A versatile approach to high-density microcrystals in lipidic cubic phase for room-temperature serial crystallography

A versatile approach for the preparation of high-density membrane protein microcrystals in lipidic cubic phase for serial crystallography is described.

Although SX of IMPs crystallized in LCP has been successful in overcoming many of the challenges associated with classical diffraction methods, technical and practical difficulties remain.SX methods typically consume large amounts of protein (up to several milligrams) and the crystals must be in a narrow size range (1-5 mm) and grow at sufficiently high density to maximize the efficiency of data collection.This is particularly important during time-resolved experiments, which require the collection of complete data sets for multiple time points (Bra ¨nde ´n & Neutze, 2021;Gru ¨nbein et al., 2020).In addition, crystallization of IMPs in LCP for SX requires large numbers of gas-tight glass syringes, which are also needed during the screening process for optimizing crystallization conditions (Liu et al., 2014).Failure to seal or store these syringes correctly results in sample leakage or dehydration.Furthermore, transport of samples to X-ray facilities can be associated with glass breakage and sample loss.
To address the above issues inherent in current methods, we have developed an alternative approach that does not require the use of large quantities of gas-tight syringes for the production of a sufficient amount of homogeneous crystals in LCP for SX studies.The procedure, termed VIALS (versatile approach to high-density microcrystals in lipidic cubic phase for serial crystallography), makes use of small borosilicate glass vials (sample volume of 300 ml) and an automated liquidhandler system to produce hundreds of microlitres of highdensity micrometre-sized crystals grown in LCP [Fig. 1(a)].The approach also facilitates ligand soaking without mechanically disturbing the crystals.Ligand soaking, vital to rational structure-based drug design, has been particularly challenging when growing IMP crystals in LCP within gastight glass syringes (Ishchenko et al., 2019;Weinert et al., 2017).The method introduced here enables soaking of many ligands in parallel, facilitating structure determination of ligand-receptor complexes in a single SX experimental session [Fig.1(b)].The VIALS workflow is simple, fast and economical, and is easily implemented in any standard crystallization laboratory.Finally, storage and transport of crystal samples in LCP are also facilitated by our approach [Fig.1(a), step 6].
To validate the versatility of the VIALS workflow, we used two different IMPs, archaerhodopsin-3 (AR3) and the human adenosine A 2A receptor (hA 2A R), as proof of concept.AR3, from the archaebacterium Halorubrum sodomense, is a lightdriven proton pump that undergoes a sequence of conformational changes induced by the absorption of a photon and is commonly used in optogenetic experiments (Bada Juarez et al., 2021).hA 2A R is a class A G-protein-coupled receptor that is expressed in the cardiovascular, respiratory, immune and central nervous systems, and is implicated in immunosuppression and the regulation of sleep (Fredholm et al., 2001).
Here we report two room-temperature AR3 structures: the first, by SSX, is the dark-adapted state of the protein; the second, by time-resolved SFX, is the 110 ns photocycle intermediate.We also report crystal structures of human hA 2A R in complex with two different ligands determined by SSX.

Protein production and reconstitution in LCP
The AR3 protein was produced as previously described (Bada Juarez et al., 2021).The purified protein sample was concentrated in distilled water to 20 mg ml À1 and stored in the dark at 4 C. AR3 was reconstituted in LCP at 20 C, by mixing the protein with monoolein lipid [9.9 monoacylglycerol (MAG), Nu-Chek Prep Inc.] in a 40:60 volume ratio using two 100 ml gas-tight Hamilton syringes (No. 81065, Hamilton) connected by a syringe coupler (SPTLabtech), as previously described (Caffrey & Cherezov, 2009).The reconstitution procedure was performed under dim, red light.
2.2.Production of high-density micrometre-sized crystals using the VIALS approach 2.2.1.Archaerhodopsin-3.Starting with a previously known crystallization condition for the growth of large crystals (>40 mm), a grid screen of solutions composed of 28-36%(v/v) polyethylene glycol 600 (PEG 600) (Fluka Analytical), 100 mM MES buffer pH 5.0-6.5, 150 mM NaCl and 50-200 mM CaCl 2 was prepared and dispensed by an automated liquid-handler system (Hamilton StarLet) into glass vials purchased from Thermo Scientific (catalogue No. 17324073) [Fig.1(a), step 1].Approximately 5-10 ml of protein-laden LCP was injected (using a gas-tight Hamilton syringe of 100 ml connected to a SPTLabtech mosquito LCP narrow-bore short needle of 0.15 mm internal diameter) into each small glass vial filled with 300 ml of precipitant screen solution, and each vial was sealed with a screw cap also purchased from Thermo Scientific (catalogue No. 17334043) [Fig.1(a), step 2].The crystallization procedure was performed under dim, red light and the glass vials were stored in storage boxes at 20 C in the dark.Crystal growth was periodically monitored using a stereo high-magnification microscope equipped with cross-polarizers [Fig.1(a), step 3].Occasionally, a small amount of crystalladen LCP sample was retrieved, sandwiched between two glass slides, and observed under the microscope with the help of the dark-field and cross-polarizer [Fig.S1(a) in the supporting information].Large quantities of AR3 micrometre-sized crystals in LCP appeared after 2 to 3 days, with their density and size varying as a function of PEG 600 and CaCl 2 concentration.Ahead of the SSX and SFX measurements, many glass vials containing the same optimized crystallization solution [33%(v/v) PEG 600, 100 mM MES buffer pH 5.5, 150 mM NaCl and 150 mM CaCl 2 ] were prepared, each containing $30-50 ml of protein-laden LCP thread [Fig.1(a), steps 4-6, and Fig. S1].

Room-temperature SSX data collection
SSX diffraction data were collected at 20 C on the I24 beamline at Diamond Light Source (DLS) using a highviscosity LCP extruder installed vertically at 90 to the synchrotron X-ray beam [Fig.S2(a)].The LCP extruder was operated using a high-performance liquid chromatography water pump and compressed helium gas as described by Weierstall et al. (2014).High-density AR3 and hA 2A R microcrystals of size 5-10 mm in LCP threads were retrieved from the glass vials using a clean, gas-tight Hamilton syringe plunger and transferred to a 250 ml gas-tight Hamilton syringe (No. 81120 Hamilton).This was to remove the excess mother liquor (Fig. S3) and facilitate sample loading into the LCP extruder sample reservoir of 20 ml (Weierstall et al., 2014).For the AR3 crystals, the procedure above was performed under dim, red light.The AR3 and hA 2A R crystal samples were extruded using 100 mm and 50 mm inner diameter (ID) silica capillaries, respectively, at an average pump flow rate of $50-150 nl min À1 .The extruded sample was caught in a sample catcher, connected to a pump that was regularly cleaned.Data were continuously recorded (shutterless mode) by a PILATUS3 6M detector running at 10 Hz, using an X-ray beam focused at 9 Â 6 mm (FWHM) with an energy of $12.8 keV (3.0 Â 10 12 photons s À1 ).

AR3 time-resolved SFX data collection
Time-resolved (TR) SFX diffraction data were collected at 20 C on the BL2 EH3 station at SPring-8 Angstrom Compact Free Electron Laser (SACLA) in Japan.AR3 crystals of around 5-10 mm in LCP threads were retrieved from the glass vials, using the same method described above (see Section 2.3 and Fig. S3), and loaded into a sample reservoir of 60 ml of a high-viscosity cartridge-type (HVC) extruder (Shimazu et al., 2019;Tono et al., 2015) under dim, red light.A stream of AR3 microcrystals, delivered from the HVC extruder at an average flow rate of $15-20 ml min À1 from a nozzle of 75 mm ID, was perpendicularly aligned with an XFEL beam and a nanosecond synchronized optical parametric oscillator (OPO) laser in a two-way excitation open setup (Kubo et al., 2017;Nango et al., 2016) [Fig.S2(b)].While XFEL pulses of <10 fs (440 mJ per pulse) operated at a repetition rate of 30 Hz focused into an approximately 1.5 mm (FWHM) diameter spot size, the OPO laser of 571 nm wavelength ran at 10 ns of pulse length at a frequency of 15 Hz with a focal spot of 40 mm (FWHM).The time delay between the pump laser and the X-ray pulse was 110 ns.The diffraction patterns were recorded on a 4 megapixel multiport charge-coupled device (4MPCCD) detector (Kameshima et al., 2014) at 50 mm distance from the sample.The hit rate was continuously monitored using the SACLA real-time data-processing pipeline (Nakane et al., 2016).

Data processing and structure determination
Diffraction patterns were integrated using DIALS (Winter et al., 2018).Protein structures were determined by molecular replacement using PHASER (McCoy et al., 2007) using as the search models Protein Data Bank (PDB) entries 1uaz (Enami et al., 2003) for the AR3 structures and 5mzj (Cheng et al., 2017) and 5olv (Rucktooa et al., 2018) for the hA 2A R structures in complex with theophylline and LUAA47070 ligands, respectively.Heteroatoms were removed from all search models.All structures were initially optimized through iterative cycles of manual rebuilding using Coot (Emsley et al., 2010) and refinement using PHENIX (Adams et al., 2010;Liebschner et al., 2019).Ligands, lipids, water molecules and other solvent ions were later modelled on the basis of the 2mF o À DF c difference electron-density maps and B factors.For the AR3 structures, retinal occupancy ratios for both SSX and time-resolved SFX structures were determined using several tools in parallel such as occupancy refinement in PHENIX, observation of the retinal B factors before and after refinement (using different occupancy values), observation of the calculated 2mF o À DF c and omit maps corresponding to each occupancy value, and ligand validation tools integrated in PHENIX and Coot.For the hA 2A R complex structures, TLS groups were initially identified using the TLSMD server (Painter & Merritt, 2006), with subsequent iterative cycles of restrained maximum-likelihood and TLS refinement performed using phenix.refine.All final structure models were validated with MolProbity (Chen et al., 2010) implemented in PHENIX.Data collection and refinement statistics for the AR3 and hA 2A R structures solved in this study are summarized in Table 1.Omit maps were generated in PHENIX and all the figures were prepared with PyMOL (Schro ¨dinger).Structure alignments and root-mean-square deviation (RMSD) calculations were performed using PyMOL.The atomic coordinates and structure factors have been deposited in the PDB under the following accession codes (see Table 1 for data set names): 6guy (SSX structure of AR3 dark-adapted ); 7zy3 (time-resolved SFX structure of AR3 110 ns ); 8a2o (SSX structure of hA 2A R-theophylline) and 8a2p (SSX structure of hA 2A R-LUAA47070).

Room-temperature structure of AR3 by SSX
The AR3 microcrystals of sizes 5-10 mm taken to DLS were loaded into a high-viscosity injector and passed across the X-ray beam (see Section 2.3) at an average flow rate of $5-10 ml min À1 at room temperature.From a total sample volume of $200 ml of crystal-laden LCP thread [Figs.S1(a) and S1(b)], 1438 frames were successfully indexed in the space group P2 1 2 1 2 1 .The SSX AR3 final structure was solved by molecular replacement and refined to 2.2 A ˚with an R work and R free of 22.3 and 25.2%, respectively (Table 1).The good-quality data and refinement statistics have yielded high-quality electrondensity maps [Figs. 2(a) and 2(e)] that allowed us to model 43 water molecules and ten lipid fragments.Light-driven proton transporters, such as AR3, critically depend on a coordinated network of internal water molecules to mediate proton translocation across membranes.This hydrogen-bond network   29, 48.27, 104.90 46.20, 48.30, 104.70 40.53, 182.31, 144.27 40.44, 181.84, 144 in the Schiff base region was clearly resolved in our structure [Fig.2(a)].In addition, the retinal chromophore covalently bound to the Lys226 in our SSX AR3 structure (AR3 dark-adapted ) [Fig.2(e)] was modelled with an occupancy ratio of 70% 13-cis and 30% all-trans isomers, consistent with previously solved AR3 dark-adapted structures [PDB entry 6gux (Bada Juarez et al., 2021) and PDB entry 6s63 (Axford et al., 2022)].The post-translational modification of Gln7 at the N-terminus (conversion to a pyroglutamate residue) was well resolved (Bada Juarez et al., 2021).To confirm the validity of the VIALS method, our SSX AR3 dark-adapted structure was compared and found to be in good agreement with the darkadapted structures previously determined at cryo and room temperatures (PDB entries 6gux and 6s63), with C RMSD values of 0.25 and 0.24 A ˚, respectively [Fig.3(a)].The minor differences observed were only related to the orientation of a small number of surface side chains.The mean B factor of the obtained structure is consistent with the resolution, temperature and data processing (see Table S1).
3.2.Room-temperature SFX structure of AR3 obtained 110 ns after photoexcitation AR3 microcrystals in LCP (5-10 mm) were taken to SACLA and time-resolved SFX data were collected as described in Section 2.4.The LCP thread with a high density of microcrystals was continuously ejected from the high-viscosity sample injector and a femtosecond pump-probe experiment was performed with a time delay of 110 ns.With a crystal hit rate of up to 31% (determined by the SACLA realtime data-processing pipeline; Nakane et al., 2016), a full data set was obtained within 2.5 h using only $60 ml of crystalladen LCP.A total of 11 912 frames were successfully indexed in the space group P2 1 2 1 2 1 and the high quality of the diffraction data was supported by the figures of merit R split and CC 1/2 of 16.1 and 99.2%, respectively (Table 1).The AR3 110 ns final structure was refined to 1.7 A ˚resolution with an R work and R free of 18.12 and 19.64%, respectively.The resulting electron-density maps revealed the presence of 61 water molecules and lipid fragments.Similar to its SSX counterpart structure AR3 dark-adapted (PDB entry 6guy), the pentagonal hydrogen-bonding network, formed by the side chains of Asp95 and Asp222 and W402, W401 and W406, was exceptionally well resolved [Fig. 2 S1 and S2, respectively.
respectively [Fig. 3(b)].The mean B-factor value of 36.0A ˚2 is also within the values reported for a room-temperature structure of 1.7 A ˚(Table S1).
The crystals taken to SACLA were first tested for isomorphism and diffraction quality at the I24 beamline (DLS).On the basis of previous observations and discussions by the SX community (Andersson et al., 2019;Mehrabi et al., 2021;Nango et al., 2019), we postulate that the difference in resolution observed between the SSX AR3 dark-adapted and the SFX AR3 110 ns structures might be due to the different parameters of the synchrotron beamlines compared with the XFELs.In addition, during our experiments, the extruder nozzle used at SACLA was 75 mm ID whereas that at DLS was 100 mm ID (the only size available at the time).Larger LCP extruder nozzle sizes usually result in data with higher scattering background from the LCP (Andersson et al., 2019;Kova ´csova ´et al., 2017;Kubo et al., 2017).Finally, the SACLA X-ray beam spot dimensions were 1.3 Â 1.5 mm (FWHM) while the beam at DLS was focused at 9 Â 6 mm (FWHM) for an average crystal size of 5 to 15 mm.

Room-temperature structures of hA 2A R in complex with theophylline and LUAA47070 by SSX
To demonstrate the applicability of our VIALS method to structure-based drug design, we performed ligand-soaking experiments.Glass vials containing the hA 2A R microcrystals were divided into two groups (the same number as the number of ligands) and the crystallization buffer was exchanged (using a gas-tight Hamilton syringe with a long needle) [Fig.1(b)] with new mother liquor, supplemented either with 0.5 mM theophylline, a weak non-selective hA 2A R antagonist (Segala et al., 2015), or 1 mM LUAA47070, a known hA 2A R antagonist (Sams et al., 2011).The crystals within the glass vials were left to incubate at 20 C for 3 to 6 h, to allow ligand exchange and binding, prior to SSX data collection at DLS on the I24 beamline.Using the high-viscosity LCP extruder (Weierstall et al., 2014) diffraction data were collected at an average flow rate of $3-8 ml min À1 .A total of 10 457 and 3618 frames were successfully indexed in the space group C222 1 for the hA 2A R in complex with theophylline and LUAA47070, respectively.The final refined structures of hA 2A R in complex with theophylline and LUAA47070 were refined to 3.45 and 3.50 A resolution with an R work /R free of 21.85/24.12and 21.20/25.40,respectively.Despite the resolution, our hA 2A R SSX structures in complex with theophylline and LUAA47070 were well resolved with good-quality electron-density maps.The presence of the ligands was initially observed by strong 2mF o À DF c electron densities around the binding pocket and confirmed by difference electron-density omit maps, generated from the crystallographic data [Figs.2(c), 2(g), 2(d) and 2(h)].The binding modes of theophylline and LUAA47070 are the same as those observed in the cryo structures (PDB entries 5mzj and 5olv, respectively).In the hA 2A R-theophylline Electron-density maps of hA 2A R's four disulfide bridges (Cys71-Cys159, Cys74-Cys146, Cys77-Cys166 and Cys259-Cys262).(a) and (c) show the 2mF o À DF c (blue mesh, contoured at 1.5) and mF o À F c (green and red meshes, contoured at AE3) electron-density maps around the hA 2A R four disulfide bonds in complex with theophylline and LUAA47070, respectively.(b) and (d) show the omit 2mF o À DF c (blue mesh, contoured at 1.5) and mF o À DF c (green and red meshes, contoured at AE3) electron-density maps when the disulfide bridges are omitted/broken.Here, strong positive electron density is seen where original disulfide bridges were formed.The hA 2A R residues in complex with theophylline and LUAA47070 are represented by yellow and green sticks, respectively.Pictures were prepared in PyMOL.
Finally, our structures superimposed very well with those acquired under cryo conditions [PDB entry 5mzj (Cheng et al., 2017) and PDB entry 5olv (Rucktooa et al., 2018)], with RMSD values of 0.44 and 0.33 A ˚for all C atoms, respectively.The observed higher average B factors were expected for resolutions of 3.5 A ˚and room-temperature SSX data collection (Table S2).

Summary
VIALS (versatile approach to high-density microcrystals in lipidic cubic phase for serial crystallography) is a semi-automated high-throughput method for the large-scale production of microcrystals of membrane proteins in LCP.The method allows a more efficient screening of crystallization conditions than is possible with the routinely used glass syringe method, and is suitable for ligand soaking for drug-design studies.Finally, it has the advantage of easy sample transportation compared with syringes.We have demonstrated the compatibility of the microcrystals generated with LCP injectors at both synchrotron and XFEL sources.Room-temperature structures of the AR3 photoreceptor protein in its darkadapted state and of the 110 ns photocycle intermediate were obtained, and the first room-temperature structures of hA 2A R in complex with theophylline and LUAA47070 were solved.We also anticipate that the microcrystals generated will be suitable for heavy-atom soaking in meso.

Figure 1
Figure 1Schematic representation of the VIALS workflow.(a) Flowchart depicting the semi-automated high-throughput procedure of VIALS.Step 1: design and automated preparation of a rational grid screen based on a previously known crystallization condition (e.g. for the growth of large crystals).Step 2: tiny amounts of protein-laden LCP string are manually injected into each previously prepared small glass vial.Step 3: periodic direct inspection of crystal growth using a stereo high-magnification microscope.Steps 4-5: scale-up process once the optimal microcrystal size/density condition is found.Step 6: microcrystal growth and storage for SX measurements.(b) VIALS experimental setup for ligand soaking in meso in parallel ahead of room-temperature SX data collection.

Figure 2
Figure 2 Quality of the electron-density maps.(a)-(d) 2mF o À DF c electron-density maps (blue mesh) contoured at the 1.5 level showing close-up views of AR3 dark-adapted (PDB entry 6guy, light pink), AR3 110 ns (PDB entry 7zy3, light magenta), and the hA 2A R binding site in complex with theophylline (PDB entry 8a2o, yellow) and LUAA47070 (PDB entry 8a2p, green), respectively.(e)-(h) mF o À DF c omit electron-density maps contoured at the AE3.0 level showing strong positive density (green mesh) when ligands are omitted during refinement.Protein residues and ligands are shown as sticks and water molecules as red spheres.Dashed lines in panels (a) and (b) represent the hydrogen-bond network.

Figure 3
Figure 3 Superposition of the room-temperature SX structures with their cryogenic counterparts.(a) Superposition of the SSX extruder dark-adapted AR3 (PDB entry 6guy, light green), SSX dark-adapted AR3 (PDB entry 6s63, grey) and cryo dark-adapted AR3 (PDB entry 6gux, yellow) structures.The inset right panel shows details of the retinal and the pentagonal hydrogen-bond network in AR3.(b) Superposition of the SFX 110 ns photocycle intermediate AR3 (PDB entry 7zy3, pink), SSX light-adapted AR3 (PDB entry 6guz, purple) and cryo light-adapted AR3 (PDB entry 6s6c, silver) structures.The inset right panel shows details of the retinal and the pentagonal hydrogen-bond network in AR3.(c) Superposition of the SSX hA 2A R-theophylline complex structure (PDB entry 8a2o, turquoise) and its synchrotron cryo counterpart's structure (PDB entry 5mzj, gold).The inset right panel shows details of the ligand-binding pocket for both structures.(d) Superposition of the SSX hA 2A R-LUAA47070 complex structure (PDB entry 8a2p, dark blue) and its synchrotron cryo counterpart's structure (PDB entry 5olv, light blue).The inset right panel shows details of the ligand-binding pocket for both structures.Ligands, selected lipids and amino acid side chains are in stick representation.Water molecules are shown as spheres and hydrogen bonds are represented by dashes.Pictures were prepared using PyMOL.AR3 and hA 2A R structural alignment RMSD values can be found in TablesS1 and S2, respectively.

Table 1
Crystallographic statistics for data collection.