Structure-based insights into evolution of rhodopsins

Rhodopsins, most of which are proton pumps generating transmembrane electrochemical proton gradients, span all three domains of life, are abundant in the biosphere, and could play a crucial role in the early evolution of life on earth. Whereas archaeal and bacterial proton pumps are among the best structurally characterized proteins, rhodopsins from unicellular eukaryotes have not been well characterized. To fill this gap in the current understanding of the proton pumps and to gain insight into the evolution of rhodopsins using a structure-based approach, we performed a structural and functional analysis of the light-driven proton pump LR (Mac) from the pathogenic fungus Leptosphaeria maculans. The first high-resolution structure of fungi rhodopsin and its functional properties reveal the striking similarity of its membrane part to archaeal but not to bacterial rhodopsins. We show that an unusually long N-terminal region stabilizes the protein through direct interaction with its extracellular loop (ECL2). We compare to our knowledge all available structures and sequences of outward light-driven proton pumps and show that eukaryotic and archaeal proton pumps, most likely, share a common ancestor.

M icrobial (type 1) rhodopsins are the most abundant family of light-harvesting proteins. Type 1 rhodopsins are heptahelical transmembrane (7TM) proteins that covalently bind the retinal chromophore and use the energy of light to perform different biological functions, such as ion pumping [1][2][3][4][5][6] , ion channeling 7-9 , sensoring [10][11][12] , and kinase activity 13 . The explosion of research on microbial rhodopsins, in large part, owes to the key role of these proteins in optogenetics, a methodology that caused a revolution in neuroscience 14,15 . Recently, discoveries of genomics and metagenomics show that rhodopsins are highly abundant, perform extremely diverse functions, and are present in all kingdoms of life as well as many large viruses 16,17 . Rhodopsins are considered to be the most abundant light-harvesting proteins on earth and the major light capturers in the oceans 18 . Given the ubiquity of rhodopsins and their crucial ecological role, there is little doubt that these proteins played a crucial role in the evolution of life on earth. Recently, rhodopsins have been identified in Asgard archaea 19 , the archaeal superphylum that includes the likely ancestors of eukaryotes 20,21 . Furthermore, a recent analysis of microbial rhodopsins has led to the hypothesis that retinal-based phototrophy emerged early in the evolution of life on Earth, predating the rise and profoundly impacting the evolution of photosynthesis 22 . Thus, the study of the functions and evolution of rhodopsins could yield valuable information on the origin and early evolution of life.
Among the factors critical in the evolution of early life was the ability to convert the energy of sunlight into a transmembrane proton gradient that provides for chemiosmotic coupling 23 . Light-driven proton pumps generating transmembrane gradients are the most abundant among the rhodopsins 24 . The structure and function of these proteins are exceptionally well studied, providing a rare opportunity to use a structure-based approach to explore evolutionary relationships among proteins.
Whereas structures of archaeal and bacterial proton pumps have been thoroughly characterized, this is not the case for rhodopsins of unicellular eukaryotes. To fill this gap and enable a structure-based analysis of the evolutionary relationships among the rhodopsins, we determined a high-resolution structure and performed an in-depth functional study of a light-driven proton pump LR (Mac) from the fungus Leptosphaeria maculans. Leptosphaeria maculans is a major pathogen of Brassica napus, an agricultural plant that is used as a feed source for livestock and the production of rapeseed oil. A dramatic epidemic of L. maculans occurred in Wisconsin on cabbage. The fungus destroys around 5-20% of canola yields in France 25 . The disease is also harmful in England 26 . Rapeseed oil is the preferred European oil source for biofuel due to its high yield. B. napus produces more oil per land area than other sources like soybeans. Thus, apparent from the fundamental importance of a thorough characterization of rhodopsin from a unicellular eukaryote, the study of LR (Mac) also could help understand the role of the rhodopsin in the fungus pathogenicity.
To date, high-resolution structures of multiple protonpumping archaeal and bacterial rhodopsins have been solved [27][28][29][30] . For instance, more than 100 structures of the moststudied microbial rhodopsin, bacteriorhodopsin from archaeon Halobacterium salinarum (HsBR), and its mutants, were deposited to the Protein Data Bank (PDB) since 1997 31 . By contrast, structural characterization of eukaryotic type-1 rhodopsins lags far behind. Two high-resolution structures of an H + pumping rhodopsin from unicellular eukaryotes are currently available, Acetabularia rhodopsin (AR) from the marine alga Acetabularia acetabulum 32,33 and Coccomyxa rhodopsin from Coccomyxa subellipsoidea 34 . The AR has been deemed to be closely similar to archaeal HsBR although this protein has a 200 ms long photocycle, compared to the 20 ms photocycle in HsBR 35 . High-resolution structures of eukaryotic proton-pumping rhodopsins from fungi would be crucial for understanding the evolution of this expansive protein superfamily.
The study of eukaryotic membrane proteins is generally hampered by the complications with their expression and crystallization. We used our recently developed techniques of the expression in LEXSY 36 and crystallized the proton-pumping type 1 rhodopsin LR from the unicellular fungus Leptosphaeria maculans. Previously, biophysical features of LR have been reported to resemble those of archaeal pump HsBR 37 with millisecond-length photocycle and similar active center organization 38 . However, unlike archaeal rhodopsins, LR and other eukaryotic rhodopsins have elongated N-and C-terminal regions that could be involved in the adaptation of the protein to different lipid environment 39 . Replacement of the LR N-terminal region did not impair the structural integrity and biochemical functionality of the protein, but the direct impact of modifications of the N-terminus has not been reported 38 . Recently, a full-length LR has been shown to enable efficient neural silencing by blue light in mammalian cells 40 .
Here we present the first high-resolution, at 2.2 Å, crystal structure of a light-driven H + pump from fungi and elucidate the functional role of the N-terminal region. We further report the results of a structure-based comparison of LR with light-driven H + pumps from all domains of life. This structural analysis revealed the archaeal ancestry of eukaryotic type 1 rhodopsins. Further structure-based phylogenetic analysis confirmed the archaeal affinity of eukaryotic proton-pumping rhodopsins, suggesting that the archaeal host of the proto-mitochondrial endosymbiont was capable of light-driven proton pumping. Besides that, LR is the second membrane protein after CrChR2 36 , which was expressed in LEXSY system, further crystallized, and the structure was solved at high-resolution. Therefore, LEXSY expression system might be a reasonable alternative for the expression of eukaryotic membrane proteins in structural studies and, more specifically, eukaryotic microbial rhodopsins.

Results and discussion
Overall LR structure and function. We performed a comparative functional characterization of the full-length (residues 1-313) LR and its previously partially characterized N-terminally truncated version (residues 49-313). These proteins did not show any detectable expression in E. coli, and therefore, we expressed them in Leishmania tarentolae (LEXSY) as previously described for channelrhodopsin 2 (ChR2) 36 (see "Methods" for details).
First, we describe the structure of the full-length protein, which will be the structural basis for understanding the differences in the properties of rhodopsins and their evolution. LR was crystallized with an in meso approach similar to that used previously 5,6 . The structure was solved with crystals grown at pH 7.0 by the molecular replacement method, using the coordinates of HsBR (PDB ID: 1C3W), and refined to 2.2 Å resolution ( Table 1). The crystals belong to the P21 space group and comprise two protomers in an asymmetric unit. LR (residues 42-286), all-trans-retinal (ATR), 19 lipid molecules, and 65 water molecules inside the protein are clearly resolved in the electrondensity map, whereas the N-and C-terminal domains lack resolution for 41 and 31 residues, respectively ( Supplementary  Figures 1-2).
The LR bundle consists of seven transmembrane helices-TM1-TM7, connected by three intracellular (ICL 1-3) and three extracellular (ECL 1-3) loops, with retinal chromophore covalently attached to the K270 on the TM7 (Fig. 1). The overall structure of LR aligns well with the archaeal pump HsBR 37 (PDB ID: 1C3W) and eukaryotic pumps ARII 32 (PDB ID: 5AWZ) and  Figure 3). The interior portions of LR and HsBR are virtually identical structurally, and the only major difference is observed at the extracellular sides of these rhodopsins (Fig. 1). Specifically, ECL1 of LR is much longer, consisting of 30 amino acids, whereas the corresponding loop in HsBR contains only 16 residues, which results in a major difference in length, 44 Å in LR vs 21 Å in HsBR. Interestingly, ECL1 of LR is reminiscent of the highly conservative ICL1 beta-sheet domain of heliorhodopsins, which was previously shown to play an important role in dimer formation 41,42 (Supplementary Figure 4).
Structural features of LR important for proton transport. In the LR structure, the retinal chromophore is fixed in an all-trans conformation. The protonated retinal Schiff base (RSB) points toward the extracellular side and donates a hydrogen bond to the w8 molecule (Fig. 2a). This arrangement is stabilized by the RSB counterions, D266 and D139, which are counterparts to D212 and D85 in HsBR. The counterions and water molecules (w5, w8, and w10) form a hydrogen bond pentamer, which stabilizes the RSB and the extracellular part of the retinal binding pocket. The counterions take the same positions as the corresponding amino acids of HsBR, and as in HsBR, are additionally stabilized by T143 and S97 residues. By contrast, Gloeobacter rhodopsin (GR) and other proteorhodopsins (PRs) exhibit a different conformation of the RSB region 43 . In the GR structure, D121 (counterpart to D85 in HsBR, the primary proton acceptor) forms a hydrogen bond with H57 44 . This configuration of the counterions was shown to cause pH dependence of the photocycle due to deprotonation of the H57 residue at pH higher than physiological 45 (Supplementary Figure 5). In archaeal pumping rhodopsins, the proton release is mediated by a glutamate pair (E194-E204 in HsBR). In LR and other eukaryotic proton pumps, the release group has a similar construction. The E258 residue interacts with D248 that equivalently substitutes E204 of HsBR in a configuration similar to that of HsBR. By contrast, the bacterial proton-release group consists of only one negatively charged residue, the sole glutamate in TM5 or TM6 ( Supplementary Figures 6 and 7). A common feature of bacterial proton pumps is that they contain a wateraccessible cavity that is connected with the extracellular bulk and protrudes up to R136 (Fig. 2a). By contrast, in both archaeal and eukaryotic proton-pumping rhodopsins, extracellular cavities are separated by a negatively charged pair of residues (E194-E204 in HsBR and D248-E258 in LR) into two parts, one of which is open to the bulk (Fig. 2a). The proton-releasing moiety of PRs consists of the sole glutamate residue in TM5. The configuration of this part is conserved in the entire family of bacterial proton-pumping rhodopsins (Supplementary Figure 7). The extracellular part of the protein is fully accessible from the bulk for the water molecules up to the arginine residue (R76 in Med12BPR), similarly to LR.
The intracellular portion of LR is structurally nearly identical to that of HsBR (Fig. 2b). In particular, reprotonation occurs through the D150 residue, presumably, followed by a synchronized movement of L147 and T143 residues 46 . In HsBR, the corresponding residues are D96, L93, and T90, respectively. In the GR structure, the residues responsible for the proton uptake are E132, Q129, and T125. This configuration results in nearly complete accessibility of the proton donor E132 to the cytoplasmic bulk (Supplementary Figure 8). In other proteorhodopsins, the glutamate is often replaced by lysine, causing an absorbance shift effect 47 .
Photocycle of LR proton pump. The photocycle kinetics of HsBR and LR are closely similar as suggested by the high structural similarity and are likely to undergo similar structural rearrangements under light illumination 48 (Fig. 3). Both full-length and truncated constructs of LR in DDM micelles exhibit almost similar relaxation time that suggests that N-terminus has little impact on protein photoactivation (Supplementary Figure 9). After photoactivation, the LR photocycle starts with a short-living K state followed by blue-shifted L-and M states associated with proton release in archaeal proton pumps 49 . Next, the protein relaxation into the ground state occurs through N-and O states that complete the photocycle [50][51][52][53] . Consistent with RSB configuration difference, PR photocycle, and in the particular photocycle of green-absorbing proteorhodopsin (GPR) has 4 photointermediate states under neutral pH, lacking an apparent L state 54 . Consistent with structural similarity, the photocycle of LR and HsBR demonstrates nearly similar photocycle composition that further supports a close relation between LR and archaeal proton pump HsBR.
LR oligomeric state. Almost all microbial rhodopsins exist as oligomers in the native lipid environment. The oligomeric state was previously shown to contribute to protein stability 55,56 and in some cases to function 57 . Besides that, the stoichiometry of the fundamental oligomer of microbial rhodopsins was recently found to correlate with the phylogenetic origin of rhodopsins 58 . Even though LR forms dimers in the crystal lattice (Supplementary Figure 1), the monomer-monomer interface was surprisingly low (187 Å 2 ) 59 . To assess oligomeric states of LR rhodopsin in a solution, we performed glutaraldehyde crosslinking of LR followed by SDS-PAGE analysis (Fig. 4a). This method was previously used in combination with size-exclusion chromatography (SEC) to confirm pentameric assembly of viral  . 4b). Given that the empty DDM micelle size is~50 kDa, pentameric KR2 and trimeric NsXeR elutes at 321 and 192 kDa respectively (Fig. 4c). Therefore, it is likely that LR in DDM micelles exist in monomeric and trimeric forms only. It is also noteworthy that truncation of N-terminus changes the ratio between monomers and trimers from~2:5 for LR 1-313 to~40:1 for LR 49-313 and thus impacts on detergent resistance of the protein.
Using template-based modeling GalaxyHomomer server, we proposed the trimeric state of LR using monomer template-based on archaeorhodopsin-2 model (PDB: 3WQJ) 60 . LR trimer has a higher monomer-monomer interface area (1799 Å 2 ) and therefore is a likely major oligomeric form of LR rhodopsin (Fig. 4d). Besides that, protomers in both LR and HsBR trimers interact within TM2'-TM4 helices (Fig. 4e), that further highlights the similarity between LR and HsBR. It should also be noted that despite the similarity of ECL1 of LR and ICL1 of heliorhodopsins, heliorhodopsins employ different dimer configuration with a much higher interface surface area (Supplementary Figure 11) Structural roles of ECL1 and the N-terminal domain. We showed that the LR core (membrane part) of the protein is strikingly similar to the core of HsBR and, in general, to the corresponding structures of other archaeal rhodopsin proton pumps as expected of core parts of highly conserved homologous proteins. By contrast, substantial structural variation could be expected to exist in the N-and C-terminus regions. To assess the role of the N-terminal domain in the structural stability of the protein, we searched protein sequence databases for homologs of LR using BLASTP 61 and identified multiple opsins with high sequence similarity to LR. Despite the overall high similarity, we identified at least two subgroups among fungal rhodopsins that follow different patterns in the extracellular side of the protein.
The first subgroup contained an elongated ECL1 domain, and a short beta-strand bears the N-terminus, whereas the second subgroup lacked both these features (Fig. 5a, Supplementary  Figure 12). Secondary-structure prediction with RAPTORX 62 suggests the presence of an additional beta-strand near the Nterminus of LR, but not in subgroup 2 of fungal rhodopsins or other light-driven proton pumps. Therefore, the elongated ECL1 domain and the N-terminal beta-strand appear to be signatures of the subfamily of fungal rhodopsins that includes LR rhodopsin. Consistent with the prediction, the LR crystal structure contained 8 residues from the N-terminal region (residues [42][43][44][45][46][47][48][49], that aligned with ECL1 and belonged to a 3-stranded antiparallel βsheet. This N-terminal β-strand strongly interacts with the ECL1 loop and is directly involved in its stabilization through at least 7 hydrogen bonds. In addition, the N-terminus of LR is connected with the TM α-helices via the V47-S52 and G48-D256 interactions (Fig. 5b). To verify the effect of the N-terminal truncation, we measured the thermal stability of the LR protein constructs with a nano-DSF method under different pH and detergent concentrations 63 (Supplementary Figure 13). The F350/330 ratio of truncated LR showed two inflection points (corresponding to protein-unfolding temperatures) at 49.3 and 64.8°C, whereas the full-length LR demonstrated a single inflection point at 66.5°C (Fig. 5c).
To assess the contribution of the N-terminal domain 64 to protein stabilization we used CABS-flex 2.0 coarse-grain protein modeling server 65 . We used two models, full-length LR structure (resolved residues 42-286) and LR structure without the Nterminal region (residues 49-286), that represent LR constructs used for functional tests (Fig. 5e). Consistent with previous observations, the average root mean square fluctuation (RMSF) of LR 49-286 (0.74 Å) was higher than the average RMSF of LR 42-286 (0.65 Å). In particular, the highest destabilization was predicted for the ECL1 and ECL3 domains that directly interact with the N-terminal domain in the crystal structure. Given that the F350/330 ratio represents a change in tryptophan fluorescence, the additional melting point at 49.3°C is likely to be caused by tryptophans on the extracellular side of the protein (W56, W190, W192, and W244). Increased flexibility of TM1, ECL1, and ECL3 domains can impact hydration and orientation of tryptophans on the extracellular side, with surprisingly little influence on protein kinetics.

Proton translocation experiments.
To estimate the protonpumping activity of the full-length and truncated proteins, we performed ion translocation experiments using POPC:POPS proteoliposomes (see "Methods" for details). Under weak acidic conditions (pH 6.5, 100 mM NaCl), the full-length protein demonstrated a substantially higher proton-pumping activity than the truncated derivative, 0.3 pH units for LR 1-313 against 0.1 for LR 49-313 (Fig. 5d, g).
A possible explanation of such a difference in proton pumping activity of two proteins with nearly similar photocycle is protein orientation in liposomes. Membrane protein orientation generally follows the "positive-inside" rule 66,67 . Truncation of a slightly negatively charged N-terminus (D4, E7, E8, and D40) substantially altered the charge distribution of the extracellular part of the LR protein. Because extracellular parts of membrane proteins are more negatively charged than intracellular parts, such removal should influence protein orientation in both cell membranes and liposomes. Besides, protein orientation might be also influenced by decreased stability of the extracellular portion of the truncated protein.
Structure-based phylogenetic analysis of light-driven proton pumps. Different aspects of the evolution of microbial rhodopsins have been addressed in detail, featuring both large datasets 68 and rhodopsins with putative functions 24,[69][70][71] . However, all the conclusions in these studies are derived from sequence-based phylogenetic analysis. To address different perspectives of the discussion, we performed a structure-based clustering of the lightdriven outward proton pumps using 16 structures available from the PDB (see Methods for details). To this end, we superimposed all structures of proton pumps and calculated the pairwise RMSD values. We used these values to cluster the rhodopsins with a hierarchical clustering algorithm (Fig. 6). Notably, and in agreement with both the detailed structural comparisons described here and the previous phylogenetic analyses 24,68 , the protonpumping rhodopsins from archaea and eukaryote confidently clustered together, to the exclusion of the structurally distinct bacterial rhodopsins (Fig. 6b). Rhodopsin genes appear to be subject to extensive horizontal gene transfer, resulting in mixed branches in phylogenetic trees and complicating inferences of common ancestry 72,73 . Nevertheless, the high structural similarity between archaeal and eukaryotic proton pumps including the conformation of key functional regions (Fig. 6c), such as the retinal binding pocket and the regions involved in proton release and proton uptake, along with functional similarities (pH dependence of the photocycle, the intermediate states), presents strong evidence of an archaeal origin of eukaryotic protonpumping rhodopsins (and, most likely, all other eukaryotic rhodopsins as well). Thus, rhodopsins appear to belong to a small set of genes, apart from the core components of the translation, transcription, and replication systems, that have been inherited by the protoeukaryotes from their archaeal ancestors.

Methods
Phylogenetic analysis. In total, 28 sequences of rhodopsins with H + -pumping activity were aligned using MUSCLE. Phylogenetic reconstruction was conducted by maximum likelihood (ML) using PhyML with the following parameters: Jones-Taylor-Thornton model, SH-like approximate likelihood-ratio test, and estimated gamma-distribution parameter. The following rhodopsins were used for phylogenetic tree construction and sequence alignment: HsBR, Bacteriorhodopsin from Archaea (Halobacterium salinarum; P02945); HwBR/MR, Midrhodopsin from Eubacteria (Haloquadratum walsbyi DSM 16790, Q18DH8); aR-1, Archaerhodopsin-1 from Archaea (Halorubrum chaoviator, P69051); aR-2, Archaerhodopsin-2 from Archaea (Halobacterium sp. AUS-2, P29563), aR-3, Archaerhodopsin-3 from Archaea (Halorubrum sadomense, P96787); cR-1, Cruxrhodopsin-1 from Archaea (Haloarcula argentinensis, Q57101); cR-3, Cruxrhodopsin-3 from Archaea (Haloarcula vallismortis, P94854); dR-3, Deltarhodopsin-3 from (Haloterrigena thermotolerans, I4DST7); LR, Leptosphaeria Rhodopsin from Eukaryote (Leptosphaeria maculans, Q9HGT7); ARII/Ace2, Acetabularia Rhodopsin II from Eukaryote (Acetabularia acetabulum, G1K3Q0); AR, Acetabularia Rhodopsin from Eukaryote (Acetabularia acetabulum, Structure alignment. For structure alignment, 17 known atomic structure models of H+ -pumping rhodopsins, including LR, were compared pairwise by RMSD values calculated with the PyMOL align function with parameter cycles = 2 and cutoff = 1.5. After hierarchical clustering of the RMSD correlation values, the analyzed structures were divided into two clusters -"archaeal" and "bacterial". Then the structure of the reference archaeal pump HsBR was aligned to each one from the "archaeal" cluster and the structure of the reference Med12BPR was aligned to each one from the "bacterial" cluster. To represent the conservation of Cα atom positions within the clusters, for every Cα atom of the reference structure, distance to the nearest Cα atom of the respective aligned structure was measured. The measurement was repeated for every aligned structure. RMSD of the distances corresponding to each Cα atom of each reference structure, Cα RMSD , was used as a measure of the Cα position conservation.  Cloning. The genes encoding LR (1-313 aa and 49-313 aa) from Leptosphaeria maculans (UniProt Q9HGT7) were synthesized de novo. The nucleotide sequences were optimized for Leishmania tarentolae expression with GeneOptimizer software (Thermo Fisher Scientific). Both genes in fusion with the C-terminal polyhistidine tags (H6 and H9) were introduced into the integrative inducible expression vector pLEXSY_I-blecherry3 (Jena Bioscience, Germany) through the BglII and NotI restriction sites. Plasmid vectors, optimized genes, and primers used for PCR amplification can be found in Supplementary Tables 1-3. LEXSY expression, solubilization, and purification. The proteins were expressed as described previously 36 . The Leishmania tarentolae cells of the strain LEXSY host T7-TR (Jena Bioscience) were transformed with the LR 1-313 and LR 49-313 expression plasmids linearized by the SwaI restriction enzyme. After the clonal selection, the transformed cells were grown at 26°C in the dark in shaking baffled flasks in the Brain-Heart-Infusion Broth (Carl Roth, Germany) supplemented with 5 mg ml −1 Hemin, 50 U ml −1 penicillin, and 50 mg ml −1 streptomycin (Appli-Chem). When OD 600 = 1.0 was reached, 10 mg ml −1 tetracycline was added, for LR 1-313 also, 10 μM all-trans-retinal (Sigma-Aldrich) was added, and incubation continued for a further 24 h. The collected cells were disrupted in an M-110P Lab Homogenizer (Microfluidics) at 10,000 psi in a buffer containing 50 mM NaH 2 PO 4 /Na 2 HPO 4 , pH 7.6, 0.1 M NaCl, 10% glycerol, 1 mM EDTA, 2 mM 6aminohexanoic acid (AppliChem), 50 mg L −1 DNase I (Sigma-Aldrich), and complete protease inhibitor cocktail (Roche). The membrane fraction of the cell Incorporation of the protein into liposomes. Phospholipids 1-palmitoyl-2oleoyl-glycero-3-phosphocholine and 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine at a ratio 4:1 (wt wt −1 ) (POPC:POPS) were dissolved in CHCl 3 (chloroform ultrapure, Applichem Panreac) and dried under a stream of N 2 in a glass vial. The solvent was removed by overnight incubation under vacuum. The dried lipids were resuspended in 4% (w v −1 ) sodium cholate. The mixture was clarified by sonication at 4°C and LR was added at a protein:lipid ratio of 1:20 (w w −1 ). The detergent was removed by the addition of detergent-absorbing beads (Amberlite XAD 2, Supelco) and incubation at 4°C for 2 days. The 1.5 ml of liposome mixture was dialyzed against 100 mM NaCl (pH 8.0) buffer at 4°C for 18 h (two 1.5 L changes) to adjust for the desired pH.
Measurement of proton translocation activity in proteoliposomes. The measurements were performed on 1.5 ml of stirred liposome suspension at 4°C. LRcontaining liposomes were prepared following the protocol described above. Liposomes were illuminated for 10 minutes with a halogen lamp (Intralux 5000-1, VOLPI) and then were kept in the dark for another 10 min. Changes in pH were monitored with a pH meter (LAB 850, Schott Instruments). Measurements were Nano-DSF measurements. Nano-DSF measurements were performed using a Prometheus NT.48 instrument (NanoTemper Technologies, Germany). In total, 6 samples for both full-length and truncated LR were made, having pH 6.0, 7.0, or 8.0 and DDM concentration of 0.05% and 1%. First, samples with purified LR at a concentration of 1 mg ml −1 were dialyzed against 200 mM NaCl, 0.05% DDM buffer for 18 h at 4°С. After that, the protein was introduced into 50 mM Na 2 HPO 4 /NaH 2 PO 4 , 200 mM NaCl buffer with the desired pH, and incubated at 4°C for another 18 h. About 10 μL of each sample was loaded into UV capillaries (NanoTemper Technologies). The temperature gradient was set at 1°C min −1 in a range from 15 to 98°C. Protein unfolding was observed by following the change in tryptophan fluorescence at emission wavelengths of 330 and 350 nm. The ratio between the emission intensities at 350 and 330 nm (F350/F330) was used to track the structural changes with increasing temperature. Melting-point temperatures (T m ) were calculated using the peaks in the first derivative of the signal data.
Time-resolved absorption spectroscopy. The laser-flash photolysis setup was similar to that described by Chizhov and co-workers 50 . The excitation/detection systems were composed as such: Brilliant B laser with OPO Rainbow (Quantel Inc.) was used, providing pulses of 4-ns duration at 530-nm wavelength and an energy of 2 mJ per pulse. Samples (5 × 5 mm spectroscopic quartz cuvette; Hellma, Germany) were placed in a thermostated house between two collimated and mechanically coupled monochromators (LOT MSH150). The probing light (xenon arc lamp, 75 W, Hamamatsu) passed the first monochromator sample and arrived after a second monochromator at a photomultiplier tube (PMT) detector (R12829, Hamamatsu). The current-to-voltage converter of the PMT determines the time resolution of the measurement system of ca. 50 ns (measured as an apparent pulse width of the 5-ns laser pulse). Two digital oscilloscopes (Keysight DSO-x4022A, 2 units, 25 kilobytes of buffer memory per channel, respectively) were used to record the traces of transient transmission changes in two overlapping time windows. The maximal digitizing rate was 10 ns per data point. Transient absorption changes were recorded from 10 ns after the laser pulses until the full completion of the phototransformation. At each wavelength, 25 laser pulses were averaged to improve the signal-to-noise ratio. The quasi logarithmic data compression reduced the initial number of data points per trace (~32,000) to~850 points evenly distributed in a log time scale giving~100 points per time decade. The wavelengths were varied from 330 to 700 nm in steps of 10 nm using a computer-controlled step motor. The absorption spectra of the samples were measured before and after each experiment on a standard spectrophotometer (Avantes Avaspec 2048 L).
Data treatment and global fit analysis. Each data set was independently analyzed using the global multiexponential nonlinear least-squares fitting program MEXFIT 50 . The number of exponential components was incremented until the SD of weighted residuals did not further improve. After establishing the apparent rate constants and their assignment to the internal irreversible transitions of a single chain of relaxation processes, the amplitude spectra of exponents were transformed to the difference spectra of the corresponding intermediates in respect to the spectrum of the final state. Subsequently, the absolute absorption spectra of states were determined by adding the difference spectra divided by the fraction of converted molecules to the spectra of the final states. Criteria for the determination of the fraction value were the absence of negative absorbances and contributions from the initial state to the calculated spectra of the final state.
Crystallization. The crystals were grown with an in meso approach [74][75][76] , similar to that used in our previous works 7, 8 . The solubilized protein (40 mg ml −1 ) in the crystallization buffer was mixed with premelted at 42°C monoolein (MO, Nu-Chek Prep) in 3:2 ratio (lipid: protein) to form a lipidic mesophase. The mesophase was homogenized in coupled syringes (Hamilton) by transferring the mesophase from one syringe to another until a homogeneous and gel-like material was formed 77 . About 150 nl drops of a protein-mesophase mixture were spotted on a 96-well LCP glass sandwich plate (Marienfeld) and overlaid with 400 nL of a precipitant solution using the NT8 crystallization robot (Formulatrix). The best crystals were obtained with a protein concentration of 20 mg ml −1 and 100 mM HEPES, pH 7.0, 4% PEG 3350, 1 M sodium malonate pH 7.0. The crystals were grown at 22°C and appeared in 1-4 weeks. The needle-like crystals grew to 100 μm in size. Once crystals reached their final size, crystallization wells were opened as described previously 78 , and drops containing the protein-mesophase mixture were covered with 100 μl of the respective precipitant solution. All crystals were harvested using micromounts (MiTeGen), flash-cooled, and stored in liquid nitrogen before data collection.
Data collection. X-ray diffraction data were collected at ID30b beamline at ESRF, Grenoble, France at 100 K with a PILATUS3 6 M (Dectris) detector. Diffraction images were processed with XDS software and initially scaled using Aimless from the CCP4 suite. The analysis of the anisotropy indicated that the data are anisotropic; therefore, it was also treated using Staraniso server 79 . The crystallographic data statistics are shown in Table 1.
Structure refinement. Reference model (HsbR, PDB 1C3W) was used for molecular replacement using Molrep from the CCP4 suite. The initial phases were successfully obtained in the P212121 space group. The initial model was iteratively refined using REFMAC5 80 , PHENIX 81 , and Coot 82 . Anisotropically treated data on LR improved the quality of the electron-density maps compared with the isotropically treated data. Therefore, the data from the Staraniso server were used to build the final model. However, the completeness of the anisotropically treated data is low (40% at the highest-resolution shell). Consequently, we refined the final model against the isotropically treated data to obtain the most reliable structure and the corresponding electron-density maps. Importantly, R-free flags were the same for both anisotropically-and isotropically-treated data. The final R work/R free values and refinement statistics are shown in Table 1. The R work /R free values are unexpectedly high for the 2.2 Å resolution, which is most likely due to the presence of a high level of translational non-crystallographic symmetry (tNCS) as indicated by Xtriage software from PHENIX Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The protein coordinates and atomic structure factors have been deposited in the Protein Data Bank (PDB) under accession number 7BMH (LR). Source data are provided as Supplementary Data 1. All other data are available from the corresponding author upon reasonable request.