Crystal structure of a natural light-gated anion channelrhodopsin

The anion channelrhodopsin GtACR1 from the alga Guillardia theta is a potent neuron-inhibiting optogenetics tool. Presented here, its X-ray structure at 2.9 Å reveals a tunnel traversing the protein from its extracellular surface to a large cytoplasmic cavity. The tunnel is lined primarily by small polar and aliphatic residues essential for anion conductance. A disulfide-immobilized extracellular cap facilitates channel closing and the ion path is blocked mid-membrane by its photoactive retinylidene chromophore and further by a cytoplasmic side constriction. The structure also reveals a novel photoactive site configuration that maintains the retinylidene Schiff base protonated when the channel is open. These findings suggest a new channelrhodopsin mechanism, in which the Schiff base not only controls gating, but also serves as a direct mediator for anion flux.

Of the 35 ACR homologs found in cryptophyte algae Govorunova et al., 2016;Wietek et al., 2016), GtACR1 is the best characterized in terms of its gating mechanism and photochemical reaction cycle Sineshchekov et al., 2016), and also is the only ACR for which light-gated anion conductance has been proven to be maintained in vitro in a purified state  further recommending it as the preferred ACR for crystallization. The effects of mutations of several key residues, including E68Q/R, S97E, C102A and D234N, on photocurrents and photochemical conversions of GtACR1 have been studied in detail Sineshchekov et al., 2016).
The two channelrhodopsin families exhibit large differences in their sequences and photochemistry (Govorunova et al., 2017a): (i) ACRs conduct only anions with complete exclusion of cations, even H + , for which CCRs exhibit their highest relative permeability; (ii) ACRs are generally more potent; for example GtACR1 exhibits 25-fold higher unitary conductance than CrChR2; (iii) The retinylidene Schiff base in the photoactive site deprotonates prior to channel opening in CCRs (Ló renz-Fonfría and Heberle, 2014) and, in contrast, in ACRs remains protonated throughout the lifetime of the open-channel state with deprotonation correlated with the initial phase of channel closing Wietek et al., 2016).
Here, we report the atomic structure of the dark (closed) state of GtACR1, which is essential for elucidating the mechanism of its unique natural function of light-gated anion conductance through biological membranes. Also, understanding ACR mechanisms at the atomic scale would enable rational engineering to tailor their use as optogenetic tools.

Results and discussion
Overall GtACR1 structure The GtACR1 protein was expressed in insect (Sf9) cells and purified as a disulfide-crosslinked homodimer ( Figure 1-figure supplement 1). We obtained lipidic cubic phase (LCP) crystals of GtACR1, applied the continuous grid-scan method (Wojdyla et al., 2016) for X-ray data collection, and determined the structure at 2.9 Å resolution using molecular replacement ( Figure 1, Table 1). Each asymmetric unit contains a GtACR1 homodimer molecule (Figure 1-figure supplement 2). Each monomer is composed of an extracellular cap domain, seven transmembrane helices (TM1-7), and a cytoplasmic loop at the carboxyl-terminus ( Figure 1). In the extracellular domain, two kinked a-helices from the amino-terminal fragment and a b-hairpin from the TM2-3 loop lay on the interface of the membrane domain. The GtACR1 homodimer is stabilized by TM3 and TM4 interactions between monomers and further by an intermolecular disulfide bridge formed by the C6 residues ( Figure 1A-B). Since TM5-7 are much longer than TM1-4, this dimeric arrangement creates a large funnelshaped cytoplasmic cavity (~18 Å deep and~28 Å wide). Despite the modest~24% amino acid sequence identity between GtACR1 and C1C2/CrChR2, the structure of each GtACR1 protomer can be superposed well (Figure 1-figure supplement 3) with either of the two CCR structures (r.m.s.d. 0.9 Å ) indicating that these functionally distinct channelrhodopsins share a common TM helical scaffold conformation in their closed states.

The anion conductance pathway
A continuous tunnel spanning through the protein from the extracellular to cytoplasmic surface was detected in each GtACR1 protomer by serial cross-sections ( Figure 2A). The tunnel, assembled by TM1-3 and 7, starts from an electropositive port on the extracellular surface, intersects the retinylidene Schiff base in the middle of the membrane, and ends at an intracellular port deeply embedded in the large dimeric cavity ( Figure 2B).

The extracellular port cap
A unique structural feature is found in the extracellular domain of GtACR1. In addition to the disulfide link between the two protomers, an intraprotomer disulfide bridge is formed between C21 from the amino-terminal segment and C219 within the TM6-7 loop ( Figure 3A). This intramolecular crosslink immobilizes the kinked helices to the retinal-conjugated TM7, and encaps a hydrophobic part of the segment on the extracellular tunnel entry port ( Figure 3B). Disrupting this extracellular loop conformation, either by truncation of the amino-terminal loop (D1-25) or by substituting C21 and C219 with serine to abolish the intramolecular disulfide, resulted in slowed channel closing ( Figure 3C). Both C21 and C219 are highly conserved in ACRs (Govorunova et al., 2017b), but not in CCRs, revealing a role of this intramolecular disulfide bridge specific to the ACR family.

Ion pathway constrictions
The intramolecular tunnel in GtACR1 is constricted at three positions: at the extracellular port (C1), near the photoactive retinylidene Schiff base (C2), and at the cytoplasmic side (C3) . Near the extracellular port, the C1 constriction ( Figure 3D) is stabilized by an H-bond network adjacent to the disulfide-immobilized extracellular cap and formed by the side chains of Y81, R94 and E223 ( Figure 3A). The mutation R94A nearly abolished Clconductance ( Figure 5D). To analyze the influence of mutations on channel kinetics, we used laser flash excitation (single-turnover conditions), because under continuous light a mixture of intermediates is formed, the composition of which depends on the intensity and duration of illumination that may influence the kinetics. Channel closing in the wild-type GtACR1 is biphasic . As shown in Figure 3-figure supplement 1 and reported earlier for the E223Q mutant , all three  mutations strongly slowed the slow decay phase to a similar extent as that observed in the C21S_C219S mutant. These results suggest that the combination of the H-bond network of E223 and its neighbouring intraprotomer disulfide bridge controls the rate of channel closing in the extracellular region and stabilizes the essential residue R94 in the closed state.
The narrowest constriction C2 lies at the photoactive site and is formed by the side chains of T101, L64, and M105 ( Figure 3E). Four of the five residues that form the intracellular constriction C3 (L108, A61, E60, L245 and P58) ( Figure 3F) are in corresponding positions as the residues that form the 'intracellular gate' in CCRs (Deisseroth and Hegemann, 2017), but in GtACR1 and other ACRs only E60 (E121/E82) is shared with CCRs. The GtACR1 structure that we obtained from dark-grown crystals is presumably the dark (closed) state of the channel protein. To examine the role of these contriction-forming residues in the channel open state, we scanned the tunnel constrictions with Glu substitutions and measured photocurrents in the respective mutants. We chose Glu as a substituent because two of the constriction residues, A61 and A75, correspond to the highly conserved Glu residues in CCRs (E122/E83 and E136/E97 in C1C2/CrChR2, respectively), and because neutralization of E83 was required for elimination of the residual H + permeability in Cl --conducting CCR mutants (Berndt et al., 2016;Wietek et al., 2015). We also hypothesized that the bulky negatively charged Glu side chain would block the GtACR1 channel when placed in the ion conduction pathway. Indeed, perturbation of any residues at C2 or C3 greatly reduced or eliminated the photocurrents, while effects of most mutations (except A75E) at the C1 position were negligible ( Figure 3G), suggesting that in the open conformation the channel is wider in the extracellular portion and more narrow in its central and intracellular stretches. Kinetically, the mutations of the C1 and C3 residues mostly affected the slow phase of channel closing, making it slower than that in the wild-type (Figure 3figure supplement 2). Accurate kinetic analysis of the C2 mutations was not possible because of their greatly reduced photocurrents.  To test whether the mutations of the constriction residues compromised permeability of the channel for Cl -, we partially replaced this ion in the bath with non-permeable aspartate  and measured the current-voltage relationships (IE curves) (Figure 3figure supplement 3) to determine the reversal potentials (E rev ). None of the mutants in which photocurrents near the reversal potential could be resolved from the background noise yielded E rev values that were statistically significantly different from that measured in the wild type (Figure 3figure supplement 4; full statistical results are in Table 2). Several other mutations were recently reported to shift the E rev under Clgradient conditions from the Cl -Nernst potential Kim et al., 2018). However, we found no such change when we tested seven of these single and double mutants (Figure 3-figure supplements 5 and 6; full statistical results are in Table 2). The shifts were attributed to disrupting anion selectivity in GtACR1 by the mutations resulting in cation permeability Kim et al., 2018). Therefore, we further tested these mutants in the presence of H + and Na + gradients (Figure 3-figure supplements 7 and 8, respectively) and again found no statistically significant difference from the wild type (Figure 3-figure supplement 9; full statistical results are in Tables 3 and 4), indicating that none of these mutations produced permeability for these cations in GtACR1.

The retinylidene Schiff base
Confocal near-infrared resonance Raman spectroscopy has shown that the unphotolyzed state of GtACR1 contains almost exclusively all-trans retinal (Yi et al., 2016). In the middle of the protein, alltrans-retinal covalently bound by a Schiff base linkage to K238 is found in an elongated cavity formed by conserved hydrophobic residues. While the conformations of the retinal polyene chain are nearly identical in GtACR1 and C1C2/CrChR2, the presence of F160 in GtACR1 (G224/G185 in C1C2/CrChR2, respectively) pushes the b-ionone ring towards the extracellular side by 1.2 Å (Figure 4). Despite this conformational difference, the action spectrum of photocurrents generated by the F160G mutant was almost identical to that of the wild-type GtACR1 (Figure 4-figure supplement 1).
Remarkable structural differences between GtACR1 and the two crystallized CCRs are found in the retinylidene Schiff base environment. In C1C2 and CrChR2 the protonated Schiff base (PSB) participates in a quadruple salt-bridge network formed with D292/D253, E162/E123 and K132/K93 sidechains ( Figure 5B). However, this strong network is absent in the GtACR1 structure due to the replacement of E162/E123 and K132/K93 with smaller uncharged residues S97 and T71, respectively ( Figure 5A). D234 is the only residue directly interacting with the protonated Schiff base (PSB) in  Table 2. The output of Kriskal-Wallis ANOVA analysis of the results shown in Figure 3-figure supplements 4   GtACR1, and its electrostatic interaction appears to be weakened by two H-bonds from tyrosine residues Y72 and Y207 ( Figure 5A). The proton pump bacteriorhodopsin exhibits similar tyrosinyl H-bond-weakened interactions of D212, the residue in the corresponding position as D234. The interactions prevent D212 from accepting the Schiff base proton, which is transferred instead to D85 in the proton release pathway (Luecke et al., 1998). Resonance Raman and UV-vis absorption spectra of the D234N mutant of GtACR1 indicate that D234 is similarly neutral and not a Schiff base proton acceptor Yi et al., 2016). The dark structure therefore may explain the persistence of protonation of the Schiff base throughout the lifetime of the open channel conformation in GtACR1. Of the two tyrosine residues, Y207 appeared to be more important functionally, as its replacement with phenylalanine suppressed the photocurrents to a greater extent than that of Y72 ( Figure 5D) and caused a 12 nm blue shift of the action spectrum ( Figure 5-figure supplement 1). The Y72F mutation mostly affected the slow decay phase, while the Y207F mutation caused a strong perturbation of both phases ( Figure 5-figure supplement 2).
In CrChR2, photoisomerization of the Schiff base rapidly disrupts the strong salt-bridged network by inducing transfer of the Schiff base proton to D253 or E123 in~10 ms prior to channel opening (Ló renz-Fonfría and Heberle, 2014). In contrast, the Schiff base remains protonated throughout the lifetime of the open channel conformation in GtACR1, and deprotonation of the Schiff base proton occurs late in the photocycle (~20 ms) correlated with fast channel closing . Unlike in the salt-bridge network around the Schiff base in the CCRs ( Figure 5B), in GtACR1 no immediate proton accepting residue is available in the vicinity of the PSB and therefore later structural changes are required to enable Schiff base proton transfer, possibly to E68 ( Figure 5A).
The location of the PSB, centered within the anion path in GtACR1, suggests that it may play a direct role in anion translocation in the open channel state. Consistent with this idea is that the PSB is only partially neutralized by its D234 counterion because the counterion is weakened by its interaction with the two tyrosinyl residues. Therefore, the PSB presents a partial positive charge capable of transient reversible interaction with Clions in a channel that is largely lined by small polar and hydrophobic aliphatic residues. Supporting a possible direct role of the PSB in the channel's permeability for anions, late deprotonation of the Schiff base after channel opening occurs in all three ACRs so far examined: GtACR1 and GtACR2  and PsuACR1 , yet Schiff base deprotonation after channel opening is not known to occur in any CCR. Further indicating an essential role of the protonated Schiff base form, the mutant S97E, in which a potential Schiff base proton acceptor is placed at the corresponding position in GtACR1 as in CCRs and many other microbial rhodopsins, exhibits (i) appearance of fast Schiff base deprotonation, and (ii) a > 30 fold suppression of the amplitude of the chloride photocurrent . Furthermore, the double mutation Y207F/Y72F, expected to release inhibition of D234 as a proton acceptor, decreased the photocurrent amplitude to a negligible value ( Figure 5D). The ENS triad E68, a glutamyl residue near the Schiff base constriction in the channel, forms an H-bond network with N239 and S43 ( Figure 5C) with a geometry similar to that of a homologous triad (E129/E90, N297/N258, and S102/S63) referred to as 'the central gate' in C1C2 and CrChR2. In the CCRs, the triad blocks the cation path from the extracellular bulk phase (Deisseroth and Hegemann, 2017) and the glutamyl residue contributes to cation selectivity over anions (Wietek et al., 2014). In contrast, in GtACR1 the ENS triad does not occlude the tunnel ( Figure 5C), but E68 is functionally important in channel gating and may serve as a Schiff base proton acceptor at least at basic pH . The three residues in the ENS triad appear to have distinct roles; that is the substitution S43A had little effect on Clconductance, whereas the mutation N239A nearly eliminated the photocurrent ( Figure 5D). Both S43A and N239A mutations decreased the rate of the slow channel closing, but N239A in addition strongly accelerated the fast decay ( Figure 5-figure Table 3. The output of Kriskal-Wallis ANOVA analysis of the results shown in Figure 3-figure supplement 9A supplement 3A). Remarkably, combining the N239A mutation with the D234N mutation which alone also accelerated the fast decay, returned the channel closing kinetics almost to that of the wild type ( Figure 5-figure supplement 3B). Given its location between C2 and C3, N239 may assist moving anions between the Schiff base and the cytoplasmic port ( Figure 2C). Additionally, the distribution of apolar residues in this portion of the channel would also facilitate quick movements of anions as has been proposed for the CLC channel (Park and MacKinnon, 2018).
Despite the large phylogenetic difference between cryptophyte ACRs and chlorophyte CCRs, their helical scaffolds are little changed. However, the GtACR1 structure reveals fundamentally different chemistry built within their common scaffold. The preexisting full-length tunnel, the location of the retinylidene photoactive site directly in the ion path, the maintenance of a net positive charge on the site's Schiff base in a largely neutral tunnel, and the novel extracellular cap, provide important clues to the structural basis of light-gated anion conductance. Comparison with the recently published GtACR1 structure We report the atomic structure of the 7-helix rhodopsin domain (residues 1-295) that contains the light-gated channel activity of GtACR1. After this manuscript was prepared for submission, an article reporting a structure of the same domain with a short truncation (PDB code: 6CSM, residues 1-282) appeared from Karl Deisseroth and coworkers . The two GtACR1 structures were both determined at 2.9 Å resolution using the lipid cubic phase crystallization method. However, they were obtained in different crystallization conditions and exhibit different space groups and crystal packing. Nevertheless, the two show an extremely high similarity with a marginal RMSD value of~0.4 Å . All seven transmembrane helices are very well superimposed between the two structures ( Figure 5-figure supplement 4). The truncation or different crystal packing did not give rise to any large differences between the two protein structures. Therefore, these two structures presented by two independent groups mutually validate the conformation of GtACR1 in the closed state.
Differences between our study and that of Kim et al., 2018 are primarily in the methods used to deduce the location of the anion conduction pathway in GtACR1 and the results of testing relative ionic permeabilities of the mutants. By examining the structure by serial cross-sectioning we identified a contiguous intramolecular tunnel from the extracellular to intracellular surfaces. The CAVER algorithm confirmed the tunnel and helped identify three constrictions. Beyond the electropositive extracellular port the tunnel is lined primarily by small polar and aliphatic residues with sparse positive regions, one of them the protonated retinylidene Schiff base itself. Kim et al. described the putative ion conduction pathway in GtACR1 by identifying intramolecular cavities (vestibules) with electropositive surfaces, which led to identification of mostly residues present in the tunnel walls. Deeper understanding of the foundations of anion conductance by ACRs will likely require an atomic structure of the open state conformation.
Regarding anion selectivity of the mutants, a difference between our study and that of Kim et al.'s is that we have found no detectable cation permeability in any so far examined mutants (Figure 3-figure supplements 3-9). We attribute the strict selectivity for anions, a notable property of ACRs, to the existence of multiple highly selective portions of the tunnel and its two entry/exit ports. Consistent with this view, Kim et al. also concluded that there are multiple selective regions rather than a single selectivity filter, in contrast to CCRs, in which cation selectivity can be weakened to allow partial anion permeability by mutation of even a single residue (Wietek et al., 2014).
Overall, the independent analyses of the structures by both groups provide complimentary information creating a firm basis for elucidating the functional mechanism and for further optimization of GtACR1 as a potent neuron-silencing optogenetic tool.

Continued on next page
Data collection and processing X-ray diffraction data collections were performed on protein crystallography beamlines X06SA-PXI at the Swiss Light Source (SLS), Villigen, Switzerland. Data were collected with a 10 Â 10 mm 2 microfocused X-ray beam of 12.398 keV (1 Å in wavelength) at 100 K using SLS data acquisition software suites (DA+) (Wojdyla et al., 2018). Continuous grid-scans were used to locate crystals in frozen LCP samples (Wojdyla et al., 2016). The collection strategy was in steps of 0.1˚at a speed of 0.1 s/ step by using the EIGER 16M detector operated in continuous/shutterless data collection mode. Data were processed with XDS and scaled and merged with XSCALE (Kabsch, 2010a;Kabsch, 2010b). Four partial data sets (three with 60˚wedges and one with 30˚wedges) were collected, processed and merged to a final data set to 2.9 Å resolution. Data collection and processing statistics are provided in Table 1.

Structure determination
The GtACR1 structure was determined using the molecular replacement (MR) method with the structure of Chlamydomonas reinhardtii ChR2 (PDB entry: 6EID) (Volkov et al., 2017) as the search model. The MR solution was obtained using Phaser (McCoy et al., 2007) with the TFZ to 8.7 and LLG to 221. The initial model was built using PHENIX-autobuild (Adams et al., 2010) and further completed manually using COOT (Emsley and Cowtan, 2004). The structural refinement was performed using PHENIX (Adams et al., 2010) The final structure has R work /R free factors of 0.25/0.27. Refinement statistics are reported in

GtACR1 expression and electrophysiology
Characterization of GtACR1 mutants was performed using whole-cell photocurrent recording as previously described . Briefly, the wild-type expression construct was cloned into the mammalian expression vector pcDNA3.1 (Life Technologies, Carlsbad, CA) in frame with an EYFP (enhanced yellow fluorescent protein). Mutations were introduced using a QuikChange XL sitedirected mutagenesis kit (Agilent Technologies, Santa Clara, CA) and verified by DNA sequencing. HEK293 (human embryonic kidney) cells were transfected using the ScreenFectA transfection reagent (Waco Chemicals USA, Richmond, VA). All-trans-retinal (Sigma, St. Louis, MO) was added at the final concentration 4 mM immediately after transfection. Photocurrents were recorded 48-72 hr after transfection in whole-cell voltage clamp mode at room temperature (25˚C) with an Axopatch 200B amplifier (Molecular Devices, Union City, CA) and digitized with a Digidata 1440A using pClamp 10 software (both from Molecular Devices). Currents recorded in response to laser excitation or continuous light were filtered with a 10 or 2 kHz low-pass Bessel filter and digitized at 250 or 5 kHz, respectively. Patch pipettes with resistances of 2-5 MW were fabricated from borosilicate glass and filled with the following solution (in mM): KCl 126, MgCl 2 2, CaCl 2 0.5, EGTA 5, HEPES 25, and pH 7.4. The standard bath solution contained (in mM): NaCl 150, CaCl 2 1.8, MgCl 2 1, glucose 5, HEPES 10, pH 7.4. To test for changes in the permeability for Cl -, this ion in the bath was partially replaced with non-permeable aspartate (the final Clconcentration 5.6 mM, rounded to 6 mM in the figure legends). To test for changes in the permeability for H + , the bath pH was adjusted to 5.4, and for Na + , this ion in the bath was partially replaced with N-methyl-D-gluconate (NMDG) neutralized with H 2 SO 4 . (the final Na + concentration 1.4 mM). In this latter case, K + in the pipette was fully replaced with Na + , so that Na + was the only monovalent metal cation present in the system. A 4 M KCl bridge was used in all measurements. Series resistance was periodically checked during recording, and measurements showing >20% increase were discarded. The current-voltage relationships (IE curves) were measured near the expected E rev to eliminate its possible changes during recording.
For each cell, one value of the E rev was calculated. Before averaging, the curves for individual cells were normalized to the value obtained at the most negative holding potential in the tested range.
The holding potential values were corrected for liquid junction potentials calculated using the Clampex built-in LJP calculator (Barry, 1994). Laser excitation was provided by a Minilite Nd:YAG laser (532 nm, pulsewidth six ns, energy 12 mJ; Continuum, San Jose, CA). A laser artifact measured with a blocked optical path was digitally subtracted from the recorded traces. For further analysis, the signals were logarithmically averaged with a custom-created computer algorithm. Curve fitting and data analysis were performed using OriginPro 2016 software (OriginLab Corporation, Northampton, MA). Continuous light pulses were provided by a Polychrome V light source (T.I.L.L. Photonics GMBH, Grafelfing, Germany) at 15 nm half-bandwidth in combination with a mechanical shutter (Uniblitz Model LS6, Vincent Associates, Rochester, NY; half-opening time 0.5 ms). The maximal quantum density at the focal plane of the 40 Â objective measured with a piezo detector was 7.7 mW mm À2 at 515 nm. For measurements of the action spectra, short (25 ms) light pulses of the intensity in the linear response range were used at 10 nm half-bandwidth. The mean currents during the initial rise were calculated and corrected for the quantum density at each wavelength, which was measured with a calibrated photodiode. In each cell, a response to illumination at each wavelength was measured at least twice in a symmetrical fashion, first, scanning from the shortest to the longest wavelength, and then in the opposite direction. The spectral data sets obtained in all scans were pooled together (because the differences between individual cells in ACR expression levels or patch parameters were not expected to influence ACR spectral properties), normalized to the maximal value and averaged to produce the mean and sem values. Transfection with each tested mutant variant was repeated in at least three different batches of culture, and the results obtained in cells from all batches were pooled. Batches of culture were randomly allocated for transfection with a specific mutant; no masking (blinding) was used. Individual transfected HEK293 cells were selected for patching by inspecting their tag fluorescence; non-fluorescent cells were excluded. Cells for which we could not establish a gigaohm seal or for which a gigaohm seal was lost during recording were excluded from measurements. Current traces recorded from the same cells upon repetitive light stimulation were considered as technical replicates; results obtained from different individual cells were considered as biological replicates. In experiments with laser excitation, 10 technical replicates were averaged to yield a single mean trace for each cell; in experiments with continuous light pulses, a single trace was recorded in each cell. The baseline measured before illumination was subtracted using Clampfit software (a subroutine of pClamp). The same software was used to measure the peak current amplitude with a cursor. The raw data obtained in individual cells are shown as open diamonds and listed in the corresponding source data tables. Sample size was estimated from previous experience and published work on a similar subject, as recommended by the NIH guidelines (Dell et al., 2002). No outliers were excluded from calculation of mean values. Normality of the data was not assumed, and therefore non-parametric statistical tests were used as implemented in OriginPro 2016 software; P values > 0.05 were considered not significant. The results of statistical hypothesis testing are shown in Tables 2-4. When no specific statistical hypothesis was tested, descriptive statistical analysis was applied.

Cell lines
Only commercially available cell lines authenticated by the vendors (Sf9 from Sigma Aldrich and HEK293 from ATCC) were used; no cell lines from the list of commonly misidentified cell lines were used. The absence of micoplasma contamination was verified by Visual-PCR mycoplasma detection kit (GM Biosciences, Frederick, MD).

Data availability
Atomic coordinates and structure factors for the reported crystal structure have been deposited with the Protein Data Bank (PDB) under the accession code 6EDQ.