Blog

By Vladimir Zhemkov, Jonathon A Ditlev, Wan-Ru Lee, Mikaela Wilson, Jen Liou, Michael K Rosen, Ilya Bezprozvanny

Excerpt from the article published in eLife 2021;10:e65192 DOI: 10.7554/eLife.65192

Editor’s Highlights

• Sigma-1 receptor (S1R) modulates many physiological processes, such as cell excitability, transcriptional activity, Ca²⁺ homeostasis, stress response, and autophagy. 
• Many biological functions of the Sigma-1 receptor can be explained by its ability to organize and remodel cholesterol-enriched ER microdomains, which in turn affects activity of ER signaling proteins, stress-response, and metabolic status of the cells.
• It has been shown that MAMs are enriched in cholesterol and ceramides, and S1R can contribute to stabilization and/or formation of these cholesterol-rich lipid microdomains in the ER membrane.
• Previous studies identified two potential sites of S1R association with cholesterol – Y173 and Y201/Y206, that are dispensable for cholesterol-mediated S1R clustering.

Abstract

Sigma 1 receptor (S1R) is a 223-amino-acid-long transmembrane endoplasmic reticulum (ER) protein. S1R modulates activity of multiple effector proteins and is a well-established drug target. However, signaling functions of S1R in cells are poorly understood. Here, we test the hypothesis that biological activity of S1R in cells can be explained by its ability to interact with cholesterol and to form cholesterol-enriched microdomains in the ER membrane. By performing experiments in reduced reconstitution systems, we demonstrate direct effects of cholesterol on S1R clustering. We identify a novel cholesterol-binding motif in the transmembrane region of human S1R. Mutations of this motif impair association of recombinant S1R with cholesterol beads, affect S1R clustering in vitro and disrupt S1R subcellular localization. We demonstrate that S1R-induced membrane microdomains have increased local membrane thickness and that increased local cholesterol concentration and/or membrane thickness in these microdomains can modulate signaling of inositol-requiring enzyme 1α in the ER. Further, S1R agonists cause disruption of S1R clusters, suggesting that biological activity of S1R agonists is linked to remodeling of ER membrane microdomains. Our results provide novel insights into S1R-mediated signaling mechanisms in cells.

Introduction

Cholesterol is an essential component of cellular membranes, and levels of cholesterol in cells are tightly controlled (Brown and Goldstein, 1999; Goldstein and Brown, 2015; Radhakrishnan et al., 2008). The plasma membrane (PM) is highly enriched in cholesterol and other sterols, containing 30–40% cholesterol and 10–30% sphingolipids based on molar amounts (van Meer et al., 2008). It is widely accepted that cholesterol is not distributed homogeneously in the PM and instead forms lipid rafts, or cholesterol-enriched microdomains, that play important signaling roles in cells (Levental et al., 2020). Although cholesterol is produced in the endoplasmic reticulum (ER), the concentration of cholesterol in ER membranes is much lower than in the PM (van Meer et al., 2008). Level of ER cholesterol is maintained by an “on-and-off” switch of the Scap-SREBP pathway with a half-maximal response of SREBP-2 at about 5 molar % cholesterol in the ER (Radhakrishnan et al., 2008). There is limited information about spatial distribution of cholesterol in the ER membrane, but recent studies suggested existence of cholesterol-enriched microdomains in the ER membrane (Area-Gomez et al., 2012; Hayashi and Fujimoto, 2010; King et al., 2020; Montesinos et al., 2020). One of the most studied examples of such microdomains are mitochondria-associated membranes (MAMs), which are sites on the ER membrane in immediate proximity to the mitochondrial outer membrane (Csordás et al., 2018; Prinz et al., 2020). MAMs have been shown to play important roles in a variety of cellular functions, such as ER to mitochondria Ca²⁺ transfer (Csordás et al., 2018; Hajnóczky et al., 2002), ATP production (Hajnóczky et al., 2002), lipid metabolism (Csordás et al., 2018; Vance, 2014), and autophagy (Garofalo et al., 2016). Unsurprisingly, MAM dysregulation was observed in numerous pathophysiological conditions, including neurodegenerative diseases (Schon and Area-Gomez, 2013), cancers (Morciano et al., 2018), and lysosomal disorders (Annunziata et al., 2018; Sano et al., 2009).

Sigma 1 receptor (S1R) is a 223-amino-acid-long, single-pass transmembrane ER protein (Hayashi, 2019; Ryskamp et al., 2019; Schmidt and Kruse, 2019) that has a short cytoplasmic tail and a large luminal ligand-binding domain (Mavylutov et al., 2018; Schmidt et al., 2016). It has been suggested that S1R acts as a molecular chaperone that can stabilize a native conformation of multiple client proteins in stress conditions (Hayashi, 2019; Nguyen et al., 2017; Su et al., 2010). Recent studies also suggested that S1R can act as an RNA-binding protein (Lee et al., 2020). S1R is highly enriched in the liver and expressed in the nervous system. Analysis of S1R knockout (KO) mice revealed a number of nervous system abnormalities (Couly et al., 2020), suggesting that S1R plays an important role in neurons. In humans, mutations in S1R lead to a juvenile form of amyotrophic lateral sclerosis (ALS) (Al-Saif et al., 2011) and distal hereditary motor neuropathies (dHMNs) (Almendra et al., 2018; Gregianin et al., 2016; Horga et al., 2016; Li et al., 2015; Ververis et al., 2020). In animal models, genetic ablation of S1R exacerbates pathology of several neurological disorders (Hong et al., 2017a; Mavlyutov et al., 2013; Wang et al., 2017). S1R is considered to be a potential drug target for treatment of neurodegenerative disorders and cancer (Herrando-Grabulosa et al., 2021; Kim and Maher, 2017; Ryskamp et al., 2019). S1R binds multiple classes of drugs with nano- and sub-micromolar affinities (Maurice and Su, 2009). Based on their biological activity, S1R ligands are classified into agonists and antagonists (Maurice and Su, 2009). Signaling functions of S1R in cells are under intense investigation. The most prominent hypothesis is that under resting conditions S1R forms an inert complex with GRP78/BiP protein in the ER (Hayashi and Su, 2007). When activated by agonists or in conditions of cellular stress, S1R dissociates from BiP and is able to interact with a variety of client proteins (Hayashi and Su, 2007).

S1R preferentially localizes to MAMs (Hayashi and Su, 2007), and genetic KO of S1R in mice results in impaired MAM stability as evidenced by reduced number of contacts on electron micrographs and biochemical fractionation of MAM components (Watanabe et al., 2016). MAMs are cholesterol-enriched microdomains within the ER membrane (Area-Gomez et al., 2012; Hayashi and Fujimoto, 2010; Montesinos et al., 2020), and previous studies demonstrated that S1R can directly interact with cholesterol and ceramides in MAMs (Hayashi and Fujimoto, 2010; Hayashi and Su, 2004; Palmer et al., 2007). In the present study, we tested the hypothesis that S1R association with cholesterol plays a critical role in organization of specialized lipid microdomains in the ER, including MAMs. Using two in vitro reconstitution systems, we demonstrated cholesterol-dependent clustering of S1R in lipid bilayers. We identified a novel cholesterol-binding site in the S1R sequence and demonstrated the importance of this site for S1R clustering. We further demonstrated that S1R agonists reduced the number and size of S1R clusters in the presence of cholesterol. Based on these results, we conclude that the main biological function of S1R in cells is related to its ability to organize and remodel cholesterol-enriched microdomains in the ER. Our conclusion is consistent with MAM defects observed in the S1R KO mice (Watanabe et al., 2016).Results

S1R localizes to MAMs

To investigate the subcellular localization of S1R, we transfected HEK293 cells with a construct encoding a S1R-GFP fusion protein and performed confocal imaging experiments. Previous reports indicated that intracellular localization of the C-terminally tagged S1R is similar to that of the endogenous receptor (Hayashi and Su, 2003b). To visualize the ER, we co-transfected cells with mCherry-Sec61β (Zurek et al., 2011). In agreement with previous reports (Hayashi and Su, 2003a; Hayashi and Su, 2003b), S1R-GFP formed puncta in the ER (Figure 1A). To determine if these puncta corresponded to MAMs, we co-stained cells with a mitochondrial marker TOM20 and established that S1R-GFP puncta were frequently found in close opposition to mitochondria (Figure 1A). In contrast to wild-type (WT) S1R, an ALS-causing mutant S1R-E102Q (Al-Saif et al., 2011) was distributed uniformly in the ER and was not enriched in proximity to mitochondria (Figure 1—figure supplement 1). In order to increase the spatial resolution of our experiments, we utilized the protein retention expansion microscopy (pro-ExM) procedure that resulted in a 4.0–4.5-fold physical expansion of the specimen (Tillberg et al., 2016). HEK293 cells were co-transfected with S1R-GFP, mCherry-Sec61β, and stained with anti-GFP, anti-mCherry, and anti-TOM20 antibodies and processed according to the pro-ExM procedure for imaging (Figure 1B). Measurements of signal intensities of mCherry-Sec61b and S1R-GFP markers along the selected ER tubule indicated enrichment of S1R in the areas of mitochondrial proximity (Figure 1C). We then calculated the overlap between S1R-GFP, mCherry-Sec61β, and mitochondria. On average, 12 ± 5% of mCherry-Sec61β signal area overlapped with mitochondria, while a higher degree of overlap was observed for S1R-GFP and mitochondria, with fractional area of 31 ± 7% (n [number of cells] = 5, p-value=0.0034). As an alternative approach, we performed a biochemical fractionation of MAMs from mice liver (Figure 1—figure supplement 2), which confirmed enrichment of S1R in MAM fraction. Taken together, this data confirmed that S1R-GFP is enriched in ER subdomains in close opposition to mitochondria, consistent with previous reports (Hayashi and Su, 2007).

S1R forms clusters in giant unilamellar vesicles in vitro

To investigate mechanisms responsible for S1R enrichment at ER subdomains, we performed series of in vitro reconstitution experiments with giant unilamellar vesicles (GUVs) (Figure 2A). NBD-labeled phosphatidylcholine (NBD-PC) was included in lipid mixtures used to generate GUVs for visualization. Full-length His-tagged S1R was expressed in Sf9 cells and purified to homogeneity by affinity, anion exchange, and gel filtration chromagraphies (Figure 2B). After purification, His-S1R protein was covalently labeled with Alexa647 dye via NHS chemistry. To reconstitute S1R in the lipid membrane, large unilamellar vesicles (LUVs) were prepared by extrusion, destabilized with detergent, and mixed with purified S1R. Detergent was subsequently removed using BioBeads. S1R-containing LUVs were dehydrated on an agarose support, and GUVs were formed by rehydration as previously described (Horger et al., 2009).

Confocal imaging revealed that when S1R was reconstituted into liposomes composed of 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC), it was distributed uniformly (Figure 2C, left panel) with occasional clustering observed at the sites of contact between different GUVs (Figure 2—figure supplement 1). In contrast, when vesicles also contained 10% cholesterol (molar ratio to DOPC), S1R was often clustered (Figure 2C, right panel, marked with asterisks). As cluster formation is a dynamic process, we made sure to prepare samples simultaneously and image them side by side. On average, S1R clusters were observed in 7 ± 5% of DOPC liposomes (n = 3, 506 liposomes) and in 31 ± 3% of DOPC:10 mol % cholesterol liposomes (n = 3, 685 liposomes) (Figure 2D). Frequently, the S1R coalesced into one large domain (more examples of S1R behavior in cholesterol-free and cholesterol-containing liposomes are presented in Figure 2—figure supplement 1). Quantification of cluster size showed that S1R formed sub-micrometer-sized clusters in GUVs (Figure 2E). Cluster size was higher in the presence of cholesterol (0.16 ± 0.07 μm² for DOPC GUVs and 0.27 ± 0.19 μm² for cholesterol-containing liposomes) (Figure 2E). We used Poisson distribution to compare an average number of S1R clusters per liposome in each condition. We determined that the average number of clusters per liposome was significantly higher for GUVs containing 10 mol % cholesterol (λ = 0.15 for DOPC GUVs and λ = 0.72 for cholesterol-containing liposomes, p<0.001, N₁ = 194, N₂ = 201) (Figure 2F). Similar results were obtained when 20 mol % of cholesterol was included in the lipid mixtures used for GUVs formation (data not shown). From these results, we concluded that presence of cholesterol can promote clustering of S1R in the lipid membranes and proposed that association with cholesterol plays a direct role in S1R-induced formation of ER membrane microdomains.

Cholesterol-binding motifs in the S1R sequence

S1R was shown to interact with cholesterol, other sterols, and sphingolipids in binding studies (Hayashi and Fujimoto, 2010; Hulce et al., 2013; Palmer et al., 2007). Previous studies identified two potential sites of S1R association with cholesterol – Y173 and Y201/Y206 (Palmer et al., 2007). However, in the S1R crystal structure (Schmidt et al., 2016) Y173 localizes adjacent to the S1R ligand-binding domain and has no membrane contact, and Y201 and Y206 are located within the C-terminal membrane-adjacent amphipathic helices. Additional sequence analysis revealed tandem CARC-like (reverse sequence for the cholesterol-recognition amino acid consensus) motifs R/K-X_1-5-Y/F/W-X_1-5-L/V (PLEASE KEEP AMINO ACID SEQUENCE ON 1 LINE) (Fantini and Barrantes, 2013) in the transmembrane helix of S1R that span amino acids R7-L14 (Figure 3A). To test the importance of this motif, we generated S1R mutants by introducing a GGGG insertion within the CARC motif (S1R-4G) or by mutating W9 and W11 to leucine residues (S1R-W9L/W11L) (Figure 3A). We also generated S1R-Y173S and S1R-Y201S/Y206S mutants to test the potential importance of the previously reported cholesterol-binding motifs. In addition to the mutants describe above, we also generated a R7E/R8E mutant or deleted the double arginine motif altogether (ΔRR).

Since mutations introduced in CARC motif are located in close proximity to membrane boundary and can potentially alter protein insertion into the bilayer, we first evaluated their effects on S1R topology. Antibody accessibility studies suggested luminal (Hayashi and Su, 2007) or cytoplasmic localization of the C-terminus (Aydar et al., 2002), while recent APEX2-enchanced electron microscopy indicated luminal localization of receptor’s C-terminus (Mavlyutov et al., 2017). We expressed wild-type and mutant S1R-GFP fusions in HEK293 cells and performed selective permeabilization of the PM with digitonin. At low concentrations, digitonin permeabilizes plasma, but not the ER membrane, thus leaving ER luminal epitopes inaccessible to antibodies. When cells expressing wild-type S1R-GFP were stained with anti-GFP antibodies, we did not observe any anti-GFP antibody staining (Figure 3—figure supplement 1), indicating that the C-terminus of the S1R-GFP fusion was inaccessible and located in the ER lumen, consistent with previous findings (Mavlyutov et al., 2017). Similar results were obtained with S1R-Y173S, S1R-Y201S/Y206S, S1R-4G, and S1R-W9L/W11L mutants (Figure 3—figure supplement 1). In contrast, intense anti-GFP antibody staining was observed in experiments with the S1R-R7E/R8E and S1R-ΔRR mutants (Figure 3—figure supplement 1), indicating that these proteins had a reversed topology in the ER membrane. Therefore, we excluded these mutants from further analysis.

Wild-type and mutant S1R were expressed in HEK293 cells and their distribution was analyzed by confocal microscopy. Wild-type S1R and the S1R-Y201S/Y206S mutant each formed puncta in the ER (Figure 3B). In contrast, the S1R distribution in the ER was diffuse for the S1R-Y173S, S1R-4G, and S1R-W9L/W11L mutants (Figure 3B). By performing TOM20-staining experiments, we confirmed that the puncta observed with the wild-type S1R and Y201S/Y206S mutant corresponded to MAMs (Figure 3B). In order to obtain a quantitative measure of the S1R distribution in the ER network, we calculated Mander’s overlap coefficient between mCherry-Sec61β and S1R-GFP. Mander’s coefficient measures fractional overlap between two channels and varies from 0 (no correlation) to 1 (complete correlation). For wild-type S1R, Mander’s coefficient was M = 0.50 ± 0.02 (n = 3, 25 cells), indicating that not all mCherry-Sec61β signal colocalized with the GFP signal in S1R clusters (Figure 3C). In contrast, Mander’s coefficient was equal to 0.77 ± 0.09 (n = 2, 11 cells) for S1R-Y173S, 0.72 ± 0.04 (n = 3, 30 cells) for S1R-4G, and 0.77 ± 0.03 (n = 3, 17 cells) for S1R-W9L/W11L (Figure 3C), reflecting more diffuse distribution patterns. We have also explored the insertion of four alanine stretch (S1R-4A) instead of glycines (Figure 3C, Figure 3—figure supplement 2 for quantification). S1R-4A behaved similarly to S1R-4G mutant. Mander’s coefficient for the S1R-Y201S/Y206S mutant was equal to 0.49 ± 0.16 (n = 2, 12 cells) (Figure 3C), similar to the wild type. Taken together, we concluded that the newly identified CARC motif is important for S1R targeting to MAMs and that the Y201/Y206 motif is dispensable. Effects of Y173 mutation may be related to misfolding of S1R, as this residue has no membrane contact in the S1R crystal structure (Schmidt et al., 2016). To better showcase the distribution of S1R-4G and S1R-W9L/W11L in cells, we performed expansion microscopy of the samples, similar to the experiments that we performed with wild-type S1R (Figure 1B). We found that in contrast to the clustered distribution of wild-type S1R, S1R-4G and S1R-W9L/W11L mutants had diffuse localization inside and outside of the ER (Figure 3—figure supplement 2).

We then asked if targeting to membrane contact sites is a general property of S1R or it is specific for MAMs. Previously S1R was shown to regulate store-operated Ca²⁺ entry (Srivats et al., 2016), and we reasoned that S1R may also be present at ER-PM junctions. To test this hypothesis, we co-expressed S1R-GFP with a genetically encoded ER-PM marker mCherry-MAPPER (Chang et al., 2013) in HeLa cells. We first examined subcellular distribution of WT protein and mutants using conventional confocal microscopy near the PM (Figure 3D). As reported previously, MAPPER (shown in red) formed small punctate patterns corresponding to the ER membrane in close proximity to the PM (Figure 3D). S1R-GFP appeared enriched in these same puncta (Figure 3D), in agreement with the previous report (Srivats et al., 2016). S1R-4G and S1R-W9L/W11L were distributed diffusely in the ER (Figure 3D). When we analyzed cells using total internal reflection fluorescence (TIRF) microscopy, we confirmed enrichment of S1R-GFP in MAPPER-positive puncta compared to the mutants (Figure 3E). We quantified S1R-GFP fluorescence in MAPPER-positive puncta and normalized this value to the total GFP fluorescence in the TIRF plane in the same cell. For wild-type S1R-GFP, 16 ± 7% of the fluorescence signal was localized to the ER-PM junctions compared to 7 ± 5% for S1R-4G mutant and 5 ± 3% for S1R-W9L/W11L mutant (Figure 3F). Taken together, our results suggest that wild-type S1R localizes to and is enriched in ER contact sites such as MAMs and ER-PM junctions and that CARC motif is important for S1R targeting to these sites.

To further validate the importance of the CARC motif in S1R cholesterol-dependent clustering, we expressed and purified S1R-4G and S1R-W9L/W11L mutant proteins; and performed GUV reconstitution experiments in the presence of 10 mol % cholesterol. The S1R-W9L/W11L mutant formed less clusters (Figure 4A, B), consistent with the diffuse distribution of this mutant in cells (Figure 3B, C). However, S1R-4G still formed clusters when reconstituted in GUVs, similar to the wild-type S1R (Figure 4A, B). This was in contrast to the diffuse distribution of the S1R-4G mutant in cells (Figure 3B, C). To explain these results, we reasoned that the W9L/W11L mutation abolished S1R association with cholesterol, but the 4G mutation only weakened it as this mutation still contains the CARC consensus sequence (Fantini and Barrantes, 2013).

o confirm reduced cholesterol binding of the generated S1R mutants, we conducted a series of pulldown experiments with cholesterol agarose (Palmer et al., 2007). Cholesteryl hemisuccinate was used for coupling to CarboxyLink reactive resin to generate cholesterol agarose beads. Purified recombinant S1R was incubated with cholesterol agarose, the beads were pelleted by centrifugation, and fractions containing beads and supernantant were analyzed by western blot with anti-S1R antibodies. First, we confirmed that recombinant S1R binds to cholesterol agarose, but not to the control resin (Figure 4—figure supplement 1). We then tested S1R-4G and S1R-W9L/W11L mutants. Based on obtained results, we concluded that both mutants have reduced affinity for cholesterol (Figure 4C, D). To further confirm these findings, we expressed wild-type and mutant S1R-6His in HEK293 cells, and used cellular lysates from transfected cells in pulldown experiments with cholesterol agarose beads. In these experiments, anti-His tag antibodies were used for detection of recombinant S1R-6His. We found that S1R-W9L/W11L-6His bound to cholesterol agarose significantly weaker than the wild-type S1R (Figure 4E, F). In contrast, S1R-4G-6His behaved similar to the wild-type S1R-6His in these experiments (Figure 4E, F). Taken together, we concluded that W9L/W11L mutation disrupted cholesterol binding to a much greater extent than 4G mutation. We reason that 4G mutation leads to diffuse distribution of S1R in cells (Figure 3B) because ER cholesterol levels in cells are no more than 5 mol % (Radhakrishnan et al., 2008). Presumably, these levels of cholesterol are sufficient for association with wild-type S1R but not with S1R-4G mutant that has reduced affinity for cholesterol.

S1R localizes to, and can generate, thick lipid domains

Our results suggested that S1R is clustered in cholesterol-rich microdomains in the ER (Figure 3). It is established that cholesterol-rich phospholipid mixtures have more ordered acyl chains and larger hydrophobic thickness (de Meyer and Smit, 2009). Interestingly, according to the crystal structure and hydrophobicity analysis, the length of the S1R transmembrane (TM) domain is 24 a.a (Figure 3A; Schmidt et al., 2016), which is longer than a typical 20 a.a TM length of ER-resident proteins (Sharpe et al., 2010). The length of the S1R TM domain is estimated to be 36.9 Å and should have a positive hydrophobic mismatch when compared to a DOPC bilayer, which has a hydrophobic thickness of 26.8 Å (Kučerka et al., 2006). In contrast, a cholesterol-rich DOPC bilayer has a hydrophobic thickness of 36.0 Å, matching well with the S1R TM length (Milovanovic and Jahn, 2015). It is therefore possible that S1R clustering in ER membrane is influenced by local changes in membrane thickness, as has been described for the targeting of plasma membrane proteins to lipid rafts (Lorent et al., 2017). To test whether the length of the S1R TM also contributes to clustering, we deleted four amino acid residues in its TM region to generate the S1R-Δ4 mutant (Figure 5A). When S1R-Δ4 GFP-fusion protein was expressed in HEK293 cells, it displayed a more diffuse distribution in the ER when compared to the wild-type S1R (Figure 5A). Mander’s coefficient for this mutant is 0.72 ± 0.07 (n = 3, 27 cells) compared to Mander’s coefficient of 0.50 ± 0.02 for the wild-type protein (Figure 3C). Lack of MAM localization for S1R-Δ4 mutant was confirmed in TOM20 staining experiments (Figure 5A).

To test the role of positive hydrophobic mismatch in S1R clustering, we reconstituted purified Alexa647-labeled S1R in GUV membranes composed of 1,2-dinervonoyl-sn-glycero-3-phosphocholine (DNPC) lipids that have a hydrophobic thickness of 35.5 Å. S1R did not form clusters in DNPC GUVs neither in the absence nor in the presence of 20% cholesterol (0%: 6.8%, 92 liposomes, 20%: 2.9%, 77 liposomes) (Figure 5B). From these results, we concluded that local increase in membrane thickness contributes to S1R targeting to cholesterol-enriched microdomains in the ER. We then hypothesized that by clustering with cholesterol, S1R can also promote the formation of microdomains with thicker bilayer structure. To measure local membrane thickness in these microdomains, we utilized two designed molecular rulers that report on transmembrane thickness – MBP-TM17, with a short transmembrane helix, and MBP-TM27, with a longer transmembrane helix (Figure 5C). The amino acid sequences of TM17 and TM27 were designed based on the synthetic peptides previously used in GUV studies of membrane thickness (Kaiser et al., 2011). Similar WALP-GFP fusion proteins were successfully used for measurements of the ER lipid heterogeneity in yeast cells (Prasad et al., 2020). The hydrophobic length of MBP-TM17, calculated using 1.5 Å rise per residue (Hildebrand et al., 2004), matches well with the hydrophobic thickness of the DOPC bilayer (25.5 Å and 26.8 Å, respectively) (Kučerka et al., 2006). The calculated hydrophobic length of TM27 is longer (40.5 Å) and matches well to a lipid bilayer with longer acyl chains or a bilayer with high cholesterol content (DOPC+30–40 mol % cholesterol, 36.0 Å) (Milovanovic et al., 2015). MBP-TM17 and MBP-TM27 proteins were expressed in Escherichia coli, purified using affinity and size exclusion chromatographies, and then covalently labeled with Alexa555 dye (Figure 5—figure supplement 1). The MBP-TM17 and MBP-TM27 were first reconstituted into DOPC GUVs in the absence or presence of 10% cholesterol. We found that in the cholesterol-free conditions , MBP-TM17 was distributed largely uniformly in the GUV membrane while MBP-TM27 was enriched at the junctions between the individual GUVs (Figure 5D). Cholesterol had little effect on the distribution of MBP-TMs (Figure 5D). We hypothesized that MBP-TM27 localized to the GUV junction sites based on a hydrophobic matching mechanism. Interestingly, clustering of S1R at the sites of contacts between different GUVs was occasionally observed in the absence of cholesterol (Figure 2C, left panel), most likely due to increase in local membrane thickness at these sites. When S1R and MBP-TM proteins were co-reconstituted together, they were distributed diffusely in the membrane in the absence of cholesterol (Figure 5—figure supplement 2). When these rulers were reconstituted together with S1R in the presence of 20% cholesterol, they displayed a different behavior: while MBP-TM17 remained largely diffuse (Figure 5E, top panel), MBP-TM27 formed clusters that colocalized with S1R clusters (Figure 5E, bottom panel). Analysis of linear density of MBP-TM proteins inside and outside S1R clusters confirmed that MBP-TM27 was enriched in S1R domains, while MBP-TM17 showed no partitioning to S1R clusters (Figure 5F). Importantly, this behavior was observed for an artificially designed MBP-TM27 protein, suggesting that recruitment and clustering events were mainly driven by lipid-protein interactions. These results suggested that in the presence of cholesterol S1R induces formation of membrane microdomains with increased local thickness.

To further validate that the same protein recruitment mechanism can be observed for known S1R biological partners, we extended our analysis to an inositol-requiring enzyme 1α (IRE1α). S1R and IRE1α localize in a close proximity to each other in the ER, but do not necessarily interact directly (Rosen et al., 2019). In addition to protein-protein interactions, IRE1α oligomerization and stress-response activity can be also modulated by the surrounding lipid environment (Cho et al., 2019; Halbleib et al., 2017). Loading ER with cholesterol affects IRE1α signaling (Feng et al., 2003). First, to confirm colocalization of S1R and IRE1α in cells, we utilized a proximity-labeling approach (Hung et al., 2016). HEK293 cells were transiently transfected with S1R-APEX2 protein or control ER-targeted constructs APEX2-KDEL or Sec61β-APEX2. Following transfection, cells were incubated with biotin-phenol and exposed to H₂O₂ for a short period of time to induce biotinylation of proteins in proximity to APEX2. Cell lysates were collected, and biotinylated proteins were pulled down using streptavidin-agarose. The eluate containing biotinylated proteins was analyzed by western blot. Significantly more biotinylated IRE1α was pulled down from the cells expressed S1R-APEX2 than from the cells expressing APEX2-KDEL or Sec61β-APEX2 control constructs (Figure 5—figure supplement 3). These results suggested close proximity of IRE1α and S1R-APEX2, consistent with the previous studies (Mori et al., 2013; Rosen et al., 2019). Short-term (2 hr) stress induction with 1 μM thapsigargin (Tg) had no measurable effect on colocalization of S1R and IRE1α (Figure 5—figure supplement 3).

To test whether S1R co-assembles with IRE1α in GUV membranes, we used a similar approach that we previously utilized for the MBP-TM rulers (Figure 5C). For these experiments, we generated an expression construct MBP-IRE1α-TM that consisted of the MBP protein fused to the transmembrane helix of human IRE1α and the upstream amphipathic helix (433–464 a.a) (Figure 5C; Cho et al., 2019). MBP-IRE1α-TM protein was expressed in bacteria, purified, and labeled with Alexa555 (Figure 5—figure supplement 1). MBP-IRE1α-TM protein labeled with Alexa555 was co-reconstituted together with S1R labeled with Alexa647 in GUVs and imaged by fluorescence confocal microscopy. In the absence of cholesterol, the distributions of MBP-IRE1α-TM and S1R were uniform (Figure 5G, top panel). However, in the presence of 20% cholesterol, MBP-IRE1α-TM was recruited to S1R-positive clusters (Figure 5G, H, bottom panel). These results suggested that the transmembrane domain of IRE1α partitions into a lipid microenvironment established by S1R in the presence of cholesterol.

To confirm the functional relevance of IRE1α localization to S1R microdomains in the ER, we compared IRE1α-mediated signaling in wild-type, and S1R knockout (KO) HEK293 cells. S1R KO HEK293 cell line was generated by CRISPR approach, using the same procedure used to generate S1R KO MEF cells in our previous study (Ryskamp et al., 2017). Western blotting experiments confirmed absence of S1R in S1R KO HEK293 cells (Figure 5—figure supplement 3). Unfolded protein response (UPR) was induced in these cells by addition of 1 μM Tg and levels of IRE1α phosphorylation were examined by western blot using anti-phospho-IRE1α antibody. We determined that under resting conditions the level of phosphorylated IRE1α was lower in S1R KO cells when compared to wild-type cells (Figure 5—figure supplement 3). Under ER stress conditions, the same peak level of IRE1α phosphorylation could not be achieved in S1R KO cells (Figure 5—figure supplement 3), and total cellular response, measured as integrated area under the curve (Figure 5—figure supplement 3), was significantly smaller in S1R KO cells compared to WT, suggesting an impairement in IRE1α signaling. To further examine the activity of IRE1α, we quantified levels of the XBP1s protein, a known downstream effector of IRE1α (Calfon et al., 2002). We found that production of XBP1s was delayed in S1R KO cells when compared to wild-type cells (Figure 5—figure supplement 3). Taken together, these results suggested that recruitment of IRE1α to S1R-organized microdomains in the ER facilitated IRE1α-mediated signaling. Our conclusions are supported by previously reported inhibition of IRE1α activity in cells transfected with S1R RNAi (Mori et al., 2013).

The dynamic organization of S1R clusters in double supported lipid bilayers

Based on their biological activity, agonists and antagonists of S1R have been described (Maurice and Su, 2009). Previous research indicated that S1Rs can redistribute in the ER upon ligand stimulation (Hayashi and Su, 2003a; Hayashi and Su, 2007). In the next series of experiments, we sought to understand whether S1R ligands can analogously affect the formation and stability of S1R clusters in vitro. It is technically difficult to measure time-resolved dynamics of protein clustering in the membrane of GUVs using standard confocal microscopy. Thus, we used total internal reflection fluorescence (TIRF) microscopy and supported lipid bilayers (SLBs), a system commonly employed to measure the dynamics of membrane-associated proteins (Ditlev, 2021; Su et al., 2016b) SLBs are typically formed on a glass or mica surface, but in this format transmembrane proteins bind to the underlying glass/mica and usually become immobilized. Recently, several groups reported formation of polyethylene-glycol (PEG)-cushioned bilayers that retained mobility of transmembrane proteins (Pace et al., 2015; Richards et al., 2016; Wong et al., 2019). When we attempted to form supported or cushioned bilayers with reconstituted S1R, we observed that protein and lipids were static, presumably due to interaction with the glass surface (data not shown). To overcome these limitations, we modified the bilayer formation procedure to generate double supported lipid bilayers (DSLBs). In DSLBs, PEGylated lipids are used as supports between the two separate bilayers: a lower bilayer that contacts the glass surface is composed of DOPC and 1 mol % DSPE-PEG2000-biotin lipids and an upper bilayer that rests on the first and is composed of lipids and purified transmembrane proteins. To build the DSLB, the first bilayer is formed by fusion and spreading of small unilamellar vesicles (SUVs), prepared by freeze-thaw of DOPC and 1 mol % DSPE-PEG2000-biotin lipids, on a clean glass surface (Figure 6A). The second bilayer is formed by deposition of extruded LUVs containing 0.1% NBD-PC as a membrane dye that can be used to measure membrane lipid fluidity by TIRF microscopy (Figure 6A). Fluorescence recovery after photobleaching (FRAP) of NBD-PC indicated that lipids in the upper bilayer are highly mobile (Figure 6—figure supplement 1). To incorporate protein in the DSLB, purified S1R-Alexa647 was first reconstituted in LUVs using the same approach as for the GUV studies. Protein-loaded LUVs were then applied to the pre-formed lower DOPC bilayer and allowed to spread and form a continuous upper bilayer (Figure 6A). The second bilayer was within the TIRF penetration depth (100 nm). FRAP experiments indicated that S1R molecules in this system are mobile and can quickly diffuse laterally in the membrane (Figure 6—figure supplement 1).

We discovered that the S1R distribution in DSLBs was heterogeneous, and that small and large clusters could be identified based on their intensity (Figure 6B). Small clusters of S1R were highly mobile in the DSLB membrane, whereas large clusters were relatively static. To investigate effects of cholesterol on S1R clustering, we incorporated S1R-647 into DSLBs containing increasing amounts of cholesterol (Figure 6B). In agreement with the GUV data (Figure 2), we found that increased cholesterol in DSLBs caused S1R to cooperatively assemble into large clusters (Figure 6B). The threshold for S1R clustering was in a narrow range between 2.5 and 5.0 mol % (Figure 6C, black line), consistent with a typical ER cholesterol content (Radhakrishnan et al., 2008). To determine the number of S1R molecules in each of these clusters, we divided the integrated intensity of each cluster by the intensity of a mono-labeled His-pLAT-647 protein molecule (Su et al., 2016a). This analysis revealed that small S1R clusters most likely corresponded to S1R trimers, containing on average 2.65 ± 0.58 (n = 47) molecules, consistent with the crystal structure of the protein (Schmidt et al., 2016). The number of S1R molecules in the large clusters was widely distributed in a range between tens and hundreds of molecules, with a mean value of 38 molecules per cluster at 20 mol % cholesterol (Figure 6D). Titration of cholesterol caused the right-shift in molecular weight distribution (Figure 6—figure supplement 2) and also increased the average cluster size (Figure 6C, red line). To measure the relative lateral mobility of the lipids, we performed fluorescence recovery after photobleaching (FRAP) experiments on NBD-PC within S1R clusters and in the surrounding membrane (in the presence of 20% cholesterol). Following photobleaching, the recovery of NBD-PC signal was complete outside of S1R clusters (Y_max(out)=1.01 ± 0.06) but incomplete inside these clusters (Y_max(in)=0.74 ± 0.07, p=0.008) (Figure 6E). Higher immobile fraction indicates that a fraction of molecules was immobilized at the bleached spot due to the fact that they were trapped in rigid domains. These data suggest that lateral mobility of the lipids within S1R clusters is reduced, consistent with known properties of lipid rafts in the plasma membrane (Sezgin et al., 2017).

We next evaluated the effects of S1R ligands on S1R clustering. These experiments were performed in the presence of 10% cholesterol in the DSLB. Addition of selective S1R agonist (+)-SKF-10047 at the time of the second bilayer formation reduced the number of large S1R clusters (Figure 6F). The effects of (+)-SKF-10047 on S1R clustering were concentration-dependent, with half-maximal effect observed at 50 nM (Figure 6F). The effects of (+)-SKF-10047 were blocked by the addition of a S1R antagonist NE-100 (Figure 6F). Consistently, agonists resulted in reduced cluster size (Figure 6G). These results suggested that S1R agonists prevented formation of S1R clusters or promoted disassembly of S1R clusters in the membrane and that these effects could be blocked by S1R antagonists. Our results are consistent with previous biochemical studies that demonstrated that the proportion of S1R multimers formed in cells was decreased by the agonists (+)-pentazocine and PRE-084 but increased by the antagonists CM304, haloperidol, and NE-100 (Hong, 2020; Hong et al., 2017b). These results are also consistent with effects of (+)-pentazocine and haloperidol described in FRET experiments performed with cells transfected with fluorescently tagged S1Rs (Mishra et al., 2015). Similar conclusions were made when using receptor bioluminescence resonance energy transfer (BRET) homomer assay and non-denaturing gel electrophoresis (Yano et al., 2018). In contrast, it has been reported that S1R agonists such as (+)-pentazocine and PRE-084 stabilized the oligomeric state of purified S1R in the presence of detergents (Gromek et al., 2014), suggesting that biochemical properties of S1R in the membrane and in detergent may differ from each other. Another potential reason for this discrepancy is that the previous study (Gromek et al., 2014) was focused on the analysis of smaller tetrameric species, while our analysis measured larger assemblies of at least ten S1R molecules. Taken together, our results indicate that S1R participates in the formation of thicker cholesterol-rich lipid clusters and S1R agonists act by disassembling these membrane microdomains.Discussion

S1R modulates many physiological processes, such as cell excitability, transcriptional activity, Ca²⁺ homeostasis, stress response, and autophagy (Christ et al., 2019; Couly et al., 2020; Hayashi, 2019; Kourrich, 2017; Maurice and Goguadze, 2017; Ryskamp et al., 2019). However, S1R-mediated signal transduction differs from a canonical second messenger-coupled transmembrane receptor signaling, and S1R is often referred to as a ‘ligand-gated molecular chaperone’ (Hayashi, 2019; Nguyen et al., 2017; Su et al., 2010). In the seminal paper by Hayashi and Su, 2007, S1R was proposed to have chaperone-like properties at MAMs. However, the molecular mechanism of chaperone activity of S1R remains unexplained because S1R lacks structural similarity with known chaperones or extensive protein interaction interfaces. Based on our results, we propose that the biological activity of S1R in cells can be explained by the ability of this protein to form cholesterol-enriched microdomains in the ER, therefore acting as ER lipid scaffolding protein. A role of S1R in organization and remodeling of lipid raft microdomains in the plasma membrane has been proposed previously based on cell biological studies (Palmer et al., 2007; Vollrath et al., 2014). Our data further suggest that such microdomains have increased local membrane thickness, providing a favorable environment for recruitment of ER proteins with longer transmembrane domains (Lorent et al., 2017). Increased local cholesterol concentration and membrane thickness can modulate activity of ER proteins recruited to these microdomains. Our hypothesis may explain how a small protein such as S1R is able to modulate activity of almost a hundred effector proteins (Couly et al., 2020; Delprat et al., 2020; Kourrich et al., 2012; Ryskamp et al., 2019; Schmidt and Kruse, 2019). We reason that activity of these proteins could be affected by changes in local lipid microenvironment and not only via protein-protein interactions with the S1R.

By performing experiments using reduced reconstitution systems, we have been able to demonstrate direct effects of cholesterol on S1R clustering (Figures 2 and 5). Previous studies identified two potential sites of S1R association with cholesterol – Y173 and Y201/Y206 (Palmer et al., 2007). Our studies suggest that Y201 and Y206 residues are dispensable for cholesterol-mediated S1R clustering (Figure 3). Moreover, we identified a tandem CARC-like motif (Fantini and Barrantes, 2013) within the transmembrane region of S1R (Figure 3A). Mutations of this motif impaired association of recombinant S1R with cholesterol beads (Figure 4), affected cholesterol-dependent S1R clustering in GUV reconstitution experiments (Figure 4), and disrupted S1R targeting to MAMs and ER-PM contact sites in cells (Figure 3). The threshold for formation of S1R clusters was in a narrow range between 2.5 and 5.0 mol% of cholesterol (Figure 6C). Similar cholesterol dependence was previously described for SREBP-2 activation, with half-maximal response at 4.5 mol % of ER cholesterol (Radhakrishnan et al., 2008), indicating that S1R affinity for cholesterol is within the physiological range of ER cholesterol concentrations.

A recent study demonstrated that micrometer-sized large intracellular vesicles exhibit phase-separation behavior at contact sites between the ER and mitochondria, plasma membrane and lipid droplets (King et al., 2020), similar to phase separation observed in giant plasma membrane-derived vesicles (Levental et al., 2011). It has been shown that MAMs are enriched in cholesterol and ceramides (Area-Gomez et al., 2012; Hayashi and Fujimoto, 2010; Hayashi and Su, 2003a). Our findings suggest that S1R can contribute to stabilization and/or formation of these cholesterol-rich lipid microdomains in the ER membrane. This conclusion is consistent with MAM defects observed in S1R KO mice (Watanabe et al., 2016), with earlier analysis of S1R targeting to detergent-resistant domains in the ER (Hayashi and Fujimoto, 2010), and with previous suggestions that S1R contributes to stability of lipid rafts in the plasma membrane (Palmer et al., 2007; Vollrath et al., 2014).

The lipid regulation of PM proteins is a well-known phenomenon (Rosenhouse-Dantsker et al., 2012). It was shown in direct and indirect experiments that activities of several ER channels including ryanodine receptors, sarco/endoplasmic reticulum Ca²⁺-ATPase, and inositol-1,4,5-triphosphate receptors can be modulated by cholesterol content, lipid packing, or membrane thickness (Cannon et al., 2003; Gustavsson et al., 2011; Li et al., 2004; Madden et al., 1979; Sano et al., 2009). Components of the gamma secretase complex reside at MAMs, and the transmembrane region of amyloid precursor protein (APP) contains cholesterol-binding motifs (Montesinos et al., 2020; Pera et al., 2017), suggesting modulatory effects of cholesterol on APP processing. Several ER stress sensors, including IRE1α, can sense membrane saturation and be activated without protein unfolding in the ER lumen (Ballweg et al., 2020; Cho et al., 2019; Halbleib et al., 2017). In our GUV reconstitution experiments, we demonstrated that a minimal sensor derived from IRE1α transmembrane domain is recruited to S1R clusters (Figure 5G, H), and in agreement with the previous report (Mori et al., 2013), S1R cellular KO impaired IRE1α-mediated signaling (Figure 5—figure supplement 3).

S1R can exist in an oligomeric form (Gromek et al., 2014; Mishra et al., 2015). Originally, a photoaffinity probe labeled high-molecular weight oligomers in rat liver microsomes (Pal et al., 2007). High-molecular weight species up to 400 kDa were later detected in membrane preparations using various biochemical approaches (Yano et al., 2018; Yano et al., 2019), and a similar wide molecular weight distribution was observed for the recombinant protein (Schmidt et al., 2016). The functional relevance of the S1R oligomerization is, however, not clear. In our experiments, we observed that S1R oligomers form continuous molecular weight distribution ranging from trimers to large (10–100 subunits) clusters (Figure 6D). Formation of large clusters was facilitated by the presence of cholesterol. S1R agonists disrupted large S1R clusters in cholesterol-containing lipid bilayers in a concentration-dependent manner (Figure 6F, G). Our results are consistent with ligand effects observed by native electrophoresis from cell membranes and S1R-BiP association (Hayashi and Su, 2007; Hong, 2020; Hong et al., 2017b). Also in agreement with our findings, S1R agonist (+)-SKF-10047 resulted in destabilization of lipid rafts in the plasma membrane of cancer cells (Palmer et al., 2007).

Mutations in S1R lead to a juvenile form of ALS (Al-Saif et al., 2011) and distal hereditary motor neuropathies (Almendra et al., 2018; Gregianin et al., 2016; Horga et al., 2016; Li et al., 2015; Ververis et al., 2020). We demonstrated that an ALS-causing E102Q variant acted as a loss-of-function mutant that disrupted S1R targeting to MAMs (Figure 1—figure supplement 1). However, E102 residue is located outside of cholesterol-binding region of S1R, and this mutation likely acts by a different mechanism, such as causing S1R misfolding, aggregation, or a change in oligomerization state (Abramyan et al., 2020). S1R is considered to be a potential drug target for treatment of neurodegenerative disorders and cancer (Herrando-Grabulosa et al., 2021; Kim and Maher, 2017; Maurice and Goguadze, 2017; Maurice and Su, 2009; Nguyen et al., 2017; Ryskamp et al., 2019). S1R agonists demonstrated neuroprotective effects in a variety of neurodegenerative disease models and are currently in clinical trials for a variety of neurological disorders including Alzheimer’s, Huntington’s, Parkinson’s diseases and ALS (Brimson et al., 2020; Herrando-Grabulosa et al., 2021; Maurice and Goguadze, 2017; European Huntington’s Disease Network et al., 2019; Ryskamp et al., 2019). It has been proposed that accumulation of C99 fragment of APP leads to enhanced formation of cholesterol microdomains in the ER and upregulation of MAM activity in AD neurons (Area-Gomez et al., 2012; Montesinos et al., 2020; Pera et al., 2017). In contrast, MAM downregulation was observed in motor neurons in a genetic model of ALS (Watanabe et al., 2016). Our results suggest that S1R agonists allow remodeling of lipid microdomains in the ER membrane, which may help to normalize MAM function in AD and ALS neurons. Long-term effects of S1R activation may include a more thorough remodeling of MAMs including size and protein composition needed for long-lasting metabolic adjustments in stress conditions. Changes in ER lipid microenvironment may also affect function of a variety of channels, transporters, and other signaling proteins localized to S1R signaling microdomains. The exact functional outcome of these lipid perturbations will be unique to each particular protein based on its activity in a lipid environment established by S1R.

In conclusion, we propose that many biological functions of S1R can be explained by its ability to organize and remodel cholesterol-enriched ER microdomains, which in turn affects activity of ER signaling proteins, stress-response, and metabolic status of the cells.