Molecular basis of Climp63-mediated ER lumen spacing

The endoplasmic reticulum (ER) is a continuous membrane system with interconnected sheets and tubules (Baumann and Walz, 2001; Shibata et al., 2006). ER sheets are cisternal structures with a relatively constant width of ∼50 nm (Fawcett, 1981) and are frequently decorated by translating ribosomes (termed rough ER) (Shibata et al., 2006). Key factors determining the morphology of ER sheets have been identified by analyzing the most upregulated genes during B-cell differentiation or most abundant ER proteins in pancreatic β-cells, two cell types that are filled with rough ER (Shibata et al., 2010).

Climp63, also known as cytoskeleton associated-protein 4 (CKAP4), has been proposed to regulate ER lumen spacing (Shibata et al., 2010). Depletion of Climp63 causes the collapse of ER sheets, with an ∼50% decrease in the luminal width of the ER, and reintroduction of Climp63 with engineered luminal domains (LDs) of different lengths consistently results in ER sheets of corresponding width (Shen et al., 2019; Shibata et al., 2010).

Climp63 is a single-pass integral membrane protein with a molecular mass of 63 kDa. It possesses a relatively short N-terminus [NT, 106 amino acids (aa) for human Climp63] in the cytosol and a long C-terminal LD region (475 aa) in the ER lumen. The NT of Climp63 is modified post-translationally.

Phosphorylation of the Climp63 NT regulates its interactions with microtubules (Vedrenne et al., 2005), which subsequently affects translocon mobility (Nikonov et al., 2007) and ER positioning (Cui-Wang et al., 2012; Farías et al., 2019). Climp63 has been recently shown to preferentially bind to centrosomal microtubules via an unclear mechanism (Zheng et al., 2022). Palmitoylation in the NT of Climp63 has also been reported (Planey et al., 2009), but the physiological role remains elusive. The LD domain of Climp63 engages the ER chaperone calumenin, which cooperatively controls ER sheet morphogenesis (Shen et al., 2019). In addition, Climp63 has been found in the triad structure of skeletal muscle, which comprises a transverse (T)-tubule and sarcoplasmic reticulum (SR) (Osseni et al., 2016), and also recently in a distinct type of ER tubule (Wang et al., 2022). Super-resolution imaging analysis suggests that Climp63 plays a role in nanohole formation in ER sheets (Schroeder et al., 2019) and nanodomain organization in ER tubules (Gao et al., 2019).

The molecular architecture of Climp63 remains unclear. The purified LD of Climp63 has been reported to form 90-nm rods observed by negative-staining electron microscopy (EM) (Klopfenstein et al., 2001). However, the recombinant protein was obtained under denatured conditions, raising questions about the relevance of the observation. We reported recently that purified Climp63 undergoes moderate and complex self-association (Zhao and Hu, 2020). Given the heterogeneity of the recombinant Climp63 fragments, it would be virtually infeasible to acquire the structural information using conventional methods. Here, we predict the structure of Climp63 using recently developed deep learning methods (Baek et al., 2021; Jumper et al., 2021). By comparing potential models, we propose that the Climp63 LD adopts a helix-bundle configuration. Electrostatic interfaces were identified for trans self-association. Our findings point to a common theme for membrane tethering by a proteinaceous bridge.

To investigate the molecular basis of ER luminal spacing, we predicted the structure of the Climp63 LD using recently developed computing methods. The domain is highly α-helical, as revealed by RoseTTAFold (RF) using the modeling method ‘TrRefindRosetta’. A leading helix (α0, residues 128–169) is followed by a five-helix bundle (5HB) (Fig. 1A,B), in which its NT is close to the membrane and the C-terminus (CT) points away. The 5HB is packed by mostly hydrophobic interactions in the core. α1 and α2 are longer than the remaining three helices. Among the five candidate models generated during the prediction, the positions of α0 and α1 varied the most (Fig. S1A); α0 and α1 could form one extensive helix or two bended helices, with α1 sometimes running parallel to α2–α5 but physically detached. These observations suggest that α0 likely acts independently and that the 5HB might be loosely packed, especially the first helix.

To confirm the prediction, we analyzed Climp63 using AlphaFold (AF). The helical nature of the LD domain was consistent between the two methods (Fig. 1C). In the AF model, α0 was independent of other helices, and α1 was also separated. The remaining four helices formed a pseudo-4HB by running parallel but with marginal interactions. Similar results were obtained when the sequence of Climp63 was subjected to the updated RF server using the modeling method ‘RoseTTAFold’ (Fig. S1B), in which case α2–α5 rearranged into a 3HB in which α3 and α4 form a continuous helix. These results confirm the secondary structures of the Climp63 LD and suggest that it forms a loosely packed helix bundle, which is likely on the borderline of what can be predicted by computational modeling. Notably, the length of the original RF-predicted LD was ∼23 nm, and that of the AF-predicted structure was ∼37 nm. If the LD forms trans interactions between ER membranes by tip-to-tip association, the RF-based prediction would fit well with the observed ER sheet spacing of ∼50 nm.

To test the ability of Climp63 to regulate the width of ER sheets, we performed EM analysis of pelleted U-2 OS cells. As expected, the width of ER sheets in wild-type (WT) cells exhibited a certain degree of variation, with an average width of 55 nm, and deletion of Climp63 reduced the ER width to ∼34 nm (Fig. 2A). Reintroduction of WT Climp63 by lentiviral expression restored the luminal spacing of ER sheets to ∼50 nm, which was slightly thinner than that of the WT. The rescued ER sheets appeared to have more constant distances between parallel membrane compared to WT, likely due to higher expression of Climp63 than the endogenous level (Fig. S3A,H). Similar results were obtained when cells were directly fixed on a sapphire disc (Fig. S1C), except the overall ER lumen width was proportionally narrower with the adhesive cells.

As mentioned in the structural analysis, the inclusion of α1 in the helix bundle was relatively weak. If α1 opens up to convert the 5HB into a 4HB, the expected luminal bridge would be longer. We noted that K283 in α2 forms salt bridges with E277 and E280, stabilizing the bending between α1 and α2 (Fig. 2B). When K283 was replaced by aspartic acid (Fig. S3A,H), the resulting ER width increased to ∼60 nm (Fig. 2C), consistent with the predicted 5HB-to-4HB conversion. These results suggest that the RF-predicted 5HB, but not the AF-predicted 4HB, is probably the natural configuration of the Climp63 LD.

Next, we tested whether the Climp63 LD forms a tip-to-tip association in trans to mediate luminal spacing. In the 5HB configuration, the α1–α2 loop and CT comprise the distal end (away from the membrane) of the domain (Fig. 3A). When the electrostatic surface of the molecule was analyzed, we noticed several clusters of charged residues: K597 and K601 (K597/K601) at the CT, and K270 and K290 (K270/K290) form two positively charged patches, and E265 and E295 (E265/E295), and E278 and D285 (E278/D285) consist of two negatively charged patches (Fig. 3B). We speculated that K597/K601, and E278/D285 form reciprocal tip-to-tip interactions, with K270/K290, and E278/D285, being on either side of protruding region, mediate a side-by-side association. In addition, the entire body of the LD contains distinct regions of charge clusters, likely allowing side-by-side, cis association of Climp63 within the same membrane plane. Importantly, all of these residues are highly conserved (Fig. S1D).

Next, we purified full-length WT Climp63 or a charge reversed mutant (4CR; with K597E, K601E, E278K and D285K mutations), and assessed their self-association by in vitro pulldown assays. As expected, NT HA-tagged Climp63 was able to precipitate NT Flag-tagged Climp63 (Fig. 3C). The 4CR mutant displayed reduced interactions with WT Climp63 (Fig. 3C), but increased association with a differently tagged 4CR (Fig. 3D), possibly with binding restored to the level of the WT–WT pair. The changes seen with the 4CR were small but reproducible because the assay measures all types of self-association, not just the tip-based interactions. Notably, a CT-tagged mouse Climp63 had stronger self-association with NT-tagged Climp63 (Fig. S1E) because the CT is directly involved in homotypic interactions. As previously reported, the self-association of Climp63 was not affected by the presence of Ca²⁺ ions (Fig. S1F), which are abundant in the ER lumen. Collectively, these results suggest that Climp63 uses the α1–α2 loop to engage the CT of a pairing protomer for luminal bridging.

We then tested distal pairing of Climp63 using cysteine-based cross-linking. The soluble region of Climp63 contains one cysteine in the NT. We replaced it with a serine residue (C100S) to avoid background cross-linking (Fig. S2A). When K601 and D285 were individually mutated to cysteine, dimer formation was readily detected, although with different mobility in SDS-PAGE, likely due to varied shape of the cross-linked products (Fig. 3E; Fig. S2B). When K601C and D285C Climp63 forms were co-transfected into Climp63 KO cells, an additional population of cross-linked products, likely heterodimers, was reproducibly detected (lane 5 versus lanes 3 and 4, Fig. S2B). This population could be detected by either anti-Flag or anti-HA antibodies when HA–K601C and Flag–D285C were co-expressed and cross-linked (lanes 5 and 12, Fig. 3E), confirming the existence of hetero-dimers formed by the two cysteine mutants and tip-to-tip interactions between Climp63. Consistent with these results, no heterodimers were observed when Climp63 forms with S140C, a remote site in α0, and D285C mutations were co-transfected (Fig. S2C).

We also performed pulldown assays to test whether α0 acts independent of the 5HB. HA–Climp63 was precipitated by anti-HA-conjugated beads, and we compared the co-precipitation of either GST–LD or GST–5HB. GST alone was used as a negative control. We found that GST–LD interacted with full-length Climp63 as previously reported (Zhao and Hu, 2020), with a stronger interaction when α0 was deleted (Fig. S2D). These results indicate that α0 is dispensable for Climp63 self-association. When Climp63 Δα0 was expressed in Climp63-knockout (KO) cells, the ER width was maintained at the WT level (Fig. S2E). These results confirm that α0 plays a minor role in luminal spacing, and that the helix is flexible in connecting the transmembrane domain (TM) and the 5HB.

We further tested the existence of the 4HB configuration by measuring trans association. If α2–α5 forms a 4HB, it presents the α2–α3 loop and α4–α5 loop at the distal end of the LD, which allows E389 and D390 (E389/D390) in the α2–α3 loop to reach R523 in the α4–α5 loop of the pairing protomer (Fig. S3A). In contrast, this salt bridge would not be possible in the 5HB setting, because these sites face their own membrane (Fig. S3B). When E389/D390 or R523 was charge-reversed and purified, the same amounts of mutants were pulled down by WT Climp63 (Fig. S3C), suggesting a marginal role in self-association. Consistently, both Climp63 with E389K, D390K and R523E mutations and Climp63 with E389A, D390A and R523A mutations rescued the ER thickness to that of the WT (Figs S3D and S4G,J). These results support Climp63 LD folding into a 5HB.

Finally, we analyzed the charged patches in mediating luminal spacing using the rescue assay. When K270 and K290 were replaced by negatively charged aspartic acid, the double mutant (K270D/K290D) failed to restore the collapsed ER sheets seen in the KO cells (Fig. 4A). In contrast, a similar mutant, with K270A and K290A mutations (K270A/K290A), was effective in maintaining the ER thickness (Fig. 4A). The discrepancy in these mutants is likely due to a stronger repulsion to pairing molecules by the charge reversion. Notably, K270D/K290D, but not K270A/K290A Climp63, densely accumulated in the perinuclear region (Fig. S4B), even though they were expressed to equivalent levels (Fig. S4I). Consistent with this, we found perinuclear clustering of ER sheets in K270D/K290D-expressing cells after EM analysis (Fig. 4A). These results suggest that K270D/K290D might also affect cis association and, in this case, probably enhances it. We also tested the positively charged CT patch. A K597E and K601E double mutation (K597E/K601E) caused slightly thinner ER sheets, whereas K597A and K601A double mutation (K597A/K601A) further decreased the ER width (Fig. 4A); both of these proteins were distributed normally in the cell compared to the WT (Fig. S4C,I). We then combined the two sets of lysine mutations. K270D/K290D/K597E/K601E exhibited a cellular pattern similar to K270D/K290D (Fig. S4D,I) and caused a large variation in the size of the ER lumen, with some budging (up to 70 nm) and some narrowing (∼30 nm, close to that in KO cells) (Fig. 4A). K270A/K290A/K597A/K601A acted similarly to K597A/K601A, as it partially rescued the ER collapse seen in the KO cells (Fig. 4A). Next, we tested negatively charged patches. E278A/D285A presented a thinner ER phenotype, similar to that of K597A/K601A (Fig. 4B). E265A/E295A had an even thinner ER, but with a large variation in size (Fig. 4B). The combination of these mutations caused further defects in the regulation of luminal spacing (Fig. 4B). Notably, E265A/E295A caused perinuclear condensation, similar to that of K270D/K290D, and E278A/D285A/E265A/E295A was even severer in that regard (Fig. S4E,I), suggesting an alteration of cis-trans coordination. We also combined mutations of different charges (Fig. S4F,J) and found that K270D/K290D/E265R/E295R was most defective in regulating ER width, equivalent to the K270D/K290D mutant but with a large variation in size (Fig. 4C). Taken together, these results confirm that at least four groups of residues in the distal end of the 5HB are involved in trans association, likely in different combinations.

We have previously shown that Climp63 serves as an ER luminal spacer by forming moderate self-associations (Zhao and Hu, 2020). Taking advantage of recently developed structural prediction methods, in this case the original RF program, we found that the LD domain of Climp63 consists of a flexible leading helix followed by a 5HB. Variations in modeling results indicate that the strength of 5HB assembly is moderate, leaving the first helix readily displaced. Nevertheless, a point mutation in the α1–α2 loop and probing of the trans association all strongly support the existence of the 5HB, but not the 4HB, as AF elucidated. These analyses suggest that deep learning-based structural prediction is capable of unveiling subtle features of protein folding even though a direct answer is not offered. It is also uniquely beneficial when the targeted protein, such as Climp63, tends to form heterogeneous oligomers.

Our findings suggest that Climp63 utilizes multiple interfaces for complex self-association, most of which are electrostatic interactions. The trans association essential for ER luminal spacing is mediated by at least four sets of charged residues, which are all localized to the distal end of the 5HB. Cysteine-based cross-linking confirms the inter-molecular pairing of K601 and D285. Mutations in these residues reduce homotypic assembly in vitro and compromise the maintenance of ER width to different extents in cells. Interestingly, mutants such as K270D/K290D accumulate in perinuclear ER membranes, indicative of elevated cis association, but fail to form normal trans associations. These observations imply regulated coordination between cis and trans assemblies. Membrane shaping and tethering often involve protein oligomerization. The multifaceted assembly of Climp63 (i.e. a complicated combination of weak interfaces) is fundamentally important for ER lumen spacing. Such an assembly mode likely offers the necessary plasticity and fine-tuned regulation of membrane tethering and ER morphogenesis.

For protein purification constructs, HA- or Flag-tagged full-length protein, the 5HB, and LD of human Climp63 were PCR amplified and inserted into pGEX-6p-3 (27-4599-01, GE Healthcare; BamHI and XhoI sites) by in-fusion cloning (C112, Vazyme). Point mutations in Climp63 were generated by site-directed mutagenesis of WT Climp63. For mammalian cell stable expression, N-terminal HA-tagged human Climp63 mutations were PCR amplified and inserted into pLenti-CMV-GFP-puro (a kind gift from Quan Chen, Department of Genetics and Cell Biology, College of Life Sciences, Nankai University, Tianjin, China; XbaI and SalI sites) by removing GFP using similar strategies to above. For constructs used in cross-linking experiments, N-terminal HA- or Flag-tagged human mutant Climp63 forms were PCR amplified and inserted into 4TO (V1030-20, Invitrogen; BamHI and XhoI sites). All constructs were confirmed by sequencing.

GST-tagged full-length Climp63 (residues 1–602) and the LD of Climp63 (residues 128–602 and residues 170–602 for GST–LD and GST–5HB, respectively) were expressed in Escherichia coli strain BL21 (DE3; CD801, TransGen Biotech). Transformed bacteria were cultured at 37°C in LB medium (214906, BD Biosciences) containing 100 μg/ml ampicillin to an optical density (OD) at 600 nm of 0.8, and then 0.3 mM isopropyl-β-D-thiogalactoside (IPTG; BN30014, Biorigin) added for 20 h at 16°C to induce protein expression. The bacteria were harvested, resuspended in protein purification buffer (500 mM NaCl, 25 mM HEPES pH 7.4, 10% glycerol for full-length Climp63; 500 mM NaCl, 25 mM HEPES pH 7.4 for Climp63 LD), and lysed by ultrasonication for 30 min on ice. The lysates were centrifuged at 198,000 g for 1 h using the type 45 Ti rotor (Beckman Coulter). For full-length Climp63, the membrane pellet was solubilized in protein purification buffer containing 1% Fos-choline-12 (Anatrace) and insoluble components cleared by centrifugation at 198,000 g for 1 h. The recombinant protein was isolated with glutathione–Sepharose (GE Healthcare). After that beads were washed twice with protein purification buffer (for full-length Climp63, with 0.1% Triton X-100). The GST tag was cleaved from full-length Climp63 by 3C protease (purified in-house) overnight at 4°C. The Climp63 LD or 5HB with GST tag was eluted with buffer containing 10 mM glutathione. The protein was further purified by gel filtration chromatography (Superdex-200 Increase 10/300 GL; GE Healthcare). Samples were verified by Coomassie Brilliant Blue staining.

For pulldown assays using full-length Climp63, HA-tagged Climp63 was immunoprecipitated with anti-HA agarose beads (A2095, Sigma) at 4°C for 1 h. After the beads were washed twice, Flag-tagged Climp63 was added and incubated at 4°C for 1.5 h. The beads were washed three times. For the pulldown assay between Climp63 full-length protein and LD or 5HB, HA-tagged Climp63 full-length protein was immunoprecipitated with anti-HA agarose beads at 4°C for 1 h. After the beads were washed twice, the GST-tagged Climp63 LD or 5HB was added and incubated at 4°C for 1.5 h. The beads were washed three times. Precipitated proteins were eluted with 2×SDS-PAGE sample buffer and detected by western blotting using the indicated antibodies (anti-HA antibody, 1:1000, H6908, Sigma; anti-Flag antibody, 1:1000, F7425, Sigma).

Climp63-deleted U-2 OS cells (HTB-96, ATCC) were generated by CRISPR Cas9 using pX330 vector (#42230 Addgene). The guide RNA (gRNA) sequence was 5′-CGCCGCGCCCGCCATGCCCT-3′. Climp63-deleted COS-7 cells were generated by CRISPR Cas9 using pX458 vector (#48138 Addgene). The guide RNA (gRNA) sequence was 5′-ATGCCCTCGGCCAAACAAAG-3′. 293T cells (CRL-3216, ATCC) and U-2 OS cells were cultured in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum (10099-141, Gibco) at 37°C in a 5% CO₂ atmosphere. To generate stable expression cell lines, 293T cells were co-transfected with Climp63 constructs, pSPAX2 (#12260 Addgene) and pMD.2G (#12259 Addgene) using PEI reagent (23,966, Polysciences) according to the manufacturer's instructions. The medium containing retrovirus was collected and passed through a 0.45-μm filter to remove cellular debris, and then added to the Climp63 KO cells with 8 μg/ml polybrene.

For electron microscopy, cells were collected and fixed with a combination of 4% paraformaldehyde and 2% glutaraldehyde. Images were collected using a FEI spirit 100 kV TEM.

For immunofluorescence, cells were fixed with 4% paraformaldehyde in PBS for 20 min, and blocked with 3% BSA for 1 h. The cells were then immunostained for 2 h at room temperature with mouse anti-Climp63 primary antibody (1:1000, ENZ-ABS669-0100, Enzo), and then incubated with fluorophore-conjugated secondary antibody (Alexa Fluor 488-conjugated anti-mouse, Invitrogen). The images were captured on a confocal microscope (LSM710, ZEISS) with a 63× objective lens.

Climp63-deleted COS-7 (CRL-1651, ATCC) cells were transfected with indicated Climp63 constructs for 24 h, and 2 mM diamide (Sigma) was added to the medium and incubated for 40 min. 20 mM N-ethylmaleimide (NEM, Thermo Fisher Scientific) was then added to quench the reaction for 25 min. Cells were harvested with lysis buffer (150 mM NaCl, 50 mM Tris-HCl pH 7.4, 0.5% Triton X-100) for 20 min on ice. The lysates were centrifuged at 13,000 g for 15 min. Loading buffer with or without 25 mM DTT was added to the lysates and samples were analyzed by western blotting with indicated antibodies (anti-HA antibody, 1:1000, H6908, Sigma; anti-Flag antibody, 1:1000, F7425, Sigma).

Protein samples from pulldown assays and cross-linking assays were separated by SDS-PAGE, and transferred to PVDF membranes (03010040001, Roche). The membranes were blocked with 5% skim milk (P0216, Beyotime) for 1 h at room temperature, and incubated with anti-HA or anti-Flag antibody for 1 h at room temperature. The membranes were then washed three times for 5 min each time with PBST, incubated with secondary antibody (A0545, Sigma) for 40 min at room temperature, and washed three times for 5 min each time with PBST. The bands on membranes were visualized using HRP substrate (WBKLS0500, Millipore) and captured by imaging system (5500, Tanon). Uncropped images of western blots are presented in Fig. S5.

The structure of full-length human Climp63 (residues 1–602) was predicted by RoseTTAFold and AlphaFold v 2.0. Comparisons and illustrations were prepared using the program PyMOL (https://pymol.org).

The results are presented as mean±s.e.m. Immunoblot bands, ER luminal width and fluorescence intensities were quantified using ImageJ software and analyzed by means of an unpaired two-tailed Student's t-test. At least three independent repeats were performed for each experiment. Calculations were performed using GraphPad Prism 8 software.

We thank Zhongshuang Lv, Xueke Tan, Can Peng, and Xixia Li for helping with electron microscopy sample preparation and TEM imaging at the Center for Biological Imaging (CBI), Institute of Biophysics, Chinese Academy of Science.

The peer review history is available online at https://journals.biologists.com/jcs/lookup/doi/10.1242/jcs.260976.reviewer-comments.pdf