Serial Femtosecond Crystallography of G Protein-Coupled Receptors

G protein-coupled receptors (GPCRs) represent a highly diverse superfamily of eukaryotic membrane proteins that mediate cellular communication. In humans, about 800 GPCRs respond to a variety of extracellular signaling molecules and transmit signals inside the cell by coupling to heterotrimeric G proteins and other effectors. Their involvement in key physiological and sensory processes in humans makes GPCRs prominent drug targets. Despite the high biomedical relevance and decades of dedicated research, knowledge of the structural mechanisms of ligand recognition, receptor activation and signaling in this broad family remains limited. Challenges for GPCR structural studies include low expression yields, low receptor stability after detergent extraction from native membranes, and high conformational heterogeneity. Many years of developments aimed at receptor stabilization, crystallization and microcrystallography culminated in a series of breakthroughs in GPCR structural biology leading to the structure determination of 22 receptors, some of which were solved in several conformational states and one in complex with its G protein partner (1–5).

Nonetheless, crystallographic studies of GPCRs remain difficult, as many of them produce only microcrystals. Most GPCR structures to date have been obtained using crystallization from the membrane-mimetic environment of a lipidic cubic phase (LCP) (6, 7). LCP crystallization has proven successful for obtaining high-resolution structures of a variety of membrane proteins including ion channels, transporters, and enzymes, in addition to GPCRs (8, 9). This method leads to highly ordered crystals that are, however, often limited in size. Microfocus x-ray beams of high intensity (~10⁹ photons/s/µm²) and long exposures (~5 s) are typically required to obtain sufficient intensity for high-resolution data from weakly diffracting microcrystals. The high radiation doses induce severe radiation damage and require merging data from multiple crystals to obtain complete datasets of sufficient quality. Accordingly, sub-10 µm GPCR crystals are currently not suitable for high-resolution data collection even at the most powerful synchrotron microfocus beamlines (7, 10).

Serial femtosecond crystallography (SFX) (11), which takes advantage of x-ray free-electron lasers (XFEL), has recently demonstrated great promise for obtaining room temperature high-resolution data from micrometer- and sub-micrometer size crystals of soluble proteins with minimal radiation damage (12, 13). The highly intense (~2 mJ, 10¹² photons per pulse) and ultrashort (<50 fs) x-ray pulses produced by XFELs enable recording high-resolution diffraction snapshots from individual crystals at single orientations before their destruction. SFX data collection, therefore, relies on a continuous supply of small crystals intersecting the XFEL beam in random orientations, typically provided by a fast-running liquid microjet (12), which is incompatible with streaming highly viscous gel-like materials, such as LCP, and requires tens to hundreds milligrams of crystallized protein for data collection (11). For many membrane proteins, including most human membrane proteins, obtaining such quantities is not practical.

Here we have modified the SFX data collection approach (Fig. 1) and obtained a room-temperature GPCR structure at 2.8 Å resolution using only 300 µg of protein crystallized in LCP. SFX experiments were performed at the Coherent X-ray Imaging (CXI) instrument of the Linac Coherent Light Source (LCLS) (14). LCP-grown microcrystals (average size 5×5×5 µm³) (fig. S1) (15) of the human serotonin 5-HT_2B receptor (16) bound to the agonist ergotamine were continuously delivered across a ~1.5 µm diameter XFEL beam, using a specially designed LCP injector. LCP with randomly distributed crystals was extruded through a 20–50 µm capillary into a vacuum chamber (10⁻⁴ Torr) at room temperature (21 °C) (17) and a constant flow-rate of 50 – 200 nL/min, and was stabilized by a co-axial flow of helium or nitrogen gas supplied at 300–500 psi. Single-pulse diffraction patterns (fig. S2) were recorded using 9.5 keV (1.3 Å) x-ray pulses of 50 fs duration at a 120 Hz repetition rate by a Cornell-SLAC pixel array detector (CSPAD) (18) positioned at a distance of 100 mm from the sample. The XFEL beam was attenuated to 3–6% to avoid detector saturation. The average x-ray pulse energy at the sample was 50 µJ (3·10¹⁰ photons/pulse), corresponding to a radiation dose of up to 25 MGy per crystal. A total of 4,217,508 diffraction patterns were collected within 10 h using ~100 µL of crystal-loaded LCP, corresponding to about 0.3 mg of protein. Of these patterns, 152,651 were identified as crystal hits (15 or more Bragg peaks) by the processing software Cheetah (19), corresponding to a hit rate of 3.6 %. Of these crystal hits, 32,819 patterns (21.5 %) were successfully indexed and integrated by CrystFEL (20) at 2.8 Å resolution (table S1). The structure was determined by molecular replacement and refined to R_work/R_free = 22.7%/27.0%. Overall, the final structure (fig. S3) has a well-defined density for most residues including the ligand ergotamine (fig. S4).

The XFEL structure of the 5-HT_2B receptor/ergotamine complex (5-HT_2B-XFEL) was compared to the recently published structure of the same receptor/ligand complex obtained by traditional microcrystallography at a synchrotron source (PDB ID: 4IB4; 5-HT_2B-SYN) (21). Synchrotron data were collected at 100 K on cryo-cooled crystals of a much larger size (average volume ~10⁴ µm³) than those used for the XFEL structure (average volume ~10² µm³) (fig. S1). Other differences between data collection protocols are listed in table S1. Both datasets were processed in the same spacegroup C222₁, which is expected given the very similar crystallization conditions. However, the lattice parameters for the room temperature XFEL crystals are slightly longer in the a and b directions and slightly shorter in the c direction, resulting in a 2.1% larger unit cell volume. Concomitant with these lattice changes we observe a ~2.5° rotation of the BRIL fusion domain with respect to the receptor (fig. S5). Otherwise, the receptor domains of the 5-HT_2B-XFEL and 5-HT_2B-SYN structures are very similar (receptor Cα root mean square deviation (rmsd) = 0.45 Å, excluding flexible residues at the N-terminus, 48–51, and in the extracellular loop (ECL) 2, 195–205) (Fig. 2). The ligand ergotamine has indistinguishable electron density and placement (total ligand rmsd = 0.32 Å) in both structures (Fig. 2, fig. S4B). The largest backbone deviations are observed in the loop regions, especially in the stretch of ECL2 between helix IV and the Cys128 – Cys207 disulfide bond, which is apparently very flexible. An unexpected backbone deviation is observed at the extracellular tip of helix II (Fig. 2C), which adopts a regular α-helix in the 5-HT_2B-XFEL structure with Thr114 forming a stabilizing hydrogen bond with the main chain hydroxyl of Ile110. In the 5-HT_2B-SYN structure, however, a water stabilized kink is found at this location, which results in the two structures deviating by 2.0 Å (at Cα atom of Thr114) at the tip of helix II and up to 3.4 Å (at O atom of Phe117) in ECL1.

Although absolute B- (or temperature) factor values can be affected by errors associated with experimental conditions, their distribution generally represents the relative static and dynamic flexibility of the protein in the crystal (22). Since both structures were obtained from similar samples and at similar resolutions, we analyzed their B-factor distributions to study the effect of the different temperatures on the thermal motions of the receptor. The average B-factor for the receptor part in the room temperature 5-HT_2B-XFEL structure (88.4 Ų) is 21 Ų larger than that in the cryo 5-HT_2B-SYN structure (67.2 Ų), consistent with larger thermal motions at higher temperature and possible effects of Bragg termination during the XFEL pulse (20). The distribution of B-factors highlights a more rigid core of the seven transmembrane helices in comparison to loops, with more pronounced B-factor deviations observed in the room temperature 5-HT_2B-XFEL structure (Fig. 3, fig. S6). N-terminus, intracellular loop (ICL) 2, ECL1, and part of ECL2 between helix IV and the Cys128 – Cys207 disulfide bond show much larger deviations in B-factors (50–100 Ų) between the two structures compared to the average difference of 21 Ų. These parts of the structure are not involved in direct interactions with the ligand ergotamine, but their mobility may affect the kinetics of ligand binding and interactions with intracellular binding partners (23). In contrast, ICL1, part of ECL2 between the Cys128 – Cys207 disulfide bond and helix V, and ECL3 display just an average increase in the B-factors, suggesting that the relative range of their thermal fluctuations was adequately captured in the cryo structure. As previously established by cryocrystallography, one of the most pronounced differences between the two subtypes of serotonin receptors, 5-HT_2B and 5-HT_1B, occurs at the extracellular tip of helix V and ECL2, which forms an additional helical turn stabilized by a water molecule in 5-HT_2B (21). This additional turn pulls the extracellular tip of helix V toward the center of the helical bundle, and was suggested to be responsible for the biased agonism of ergotamine at the 5-HT_2B receptor. The 5-HT_2B-XFEL structure confirms the rigid structured conformation of ECL2, stabilized by a comprehensive network of hydrogen bonds, involving residues Lys193, Glu196, Arg213, Asp216 and a lipid OLC (monoolein) (fig. S7), however, no ordered water molecule is observed, emphasizing that water is more disordered and probably does not play a substantial structural role at this location.

Several side chains have partly missing electron density in both room temperature and cryo structures (table S2). Such lack of density is most likely related to disorder of the corresponding side chains (such as residues at the N-terminus, ECL2 and ICL2) (Fig. 2A). Two disulfide bonds, Cys128 – Cys207 and Cys350 – Cys353, are intact and well resolved in both structures, however the B-factor increase in the 5-HT_2B-XFEL structure compared to 5-HT_2B-SYN for each of these disulfide bonds (11.1 Ų and 5.7 Ų, respectively) is lower than the average B-factor increase (21 Ų). Several side chains have different rotamer conformations between the two structures (table S3, Fig. 2D), consistent with a partial remodeling of the side chain conformational distribution upon cryo-cooling observed in soluble proteins (24). Interestingly, several interactions involving charged residues appear stronger and better defined in 5-HT_2B-XFEL compared to the 5-HT_2B-SYN structure (table S4). This strengthening of the charged interactions at higher temperatures potentially can be explained by a decrease in the dielectric constant of water with temperature, reducing the de-solvation penalty (25, 26). In particularly, the salt bridge between Glu319 and Lys247 is well defined in the 5-HT_2B-XFEL structure, but appears broken in the cryo 5-HT_2B-SYN structure (Fig. 2B). Since GPCR activation has been associated with large-scale structural changes in the intracellular parts of helices V and VI, this salt bridge may play a role in the receptor function and is likely to be more accurately resolved and represented in the 5-HT_2B-XFEL structure recorded at room temperature.

Overall, the observed differences likely originate from effects related to thermal motions, cryo-cooling (24) and radiation damage (27). Thus, the XFEL source enables access to a room temperature GPCR structure, which more accurately represents the conformational ensemble for this receptor under native conditions. Since dynamics are an integral part of GPCR biology, the use of SFX to accurately determine GPCR structural details at room temperature can make an important contribution to understanding the structure-function relationships in this superfamily.