An open-like conformation of the sigma-1 receptor reveals its ligand entry pathway

The sigma-1 receptor (σ1R) is a non-opioid transmembrane receptor which has been implicated in many diseases, including neurodegenerative disorders and cancer. After more than forty years of research, substantial progress has been made in understanding this unique receptor, yet the molecular mechanism of its ligand entry pathway remains uncertain. Published structures of human σ1R reveal its homotrimeric organization of a cupin-fold β-barrel body that contains the ligand binding site, a carboxy-terminal V-shaped two-helix bundle, and a single amino-terminal transmembrane helix, while simulation studies have suggested a ligand entry pathway that is generated by conformational rearrangements of the cupin-fold domain. Here, we present multiple crystal structures, including an open-like conformation, of σ1R from Xenopus laevis. Together with functional binding analysis our data suggest that access to the σ1R ligand binding site is likely achieved by protein conformational changes that involve the carboxy-terminal two-helix bundle, rather than structural changes in the cupin-fold domain.

Recent crystal structures of human σ 1 R (hσ 1 R) show that hσ 1 R adopts a homotrimeric configuration, with each protomer comprising a single transmembrane helix (α1) at the amino-terminus, followed by a cupin-fold 19 β-barrel body containing the ligandbinding site, and a membrane-adjacent V-shaped two-helix bundle (α4/α5) at its carboxy-terminus covering the cupin-fold domain like a lid 17,20 (Fig. 1a). Currently, there are mainly two pathways proposed for ligand entry in σ 1 R 17,20,21 . The first proposed pathway (PATH1) would allow a ligand to enter σ 1 R through structural rearrangements of the cupin-fold domain, which involves unfolding and refolding of the β barrel; the second proposed pathway (PATH2) would allow a ligand to access its binding site in σ 1 R through the opening between α4 and α5 (Fig. 1b).
Recent molecular dynamics simulations support PATH1, as the simulations revealed two drastic conformational changes in hσ 1 R that break hydrogen bonds of multiple β sheets of the cupin-fold domain to expose the ligand-binding site 20 (Fig. 1a). However, considering the energy expenditure required to disrupt the β barrel, and the entropic penalty incurred when exposing the largely hydrophobic interior of the cupin-fold domain to the aqueous environment, PATH1 seems rather energetically unfavorable. The results of another computational study involving steered molecular dynamics revealed that pulling a ligand (PD144418) out of the binding pocket of hσ 1 R required less initial force in PATH1 21 . However, such a simulated conformational change of σ 1 R may not represent a physiological condition. Furthermore, the unfavorable energy expense would also accompany the pulling process in PATH1 when it breaks the β barrel.
To provide experimental evidence to assess ligand entry to σ 1 R we present multiple crystal structures of σ 1 R from Xenopus laevis (xlσ 1 R). Together with mutagenesis and binding analysis, the structural data support the notion that ligand entry to σ 1 R is achieved through PATH2.

Results
The closed conformation of xlσ 1 R. To better define the ligand entry pathway in σ 1 R, our initial strategy was to capture a σ 1 R structure in its open conformation. To achieve this goal, we screened expression for ten σ 1 R homologs from different species and crystallized well-behaved clones without adding any known  Fig. 1 The structure of xlσ1R closed-endo compared to hσ1R PD144418 . a The structure of the hσ1R PD144418 (PDB: 5HK1) homotrimer is viewed parallel to the membrane, with two protomers rendered in the surface mode and one in the cartoon. The helices are rendered as cylinders, and are labeled as α1-α5 from the amino-(N-) to the carboxy-(C-) terminus throughout the paper. The cupin-fold domain is rendered as a blue β barrel, with the red sheets indicating the hydrogen bonds to disrupt for ligand entry as suggested previously 20 . The C-terminal two-helix bundle (α4/α5) is colored in green, while the N-terminal part is orange. The relative position of the ER membrane is indicated by two gray lines. b The close-up view of one protomer from panel (a). The bound ligand, PD144418, is displayed as magenta sticks. The two pathways proposed for ligand entry are indicated by two magenta arrows, PATH1 and PATH2. c Left, superposition of the xlσ1R closed-endo trimer (in green) and the hσ1R PD144418 trimer (in brown) viewed perpendicular to the membrane from the cupin-fold side. Right, superposition of one xlσ1R closed-endo protomer and one hσ1R PD144418 protomer viewed parallel to the membrane. d One xlσ1R closed-endo protomer is rendered in both cartoon and surface modes in a "slab" view, in which the internal cavity is displayed as a gray shadowy compartment within the β barrel. The purple mesh shows the simulated annealing F o -F c map contoured at 3.0 σ level corresponding to an unidentifiable molecule.
ligand for structure determination. In the end, we solved a structure for the wild-type σ 1 R from Xenopus laevis (xlσ 1 R), which shares 67% sequence identity and 89% sequence homology with human σ 1 R (hσ 1 R) ( Supplementary Fig. 1), to 3.20 Å, and termed it xlσ 1 R closed-endo . The xlσ 1 R closed-endo structure was solved in the space group P2 1 with each asymmetric unit containing 12 protomers (Supplementary Fig. 2a), which deviate little from each other with all-atom RMSDs of only~0.2 Å between the protomers. Like all reported structures for hσ 1 R, xlσ 1 R also crystallizes as a homotrimer in xlσ 1 R closed-endo , which resembles closely the hσ 1 R structures, especially the ones complexed with an antagonist such as PD144418 (hσ 1 R PD144418 , PDB 5HK1), except the orientation of transmembrane helix α1 (Fig. 1c). Superposition of individual protomers of xlσ 1 R closed-endo and hσ 1 R PD144418 excluding α1 yielded an all-Cα RMSD of 0.35 Å, indicating a high degree of structural conservation among σ 1 Rs from different species (Fig. 1c). However, xlσ 1 R closed-endo failed to achieve our initial goal as it was captured in a closed conformation similarly to hσ 1 R PD144418 (Fig. 1d). Surprisingly, though no known ligand was added during purification or crystallization of xlσ 1 R, a clear electron density within the ligandbinding pocket indicates that xlσ 1 R closed-endo was bound by an unknown molecule ( Fig. 1d and Supplementary Fig. 2b), which may have an endogenous origin. Unfortunately, the current resolution of xlσ 1 R closed-endo (3.20 Å) provides only limited details about the shape of the electron density ( Fig. 1d and Supplementary Fig. 2b), making it impossible to reveal readily the molecular identity of this unknown molecule, and therefore the electron density was not modeled in xlσ 1 R closed-endo .
The crystal packing of xlσ 1 R closed-endo prohibits the PATH1 formation during ligand binding. In each asymmetric unit of xlσ 1 R closed-endo there are four homotrimers of xlσ 1 R, which are organized in a diamond shape with the cupin-fold domains packing against each other, whereas the transmembrane helices (α1s) mediate crystal contacts between adjacent asymmetric units ( Supplementary Fig. 2a). In such a packing mode, each trimer has one protomer whose cupin-fold domain is tightly buried by two adjacent cupin-fold domains (Fig. 2a). For example, the tip region of the cupin-fold domain (Trp133-Tyr144) of Protomer C is in close contact with Protomer A (Thr106-Thr110) and Protomer D (Lys139-Tyr144) (Fig. 2b). As a result, the steric hindrance from Protomers A and D would prevent the tip region of Protomer C from moving away from the rest of its cupin-fold body, a conformational change that is required for the PATH1 formation in σ 1 R as suggested in previous simulations 17,21 . Therefore, one protomer in every trimer would not be able to open in this packing mode if PATH1 is the ligand entry pathway in σ 1 R. Then if a ligand was added to these packed crystals, it would not be able to access the ligand-binding site in one-third of the protomers.
To test this hypothesis experimentally, we soaked two known ligands of σ 1 R, PRE084 (agonist) or S1RA (antagonist), directly into the xlσ 1 R closed-endo crystals, and solved both structures, xlσ 1 R closed-PRE084 and xlσ 1 R closed-S1RA , respectively. Soaking of either ligand changed little on the xlσ 1 R crystal packing. Both xlσ 1 R closed-PRE084 and xlσ 1 R closed-S1RA are almost identical to xlσ 1 R closed-endo with all-atom RMSDs < 0.25 Å when aligning the three structures ( Supplementary Fig. 2c). However, in both xlσ 1 R closed-PRE084 and xlσ 1 R closed-S1RA , each of the three protomers in the trimeric assembly features a ligand (PRE084 or S1RA) bound in the ligand-binding pocket ( Fig. 2c and Supplementary  Fig. 2d, e). This result demonstrates that soaking of the tightly packed crystals of xlσ 1 R closed-endo with the ligands led to the replacement of the unidentified molecule in each protomer, indicating that ligand access via proposed PATH1 is highly unlikely and that the conformational changes proposed for PATH2 seem more likely to facilitate ligand entry to each individual binding site.
An open-like conformation of xlσ 1 R that is consistent with PATH2. In our continued search for an open structure of xlσ 1 R, crystals with an appearance (cube-like) different from the xlσ 1 R closed-endo crystals (cuboid-like) were obtained and their structure was determined to 3.56 Å (xlσ 1 R open-endo ). The xlσ 1 R open-endo structure was solved in the P2 1 2 1 2 1 space group, with each asymmetric unit containing 12 protomers. These protomers are nearly identical to each other with all-atom RMSDs < 0.2 Å between the protomers. Similar to all σ 1 R structures, xlσ 1 R open-endo also assembles as homotrimers (Fig. 3a). Each asymmetric unit contains four trimers, which are packed in a tetrahedron shape through the α1-α1 contacts ( Supplementary  Fig. 3a).
Structural comparison between xlσ 1 R open-endo and xlσ 1 R closed-endo shows that the two structures are very similar, with the exception of two regions. One region consists of the transmembrane helix α1, which swings~18°away from the trimer center in xlσ 1 R open-endo compared to xlσ 1 R closed-endo (Fig. 3a). Structural alignment between the two structures excluding the α1 region yielded an all-atom RMSD of 0.34 Å. This orientation difference of α1 may be due to crystal packing, as α1 in the hσ 1 R structures exhibits different orientations even between different protomers of the same structure 17,20 , e.g., hσ 1 R PD144418 (Fig. 3b). The other major difference lies in the carboxy-terminal two-helix bundle, α4/α5. In xlσ 1 R open-endo , the α4 helix rotates slightly away from α5, and the side chain of Tyr203 from α5 adopts a gauche + χ 1 rotamer to point to the cupin-fold domain (Fig. 3c), thus creating an opening between α4 and α5 that generates a direct pathway for the ligand from the outside milieu to the ligand-binding site (Fig. 3d). Such an access pathway is reminiscent of that proposed in PATH2 (Fig. 1b). This opening spans an area of more than 10.3 Å × 5.5 Å (Fig. 3d), which is sufficiently large for the σ 1 R ligands such as PRE084 (~7.5 Å × 5.5 Å measured transversely) and S1RA (~7.6 Å × 5.2 Å transversely) to pass through. Therefore, the xlσ 1 R open-endo structure may represent an open-like conformation for σ 1 R and is consistent with PATH2 that predicts the ligand entry between α4 and α5.
Like xlσ 1 R closed-endo , the xlσ 1 R open-endo crystals were grown in the absence of any known ligand, but the xlσ 1 R open-endo structure also has an unidentifiable electron density within its ligand-binding pocket ( Supplementary Fig. 3b). To investigate if the identified opening between α4 and α5 was induced randomly by the unknown molecule, or whether it is indeed relevant for ligand access to the xlσ 1 R binding site, we further determined two structures for the xlσ 1 Supplementary Fig. 3c), and one PRE084 molecule binds in every protomer of the two structures ( Supplementary Fig. 3d). Interestingly, we also observed some dynamics in the Tyr203 side chain in these two PRE084-containing structures, possibly due to their relatively high resolutions. In xlσ 1 R open-PRE084-co , some protomers have Tyr203 in a gauche + χ 1 rotamer as in xlσ 1 R open-endo , whereas others have Tyr203 in a trans χ 1 rotamer as in xlσ 1 R closed-endo (Fig. 3e). In xlσ 1 R open-PRE084-soak , while most protomers have Tyr203 adopt a gauche + χ 1 rotamer, a few protomers have Tyr203 in both gauche + and trans χ 1 rotamers with a partial occupancy for each (Fig. 3f). This result indicates that the σ 1 R ligand PRE084 can bind σ 1 R concurrently with an entrance formation between α4 and α5, consistent with the PATH2 hypothesis as a mechanism of ligand entry.
Blocking the entrance in PATH2 hinders ligand binding in σ 1 R. To functionally validate the PATH2 hypothesis of ligand entry to the σ 1 R binding site, we sought ways to either block the opening between α4 and α5 or prevent its formation, and assess whether either approach would impair ligand binding. The identified opening between α4 and α5 is surrounded by residues mainly from α4/α5, including Pro176, Leu179, Leu183 from α4 and Leu196, Val200, Tyr203, Leu207 from α5, as well as Leu97 from a loop region of the cupin-fold domain (Fig. 4a). These residues outline an 'egg' shape, with Leu183, Leu196, and Val200 at the narrower pointed end, Leu179 and Tyr203 in the middle, and Leu97, Pro176, and Leu207 at the wider rounded end (Fig. 4a). We reasoned that alterations of the middle residues (Leu179 and Tyr203) would probably have the most profound effect on the entrance formation. Therefore, we designed two experiments targeting Leu179 and Tyr 203 to obstruct PATH2hypothesized ligand entry.
The first idea was to create a disulfide bridge between residues 179 and 203 to prevent the entrance from opening (Supplementary Fig. 4a), leading to a reduction in available binding sites. Therefore, only a fraction of the xlσ 1 R molecules would bind its ligand (e.g., PRE084), which can be evaluated by stoichiometry measurement using isothermal titration calorimetry (ITC). After introducing a pair of cysteine mutations at residue positions 179 and 203 (xlσ 1 R-C179/C203), the formation of the disulfide bond was catalyzed by oxidation. All ITC isotherms measured in this study were best fitted by a one-site model. As expected, the number of the available binding sites per protomer (stoichiometry) in xlσ 1 R-C179/C203 decreased to 0.18 ± 0.02 after oxidation compared to 0.92 ± 0.01 before oxidation ( Fig. 4b and Supplementary Fig. 4b, c), indicating an~78% efficiency for the disulfide bond formation. Furthermore, the stoichiometry was partially reverted to 0.52 ± 0.02 by re-reducing the oxidized sample of xlσ 1 R-C179/C203 with β-mercaptoethanol to break the disulfide bond ( Fig. 4b and Supplementary Fig. 4d). A partial recovery in the available binding site number after the rereduction may result from the limited accessibility of βmercaptoethanol to the disulfide bond between α4 and α5, which are likely covered by detergents in the protein sample due to the hydrophobic exterior face of α4/α5 17 . This data supports that the opening between α4 and α5 mediates ligand entry and binding. Interestingly, the equilibrium dissociation constant (K d ) of xlσ 1 R-C179/C203 for PRE084 also increased by~9-fold to 6.71 ± 1.44 μM after oxidation compared to 0.68 ± 0.24 μM before oxidation, and was partially reverted to 1.54 ± 0.16 μM after rereduction ( Fig. 4c and Supplementary Fig. 4b-d). These data indicate that the disulfide-bonded xlσ 1 R-C179/C203 protomers, in which the movement of α4 and α5 is constrained by the disulfide bridge, negatively affected ligand binding in the other, unbonded xlσ 1 R-C179/C203 protomers. This result is consistent with a previous report that suggests a cooperative ligand binding mechanism in σ 1 R 20 .
To assess the disulfide formation in xlσ 1 R-C179/C203 more directly, we utilized a bulky reagent, methoxypolyethylene glycol maleimide 5000 (mPEG-Mal-5K), to modify the free thiol groups of Cys179 and Cys203. The only native cysteine (Cys91) was . The residues participating in the close contacts are labeled and are displayed in spheres. c Left, one xlσ1R closed-PRE084 trimer in the cartoon mode viewed perpendicular to the membrane from the cupin-fold side. Right, the close-up view of the same trimer is viewed parallel to the membrane. The ligand PRE084 is displayed in sticks, and the purple mesh shows the simulated annealing F o -F c omit map contoured at 3.0σ level corresponding to PRE084. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-28946-w mutated to serine to facilitate this analysis (see "Methods" for details). Modification of mPEG-Mal-5K adds 5 kDa per modification to the modified protein, which causes a shift of the relative mobility of the protein in a sodium dodecyl sulfatepolyacrylamide gel electrophoresis (SDS-PAGE) gel. Considering that the access of mPEG-Mal-5K to Cys179 or Cys203 would likely be impeded by the nearby detergents that cover the hydrophobic exterior face of α4/α5, we expected a relatively limited efficiency for modification, and therefore this result was only used for correlative analysis. As expected, the modification percentage of xlσ 1 R-C179/C203 before oxidation reached merely 21.9% (Fig. 4d), though the majority of the receptors are thought to be disulfide-free (Fig. 4b). However, consistent with the disulfide bond formation and breakage, the modification percentage of xlσ 1 R-C179/C203 decreased to 3.7% after oxidation and returned to 15.9% after re-reduction with β-mercaptoethanol ( Fig. 4d). This result correlates well with the ITC data for xlσ 1 R-C179/C203. The second idea was to modify the residue 179 or 203 with a bulky reagent to sterically block the opening between α4 and α5, which would also decrease the number of the available ligand binding sites. To do so, we generated a Leu179-to-cysteine mutant (xlσ 1 R-C179) and a Tyr203-to-cysteine mutant (xlσ 1 R-C203), and chose mPEG-Mal-5K as the blocking modifier. This reagent is bulky and has been shown to successfully modify the free thiol group of Cys179 or Cys203 in xlσ 1 R (Fig. 4d), and has the advantage to allow quick assessment of the relative modification levels for the ITC samples. Since xlσ 1 R contains a native cysteine (Cys91), we first tested the effect of mPEG-Mal-5K modification of Cys91 on PRE084 binding. The stoichiometry (number of available binding sites per protomer) and affinity (K d ) changed little for wild-type xlσ 1 R after mPEG-Mal-5K treatment (Fig. 4e, f), indicating that the mPEG-Mal-5K modification of Cys91 would not interfere with the following analysis for Cys179 or Cys203. Therefore, Cys91 was not mutated in this analysis. We then measured PRE084 binding by ITC for xlσ 1 R-C179 or xlσ 1 R-C203 after mPEG-Mal-5K modification. As expected, the number of the available binding sites per protomer (stoichiometry) decreased to 0.48 ± 0.07 from 0.85 ± 0.09 for xlσ 1 R-C179, and to 0.69 ± 0.04 from 0.87 ± 0.01 for xlσ 1 R-C203 (Fig. 4e). The blockade percentage was~44% for xlσ 1 R-C179 and~21% for xlσ 1 R-C203, which correlates with the SDS-PAGE analysis with more modification in xlσ 1 R-C179 (~23%) than in xlσ 1 R-C203 (~13%) (Supplementary Fig. 4e), possibly due to the location of Cys203 being in a more occluded region compared to Cys179 when the receptor adopts an open conformation (Fig. 4a). Similarly, the equilibrium dissociation constant (K d ) increased from 1.02 ± 0.05 μM to 1.95 ± 0.05 μM for the modified xlσ 1 R-C179, and from 0.99 ± 0.03 μM to 1.61 ± 0.09 μM for the modified xlσ 1 R-C203 (Fig. 4f), again indicating that the modified xlσ 1 R protomers impaired ligand binding in the other, unmodified protomers. Meanwhile, we are aware of the limitations associated with the modification experiment, such as the limited accessibility to Cys179 or Cys203, the relatively low efficiency of modification by mPEG-Mal-5K, and the extended shape of the mPEG-Mal-5K molecule that might have affected more xlσ 1 R protomers than the Right, the surface representation shows an "egg" shape of the entrance residues in three parts: the rounded end (in blue), the middle (in yellow), and the pointed end (in red). b, c The stoichiometry (available binding site number per protomer, panel b) and the equilibrium dissociation constant (K d , panel c) determined by ITC for the xlσ1R-C179/C203 protein in the indicated conditions. All ITC measurements were repeated with biologically independent samples. "Before oxidation": n = 5. "After oxidation": n = 3. "Re-reduction": n = 4. Two-tailed Student's t-test was performed between the conditions of "Before oxidation" and "After oxidation", and between the conditions of "After oxidation" and "Re-reduction". The p values are provided in a Source Data file. **p < 0.01. d Modification of mPEG-Mal-5K to the xlσ1R-C179/C203 protein in the indicated conditions analyzed by SDS-PAGE and Coomassie blue staining. The wild-type (WT) xlσ1R served as the control. The pound sign ( # ) indicates that the protein contains the extra C91S mutation (see "Methods" for details). The black arrowheads labeled with 0, +1, and +2 indicate the band positions without, with one, and with two mPEG-Mal-5K modification(s). The red arrowhead indicates a minor impurity band near 50 kDa, which does not affect the gel analysis. The modification experiment was repeated three times independently with similar results. e, f The stoichiometry (available binding site number per protomer, panel e) and the equilibrium dissociation constant (K d , panel f) d etermined by ITC for the indicated xlσ1R proteins before and after the mPEG-Mal-5K modification. All ITC measurements were repeated with biologically independent samples. "Before modification": n = 4 for xlσ1R WT, n = 4 for xlσ1R-C179, n = 3 for xlσ1R-C203. "After modification": n = 6 for xlσ1R WT, n = 4 for xlσ1R-C179, n = 4 for xlσ1R-C203. Two-tailed Student's t-test was performed between the groups of "Before modification" and "After modification". The p values are provided in a Source Data file. **p < 0.01. For panels b, c, e, f, data are presented as mean + SD in the bar graphs, which are overlaid with corresponding data points shown as white dots. Source data are provided as a Source Data file.
one being modified. Therefore, this result needs to be interpreted with caution. Nevertheless, the simplest explanation for the xlσ 1 R-C179 or xlσ 1 R-C203 data is that the mPEG-Mal-5K modification physically blocks the entrance between α4 and α5, providing supplementary support for the PATH2 hypothesis of ligand access.
A structure for the xlσ 1 R-C179/C203 double mutant. To further validate the structural integrity of the cysteine mutants of xlσ 1 R, we solved a structure for the xlσ 1 R-C179/C203 double mutant (xlσ 1 R C179/C203-S1RA , 3.80 Å) to ensure that the folding of xlσ 1 R was not disrupted by the cysteine mutation(s). The xlσ 1 R C179/C203-S1RA structure was solved in the P2 1 space group, with 24 protomers in each asymmetric unit ( Supplementary Fig. 5a). The same homotrimeric organization was observed in xlσ 1 R C179/C203-S1RA , which resembles closely other xlσ 1 R structures. Superposition of the xlσ 1 R C179/C203-S1RA homotrimer and other xlσ 1 R structures excluding α1 yielded all-atom RMSDs of 0.2-0.3 Å (Fig. 5a). The α4/α5 region of xlσ 1 R C179/C203-S1RA appeared also similar to the closed and the open-like conformations (Fig. 5a). The xlσ 1 R C179/C203-S1RA crystals were grown in the presence of PRE084, and S1RA was later soaked in after the crystals had formed. Each protomer of the xlσ 1 R C179/C203-S1RA structure contained one S1RA molecule in the ligand-binding pocket, which adopts a similar pose to the ligand in the xlσ 1 R closed-S1RA structure (Fig. 5b). As expected, no disulfide bond was formed between Cys179 and Cys203 in this structure (Fig. 5b), thus allowing S1RA access to and binding in the ligandbinding site. This result indicates that substitution of the residues 179 and 203 with cysteine did not perturb the structural integrity of xlσ 1 R, and the resulting xlσ 1 R mutants retained the ligand-binding activity.

Discussion
The combinatory approach of structural and functional studies has gained critical insight into the molecular underpinnings of σ 1 R function. However, how ligands access the ligand-binding site in σ 1 R remains elusive. In this study, we focused on this important question and interrogated two ligand entry pathways proposed for σ 1 R, PATH1 and PATH2, and our structural and functional data favor the PATH2 hypothesis for ligand access. Moreover, our data offered some new information for further discussion. First, within the ligand-binding pocket of xlσ 1 R closed-PRE084 or xlσ 1 R closed-S1RA , the ligand is coordinated through a salt bridge between its cationic nitrogen and the carboxy group of Glu169 (the equivalent of Glu172 in hσ 1 R), and through hydrophobic interactions with Val81, Trp86, Met90, Tyr100, Leu102, Phe104, Tyr117, Ile121, Trp161, Ile175, Leu179, Phe181, Ala182, and Tyr203 ( Supplementary Fig. 6), which is nearly identical to the ligand-binding pattern in hσ 1 R 17 . The highly conserved structures of σ 1 R and their ligand binding sites from two different species further support the single-transmembrane-helix model 16 for σ 1 R as revealed by the crystal structures 17,20 .
Second, the structures of the xlσ 1 R-PRE084 complex in both closed and open-like conformations provide a unique opportunity to discuss how a ligand may enter and exit σ 1 R, which are likely reversed processes. Superposition of xlσ 1 R closed-PRE084 and xlσ 1 R open-PRE084-co indicates that PRE084 adopts slightly different poses in these two states ( Supplementary Fig. 7), and it is feasible to propose the following exit process. From the closed state to the open-like state, the phenyl ring of PRE084 moves~2 Å closer to α4 to avoid a steric clash with the side chain of Tyr203 (α5), which points toward the ligand-binding site. As a result, the PRE084 phenyl ring pushes against part of α4, causing it to rotate and move slightly away from α5. Interestingly, the steric pressure between PRE084 and α4 has also been observed in previous simulations 20,22 . Afterward, PRE084 may exit σ 1 R through the opening formed between α4 and α5. Similarly, PRE084 may enter σ 1 R in an inversed process. However, though we have observed both closed and an open-like conformation for σ 1 R, the mechanism that regulates the opening and closing of σ 1 R remains unknown. Meanwhile, the conformational change of α4 has also been suggested in σ 1 R agonism 20 , but its relationship with the conformational changes that lead to the opening of the receptor remains to be investigated.
Third, the PATH2 access hypothesis infers that the σ 1 R ligand enters from, and exits to, the membrane rather than an aqueous environment (Figs. 1a, 1b). Given the largely hydrophobic nature of most synthetic σ 1 R ligands 2 , it would be reasonable to assume that the ligand would partition into the membrane, or at the surface of the membrane (as the ligand becomes amphipathic when protonated). Since σ 1 R is primarily localized within cells, it has been suggested that a synthetic ligand would need to partition into and pass through, at least the plasma membrane 22 . Drug delivery studies also show that hydrophobic molecules do partition and enrich into the cellular membranes 23 . Therefore, a membrane pathway would not be a prohibitory factor for ligand entry and exit in σ 1 R.
which is like a plug between α4 and α5 and is one of the residues surrounding the putative ligand entrance (Fig. 4a). Since α1 pulls away from Leu97 in the open-like conformation of xlσ 1 R (Fig. 3a), it is tempting to postulate that the Leu97-containing loop may have space to swing slightly away from the membrane ( Supplementary Fig. 8b), so that a ligand dissolved at the membrane surface may slide in and out between α4 and α5 to access the ligand-binding site from within the membrane (Supplementary Fig. 8b). Unfortunately, this structural feature may be a very dynamic process and was not captured in the open-like xlσ 1 R structures in this study.
Finally, it is worth mentioning another structural effort that we attempted to capture an open conformation of σ 1 R by soaking a ligand of extended length into the xlσ 1 R crystals. We anticipated that the ligand would extend through the ligand entrance and cause the receptor to adopt an open state. Such an extended ligand, N 1 ,N 3 -bis(1-((R)-3-([1,1′-biphenyl]-4-yl)butyl)-piperidin-4-yl)malonamide (DIM-3C) 21 , was kindly provided by Dr. Collina and Dr. Rossino for soaking the xlσ 1 R crystals. Unfortunately, the highly hydrophobic DIM-3C could only be dissolved with organic solvents such as methanol and dimethyl sulfoxide, which damaged the xlσ 1 R crystals, and no structure was obtained for the DIM-3C soaking experiment. However, interestingly, in one of the 24 protomers of the xlσ 1 R C179/C203-S1RA structure, an extra electron density resembling the shape of S1RA was observed near the ligand entrance in addition to the S1RA molecule in the ligand-binding site ( Supplementary Fig. 5b). The density was readily and best modeled with an S1RA molecule. Its morpholine ring sits between α4 and α5 and is surrounded by residues 179, 203, and Leu97, whereas the rest of the ligand remains outside of the receptor, possibly representing an entering (or exiting) pose for the ligand. Meanwhile, due to the relatively low resolution (3.80 Å) of the xlσ 1 R C179/C203-S1RA structure, this extra density needs to be interpreted with caution, and we used it only for supplementary discussion.
Protein expression and purification. The gene encoding the full-length wild-type σ 1 R from Xenopus laevis (xlσ 1 R, NCBI accession NP_001087013.1) was synthesized by Genewiz (Suzhou, China) in a modified pPICZ plasmid (Thermo Fisher Scientific, Waltham, MA) containing an amino-terminal tag of decahistidine and Tobacco etch virus (TEV) protease recognition site following the hemagglutinin signal peptide. Cysteine mutations were introduced by site-directed mutagenesis using QuikChange II system (Agilent, Santa Clara, CA) according to the manufacturer's recommendation, and all mutations were verified by sequencing. The wild-type xlσ 1  The xlσ 1 R C179/C203-S1RA crystals grew in 0.05 M ADA pH 6.5, 24% PEG 400, 1 mM octyl-maltoside fluorinated, 0.4 mM PRE084 hydrochloride, and were soaked with 2 mM S1RA hydrochloride. The xlσ 1 R crystals usually appear within a week, and reach full-size in two weeks. The crystals were cryo-protected by raising the precipitant concentrations (final 36% for PEG 400 and 32% for PEG 300) with a 2% (v/v) incremental step, and flash-frozen in liquid nitrogen.
Data collection, structure solution, and structural analysis. Diffraction data were collected on beamlines BL18U1 and BL19U1 24 of the National Facility for Protein Science in Shanghai (NFPS) at the Shanghai Synchrotron Radiation Facility (SSRF). The data were indexed, integrated, and scaled using the autoPROC pipeline package (Global Phasing Limited) 25 , which includes XDS 26 and AIMLESS (CCP4 package) 27 . All xlσ 1 R structures were solved by molecular replacement with Phaser 28 using the published hσ 1 R PD144418 structure (PDB entry 5HK1) as a template. Manual model building and refinement were carried out using Coot 29 and phenix.refine 30 , and Molprobity 31 was used to monitor and improve protein geometry. The asymmetric unit of all xlσ 1 R structures contains 12 or 24 protomers, and so the non-crystallographic symmetry (NCS) was applied throughout the refinement to improve the map, except that for xlσ 1 R open-PRE084-co (3.10 Å) and xlσ 1 R open-PRE084-soak (2.85 Å), the NCS restraints were relaxed in the last a few rounds of refinement. The data collection and refinement statistics were generated using phenix.table_one 30 and the values are listed in Table 1. All structural figures, RMSD calculations, and length measurements were performed in PyMOL (Schrӧdinger, LLC). Accessibility analysis was performed using the volume-filling program HOLLOW 32 with default settings.
Oxidized disulfide formation and re-reduction. The disulfide bond formation between Cys179 and Cys203 was catalyzed by adding a freshly prepared CuSO 4 stock in the xlσ 1 R-C179/C203 protein sample, which was SEC-purified in the absence of any reducing reagent, to a final concentration of 200 μM for 1 h at the room temperature. The oxidation reaction was stopped by the addition of 10 mM EDTA pH 8.0 for 10 min at room temperature, and the protein sample was repurified by SEC. For the re-reduction experiment, the oxidized xlσ 1 R-C179/ C203 sample was incubated with 60 mM β-mercaptoethanol at 4°C for 12 h, and the protein sample was then re-purified by SEC. The oxidized or re-reduced xlσ 1 R-C179/C203 samples were subjected to ITC or SDS-PAGE analysis.
Cysteine modification. Modification of wild-type and cysteine mutants of xlσ 1 R was performed with a bulky thiol-modifying reagent, methoxypolyethylene glycol maleimide 5000 (mPEG-Mal-5K) 33 . Briefly, the SEC-purified xlσ 1 R protein in the absence of β-mercaptoethanol were incubated with 1 mM mPEG-Mal-5K at room temperature for 1 h, and the modification reaction was stopped by addition of β-mercaptoethanol to a final concentration of 10 mM before another SEC purification to remove any free mPEG-Mal-5K. For the entrance blocking experiment, the modified xlσ 1 R-C179 or xlσ 1 R-C203 proteins were subjected to ITC or SDS-PAGE analysis. For the assessment of disulfide formation in xlσ 1 R-C179/C203, the modified sample was analyzed by SDS-PAGE. To reduce background modification and facilitate gel analysis, the only native cysteine (Cys91) was mutated to serine in all samples for the disulfide analysis.
Band intensity analysis of SDS-PAGE gels. Protein bands in SDS-PAGE gels were visualized by Coomassie blue staining, and the band intensities were quantified using the ImageJ software 34 . Briefly, the gel image was first converted to grayscale, and equal-sized rectangles were drawn to encompass the bands of interest to measure their intensities. The intensity of a background rectangle was also measured for background correction. The modification ratio of the xlσ 1 R variants was calculated by the following equation: modification ratio (%) = (modified band intensity)/(modified + unmodified band intensities) × 100.
Isothermal titration calorimetry (ITC). The ITC measurement for the binding of PRE084 to xlσ 1 R was performed using a MicroCal iTC200 microcalorimeter (Malvern Instruments, Malvern, UK). The xlσ 1 R samples and PRE084 were prepared in the titration buffer containing 20 mM sodium HEPES pH 7.5, 150 mM NaCl, 1 mM DDM, and 0.005% (w/v) CHS. The first injection was 0.2 µl in volume and the subsequent injections were 2.0 µl. Before the data collection, the system was equilibrated to 25°C with the stirring speed set to 750 rounds per min. The titration curve of PRE084 was generated with nineteen 2.0-µl injections at 90-s intervals. All the xlσ 1 R titration data were best fitted by a one-site binding model using ORIGIN 7, and calculations were performed using Microsoft Excel for Mac 2016. The values of the available binding site number per protomer (stoichiometry) and the equilibrium dissociation constant (K d ) were determined from the average of three to six biologically independent ITC measurements and were expressed as mean ± SD in the text. Two-tailed Student's t-test was performed for statistical analysis, and details were described in relevant figure legends.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.