-2.
Partial unwinding of the short α-helix elevates ECL1 to an extended conformation that stands upwards in line with TM2 and TM3, thereby engaging the C-terminal half of GLP-2 through an extensive interface spanning from Leu 14P to Trp 25P (Figs. 2a, 3). Specifically, Asp 21P forms a salt bridge with R242 ECL1 , while Trp 25P forms a hydrogen bond with D244 ECL1 and pi-pi stacking with W249 ECL1 , which was stabilized by Phe 22P . Besides, Leu 14P and Ala 18P pack against Y239 ECL1 , R242 ECL1 and Y252 ECL1 . These interactions strengthen the binding and function of GLP-2, consistent with the decreased potencies of R242A (by 5-fold), R242E (by 20-fold), W249A (by 1552-fold) and Y252A (by 5-fold) (Fig. 2d). Similar effects of mutations were observed with the binding of teduglutide (Table 1). Together, these results highlight the importance of ECL1 in GLP-2 recognition and receptor activation. GPCRs reveals substantial similarity in the G protein-binding interface, consistent with a common mechanism of G s protein engagement. However, many receptor-specific interactions were observed. Y391 GαH5 is an essential residue in determining G protein coupling specificity by forming H-bond in class A GPCRs, 30 however, this interaction was not observed among available class B GPCR-G protein complexes. Most strikingly, Y391 GαH5 formed a tight hydrogen bond with a water molecule, which connects to E281 3.50b and H214 2.50b in the GLP-2R-G s complex (Supplementary information, Fig. S6), a phenomenon that was not observed among other class B GPCRs. Polar interactions also occurred between K368 5.64b and Q384 GαH5 , H372 ICL3 and R385 GαH5 as well as R382 6.37b and E392 GαH5 . The ICL2 of GLP-2R formed additional hydrophobic interactions with GαH5. There were also limited interactions between GαH5 subunit and helix 8 of the receptor (E392 GαH5 , N440 8.47b and G441 8.48b ). These receptor-specific interactions may contribute to the G protein specificity of GLP-2R.
Ligand specificity GLP-2 and GLP-1 share high sequence similarity (69.7%) and have almost identical residues in two regions: the N-terminus (first six residues in GLP-2) that penetrates into the TMD core, and the C-terminus (residues 18-33) that is recognized by the ECD and ECL1 of the receptor. On the contrary, the central region of the peptide mainly interacts with TMs 1, 2, 7, ECL1 and ECL2 of the receptor and differs to that of GLP-1 (   information, Fig. S7c). GLP-1 faces toward TM5 to form cation-pi stacking and hydrogen bond with the rotated down side-chain of R299 ECL2 , while that of W306 5.36b and R310 5.40b switch out from the receptor core to accommodate the approaching of histidine. However, by replacing arginine with threonine, the ECL2 of GLP-2R is no longer able to dip into the receptor core and becomes distant from the N-terminal histidine of GLP-2. Therefore, W340 5.36b and R344 5.40b of GLP-2R rotate inwards. Together with the inward movement of the extracellular part of TM5, they push the N-terminus of GLP-2 away from TM5 to reside closely with TM3, where H268 3.37b contributes a hydrogen bond to stabilize the conformation (Supplementary information, Fig. S7c). Compared to GLP-1, the middle region of GLP-2 has more extensive contacts with the ligand-binding pocket (Fig. 4b).
The interface areas between TMD and GLP-2 contributed by TM1 and TM7 are 864 Å 2 and 432 Å 2 , respectively, markedly larger than those between TMD and GLP-1 (330 Å 2 and 303 Å 2 , respectively). Specifically, in the extracellular half of GLP-2R, TM1 forms at least one additional helical turn, thereby moving toward TM2 (5 Å, measured by Cα of the residue at 1.33b) and bending down toward GLP-2. This ultimately leads to a much larger interaction pattern for GLP-2 (ranging from Asp 3P to Asn 16P ) than for GLP-1, where only two residues of GLP-1 interact with TM1.
Meanwhile, side-chain orientations of TM1 residues are reorganized (Supplementary information, Fig. S7d). For example, Y182 1.43b is reoriented~90°from an outside facing position to a position pointing to TM2 and engages T-shaped stacking with    of GLP-2R is unavailable, its impact on the structure and dynamics remains elusive.
Considering the uneven sequence conservation and distinct binding modes of GLP-2 and GLP-1, we designed chimeric GLP-2/ GLP-1 peptides to explore the molecular basis of ligand specificity. As shown in Fig. 4c, replacement of the first six residues of GLP-2 by those of GLP-1 (i.e., double mutant D 3P E and S 5P T) decreased cAMP accumulation by 12-fold compared to the native peptide, suggesting that subtle substitution may directly alter receptor activation. In addition, chimeric GLP-2 whose C-terminal fragment (residues 20-33) was replaced with that of GLP-1 displayed a 17fold potency reduction in GLP-2R signaling but successfully elicited GLP-1R-mediated cAMP accumulation (Fig. 4c). Replacement of the GLP-2 middle region (residues 7-19) that mainly interacts with TM1, TM7 and ECL1 by the corresponding part of GLP-1, almost abolished GLP-2R-mediated cAMP accumulation but fully rescued that induced by GLP-1R (Fig. 4c), indicative of a fundamental role of this region in determining the specificity for ligand recognition.
Conformational changes Triggered by GLP-2 binding and G protein coupling, GLP-2R would presumably undergo significant conformational changes. It shares a high degree of sequence similarity in the TMD region with GLP-1R (61%) that has been studied extensively both structurally and functionally. 31 Thus, inactive, intermediate and active GLP-1R structures published previously 13,17,32-35 provide good references for the present study. Superimposing the TMD of GLP-2R with that of GLP-1R reveals that activated GLP-2R shows a similar conformation to that of GLP-1-bound or ExP5-bound GLP-1R-G s complexes 13,17 (Cα RMSD = 1.41 and 0.91 Å, respectively), distinct from inactive GLP-1R 33 (Cα RMSD = 1.79 Å). Further comparison demonstrates that diversified peptide-receptor interactions converge to induce conserved and extensive conformational transition of the TMD of GLP-2R and GLP-1R (Fig. 5).
In the extracellular half of GLP-2R, accompanying the unwinding of the last two turns of TM6, TM7 moves outward and bends toward TM6 via a conserved pivot point G429 7.50b , which expands the TMD-binding pocket to accommodate peptide entrance and connection. Meanwhile, TM1 moves inwards and folds down toward GLP-2, which contributes both polar and hydrophobic contacts to stabilize the binding. Furthermore, different from the ordered α-helical ECL1 in inactive GLP-1R 33 and β-hairpin in inactive GCGR 36 that cover the binding pocket and probably block peptide insertion, ECL1 of active GLP-2R, GLP-1R and GCGR simultaneously move outward and stand upwards in line with TM2 and TM3. Together with inward movement of α-helical extended TM2, raised ECL1 of GLP-2R forms extensive interaction with the Cterminal half of GLP-2 (Fig. 4a, b) and was shown to be essential for peptide binding and receptor activation (Fig. 4c). As far as the intracellular half is concerned, the positions of TM1-2-3 of active GLP-2R resemble that of active GLP-1R (Fig. 5). The most profound conformational change is the sharp kink in the middle of TM6, leading to an outward movement of the intracellular portion of TM6 as measured by Cα carbon of Lys 6.35b (19.0 Å, similar to that of other G s -coupled class B receptors). In contrast, in the absence of G protein coupling, the binding of peptide 5 promotes a rigidly outward shift of the intracellular tip of TM6 of GLP-1R by 12.1 Å without a sharp kink, 34 thereby creating an intracellular crevice for G protein coupling. For PTH1R 37 and GCGR 38 (whose crystal structures were determined with PGS domain fusion in ICL3 and T4L fusion at ICL2, respectively), agonist binding failed to promote the outward movement of the intracellular half of TM6 as seen in the inactive state, probably due to the fusion. With more structural and dynamics studies on class B GPCRs, the activation process from inactive to active states via an intermediate state, i.e., allosteric communication between extracellular peptide binding and intracellular G protein coupling, will be better understood.

Role of ECL1
Relative to ECL2 and ECL3, ECL1 in class B GPCRs is the most divergent extracellular loop with flexible lengths, ranging from 8 (CRF2R) to 27 (PTH1R) residues. Inspired by the intriguing ECL1 conformation of GLP-2R, we performed structural comparison of ECL1 across class B GPCRs to unveil its functionality (Fig. 6). Among 20 agonist-bound structures from 9 receptors, we calculated the height of ECL1 (defined as the maximum vertical distance of Cα carbon relative to the membrane layer) and the ECL1-agonist interface area. We found that GLP-2R has the most protruded ECL1 with a height of 22.6 Å and this ECL1 closely attaches to GLP-2 with the largest interface area of 791 Å 2 . The ECL1 of all activated GLP-1R and GCGR structures consistently adopt α-helical conformation bridging extended TM2 and TM3 helices by two short linkers to form extensive interactions with bound agonists. Remarkably, significant conformational changes of ECL1 upon receptor activation were observed relative to their inactive conformations: (i) in the inactive GCGR, ECL1 forms a βhairpin conformation and further compacts β-sheet with the stalk region; and (ii) in the inactive GLP-1R, ECL1 moves toward TM1 and connects ECD in its closed TMD-interacting conformation. As far as CRF1R, CRF2R, AM1R, AM2R and CLR are concerned, their ECL1 are relatively short (consisting of 8-11 residues) and can only form one-turn α-helix to connect the adjacent transmembrane helices TM2 and TM3, contributing limited contacts with the peptides. PTH1R has the longest ECL1, but it is unstructured in both LA-PTH-PTH1R-G s 19 and ePTH-PTH1R 37 structures, probably owing to the great mobility of PTH1R's ECD and specific binding mode, consistent with the mutagenesis studies. 37,39,40 To date, four PAC1R structures were reported, namely, PACAP38-PAC1R-G s (3.0 Å), 20 PACAP-PAC1R-mini-G s (3.9 Å), 41 PACAP38-PAC1R-G s (3.5 Å) 42 and maxadilan-PAC1R-G s (3.6 Å). 42 The ECL1 was not resolved in the first two structures, and the third map showed that the ECL1 closely interacted with PACAP38 (pituitary adenylate cyclase activating polypeptide-38) without detailed residue contact information due to limited resolution. 42 Interestingly, maxadilan, a peptidic PAC1R agonist, reorganized the ECD orientation and reshaped the ligand-binding pocket conformation, including ECL1 42 (Fig. 6b). Collectively, these data demonstrate the diversity and flexibility of ECL1 among class B GPCRs.
To further explore the functional importance of ECL1 in ligand binding and receptor activation, we performed mutagenesis studies on this part of the receptor (Table 1; Supplementary information, Fig. S8 and Table S4). Replacement of GLP-2R ECL1 by poly-alanine abolished the ability of GLP-2 to elicit cAMP response, while replacement by that of GLP-1R diminished its potency by 196-fold. Meanwhile, single-point mutations on deeply buried residues (R242, W249 and Y252) in the interface between GLP-2 and ECL1 also demonstrated negative impact on ligand binding (Supplementary information, Table S5).
In conclusion, the structure of the GLP-2-GLP-2R-G s complex delineates detailed interactions between GLP-2 and GLP-2R that account for the peptide binding specificity, and provides new insights into the structural reorganization of class B GPCRs upon Fig. 6 Family-wide comparison of ECL1 conformations. a Scatter plot of ECL1 from 9 class B GPCRs. x axis shows the interface area between ECL1 and agonist, and y axis is the height of ECL1. b Structural comparison of diversified ECL1 conformations among class B GPCRs. All structures are superimposed on the GLP-1-bound GLP-1R (PDB: 5VAI) 13 using the Cα carbons of the residues in the TM2-3-4. The locations of the membrane planes predicted by OPM (Orientations of Proteins in Membranes) 54 are shown as black dashed lines. The height of ECL1 is defined as the maximum vertical distance of Cα carbon relative to the membrane layer, while the interface areas were calculated using freeSASA. 53 activation. Combined with previous research on the GLP-1R, our results reveal that subtle differences in the sequence of TMD binding pocket and the consequent diversified conformational changes of the extracellular half accommodate peptide binding and empower the receptor to be specifically trigged by the cognate hormone. This complex structure points to an essential role of ECL1 in peptide recognition and receptor activation, in addition to that of ECD. This work also highlights the applicability of the NanoBiT tethering strategy to stabilize GPCR-G protein complexes for structural studies.

Constructs
WT human GLP-2R was modified to include a HA signal peptide at the N-terminus. LgBiT and an OMBP-MBP tag were added to the C-terminus of the receptor with a TEV site between them. A bovine Gα s was used with G226A and A366S mutations to stabilize the complex; 43 rat Gβ1 was attached to HiBiT with a 15-amino acid linker between them; bovine Gγ2 was also used in cloning. All of the above constructs were generated in pFastBac and pcDNA3.1 vectors for protein expression in insect cells and functional assays in mammalian cells, respectively. Nb35 production E. coli BL21 cells were transformed with the plasmid containing Nb35 target gene and were grown in TB medium with 100 μg/mL ampicillin, 1 M MgCl 2 , 2% (w/v) glucose at 37°C, 180 rpm for 3 h. IPTG (1 M) was added to induce the expression when OD600 reached 0.7-1.2. After 8 h expression, the cell pellet was collected and stored at −80°C until use. Nb35 was purified by nickel affinity chromatography as previously described, 25 followed by size-exclusion chromatography using a HiLoad 16/600 Superdex 75 column (GE Healthcare). The column was pre-equilibrated with 20 mM HEPES, pH 7.4 and 100 mM NaCl. Glycerin 30% (v/v) was added to collect Nb35 which was frozen in liquid nitrogen and stored at −80°C.
Negative-staining electron microscopy Purified GLP-2R-G s complex was diluted to 0.012 mg/mL with 20 mM HEPES, pH 7.4, 100 mM NaCl, 5 μM GLP-2, 100 μM TCEP, 2 mM MgCl 2 , 0.00075% (w/v) LMNG, 0.00025% (w/v) GDN and 0.0002% (w/v) CHS. Eight microliters of protein complex and 8 μL uranium formate were applied to a glow-discharged 300-mesh cooper grid (EMCN, Beijing, China) for 1 min and then sucked with filter paper. After grilling the grid with a lamp, the samples were imaged at RT with a Talos L120C electron microscope (FEI, Hillsboro, USA) operated at 120 kV. Images were recorded at magnification of 57,000× and a defocus value of −1 μm. Data processing and 3D reconstructions Dose-fractionated image stacks were subjected to beam-induced motion correction using MotionCor2.1. 44 A sum of all frames, filtered according to the exposure dose, in each image stack was used for further processing. Contrast transfer function parameters for each micrograph were determined by Gctf v1.06. 45 Particle selection, 2D and 3D classifications were performed on a binned dataset with a pixel size of 2.028 Å using RELION-3.0-beta2. 46 Auto-picking yielded 1,498,893 particle projections were subjected to reference-free 2D classification to discard false positive particles or particles categorized in poorly defined classes, producing 485,548 particle projections for further processing. This subset of particle projections was subjected to a round of maximumlikelihood-based three dimensional classifications with a pixel size of 2.028 Å, resulting in one well-defined subsets with 162,932 projections. Further 3D classifications with mask on the complex produced two good subsets accounting for 332,946 particles, which were subsequently subjected to a round of 3D classifications with mask on the receptor. A selected subset containing 284,669 projections was then subjected to 3D refinement and Bayesian polishing with a pixel size of 1.014 Å. After last round of refinement, the final map has an indicated global resolution of 3.0 Å at a Fourier shell correlation (FSC) of 0.143. Local resolution was determined using the Bsoft package with half maps as input maps. 47 Model building The initial template of the TMDs in GLP-2R was generated using SWISS-MODEL. 48 The models of G s protein and Nb35 from PTH1R complex structure (PDB: 6NBF) 19 were used as the relevant components of the GLP-2R-G s complex models. While the whole TMDs, G s protein, and Nb35 showed decent density, the ECDs of GLP-2R complex structures displayed very poor density in the whole density maps and were thus omitted in the final structural models. All the models were docked into the EM density map using UCSF Chimera. 49 Model adjustment and rebuilding were performed in COOT. 50 Real space and Rosetta refinements were carried out using Phenix software package. 51 All residues were checked for fitting in electron density, and Ramachandran and rotamer restraints. The refined models exhibited good model-density fitting and model geometry (see Supplementary information, Table S1 for data and model refinement statistics). cAMP accumulation HEK 293T cells were cultured in Dulbecco's Modified Eagle's Medium (DMEM) (Gibco, Waltham, USA) supplemented with 10% (v/v) fetal bovine serum (FBS) (Gibco) and 1% (v/v) sodium pyruvate (Gibco) at 37°C, 5% CO 2 in a humidified incubator. They were seeded onto 6-well cell culture plates and transiently transfected with different GLP-2R constructs using Lipofectamine 2000 transfection reagent (Invitrogen, Carlsbad, CA, USA). After 24 h, the transfected cells were seeded onto 384-well microtiter plates (3000 cells per well) in HBSS supplemented with 5 mM HEPES, 0.1% (w/v) bovine serum albumin (BSA) and 0.5 mM 3isobutyl-1-methylxanthine, and stimulated with increasing concentrations of GLP-2 for 30 min at RT. Eu and Ulight were diluted by cAMP detection buffer and added to the plates separately to terminate the reaction. Plates were incubated at RT for 60 min before measuring the fluorescence intensity at 620 nm and 650 nm by an EnVision multilabel plate reader (PerkinElmer, Boston, USA).

Cryo-EM
NanoBiT assay HEK 293T cells were seeded onto 6-well cell culture plates in DMEM containing 10% FBS and 1% sodium pyruvate at 37°C, 5% CO 2 before transfection. They were transiently transfected with truncated GLP-2 receptors attached to LgBiT and Gβ-SmBiT (or Gβ-HiBiT) in a ratio of 1:1 using Lipofectamine 2000 transfection reagent (Invitrogen) and incubated for 24 h. After seeding onto 96-well microtiter plates (48,000 cells per well), the medium was replaced by HBSS followed by 30 min incubation at 37°C. Coelenterazine H (5 μM) was then added to the plates and luminescence signals measured 1 h thereafter using an EnVision multilabel plate reader (PerkinElmer) at 18 s intervals before and after ligand addition (25°C).

Oxidative iodination
The GLP-2 radioligand was created by oxidative iodination using ChloramineT to incorporate the iodine isotope [ 125 I] at position Tyr10 of human GLP-2(1-33, M10Y). A stepwise oxidation reaction was performed by addition of 6 aliquots of 5 µL ChloramineT (30 µL/mL) with a 1-min interval during constant stirring. The reaction was terminated by the addition of 400 µL phase A (0.1% trifluoracetic acid) before being fractionated into 1 mL volume by reverse-phase high-performance liquid chromatography (RP-HPLC, Waters 600 SimiPrep system) on a C18 column. The iodinated protein was assembled by increasing concentrations of phase B (Acetonitrile + 0.1% TFA) and tested for competitive binding ability.

Structural analysis
The interface area was calculated by FreeSASA 53 using the Sharke-Rupley algorithm with a probe radius of 1.2 Å. Structural figures were prepared using UCSF Chimera 49 and PyMOL (The PyMOL Molecular Graphics System v.2.1). RMSD analysis was performed in PyMOL.

DATA AVAILABILITY
The atomic coordinates and the electron microscopy maps have been deposited in the Protein Data Bank (PDB) under accession number 7D68 and Electron Microscopy Data Bank (EMDB) under accession number EMD-30590, respectively. All relevant data are available from the authors upon request and/or included in the manuscript or Supplementary information.