research letters\(\def\hfill{\hskip 5em}\def\hfil{\hskip 3em}\def\eqno#1{\hfil {#1}}\)

IUCrJ
Volume 6| Part 6| November 2019| Pages 996-1006
ISSN: 2052-2525

Mutagenesis facilitated crystallization of GLP-1R

aShanghai Key Laboratory of Regulatory Biology, Institute of Biomedical Sciences and School of Life Sciences, East China Normal University, Shanghai 200241, People's Republic of China, biHuman Institute, ShanghaiTech University, 393 Middle Huaxia Road, Shanghai 201210, People's Republic of China, cComplex Systems Division, Beijing Computational Science Research Center, Beijing 100193, People's Republic of China, and dSchool of Life Science and Technology, ShanghaiTech University, Shanghai 201210, People's Republic of China
*Correspondence e-mail: gjsong@bio.ecnu.edu.cn

Edited by J. Trewhella, University of Sydney, Australia (Received 26 July 2019; accepted 1 October 2019; online 17 October 2019)

The class B family of G-protein-coupled receptors (GPCRs) has long been a paradigm for peptide hormone recognition and signal transduction. One class B GPCR, the glucagon-like peptide-1 receptor (GLP-1R), has been considered as an anti-diabetes drug target and there are several peptidic drugs available for the treatment of this overwhelming disease. The previously determined structures of inactive GLP-1R in complex with two negative allosteric modulators include ten thermal-stabilizing mutations that were selected from a total of 98 designed mutations. Here we systematically summarize all 98 mutations we have tested and the results suggest that the mutagenesis strategy that strengthens inter-helical hydro­phobic interactions shows the highest success rate. We further investigate four back mutations by thermal-shift assay, crystallization and molecular dynamic simulations, and conclude that mutation I1962.66bF increases thermal stability intrinsically and that mutation S2714.47bA decreases crystal packing entropy extrinsically, while mutations S1932.63bC and M2333.36bC may be dispensable since these two cysteines are not di­sulfide-linked. Our results indicate intrinsic connections between different regions of GPCR transmembrane helices and the current data suggest a general mutagenesis principle for structural determination of GPCRs and other membrane proteins.

1. Introduction

Type 2 diabetes is a long-term metabolic disorder that is predicted to affect ∼10% of the adult population by 2030 (Shaw et al., 2010[Shaw, J. E., Sicree, R. A. & Zimmet, P. Z. (2010). Diabetes Res. Clin. Pract. 87, 4-14.]). Major causes of type 2 diabetes are a lack of insulin or insulin resistance, both of which cause high blood sugar. Current treatments of type 2 diabetes include injection of insulin or peptide agonists of glucagon-like peptide-1 receptor (GLP-1R) that provoke the synthesis and release of insulin (Gutniak et al., 1992[Gutniak, M., Ørkov, C., Holst, J. J., Ahrén, B. & Efendić, S. (1992). N. Engl. J. Med. 326, 1316-1322.]). GLP-1R belongs to the class B family of G-protein-coupled receptors (GPCRs), a group characterized by a 120–160 residue folded extracellular domain (ECD) followed by a canonical transmembrane domain (TMD) (Graaf et al., 2016[Graaf, C. de, Donnelly, D., Wootten, D., Lau, J., Sexton, P. M., Miller, L. J., Ahn, J. M., Liao, J., Fletcher, M. M., Yang, D., Brown, A. J., Zhou, C., Deng, J. & Wang, M. W. (2016). Pharmacol. Rev. 68, 954-1013.]). The widely accepted two-domain binding model suggests the ECDs of class B receptors recognize the C-terminal helices of their hormone peptide ligands, thus facilitating binding of the N-terminal region of the peptides to the TMDs for downstream signaling (Hoare, 2005[Hoare, S. R. (2005). Drug Discov. Today, 10, 417-427.]). Currently, discovery of anti-diabetes drugs is limited to peptide agonists including GLP-1, extendin-4 and their analogs (Pabreja et al., 2014[Pabreja, K., Mohd, M. A., Koole, C., Wootten, D. & Furness, S. G. (2014). Br. J. Pharmacol. 171, 1114-1128.]), while previously the development of small-molecule drugs to target this receptor was extremely challenging because of the lack of structural information about druggable small-molecule binding sites.

Two TMD structures of class B receptors [corticotropin-releasing factor 1 (CRF1R) and glucagon receptor (GCGR)] (Hollenstein et al., 2013[Hollenstein, K., Kean, J., Bortolato, A., Cheng, R. K., Doré, A. S., Jazayeri, A., Cooke, R. M., Weir, M. & Marshall, F. H. (2013). Nature, 499, 438-443.]; Siu et al., 2013[Siu, F. Y., He, M., de Graaf, C., Han, G. W., Yang, D., Zhang, Z., Zhou, C., Xu, Q., Wacker, D., Joseph, J. S., Liu, W., Lau, J., Cherezov, V., Katritch, V., Wang, M. W. & Stevens, R. C. (2013). Nature, 499, 444-449.]) were solved in 2013 and these structures revealed interesting diversities within this subfamily. The GCGR structure features an extraordinary extended helix (called the stalk region) at the N-terminus of transmembrane helix 1 (TM1) and a tilted helix 8 region at the C-terminal end, whereas the crystallized CRF1R is truncated before helix 8 making it impossible to reveal its conformation. In 2016, a second structure of thermal-stabilized GCGR was solved but the construct lacked the C-terminal three-helical turns of helix 8 and the N-terminus of the stalk region in TM1 (Jazayeri et al., 2016[Jazayeri, A., Doré, A. S., Lamb, D., Krishnamurthy, H., Southall, S. M., Baig, A. H., Bortolato, A., Koglin, M., Robertson, N. J., Errey, J. C., Andrews, S. P., Teobald, I., Brown, A. J., Cooke, R. M., Weir, M. & Marshall, F. H. (2016). Nature, 533, 274-277.]). The recently published crystal structures of GLP-1R TMD (Song et al., 2017[Song, G., Yang, D., Wang, Y., de Graaf, C., Zhou, Q., Jiang, S., Liu, K., Cai, X., Dai, A., Lin, G., Liu, D., Wu, F., Wu, Y., Zhao, S., Ye, L., Han, G. W., Lau, J., Wu, B., Hanson, M. A., Liu, Z. J., Wang, M. W. & Stevens, R. C. (2017). Nature, 546, 312-315.]), full length GCGR (Zhang, Qiao et al., 2017[Zhang, H., Qiao, A., Yang, D., Yang, L., Dai, A., de Graaf, C., Reedtz-Runge, S., Dharmarajan, V., Zhang, H., Han, G. W., Grant, T. D., Sierra, R. G., Weierstall, U., Nelson, G., Liu, W., Wu, Y., Ma, L., Cai, X., Lin, G., Wu, X., Geng, Z., Dong, Y., Song, G., Griffin, P. R., Lau, J., Cherezov, V., Yang, H., Hanson, M. A., Stevens, R. C., Zhao, Q., Jiang, H., Wang, M. W. & Wu, B. (2017). Nature, 546, 259-264.]), full length GLP-1R (Jazayeri et al., 2017[Jazayeri, A., Rappas, M., Brown, A. J. H., Kean, J., Errey, J. C., Robertson, N. J., Fiez-Vandal, C., Andrews, S. P., Congreve, M., Bortolato, A., Mason, J. S., Baig, A. H., Teobald, I., Doré, A. S., Weir, M., Cooke, R. M. & Marshall, F. H. (2017). Nature, 546, 254-258.]), and cryoEM structures of Gs-protein-bound GLP-1R (Zhang, Sun et al., 2017[Zhang, Y., Sun, B., Feng, D., Hu, H., Chu, M., Qu, Q., Tarrasch, J. T., Li, S., Sun Kobilka, T., Kobilka, B. K. & Skiniotis, G. (2017). Nature, 546, 248-253.]), calcitonin receptor (CTR) (Shihoya et al., 2016[Shihoya, W., Nishizawa, T., Okuta, A., Tani, K., Dohmae, N., Fujiyoshi, Y., Nureki, O. & Doi, T. (2016). Nature, 537, 363-368.]) and parathyroid hormone receptor 1 (PTH1R) (Zhao et al., 2019[Zhao, L. H., Ma, S., Sutkeviciute, I., Shen, D. D., Zhou, X. E., de Waal, P. W., Li, C. Y., Kang, Y., Clark, L. J., Jean-Alphonse, F. G., White, A. D., Yang, D., Dai, A., Cai, X., Chen, J., Li, C., Jiang, Y., Watanabe, T., Gardella, T. J., Melcher, K., Wang, M. W., Vilardaga, J. P., Xu, H. E. & Zhang, Y. (2019). Science, 364, 148-153.]) have extended our understanding of the structural determinants of class B GPCR function and modulation by small molecules, peptide ligands and functional antibodies.

In the past 12 years, the determination of more than 40 novel GPCR crystal structures has provided valuable information for understanding GPCR signaling and drug discovery (Thal et al., 2018[Thal, D. M., Vuckovic, Z., Draper-Joyce, C. J., Liang, Y. L., Glukhova, A., Christopoulos, A. & Sexton, P. M. (2018). Curr. Opin. Struct. Biol. 51, 28-34.]). However, crystallization is still hampered by the highly dynamic nature of GCPRs in solution and the existence of multiple metastable states. To enhance the stability of GPCRs, thermal-stabilizing mutations were introduced to the expression construct for crystallization. Structures solved by this approach include the β1-adrenergic receptor (Warne et al., 2008[Warne, T., Serrano-Vega, M. J., Baker, J. G., Moukhametzianov, R., Edwards, P. C., Henderson, R., Leslie, A. G., Tate, C. G. & Schertler, G. F. (2008). Nature, 454, 486-491.]), agonist-bound adenosine A2A (Lebon & Tate, 2011[Lebon, G. & Tate, C. G. (2011). Med. Sci. 27, 926-928.]), neurotensin receptor 1 (White et al., 2012[White, J. F., Noinaj, N., Shibata, Y., Love, J., Kloss, B, Xu, F., Gvozdenovic Jeremic, J., Shah, P., Shiloach, J., Tate, C. G. & Grisshammer, R. (2012). Nature, 490, 508-513.]), metabotropic glutamate 5 (mGluR5) (Doré et al., 2014[Doré, A. S., Okrasa, K., Patel, J. C., Serrano-Vega, M., Bennett, K., Cooke, R. M., Errey, J. C., Jazayeri, A., Khan, S., Tehan, B., Weir, M., Wiggin, G. R. & Marshall, F. H. (2014). Nature, 511, 557-562.]), CRF1R (Hollenstein et al., 2013[Hollenstein, K., Kean, J., Bortolato, A., Cheng, R. K., Doré, A. S., Jazayeri, A., Cooke, R. M., Weir, M. & Marshall, F. H. (2013). Nature, 499, 438-443.]), free fatty-acid receptor 1 (FFAR1, also known as GPR40) (Srivastava et al., 2014[Srivastava, A., Yano, J., Hirozane, Y., Kefala, G., Gruswitz, F., Snell, G., Lane, W., Ivetac, A., Aertgeerts, K., Nguyen, J., Jennings, A. & Okada, K. (2014). Nature, 513, 124-127.]), GCGR (Jazayeri et al., 2016[Jazayeri, A., Doré, A. S., Lamb, D., Krishnamurthy, H., Southall, S. M., Baig, A. H., Bortolato, A., Koglin, M., Robertson, N. J., Errey, J. C., Andrews, S. P., Teobald, I., Brown, A. J., Cooke, R. M., Weir, M. & Marshall, F. H. (2016). Nature, 533, 274-277.]), CC chemokine receptor 9 (CCR9) (Oswald et al., 2016[Oswald, C., Rappas, M., Kean, J., Doré, A. S., Errey, J. C., Bennett, K., Deflorian, F., Christopher, J. A., Jazayeri, A., Mason, J. S., Congreve, M., Cooke, R. M. & Marshall, F. H. (2016). Nature, 540, 462-465.]), protease-activated receptor 2 (PAR2) (Cheng et al., 2017[Cheng, R. K. Y., Fiez-Vandal, C., Schlenker, O., Edman, K., Aggeler, B., Brown, D. G., Brown, G. A., Cooke, R. M., Dumelin, C. E., Doré, A. S., Geschwindner, S., Grebner, C., Hermansson, N. O., Jazayeri, A., Johansson, P., Leong, L., Prihandoko, R., Rappas, M., Soutter, H., Snijder, A., Sundström, L., Tehan, B., Thornton, P., Troast, D., Wiggin, G., Zhukov, A., Marshall, F. H. & Dekker, N. (2017). Nature, 545, 112-115.]), GLP-1R TMD (Song et al., 2017[Song, G., Yang, D., Wang, Y., de Graaf, C., Zhou, Q., Jiang, S., Liu, K., Cai, X., Dai, A., Lin, G., Liu, D., Wu, F., Wu, Y., Zhao, S., Ye, L., Han, G. W., Lau, J., Wu, B., Hanson, M. A., Liu, Z. J., Wang, M. W. & Stevens, R. C. (2017). Nature, 546, 312-315.]) and the recently solved melatonin receptor 1 and 2 (Stauch et al., 2019[Stauch, B., Johansson, L. C., McCorvy, J. D., Patel, N., Han, G. W., Huang, X. P., Gati, C., Batyuk, A., Slocum, S. T., Ishchenko, A., Brehm, W., White, T. A., Michaelian, N., Madsen, C., Zhu, L., Grant, T. D., Grandner, J. M., Shiriaeva, A., Olsen, R. H. J., Tribo, A. R., Yous, S., Stevens, R. C., Weierstall, U., Katritch, V., Roth, B. L., Liu, W. & Cherezov, V. (2019). Nature, 569, 284-288.]; Johansson et al., 2019[Johansson, L. C., Stauch, B., McCorvy, J. D., Han, G. W., Patel, N., Huang, X. P., Batyuk, A., Gati, C., Slocum, S. T., Li, C., Grandner, J. M., Hao, S., Olsen, R. H. J., Tribo, A. R., Zaare, S., Zhu, L., Zatsepin, N. A., Weierstall, U., Yous, S., Stevens, R. C., Liu, W., Roth, B. L., Katritch, V. & Cherezov, V. (2019). Nature, 569, 289-292.]). Wild-type TMD of GLP-1R (residue range 128–431) is extremely unstable as suggested by size-exclusion chromatography (SEC) and thermal-shift assay (Song et al., 2017[Song, G., Yang, D., Wang, Y., de Graaf, C., Zhou, Q., Jiang, S., Liu, K., Cai, X., Dai, A., Lin, G., Liu, D., Wu, F., Wu, Y., Zhao, S., Ye, L., Han, G. W., Lau, J., Wu, B., Hanson, M. A., Liu, Z. J., Wang, M. W. & Stevens, R. C. (2017). Nature, 546, 312-315.]). We have previously described that ten thermal-stabilizing mutations were introduced in the TMD structure of GLP-1R in complex with two negative allosteric modulators (NAMs), PF-06372222 and NNC0640, and we showed that a di­sulfide bond (I3175.47bC—G3616.50bC) and a GCGR mimicking mutation C3476.36bF are indispensable for crystallization (Song et al., 2017[Song, G., Yang, D., Wang, Y., de Graaf, C., Zhou, Q., Jiang, S., Liu, K., Cai, X., Dai, A., Lin, G., Liu, D., Wu, F., Wu, Y., Zhao, S., Ye, L., Han, G. W., Lau, J., Wu, B., Hanson, M. A., Liu, Z. J., Wang, M. W. & Stevens, R. C. (2017). Nature, 546, 312-315.]). Here, we further summarize all of the 98 mutations that we have tested, analyze the effects of thermal-stabilizing mutations and describe attempts to crystallize several constructs with back mutations. Our experimental data combined with molecular dynamic (MD) simulations suggest general principles for mutagenesis design to increase thermal stability and crystallization success rates of GPCRs and other membrane proteins.

2. Results

2.1. Mutation overview

We built a model of GLP-1R based on a previously solved GCGR structure (PDB entry 4l6r; Siu et al., 2013[Siu, F. Y., He, M., de Graaf, C., Han, G. W., Yang, D., Zhang, Z., Zhou, C., Xu, Q., Wacker, D., Joseph, J. S., Liu, W., Lau, J., Cherezov, V., Katritch, V., Wang, M. W. & Stevens, R. C. (2013). Nature, 499, 444-449.]) that we used as a template for mutagenesis design to stabilize the transmembrane bundle, especially the thermodynamic regions revealed in the homologous GCGR structure (e.g. the extracellular halves of TM3–6). The principle for mutagenesis design was to strengthen inter-helical interaction patterns by predicting hydrogen bonds (salt bridges), hydrophobic interactions and di­sulfide bonds, or to strengthen ligand–receptor interaction patterns by covalent bonds or other interactions. Besides manual prediction based on modeling, the prediction of di­sulfide bonds was also assisted by a di­sulfide prediction algorithm (Pu et al., 2018[Pu, M., Xu, Z., Peng, Y., Hou, Y., Liu, D., Wang, Y., Liu, H., Song, G. & Liu, Z. J. (2018). Protein Cell, 9, 659-663.]). In each round beneficial mutations were passed on to the next round after consideration of monodispersity (the percentage of monomeric fraction in total fractions in SEC), protein yield (the height of monomeric fraction in SEC) and thermal stability (the melting temperature). At earlier stages we mainly selected the mutations with improved homogeneities (using protein yield as another reference), while at later stages we mostly considered the thermal stabilities of mutations since monodispersities were already sufficiently high (see the Methods section[link]).

Briefly, the results show that while most of these double-cysteine mutations were unfavorable in the constructs, we successfully screened two pairs (I3175.47bC—G3616.50bC, S1932.63bC—M2333.36bC) that significantly increased protein homogeneity [see Figs. S1(a), S1(b) and Table S1 in the Supporting information], correlating with the strict restriction of the Cβ–Cβ distance and dihedral angle for di­sulfide bonds. In contrast, most of the single point mutations showed moderate effects, and a relatively higher proportion of mutations increased protein homogeneity [Figs. S1(c)–S1(e) and Table S2]. Specifically, we observed eight single point mutations that aided protein homogeneity through hydrophobic interactions, whereas mutations by predicting inter-helical hydrogen bonds or ligand–receptor covalent bonds are yet to be successful. Actually, all six single point mutations in the final crystallization construct were mutated either from hydrophilic to hydrophobic residues (S2253.28bA, S2714.47bA, G3185.48bI, K3466.35bA), or from hydrophobic to bulky hydrophobic residues (I1962.66bF, C3476.36bF). The design rationale for these ten mutations is further analyzed below, while other mutations are summarized in Tables S1, S2 and Fig. S1.

2.2. Construct design

Since glycine residues usually provide flexibility for conformational equilibrium of GPCRs, we mutated Gly3185.48b to isoleucine (G3185.48bI) to facilitate crystallization based on sequence alignment within class B receptors [Fig. 1[link](a)]. Furthermore, Gly3616.50b in TM6 was mutated to cysteine to form a di­sulfide bond with I3175.47bC for linking the middle region of TM5 and TM6, the most thermodynamic region in the previously solved GCGR structure [Fig. 1[link](a)]. The mutation S2253.28bA fitted the nearby hydrophobic environment provided by Ile1962.66b, Leu2243.27b, Leu2283.31b and Val2293.32b, while I1962.66bF further stabilized the orientations of TM2 and TM3 by forming a patch of hydro­phobic interactions with TM3 residues [Fig. 1[link](b)]. Ser2714.47b was mutated to alanine as it was facing the lipid bilayer [Fig. 1[link](a)]. Lys3466.35b is located at the intracellular tip of TM6, and the K3466.35bA mutation makes it stay in a close position with the corresponding residues Leu2543.57b, Leu2553.58b and Lys3345.64b that mimic the inactive conformation of GPCRs [Fig. 1[link](c)]. C3476.36bF is a GCGR mimicking mutation used to strengthen the hydrophobic interaction with NAMs and Lys3516.40b based on previous MK0893-bound GCGR structure [Fig. 1[link](d)]. Besides these two cysteine mutations that successfully formed a di­sulfide bond (I3175.47bC—G3616.50bC), our attempts to introduce another di­sulfide bond pair (S1932.63bC—M2333.36bC) was not successful according to their densities in the solved crystal structure. This finding indicated that these two cysteines may function independently to improve the monodispersity or thermal stability at that stage of construct optimization. Interestingly, in the solved structure we found that the Cys2333.36b was oxidized into sulfinic acid (CSD2333.36b) during crystallization [Fig. 1[link](b)], a modification that was frequently observed, e.g. in the structure of Medicago sativa chalcone synthase (Ferrer et al., 1999[Ferrer, J. L., Jez, J. M., Bowman, M. E., Dixon, R. A. & Noel, J. P. (1999). Nat. Struct. Biol. 6, 775-784.]).

[Figure 1]
Figure 1
Locations of the ten thermal-stabilized mutations. (a)–(d) The original structure of inactive GLP-1R in complex with PF-06372222 (PDB entry 5vew; Song et al., 2017[Song, G., Yang, D., Wang, Y., de Graaf, C., Zhou, Q., Jiang, S., Liu, K., Cai, X., Dai, A., Lin, G., Liu, D., Wu, F., Wu, Y., Zhao, S., Ye, L., Han, G. W., Lau, J., Wu, B., Hanson, M. A., Liu, Z. J., Wang, M. W. & Stevens, R. C. (2017). Nature, 546, 312-315.]) is shown as a gray cartoon, with mutations as green sticks and other interacting residues as gray sticks. Mutations analyzed in the current study are underlined.

To study the effects of these sites we made further constructs in which we mutated back single point (C2333.36bM, named M9 hereafter), double points (C1932.63bS/C2333.36bM, named M8 hereafter) and quadruple points (C1932.63bS/C2333.36bM/A2714.47bS/F1962.66bI, named M6 hereafter). These mutants covered the sites where the engineered mutations seemed unnecessary since a di­sulfide bond was not formed (C1932.63bS/C2333.36bM), as well as the sites that affect either intramolecular interactions (F1962.66bI) or intermolecular interactions during crystal packing (A2714.47bS). These constructs were purified, characterized, crystallized and compared with the construct (named M10 hereafter) in the previously published crystal structure (Song et al., 2017[Song, G., Yang, D., Wang, Y., de Graaf, C., Zhou, Q., Jiang, S., Liu, K., Cai, X., Dai, A., Lin, G., Liu, D., Wu, F., Wu, Y., Zhao, S., Ye, L., Han, G. W., Lau, J., Wu, B., Hanson, M. A., Liu, Z. J., Wang, M. W. & Stevens, R. C. (2017). Nature, 546, 312-315.]).

2.3. Thermal stability

All mutants were expressed in insect cells and purified to high homogeneity [Fig. 2[link](a)]. We then measured the thermal stabilities of these proteins in complex with PF-06372222 by a fluorescence-based thermal-shift assay [Fig. 2[link](b)]. As a control, we also measured the thermal-shift data of apo proteins whose melting temperatures decreased by 5–8°C, suggesting that the PF-06372222 indeed stabilized the receptor in each mutant. Fig. 2[link](b) shows that compared with M10, the M9 and M8 constructs have similar and slightly lower melting temperatures, respectively, which is consistent with our finding that residues Cys1932.63b and Cys2333.36b are not di­sulfide linked. In contrast, further removal of the other two cysteines, which are di­sulfide linked (C1932.63bS/C2333.36bM/C3175.47bI/C3616.50bG, named M6-ND hereafter), induced a decrease of ∼9°C in the melting temperature, and the M6-ND protein could not be crystallized in our crystallization trials. The similar melting temperatures of M9 and M10 suggest that the endogenous M233 residue and the modified cysteine residue (CSD233) contributed similarly to thermal stability, while the ∼2°C decrease of M8 and M6 indicates that the back mutations of C1932.63bS, A2714.47bS and F1962.66bI slightly affected the protein stability in solution.

[Figure 2]
Figure 2
Thermal-shift assay of GLP-1R mutants, and potential mechanisms of I1962.66bF and S2714.47bA in the thermal stability and crystallization of GLP-1R. (a) SEC of GLP-1R fusion proteins suggests that the GLP-1R mutants are mostly monomeric and of similar homogeneity. (b) Thermal-shift assay of GLP-1R mutants in apo state or in complex with ligand PF-06372222. The different melting temperatures of the GLP-1R mutants indicating these mutations significantly affect thermal stability. (c) Densities of representative mutations in mutant structures. (d) B factor and distribution map of four constructs. (e) I1962.66bF mutation stabilized GLP-1R through its connections with the central polar network and other regions. (f) Packing and asymmetric unit of the four crystallized constructs. The back mutation of A2714.47bS induced significant change in the cell contents of M6 (yellow) compared with the other three back mutations and disturbed the packing. The four structures are superimposed through chain b. In Figs. 2[link](e) and 2[link](f) the color codes are shown in the key at the center.

2.4. Crystallization

The three new constructs were crystallized in complex with PF-06372222 in the same condition as the M10 crystals (Fig. S2). GPCR crystals are prone to radiation damage and decay easily under a synchrotron X-ray source, and so one usually has to collect and merge data from many high-diffraction-quality crystals. Furthermore, the non-symmetric space group of P1 makes the data collection of GLP-1R crystals challenging and time consuming. While the previous M10 crystals were collected to a completeness of 95%, we could only collect data of the three new constructs to a relatively low level of completeness (M9, 88.8%; M8, 77.9% and M6, 79.9%) because of limited resources. Despite fewer measured crystals, the available data show comparable statistics with the M10 crystals, e.g. similar Rmerge and I/σ values (Table 1[link]). The new crystals were processed to the same space group and structures were determined by molecular replacement successfully. The densities of key side chains indicated successful mutagenesis of specified residues [Fig. 2[link](c)]. The cell contents of M9 and M8 are within 2% variation compared with M10, whereas the differences between M6 and the other three are larger. For example, the b axis of M6 (71.1 Å) is 4.7 Å longer than that of M10 (66.4 Å), while the c axis of M6 is 2.4 Å shorter (Table 1[link]). The crystals of M9 and M8 were both processed to 2.8 Å resolutions and their structures were refined to an Rwork/Rfree of 0.247/0.280 and 0.244/0.290, respectively. Notably, the M6 crystals were apparently worse than the others and the collected M6 data were cut to 3.1 Å using the same criteria (CC1/2 > 0.6). Indeed, our attempts to include higher resolution (<3.1 Å) data of M6 generated a bad density map and an even worse Rwork/Rfree. The 3.1 Å M6 structure was finally refined to 0.257 (Rwork) and 0.303 (Rfree); in line with this, the B factor of M6 (107.9 Å2) is also significantly higher than the other three crystallized constructs (87–97 Å2) [Table 1[link] and Fig. 2[link](d)].

Table 1
Data-collection and refinement statistics of GLP-1R mutants

Data collection        
Construct M10 M9 M8 M6
Mutations S193C, I196F, M233C, S271A, S225A, G318I, K346A, C347F, I317C—G361C S193C, I196F, S271A, S225A, G318I, K346A, C347F, I317C—G361C I196F, S271A, S225A, G318I, K346A, C347F, I317C—G361C S225A, G318I, K346A, C347F, I317C—G361C
Number of crystals 25 17 18 15
Space group P1 P1 P1 P1
Cell dimensions        
a, b, c (Å) 64.8, 66.4, 83.4 65.0, 68.3, 83.4 64.9, 67.4, 83.7 65.2, 71.1, 81.0
α, β, γ (°) 90.5, 90.2, 107.7 91.5, 90.3, 106.5 91.07, 90.10, 107.9 92.5, 92.6, 105.1
Total reflections 133127 74289 80566 43577
Unique reflections 34615 30099 25859 20302
Resolution (Å) 50.0–2.7 (2.85–2.7) 45.2–2.8 (2.95–2.80) 49.50–2.80 (2.95–2.80) 41.20–3.10 (3.27–3.10)
Rmerge 0.12 (0.51) 0.12 (0.57) 0.13 (0.40) 0.11 (0.46)
Mean I/σ(I) 6.2 (1.4) 5.3 (1.4) 5.3 (2.2) 4.6 (1.6)
Completeness (%) 95.2 (84.2) 88.8 (79.7) 77.9 (70.9) 79.9 (74.5)
Redundancy 3.8 (1.9) 2.5 (1.8) 3.1 (2.6) 2.1 (1.9)
CC1/2 0.99 (0.61) 0.98 (0.62) 0.98 (0.76) 0.99 (0.61)
         
Refinement        
Resolution (Å) 30.0–2.7 29.8–2.8 49.5–2.8 40.4–3.1
Number of reflections (test) 34567 (1743) 30036 (1324) 25796 (1136) 20218 (955)
Rwork/Rfree (%) 22.8/24.6 24.7/28.0 24.4/29.0 25.7/30.3
Average protein B factor (Å2) 97 87.9 90.7 107.9
Number of atoms (A, B)        
 Protein 3302, 3305 3302, 3305 3302, 3315 3300, 3290
 Ligand 37, 37 37, 37 37, 37 37, 37
 Lipid and other 96, 72 31, 27 4, 3 0, 0
RMS deviation        
 Bond lengths (Å) 0.01 0.004 0.009 0.013
 Bond angles (°) 0.90 0.73 1.441 1.557
Ramachandran plot (%)§        
 Favored regions 94.0 95.7 95.0 91.6
 Allowed regions 6.0 4.3 5.0 8.3
 Disallowed regions 0.0 0.0 0.0 0.1
PDB entry 5vew 6kjv 6kkl 6kk7
†Bold indicates the sites where we carried out back mutation in this study.
‡The highest resolution shell is shown in parentheses.
§As defined in MolProbity.
¶This PDB entry has been reported previously (Song et al., 2017[Song, G., Yang, D., Wang, Y., de Graaf, C., Zhou, Q., Jiang, S., Liu, K., Cai, X., Dai, A., Lin, G., Liu, D., Wu, F., Wu, Y., Zhao, S., Ye, L., Han, G. W., Lau, J., Wu, B., Hanson, M. A., Liu, Z. J., Wang, M. W. & Stevens, R. C. (2017). Nature, 546, 312-315.]).

Compared with M8, M6 shared a similar thermal stability but much worse crystal diffraction, which must be a consequence of the two back mutations in M6, F1962.66bI and A2714.47bS. Reflecting on the positions of these two residues, we reasoned that the mutations F1962.66bI and A2714.47bS affected intrinsic thermal stability and extrinsic crystal packing, respectively. Phe1962.66b was located in the extracellular half of TM2, and the bulky phenyl group formed stronger hydrophobic interactions with nearby residues Tyr220ECL1, Ala2253.28b, Val2293.32b, Ala1992.69b and Ala2002.70b [Fig. 2[link](e)], thus the back mutation of F1962.66bI weakened the inter-helical interactions and thermal stability of the TM2–ECL1–TM3 region. In contrast, Ala2714.47b sat in the interface of TM4 with the C-terminal tail of TM6 in the symmetry-related molecule [Fig. 2[link](f)]. Compared with alanine, the additional hydroxyl group of A2714.47bS pushed the symmetry-related molecule away mostly along the b axis, and the opposing residue Ile3666.55b moved by 2.7 Å. This explained the apparent distinct cell contents of M6 crystals mentioned above. Moreover, the incompatibility between Ser2714.47b and opposing hydrophobic residues including Ile3666.55b generated high entropy that was unfavorable for crystal packing. Nevertheless, the effects of A2714.47bS in solution are very limited, as suggested by M6's slightly decreased thermal stability compared with other mutants [Fig. 2[link](b)]. To study further the effects of these four mutations we conducted MD simulations to investigate the dynamics of GLP-1R mutants in aqueous environments.

2.5. MD simulations

MD simulations were carried out to investigate the effects of mutations on the stability of the GLP-1R TMD. Since the fusion partner T4 lysozyme (T4L) was included in each expression construct for thermal-stability measurement, we also included T4L in the MD simulations so the results are comparable to each other. To perturb the system minimally, three new mutants were built based on the M10 structure and the rotamers of these mutated residues were chosen based on the densities in their corresponding crystal structures (M9, M8 and M6). In the M10 structure, the special residue CSD2333.36b, whose force field parameters were not available in the database of CHARMM36, was substituted by wild-type cysteine residue during simulations.

According to root-mean-square deviation (RMSD) analysis, the structure variations for T4L were within 2 Å compared with the initial structures for all four systems, indicating that the T4L domain was highly stable in all constructs (Fig. S3). While the TMDs of the three constructs (M10, M9 and M8) exhibited similar conformational fluctuations, the structural changes for TMs of M6 were relatively large (3 Å), indicating that the back mutations in M6 destabilized the TMD of the receptor (Figs. 3[link] and S3). The 2D RMSD cluster analysis revealed that the top five clusters of M6 covered only 60% of total populations during the 200 ns simulation, whereas in the other mutants, top five clusters could cover >80% of total populations (Fig. S4). These simulations suggest larger conformational fluctuations of M6 compared with the other three constructs, thus providing a possible explanation for the low diffraction quality of M6. The structural changes of M6 mainly occurred in TMs 6 and 7 [Figs. 3[link](f) and 3[link](g)], while TMs 1–5 [Figs. 3[link](a)–3(e)] remained stable throughout the simulation; this is seemingly in contrast with the fact that these mutations were located at TM2 (C1932.63bS/F1962.66bI), TM3 (C2333.36bM) and TM4 (A2714.47bS). However, the structure of GLP-1R revealed strong intrinsic interactions between TM1–5 and TM6–7, and these interactions were necessary for GPCR dynamics and for the signal communications between the extracellular and intracellular sides of GPCRs. Considering Phe1962.66b for example [Fig. 2[link](e)], the hydrophobic network around Phe1962.66b strengthened the stability of the TM2–ECL1–TM3 region and the rigidity could be expanded to the nearby ECL2 region through the conserved TM3–ECL2 di­sulfide bond (Cys2263.29b—Cys296ECL2). Furthermore, this hydrophobic network also interacted with TM1 through multiple inter­actions, including a hydrogen bond between Asp1982.68b and Tyr1451.40b. Therefore, the I1962.66bF mutation indirectly stabilized the network in the TM1–TM7–TM6 region that has been suggested to function as the conformational switch (de Graaf et al., 2017[Graaf, C. de, Song, G., Cao, C., Zhao, Q., Wang, M. W., Wu, B. & Stevens, R. C. (2017). Trends Biochem. Sci. 42, 946-960.]) of class B receptors.

[Figure 3]
Figure 3
MD simulation results of the four crystallized constructs. (a)–(g) The RMSD of TMs 1–7 individually and (h) the RMSD of overall TMD. The RMSD was computed with respect to the equilibrated structures.

3. Discussion

The GPCR TMD is highly dynamic because different conformational states co-exist physiologically in the cell membrane. To solve a GPCR crystal structure one needs to unify the receptor into a single state. To accomplish that, a high-affinity GPCR ligand (agonist or antagonist) is usually added to the extracellular side to shift the equilibrium to either the active or inactive state. At the intracellular side, G protein/mini-G protein/nanobody has recently been used to stabilize the GPCR active conformation, which is especially suitable for cryoEM structural determination because the complex with a G protein/mini-G protein/nanobody increases molecular size. However, the current cryoEM method cannot be used for the determination of a crucial state of GPCR, i.e. the antagonist/NAM-stabilized inactive conformation. Hence, both crystallography and cryoEM have their own advantages and drawbacks, and they can complement each other in determining macromolecular structures and in understanding their physiological functions. In GPCR structural biology studies, more and more mutations have been introduced to improve the expression level and thermal stability. In some cases alanine-scanning mutagenesis was used to search for thermal-stabilizing point mutations (Kean et al., 2015[Kean, J., Bortolato, A., Hollenstein, K., Marshall, F. H. & Jazayeri, A. (2015). Sci. Rep. 5, 11954.]), while in other situations mutations were designed by homolog modeling and computational approaches, for example, the CompoMug program that employs sequence-based analysis, structural information and a derived machine-learning predictor (Popov et al., 2018[Popov, P., Peng, Y., Shen, L., Stevens, R. C., Cherezov, V., Liu, Z. J. & Katritch, V. (2018). Elife, 7, e34729.]). We designed 98 mutations in total during construct optimization and selected ten thermal-stabilizing mutations to assist the crystallization of inactive GLP-1R in complex with NAMs. These design strategies may be applicable for structure determination of class B or other families of GPCRs.

In the GLP-1R, as well as previous class B CRF1R and GCGR structures, the inactive conformation is stabilized by the central polar network around TM3, 5, 6 and 7, as well as the binding of ligands (de Graaf et al., 2017[Graaf, C. de, Song, G., Cao, C., Zhao, Q., Wang, M. W., Wu, B. & Stevens, R. C. (2017). Trends Biochem. Sci. 42, 946-960.]). Specifically, the conserved Asn5.50b forms polar contacts with the backbone of Tyr3.44b and/or Met/Leu3.47b and Phe5.51b that packs against the Pro6.47bXXGly6.50b bulge in TM6. In the active GLP-1R and CTR structures (Zhang, Sun et al., 2017[Zhang, Y., Sun, B., Feng, D., Hu, H., Chu, M., Qu, Q., Tarrasch, J. T., Li, S., Sun Kobilka, T., Kobilka, B. K. & Skiniotis, G. (2017). Nature, 546, 248-253.]; Liang et al., 2017[Liang, Y. L., Khoshouei, M., Radjainia, M., Zhang, Y., Glukhova, A., Tarrasch, J., Thal, D. M., Furness, S. G. B., Christopoulos, G., Coudrat, T., Danev, R., Baumeister, W., Miller, L. J., Christopoulos, A., Kobilka, B. K., Wootten, D., Skiniotis, G. & Sexton, P. M. (2017). Nature, 546, 118-123.]), the Pro6.47bXXGly6.50b bulge is unwound and stabilized by hydrogen-bond interactions with Asn5.50b (Leu6.49b backbone) and Gln7.49b and His6.52b (Pro6.47b backbone). In the inactive GLP-1R structure, the di­sulfide linkage between I3175.47bC and G3616.50bC locks the conformation of the bulge in TM6 and, therefore, neither unwinds from the bulge nor swings out of the intact TM6 are possible in the active conformation. The strategy of introducing di­sulfide bonds has previously been demonstrated as useful for crystallization of GPCRs, for example, in the lysophosphatidic acid receptor 1 (LPA1) structure to link the extracellular tips of TM5 and TM6 (Chrencik et al., 2015[Chrencik, J. E., Roth, C. B., Terakado, M., Kurata, H., Omi, R., Kihara, Y., Warshaviak, D., Nakade, S., Asmar-Rovira, G., Mileni, M., Mizuno, H., Griffith, M. T., Rodgers, C., Han, G. W., Velasquez, J., Chun, J., Stevens, R. C. & Hanson, M. A. (2015). Cell, 161, 1633-1643.]), but to our knowledge it has not been utilized before to lock a GPCR into a specific conformation. We want to highlight that a di­sulfide linkage such as I3175.47bC—G3616.50bC may be especially suitable for class B receptors as this conserved glycine residue (only found in class B) is located precisely at the bulge region, and its helix-breaker feature usually provides flexibility for the equilibrium of different conformations. The present data show that without this di­sulfide, the protein is unstable and not crystallizable. In contrast, removal of unlinked cysteines, as in the M8 construct, only slightly affects the quality of crystal diffraction.

Alternatively, replacing glycine by non-glycine residues is also expected to provide rigidity for the transmembrane helices and thus increases thermal stability, e.g. G1634.60N in the CCR5–maraviroc structure (Tan et al., 2013[Tan, Q., Zhu, Y., Li, J., Chen, Z., Han, G. W., Kufareva, I., Li, T., Ma, L., Fenalti, G., Li, J., Zhang, W., Xie, X., Yang, H., Jiang, H., Cherezov, V., Liu, H., Stevens, R. C., Zhao, Q. & Wu, B. (2013). Science, 341, 1387-1390.]), G1033.49A in the GPR40–TAK875 structure (Srivastava et al., 2014[Srivastava, A., Yano, J., Hirozane, Y., Kefala, G., Gruswitz, F., Snell, G., Lane, W., Ivetac, A., Aertgeerts, K., Nguyen, J., Jennings, A. & Okada, K. (2014). Nature, 513, 124-127.]), G6753.60M in the mGlu5–mavoglurant structure (Doré et al., 2014[Doré, A. S., Okrasa, K., Patel, J. C., Serrano-Vega, M., Bennett, K., Cooke, R. M., Errey, J. C., Jazayeri, A., Khan, S., Tehan, B., Weir, M., Wiggin, G. R. & Marshall, F. H. (2014). Nature, 511, 557-562.]), G215ECL2A in the NTSR1–NTS8–13 structure (White et al., 2012[White, J. F., Noinaj, N., Shibata, Y., Love, J., Kloss, B, Xu, F., Gvozdenovic Jeremic, J., Shah, P., Shiloach, J., Tate, C. G. & Grisshammer, R. (2012). Nature, 490, 508-513.]), G3616.50bA in the GLP-1R−peptide 5 structure (Jazayeri et al., 2017[Jazayeri, A., Rappas, M., Brown, A. J. H., Kean, J., Errey, J. C., Robertson, N. J., Fiez-Vandal, C., Andrews, S. P., Congreve, M., Bortolato, A., Mason, J. S., Baig, A. H., Teobald, I., Doré, A. S., Weir, M., Cooke, R. M. & Marshall, F. H. (2017). Nature, 546, 254-258.]) and, in our case, G3185.48bI in GLP-1R–NAM structures (Song et al., 2017[Song, G., Yang, D., Wang, Y., de Graaf, C., Zhou, Q., Jiang, S., Liu, K., Cai, X., Dai, A., Lin, G., Liu, D., Wu, F., Wu, Y., Zhao, S., Ye, L., Han, G. W., Lau, J., Wu, B., Hanson, M. A., Liu, Z. J., Wang, M. W. & Stevens, R. C. (2017). Nature, 546, 312-315.]). Apart from in CCR5 and GLP-1R where the mutations were specifically designed, mutations in other receptors were chosen by a random screening method, and the conformational tendencies were not described well in these studies.

In addition, mutations can also be considered to fit the local hydrophobic or hydrophilic environment, for example, the S2253.28bA, I1962.66bF and K3466.35bA mutations described above, and the A2336.33D mutation applied in the structure determination of CCR5–maraviroc (Tan et al., 2013[Tan, Q., Zhu, Y., Li, J., Chen, Z., Han, G. W., Kufareva, I., Li, T., Ma, L., Fenalti, G., Li, J., Zhang, W., Xie, X., Yang, H., Jiang, H., Cherezov, V., Liu, H., Stevens, R. C., Zhao, Q. & Wu, B. (2013). Science, 341, 1387-1390.]). An atypical case is the S2714.47bA mutation which we included in our constructs to fit the hydrophobic lipid bilayer environment, a strategy rarely used before in the structure determination of GPCRs.

Finally, one unique feature in the crystallization of GLP-1R is the C3476.36bF mutation introduced at the binding interface with the NAMs. Designed based on the sequence alignment within class B GPCRs, this mutation was predicted to increase the binding area with the hydrophobic arms of the NAMs while not affecting the binding modes. Our previous assay showed that the mutation could enhance the inhibitory potency of the NAM, as well as the thermal stability of the NAM-bound receptor. As a control, the mutation does not affect the potency of GLP-1 from the orthosteric pathway (Song et al., 2017[Song, G., Yang, D., Wang, Y., de Graaf, C., Zhou, Q., Jiang, S., Liu, K., Cai, X., Dai, A., Lin, G., Liu, D., Wu, F., Wu, Y., Zhao, S., Ye, L., Han, G. W., Lau, J., Wu, B., Hanson, M. A., Liu, Z. J., Wang, M. W. & Stevens, R. C. (2017). Nature, 546, 312-315.]). However, caution should be taken when making mutations on the ligand-binding interface, which might change the original binding pose. Thus, this method relies on accurate prediction of the ligand-binding interface.

In conclusion, we have summarized all 98 tested mutations in the structural determination of GLP-1R in complex with NAMs. We specifically investigated four back mutations in this study and confirmed that I1962.66bF and S2714.47bA contributed differently to the receptor thermal stability and crystallization probability, while the other two mutations, S1932.63bC and M2333.36bC, may be dispensable. The mutagenesis strategies we successfully applied for crystallization of GLP-1R include mutations to: lock the receptor in a specific conformation (e.g. I3175.47bC—G3616.50bC), increase intramolecular (e.g. S2253.28bA, K3466.35bA, I1962.66bF) and intermolecular interactions for crystal packing (e.g. S2714.47bA), enhance the rigidity of a transmembrane helix (e.g. G3185.48bI), and strengthen the ligand–receptor binding interface (e.g. C3476.36bF). Our mutagenesis data indicate that hydrophobic interactions, although less specific, are more flexible, making it easier to achieve the initial design. Conversely, more specific interactions such as di­sulfide and hydrogen bonds are more challenging to implement because of their strict restraints on bond geometries. The mutagenesis principle described above can be used not only to crystallize GPCRs and other membrane proteins, but also to lock proteins into specific conformations for functional or pharmacological studies.

4. Methods

4.1. Mutation screening

The double-cysteine mutation screening (Table S1) was based on a construct of the TMD that included a thermo-stabilized apocytochrome b562RIL (BRIL) at the N-terminus and a T4L inserted into the intracellular loop 2. Sixteen pairs of double-cysteine mutants were screened in the first round and purified with ligand NNC0640 to stabilize the receptor during purification and generate more monomeric proteins. In the second round, no ligand was included since the proteins were relatively stable compared with the first round. One pair, I3175.47bC—G3616.50bC, immediately stood out from the list and was included in the second round (mutations L2453.48bC—N3205.50bC were included in the second round erroneously). A second mutation, S1932.63bC—M2333.36bC, also helped protein homogeneity and was included in the final crystallization construct, though we found that this pair was not di­sulfide linked after solving the structure, which indicated that the two cysteines may function independently to improve the monodispersity or thermal stability at that stage of construct optimization.

The screening of single point mutations (Table S2) was conducted at different stages: the first 23 were from an earlier stage without double-cysteine mutations, while the remaining 17 mutations were conducted based on the two pairs of double-cysteine mutants. We made two mutations (1 and 2) to break the potential dimerization as suggested previously (Harikumar et al., 2012[Harikumar, K. G., Wootten, D., Pinon, D. I., Koole, C., Ball, A. M., Furness, S. G., Graham, B., Dong, M., Christopoulos, A., Miller, L. J. & Sexton, P. M. (2012). Proc. Natl Acad. Sci. USA, 109, 18607-18612.]). Afterwards, we made five mutations (3–7) to increase the binding interactions with the ligand NNC0640, and 11 mutations (8–18) to form covalent interactions with a modified NNC0640 ligand based on docking models of NNC0640. The mutations 8–15 were based on a model where the ligand was docked to the orthosteric pocket that was proven wrong according to a later GCGR–MK0893 structure. The remaining mutations (19–40) were designed to strengthen local interactions between corresponding residues by forming hydrogen bonds (salt bridges), hydrophobic interactions and so on. At earlier stages including double-cysteine screening, we mainly selected the mutations with improved homogeneities since thermal-stability data were not reliable for proteins with high proportions of aggregation, while at later stages we mainly considered the thermal-stability data. Protein-yield data usually correlated with the monodispersity data so they were only taken into consideration in special cases. For example, mutant G3185.48bI did not increase homogeneity but its protein yield was much improved so we included it in the crystallization construct. It is worth pointing out that in each round the values of monodispersity/yield were compared with the controls rather than the wild type. In rare cases we comprehensively considered all three characteristics. For example, mutant S3897.44bL showed better monodispersity compared with the control, whereas the thermal stability was significantly lower. Therefore, we did not include it in the M10 crystallization construct. Furthermore, at that stage we already had one mutation (C3476.36bF) in the ligand–receptor interface that helped thermal stability.

At mutation screening stages, we mainly used NNC0640 for co-purification and thermal-stability measurement because of its availability. Both NNC0640 and PF-06372222 were used in the previous co-crystallization with the M10 construct, while only PF-06372222 was used in the crystallization of the three new constructs (M9, M8, M6) in the current study since PF-06372222 generated higher resolution data than NNC0640 in the M10 construct.

4.2. Purification of new GLP-1R mutants

The new GLP-1R constructs (M9, M8, M6) containing the same expression cassette as M10 were purified and crystallized as previously described (Song et al., 2017[Song, G., Yang, D., Wang, Y., de Graaf, C., Zhou, Q., Jiang, S., Liu, K., Cai, X., Dai, A., Lin, G., Liu, D., Wu, F., Wu, Y., Zhao, S., Ye, L., Han, G. W., Lau, J., Wu, B., Hanson, M. A., Liu, Z. J., Wang, M. W. & Stevens, R. C. (2017). Nature, 546, 312-315.]). Briefly, the Bac-to-Bac Baculovirus System (Invitrogen) in Spodoptera frugiperda cells was used for expression. Cells were infected at a density of 2–3 × 106 cells ml−1, grown at 27°C, and harvested 48 h after infection. After two washes of low-salt buffer and three washes of high-salt buffer, the cell membrane was solubilized in the presence of 200 µM ligand (PF-06372222), 2 mg ml−1 iodo­acetamide and (EDTA)-free protease inhibitor cocktail (Roche). 1.0%(w/v) DDM (n-do­decyl-β-D-malto­pyran­oside, Affymetrix) and 0.2%(w/v) CHS (cholesteryl hemisuccinate, Sigma-Aldrich) were used for the solubilization, and the supernatant was then collected and incubated with TALON IMAC resin (Clontech, Palo Alto, USA) at 4°C, overnight. After washing, the resin was resuspended and treated with tobacco etch virus protease (TEV), and the receptor protein was harvested in the flow through with buffer [25 mM HEPES, pH 7.5, 500 mM NaCl, 5%(v/v) glycerol, 0.01%(w/v) DDM, 0.002%(w/v) CHS, 30 mM imidazole and 100 µM PF-06372222]. The protein was concentrated to ∼30 mg ml−1 with a 100 kDa molecular weight cutoff concentrator.

4.3. Thermal-stability assay

CPM dye {N-[4-(7-di­ethyl­amino-4-methyl-3-coumarinyl)phenyl]­male­imide} was dissolved in DMSO (di­methyl sulfoxide) at 4 mg ml−1 as stock solution and diluted 20 times in CPM buffer [25 mM HEPES, pH 7.5, 500 mM NaCl, 5%(v/v) glycerol, 0.01%(w/v) DDM, 0.002%(w/v) CHS] before use. 1 µl of diluted CPM was added to the same buffer with ∼0.5–2 µg receptor in a final volume of 50 µl. The thermal-denaturation assay was performed in a Rotor-gene real-time PCR cycler (Qiagen). The excitation wavelength was 365 nm and the emission wavelength was 460 nm. All assays were performed over a temperature range of 25–85°C. The stability data were processed with GraphPad Prism.

4.4. Crystallization, data collection and structure refinement

The protein sample was reconstituted into lipidic cubic phase (LCP) by mixing 40% of ∼30 mg ml−1 protein with 60% lipid [10%(w/w) cholesterol, 90%(w/w) monoolein]. Crystallization trials were performed using a syringe lipid mixer and the protein–lipid mixture was dispensed in 40 nl drops onto glass sandwich plates and overlaid with 800 nl precipitant solution using an NT8 (Formulatrix). For M9, M8 and M6, crystals appeared after 1–2 weeks in 0.4–0.45 M ammonium acetate, 0.1 M sodium cacodylate, pH 6.2–6.6, 35–38% PEG 400, 3%(w/v) amino­hexanoic acid, and reached their full size (150 × 50 × 10 µm) within 2–3 weeks. Crystals were harvested directly from LCP using 50–150 µm micromounts (M2-L19-50/150, MiTeGen), flash frozen and stored in liquid nitrogen. Initial diffractions were tested at the Shanghai Synchrotron Radiation Facility in China.

X-ray diffraction data were collected at the SPring-8 beamline 41XU, Hyogo, Japan, using a Rayonix MX225HE detector (X-ray wavelength 1.0000 Å). The crystals were exposed with a 10 µm minibeam for 0.2 s and 0.2° oscillation per frame, and a rastering system was applied to find the best-diffracting parts of single crystals (Cherezov et al., 2009[Cherezov, V., Hanson, M. A., Griffith, M. T., Hilgart, M. C., Sanishvili, R., Nagarajan, V., Stepanov, S., Fischetti, R. F., Kuhn, P. & Stevens, R. C. (2009). J. R. Soc. Interface, 6, S587-S597.]). XDS (Kabsch, 2010[Kabsch, W. (2010). Acta Cryst. D66, 125-132.]) was used for integrating and scaling data from the best-diffracting crystals. M9, M8 and M6 structures were solved by molecular replacement with Phaser using previous GLP-1R (PDB entry 5vew; Song et al., 2017[Song, G., Yang, D., Wang, Y., de Graaf, C., Zhou, Q., Jiang, S., Liu, K., Cai, X., Dai, A., Lin, G., Liu, D., Wu, F., Wu, Y., Zhao, S., Ye, L., Han, G. W., Lau, J., Wu, B., Hanson, M. A., Liu, Z. J., Wang, M. W. & Stevens, R. C. (2017). Nature, 546, 312-315.]) as the search model. The structure was refined iteratively with Phenix (Adams et al., 2010[Adams, P. D., Afonine, P. V., Bunkóczi, G., Chen, V. B., Davis, I. W., Echols, N., Headd, J. J., Hung, L.-W., Kapral, G. J., Grosse-Kunstleve, R. W., McCoy, A. J., Moriarty, N. W., Oeffner, R., Read, R. J., Richardson, D. C., Richardson, J. S., Terwilliger, T. C. & Zwart, P. H. (2010). Acta Cryst. D66, 213-221.]; Liebschner et al., 2019[Liebschner, D., Afonine, P. V., Baker, M. L., Bunkóczi, G., Chen, V. B., Croll, T. I., Hintze, B., Hung, L.-W., Jain, S., McCoy, A. J., Moriarty, N. W., Oeffner, R. D., Poon, B. K., Prisant, M. G., Read, R. J., Richardson, J. S., Richardson, D. C., Sammito, M. D., Sobolev, O. V., Stockwell, D. H., Terwilliger, T. C., Urzhumtsev, A. G., Videau, L. L., Williams, C. J. & Adams, P. D. (2019). Acta Cryst. D75, 861-877.]) with manual modification using Coot (Emsley et al., 2010[Emsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. (2010). Acta Cryst. D66, 486-501.]). Structures were checked by MolProbity (Chen et al., 2010[Chen, V. B., Arendall, W. B., Headd, J. J., Keedy, D. A., Immormino, R. M., Kapral, G. J., Murray, L. W., Richardson, J. S. & Richardson, D. C. (2010). Acta Cryst. D66, 12-21.]) and statistics are provided in Table 1[link].

4.5. MD simulation

The missing residues in ECL2 (Trp203–Leu218) and ECL3 (Asp372–Arg380) of M10 were fixed by the FREAD server while the missing heavy atoms in the M10 structure were fixed by tleap in Amber (Case et al., 2005[Case, D. A., Cheatham, T. E., Darden, T., Gohlke, H., Luo, R., Merz, K. M., Onufriev, A., Simmerling, C., Wang, B. & Woods, R. J. (2005). J. Comput. Chem. 26, 1668-1688.]; Choi & Deane, 2010[Choi, Y. & Deane, C. M. (2010). Proteins, 78, 1431-1440.]). M10 also contains the ligand (PF-06372222) and lipids OLC and OLA. The CHARMM36 force field parameters of ligand (PF-06372222) and detergents OLC and OLA were generated by CGenFF (Vanommeslaeghe & MacKerell, 2012[Vanommeslaeghe, K. & MacKerell, A. D. Jr (2012). J. Chem. Inf. Model. 52, 3144-3154.]).

The PPM server was used to reorient the GLP-1R-T4L systems to ensure that the transmembrane domain of GLP-1R was well located in the POPC (1-palmitoyl–2-oleoyl-sn-glycero-3-phospho­choline) lipid bilayer (Lomize et al., 2012[Lomize, M. A., Pogozheva, I. D., Joo, H., Mosberg, H. I. & Lomize, A. L. (2012). Nucleic Acids Res. 40, D370-D376.]). For each system, 200 ns MD simulations were performed. Before the production ran, the systems were equilibrated in a POPC lipid bilayer and solvated in a water box. The CHARMM graphical user interface was used to generate topology and the parameter files for the GLP-1R-T4L systems (Wu et al., 2014[Wu, E. L., Cheng, X., Jo, S., Rui, H., Song, K. C., Dávila-Contreras, E. M., Qi, Y., Lee, J., Monje-Galvan, V., Venable, R. M., Klauda, J. B. & Im, W. (2014). J. Comput. Chem. 35, 1997-2004.]). In addition to the protein complex, ∼255 POPC molecules, ∼42 000 water molecules (TIP3) and excess sodium chloride ions were added to maintain an ion concentration of 150 mM; thus, the entire system contained ∼160 000 atoms. The systems were modeled using the force parameters of CHARMM36. The NPT ensemble (constant particle number, pressure and temperature) MD simulations were generated using GROMACS 5.1.2 (Hess et al., 2008[Hess, B., Kutzner, C., van der Spoel, D. & Lindahl, E. (2008). J. Chem. Theory Comput. 4, 435-447.]). The REDUCE program in Amber was used to add hydrogens to the original PDB files and to determine the protonation state of histidine residues (Word et al., 1999[Word, J. M., Lovell, S. C., Richardson, J. S. & Richardson, D. C. (1999). J. Mol. Biol. 285, 1735-1747.]).

The initial energy minimizations were achieved using the steepest descent algorithm and this process was followed by two steps of equilibrations, i.e. 20 ns NVT (constant particle number, volume and temperature) dynamics with large restraint force constants and 40 ns NPT dynamics with gradually decreased restraint force constants. These restraints included harmonic restraints on heavy atoms of the protein and planar restraints to hold the position of lipid head groups of membranes along the Z axis. The system temperature was set to 303 K. Once all the equilibration steps were completed, the constraints were removed and each system propagated under constant temperature and pressure for 200 ns with a time step of 2 fs.

To evaluate the stabilities, the structure changes caused by mutation were quantified using the RMSD of the atomic position of the TMD of a receptor with respect to the equilibrated structures. For the stability of fusion partner T4L, the RMSD was computed using the transformation matrix obtained by aligning the TMD of the receptor. This described the changes in relative motion of the T4L. The CPPTRAJ package in Amber tools was used to compute the 2D RMSD of the ligand–receptor complex to check the overall stability of the complex structure (Case et al., 2005[Case, D. A., Cheatham, T. E., Darden, T., Gohlke, H., Luo, R., Merz, K. M., Onufriev, A., Simmerling, C., Wang, B. & Woods, R. J. (2005). J. Comput. Chem. 26, 1668-1688.]). For the 2D RMSD case, RMSD was calculated between pairwise snapshots obtained from the 200 ns simulation trajectories. The RMS fluctuation of each residue in the fusion partner and receptor was calculated to compare the thermal fluctuations of the systems.

We carried out clustering analysis to group these mutant structures based on their similarity (Wang et al., 2019[Wang, Y., Park, J. H., Lupala, C. S., Yun, J. H., Jin, Z., Huang, L., Li, X., Tang, L., Lee, W. & Liu, H. (2019). Sci. Rep. 9, 5317.]). First, the pairwise RMSD was computed based on the TMD of GLP-1R, then the Javis–Patrick method was used for clustering. Secondly, an RMSD of 1.5 Å was used as the difference cutoff to define structure clusters. If the RMSD value was smaller than 1.5 Å, the two structures were considered to be in the same cluster. These clustering results were used to gauge the stability of the ligand–receptor complex. A new structure was joined to an existing cluster if at least one structure in the cluster shared three common neighbors with the new structure, where the neighbors were the ten most similar structures or all the structures within the RMSD cutoff of 1.5 Å. The clustering program used for this analysis was implemented in the GROMACS package. The size of each cluster was correlated to the stability of the corresponding cluster. The results are summarized in Fig. S2.

Footnotes

These authors contributed equally

Acknowledgements

The synchrotron radiation experiments were performed at the BL41XU of SPring-8 (Japan) and beamline BL18U1 (Shanghai Synchrotron Radiation Facility, China). We are grateful to Professor Raymond C. Stevens and Professor Zhi-Jie Liu for their mentoring and assistance on this project.

Funding information

This research work was supported by a Tianhe-2JK computing time award at the Beijing Computational Research Center. This work was also supported by the National Nature Science Foundation of China grants 11575021 (HL) and 31770898 (GS), and the National Key R&D Program of China 2018YFA0507001.

References

First citationAdams, P. D., Afonine, P. V., Bunkóczi, G., Chen, V. B., Davis, I. W., Echols, N., Headd, J. J., Hung, L.-W., Kapral, G. J., Grosse-Kunstleve, R. W., McCoy, A. J., Moriarty, N. W., Oeffner, R., Read, R. J., Richardson, D. C., Richardson, J. S., Terwilliger, T. C. & Zwart, P. H. (2010). Acta Cryst. D66, 213–221.  Web of Science CrossRef CAS IUCr Journals Google Scholar
First citationCase, D. A., Cheatham, T. E., Darden, T., Gohlke, H., Luo, R., Merz, K. M., Onufriev, A., Simmerling, C., Wang, B. & Woods, R. J. (2005). J. Comput. Chem. 26, 1668–1688.  Web of Science CrossRef PubMed CAS Google Scholar
First citationChen, V. B., Arendall, W. B., Headd, J. J., Keedy, D. A., Immormino, R. M., Kapral, G. J., Murray, L. W., Richardson, J. S. & Richardson, D. C. (2010). Acta Cryst. D66, 12–21.  Web of Science CrossRef CAS IUCr Journals Google Scholar
First citationCheng, R. K. Y., Fiez-Vandal, C., Schlenker, O., Edman, K., Aggeler, B., Brown, D. G., Brown, G. A., Cooke, R. M., Dumelin, C. E., Doré, A. S., Geschwindner, S., Grebner, C., Hermansson, N. O., Jazayeri, A., Johansson, P., Leong, L., Prihandoko, R., Rappas, M., Soutter, H., Snijder, A., Sundström, L., Tehan, B., Thornton, P., Troast, D., Wiggin, G., Zhukov, A., Marshall, F. H. & Dekker, N. (2017). Nature, 545, 112–115.  Web of Science CrossRef CAS PubMed Google Scholar
First citationCherezov, V., Hanson, M. A., Griffith, M. T., Hilgart, M. C., Sanishvili, R., Nagarajan, V., Stepanov, S., Fischetti, R. F., Kuhn, P. & Stevens, R. C. (2009). J. R. Soc. Interface, 6, S587–S597.  Web of Science CrossRef PubMed CAS Google Scholar
First citationChoi, Y. & Deane, C. M. (2010). Proteins, 78, 1431–1440.  Web of Science CAS PubMed Google Scholar
First citationChrencik, J. E., Roth, C. B., Terakado, M., Kurata, H., Omi, R., Kihara, Y., Warshaviak, D., Nakade, S., Asmar-Rovira, G., Mileni, M., Mizuno, H., Griffith, M. T., Rodgers, C., Han, G. W., Velasquez, J., Chun, J., Stevens, R. C. & Hanson, M. A. (2015). Cell, 161, 1633–1643.  Web of Science CrossRef CAS PubMed Google Scholar
First citationDoré, A. S., Okrasa, K., Patel, J. C., Serrano-Vega, M., Bennett, K., Cooke, R. M., Errey, J. C., Jazayeri, A., Khan, S., Tehan, B., Weir, M., Wiggin, G. R. & Marshall, F. H. (2014). Nature, 511, 557–562.  Web of Science PubMed Google Scholar
First citationEmsley, P., Lohkamp, B., Scott, W. G. & Cowtan, K. (2010). Acta Cryst. D66, 486–501.  Web of Science CrossRef CAS IUCr Journals Google Scholar
First citationFerrer, J. L., Jez, J. M., Bowman, M. E., Dixon, R. A. & Noel, J. P. (1999). Nat. Struct. Biol. 6, 775–784.  Web of Science PubMed CAS Google Scholar
First citationGraaf, C. de, Donnelly, D., Wootten, D., Lau, J., Sexton, P. M., Miller, L. J., Ahn, J. M., Liao, J., Fletcher, M. M., Yang, D., Brown, A. J., Zhou, C., Deng, J. & Wang, M. W. (2016). Pharmacol. Rev. 68, 954–1013.  Web of Science CrossRef PubMed Google Scholar
First citationGraaf, C. de, Song, G., Cao, C., Zhao, Q., Wang, M. W., Wu, B. & Stevens, R. C. (2017). Trends Biochem. Sci. 42, 946–960.  Web of Science PubMed Google Scholar
First citationGutniak, M., Ørkov, C., Holst, J. J., Ahrén, B. & Efendić, S. (1992). N. Engl. J. Med. 326, 1316–1322.  CrossRef PubMed CAS Web of Science Google Scholar
First citationHarikumar, K. G., Wootten, D., Pinon, D. I., Koole, C., Ball, A. M., Furness, S. G., Graham, B., Dong, M., Christopoulos, A., Miller, L. J. & Sexton, P. M. (2012). Proc. Natl Acad. Sci. USA, 109, 18607–18612.  Web of Science CrossRef CAS PubMed Google Scholar
First citationHess, B., Kutzner, C., van der Spoel, D. & Lindahl, E. (2008). J. Chem. Theory Comput. 4, 435–447.  Web of Science CrossRef CAS PubMed Google Scholar
First citationHoare, S. R. (2005). Drug Discov. Today, 10, 417–427.  Web of Science CrossRef PubMed CAS Google Scholar
First citationHollenstein, K., Kean, J., Bortolato, A., Cheng, R. K., Doré, A. S., Jazayeri, A., Cooke, R. M., Weir, M. & Marshall, F. H. (2013). Nature, 499, 438–443.  Web of Science CrossRef CAS PubMed Google Scholar
First citationJazayeri, A., Doré, A. S., Lamb, D., Krishnamurthy, H., Southall, S. M., Baig, A. H., Bortolato, A., Koglin, M., Robertson, N. J., Errey, J. C., Andrews, S. P., Teobald, I., Brown, A. J., Cooke, R. M., Weir, M. & Marshall, F. H. (2016). Nature, 533, 274–277.  Web of Science CrossRef CAS PubMed Google Scholar
First citationJazayeri, A., Rappas, M., Brown, A. J. H., Kean, J., Errey, J. C., Robertson, N. J., Fiez-Vandal, C., Andrews, S. P., Congreve, M., Bortolato, A., Mason, J. S., Baig, A. H., Teobald, I., Doré, A. S., Weir, M., Cooke, R. M. & Marshall, F. H. (2017). Nature, 546, 254–258.  Web of Science CrossRef CAS PubMed Google Scholar
First citationJohansson, L. C., Stauch, B., McCorvy, J. D., Han, G. W., Patel, N., Huang, X. P., Batyuk, A., Gati, C., Slocum, S. T., Li, C., Grandner, J. M., Hao, S., Olsen, R. H. J., Tribo, A. R., Zaare, S., Zhu, L., Zatsepin, N. A., Weierstall, U., Yous, S., Stevens, R. C., Liu, W., Roth, B. L., Katritch, V. & Cherezov, V. (2019). Nature, 569, 289–292.  Web of Science CrossRef CAS PubMed Google Scholar
First citationKabsch, W. (2010). Acta Cryst. D66, 125–132.  Web of Science CrossRef CAS IUCr Journals Google Scholar
First citationKean, J., Bortolato, A., Hollenstein, K., Marshall, F. H. & Jazayeri, A. (2015). Sci. Rep. 5, 11954.  Web of Science CrossRef PubMed Google Scholar
First citationLebon, G. & Tate, C. G. (2011). Med. Sci. 27, 926–928.  Google Scholar
First citationLiang, Y. L., Khoshouei, M., Radjainia, M., Zhang, Y., Glukhova, A., Tarrasch, J., Thal, D. M., Furness, S. G. B., Christopoulos, G., Coudrat, T., Danev, R., Baumeister, W., Miller, L. J., Christopoulos, A., Kobilka, B. K., Wootten, D., Skiniotis, G. & Sexton, P. M. (2017). Nature, 546, 118–123.  Web of Science CrossRef CAS PubMed Google Scholar
First citationLiebschner, D., Afonine, P. V., Baker, M. L., Bunkóczi, G., Chen, V. B., Croll, T. I., Hintze, B., Hung, L.-W., Jain, S., McCoy, A. J., Moriarty, N. W., Oeffner, R. D., Poon, B. K., Prisant, M. G., Read, R. J., Richardson, J. S., Richardson, D. C., Sammito, M. D., Sobolev, O. V., Stockwell, D. H., Terwilliger, T. C., Urzhumtsev, A. G., Videau, L. L., Williams, C. J. & Adams, P. D. (2019). Acta Cryst. D75, 861–877.  CrossRef IUCr Journals Google Scholar
First citationLomize, M. A., Pogozheva, I. D., Joo, H., Mosberg, H. I. & Lomize, A. L. (2012). Nucleic Acids Res. 40, D370–D376.  Web of Science CrossRef CAS PubMed Google Scholar
First citationOswald, C., Rappas, M., Kean, J., Doré, A. S., Errey, J. C., Bennett, K., Deflorian, F., Christopher, J. A., Jazayeri, A., Mason, J. S., Congreve, M., Cooke, R. M. & Marshall, F. H. (2016). Nature, 540, 462–465.  Web of Science CrossRef CAS PubMed Google Scholar
First citationPabreja, K., Mohd, M. A., Koole, C., Wootten, D. & Furness, S. G. (2014). Br. J. Pharmacol. 171, 1114–1128.  Web of Science CrossRef CAS PubMed Google Scholar
First citationPopov, P., Peng, Y., Shen, L., Stevens, R. C., Cherezov, V., Liu, Z. J. & Katritch, V. (2018). Elife, 7, e34729.  Web of Science CrossRef PubMed Google Scholar
First citationPu, M., Xu, Z., Peng, Y., Hou, Y., Liu, D., Wang, Y., Liu, H., Song, G. & Liu, Z. J. (2018). Protein Cell, 9, 659–663.  Web of Science CrossRef CAS PubMed Google Scholar
First citationShaw, J. E., Sicree, R. A. & Zimmet, P. Z. (2010). Diabetes Res. Clin. Pract. 87, 4–14.  Web of Science CrossRef PubMed CAS Google Scholar
First citationShihoya, W., Nishizawa, T., Okuta, A., Tani, K., Dohmae, N., Fujiyoshi, Y., Nureki, O. & Doi, T. (2016). Nature, 537, 363–368.  Web of Science CrossRef CAS PubMed Google Scholar
First citationSiu, F. Y., He, M., de Graaf, C., Han, G. W., Yang, D., Zhang, Z., Zhou, C., Xu, Q., Wacker, D., Joseph, J. S., Liu, W., Lau, J., Cherezov, V., Katritch, V., Wang, M. W. & Stevens, R. C. (2013). Nature, 499, 444–449.  Web of Science CrossRef CAS PubMed Google Scholar
First citationSong, G., Yang, D., Wang, Y., de Graaf, C., Zhou, Q., Jiang, S., Liu, K., Cai, X., Dai, A., Lin, G., Liu, D., Wu, F., Wu, Y., Zhao, S., Ye, L., Han, G. W., Lau, J., Wu, B., Hanson, M. A., Liu, Z. J., Wang, M. W. & Stevens, R. C. (2017). Nature, 546, 312–315.  Web of Science CrossRef CAS PubMed Google Scholar
First citationSrivastava, A., Yano, J., Hirozane, Y., Kefala, G., Gruswitz, F., Snell, G., Lane, W., Ivetac, A., Aertgeerts, K., Nguyen, J., Jennings, A. & Okada, K. (2014). Nature, 513, 124–127.  Web of Science CrossRef CAS PubMed Google Scholar
First citationStauch, B., Johansson, L. C., McCorvy, J. D., Patel, N., Han, G. W., Huang, X. P., Gati, C., Batyuk, A., Slocum, S. T., Ishchenko, A., Brehm, W., White, T. A., Michaelian, N., Madsen, C., Zhu, L., Grant, T. D., Grandner, J. M., Shiriaeva, A., Olsen, R. H. J., Tribo, A. R., Yous, S., Stevens, R. C., Weierstall, U., Katritch, V., Roth, B. L., Liu, W. & Cherezov, V. (2019). Nature, 569, 284–288.  Web of Science CrossRef CAS PubMed Google Scholar
First citationTan, Q., Zhu, Y., Li, J., Chen, Z., Han, G. W., Kufareva, I., Li, T., Ma, L., Fenalti, G., Li, J., Zhang, W., Xie, X., Yang, H., Jiang, H., Cherezov, V., Liu, H., Stevens, R. C., Zhao, Q. & Wu, B. (2013). Science, 341, 1387–1390.  Web of Science CrossRef CAS PubMed Google Scholar
First citationThal, D. M., Vuckovic, Z., Draper-Joyce, C. J., Liang, Y. L., Glukhova, A., Christopoulos, A. & Sexton, P. M. (2018). Curr. Opin. Struct. Biol. 51, 28–34.  Web of Science CrossRef CAS PubMed Google Scholar
First citationVanommeslaeghe, K. & MacKerell, A. D. Jr (2012). J. Chem. Inf. Model. 52, 3144–3154.  Web of Science CrossRef CAS PubMed Google Scholar
First citationWang, Y., Park, J. H., Lupala, C. S., Yun, J. H., Jin, Z., Huang, L., Li, X., Tang, L., Lee, W. & Liu, H. (2019). Sci. Rep. 9, 5317.  Web of Science CrossRef PubMed Google Scholar
First citationWarne, T., Serrano-Vega, M. J., Baker, J. G., Moukhametzianov, R., Edwards, P. C., Henderson, R., Leslie, A. G., Tate, C. G. & Schertler, G. F. (2008). Nature, 454, 486–491.  Web of Science CrossRef PubMed CAS Google Scholar
First citationWhite, J. F., Noinaj, N., Shibata, Y., Love, J., Kloss, B, Xu, F., Gvozdenovic Jeremic, J., Shah, P., Shiloach, J., Tate, C. G. & Grisshammer, R. (2012). Nature, 490, 508–513.  Web of Science CrossRef CAS PubMed Google Scholar
First citationWord, J. M., Lovell, S. C., Richardson, J. S. & Richardson, D. C. (1999). J. Mol. Biol. 285, 1735–1747.  Web of Science CrossRef CAS PubMed Google Scholar
First citationWu, E. L., Cheng, X., Jo, S., Rui, H., Song, K. C., Dávila-Contreras, E. M., Qi, Y., Lee, J., Monje-Galvan, V., Venable, R. M., Klauda, J. B. & Im, W. (2014). J. Comput. Chem. 35, 1997–2004.  Web of Science CrossRef CAS PubMed Google Scholar
First citationZhang, H., Qiao, A., Yang, D., Yang, L., Dai, A., de Graaf, C., Reedtz-Runge, S., Dharmarajan, V., Zhang, H., Han, G. W., Grant, T. D., Sierra, R. G., Weierstall, U., Nelson, G., Liu, W., Wu, Y., Ma, L., Cai, X., Lin, G., Wu, X., Geng, Z., Dong, Y., Song, G., Griffin, P. R., Lau, J., Cherezov, V., Yang, H., Hanson, M. A., Stevens, R. C., Zhao, Q., Jiang, H., Wang, M. W. & Wu, B. (2017). Nature, 546, 259–264.  Web of Science CrossRef CAS PubMed Google Scholar
First citationZhang, Y., Sun, B., Feng, D., Hu, H., Chu, M., Qu, Q., Tarrasch, J. T., Li, S., Sun Kobilka, T., Kobilka, B. K. & Skiniotis, G. (2017). Nature, 546, 248–253.  Web of Science CrossRef CAS PubMed Google Scholar
First citationZhao, L. H., Ma, S., Sutkeviciute, I., Shen, D. D., Zhou, X. E., de Waal, P. W., Li, C. Y., Kang, Y., Clark, L. J., Jean-Alphonse, F. G., White, A. D., Yang, D., Dai, A., Cai, X., Chen, J., Li, C., Jiang, Y., Watanabe, T., Gardella, T. J., Melcher, K., Wang, M. W., Vilardaga, J. P., Xu, H. E. & Zhang, Y. (2019). Science, 364, 148–153.  Web of Science CrossRef CAS PubMed Google Scholar

This is an open-access article distributed under the terms of the Creative Commons Attribution (CC-BY) Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original authors and source are cited.

IUCrJ
Volume 6| Part 6| November 2019| Pages 996-1006
ISSN: 2052-2525