ERLIN1/2 scaffolds bridge TMUB1 and RNF170 and restrict cholesterol esterification to regulate the secretory pathway

The ERLIN complex mediates the interaction between TMUB1 and RNF170

To shed light on the processes regulated by complexes of ERLIN1 and ERLIN2, we introduced out-of-frame deletions in exon 1 of ERLIN1 and exon 2 of ERLIN2 using CRISPR/Cas9 gene editing in HeLa cells. Double knock-out (DKO) cells showed reduced mRNA levels of ERLIN1 and ERLIN2 consistent with RNA nonsense-mediated decay (Fig 1A), and lack of protein expression by Western blot using antibodies that recognize regions downstream of the deletions (Fig 1B). To control for off-target effects of gene editing, we used retroviral transduction in DKO cells to re-express the proteins. For further experiments, we selected a clone that showed near-to-endogenous levels of ERLIN2 and the substantial re-expression of ERLIN1 (DKO^+E1/E2) (Fig 1B). Immunofluorescence using a specific antibody against ERLIN2 showed a staining consistent with localization to the ER that was lost in the DKO and restored in DKO^+E1/E2 cells (Fig 1C). Furthermore, recovery of detergent-resistant membranes (DRMs) after solubilization with a non-ionic detergent (Triton X-100) by flotation using an OptiPrep gradient revealed accumulation of both endogenous and re-expressed ERLIN1 and ERLIN2 in the same fraction as other DRM markers, such as flotillin-1 (FLOT1) and KIDINS220 (Fig 1D). This result agrees with the known accumulation of ERLINs in cholesterol-rich lipid nanodomains (Browman et al, 2006).

Currently, it is unclear whether different ERLIN complexes exist in association with distinct E3 ubiquitin ligases or whether ERLINs organize membrane domains containing more than one E3 ubiquitin ligase. A limitation of earlier studies is that they have been performed in overexpression conditions, which may promote unspecific interactions. We employed anti-ERLIN2 antibodies to pull down the endogenous complex and identify interacting proteins by label-free quantitative mass spectrometry (MS) (Fig 1E and Table S1). Among the most enriched proteins in the immunoprecipitation (IP) of WT cells were the molecular partner ERLIN1 and the E3 ubiquitin ligase RNF170. In addition, two other ERAD regulators, TMEM259 (transmembrane protein 259, also known as membralin) and TMUB1, were highly and significantly enriched, in agreement with previous studies (Fig 1E) (Jo et al, 2011; van de Weijer et al, 2020). We also identified the trans-2,3-enoyl-CoA reductase (TECR), an endoplasmic reticulum multipass protein that is involved in the fatty acid elongation cycle. In addition, ERLIN2 IP included other proteins that associate with DRMs, such as FLOT1 and stomatin (STOM), or proteins that traffic along the lipid rafts, as revealed by pathway analysis (Fig 1F). We also detected proteins belonging to other organelles, such as the mitochondrial voltage-dependent anion-selective channel protein 2 (VDAC2), the peroxisomal ATP–binding cassette subfamily D member 3 (ABCD3), and the small GTPases ARF1 and ARF4 (ADP-ribosylation factor 1 and 4), which are involved in vesicular trafficking (Adarska et al, 2021), possibly reflecting the presence of ERLINs at the contact sites between the ER and other organelles.

TMUB1 contains a ubiquitin-like (UBL) domain, and acts as an escortase stabilizing intermediates of membrane proteins during retro-translocation and delivering them to p97/VCP (valosin-containing protein) (Wang et al, 2022). A long and a short isoform of TMUB1 (from now on TMUB1-L and TMUB1-S) deriving from alternative initiation of translation have been previously reported (Fig 2A) (Castelli et al, 2014; van de Weijer et al, 2020; Wang et al, 2022). Reciprocal co-IP experiments demonstrated that ERLIN2 interacts specifically with TMUB1-L (Fig 2B and C). Furthermore, the amount of TMUB1-L recovered in the DRM fraction was decreased in DKO cells and rescued upon the re-expression of the ERLINs (Fig 2D–F), suggesting a role of ERLIN complexes to cluster TMUB1-L in specific membrane nanodomains. In contrast, the distribution of TMUB1-S and of other ERLIN2 interactors remained unaffected (Figs 2D–F and S1). Unfortunately, we could not test whether the lack of ERLINs impairs targeting of RNF170 to the DRMs because of a lack of suitable antibodies.

To gain further insights into the specific function of TMUB1-L, we generated TMUB1-L–specific KO cell lines using an inactive Cas9 fused with an adenine deaminase to mutagenize the first ATG (Gaudelli et al, 2017) (Fig S2A), and used one of the clones as control in IP experiments, followed by MS (Table S2). ERLIN1 and ERLIN2 emerged as the top proteins co-immunoprecipitated by TMUB1-L, validating our earlier findings (Fig 2G). Moreover, we confirmed the interaction of TMUB1-L with RNF170, TMEM259, RNF185, FAF2 (FAS-associated factor 2), and VCP (Wang et al, 2022). FAF2 is anchored to the ER by a hairpin domain, contains both a UBA and UBX domain, and has been previously implicated in ERAD pathways (Mueller et al, 2008; Olzmann et al, 2013). Intriguingly, AlphaFold-Multimer predicts not only the known interaction of the UBX domain with VCP (Schuberth & Buchberger, 2008), but also a possible interaction between the UBA domain of FAF2 and the UBL domain of TMUB1-L (Fig S2B–D).

We then asked whether any of the molecular interactions of TMUB1-L depend on the presence of ERLIN scaffolds. To this purpose, we used antibodies detecting both TMUB1 isoforms for IP experiments in WT and DKO cells (Figs 2H and S2E and Table S2). Many proteins enriched in the TMUB1 IP belonged to a nucleolar compartment but were likely due to a cross-reactivity of the antibody, because a nucleolar staining could not be suppressed by down-regulation of TMUB1 with siRNA (Fig S2F and G). Surprisingly, lack of ERLINs completely prevented the interaction of TMUB1 with RNF170 (Fig 2H), whereas the TMUB1-RNF185-TMEM259 interaction previously reported was still detected in DKO cells. We conclude that the ERLIN scaffolds are required to cluster TMUB1-L and RNF170 in ER membrane nanodomains.

A conserved luminal domain in TMUB1-L and RNF170 interacts with the ERLIN complex

TMUB1-L and RNF170 have a similar topology with a luminal domain, three transmembrane domains, and a cytosolic UBL and RING domain, respectively (Fig 3A). Inspection of the N-terminal luminal domain of TMUB1-L, which is required for interaction with ERLINs, revealed the presence of a stretch of seven amino acids that was conserved in the N-terminal domain of RNF170 (Fig 3B). We used AlphaFold-Multimer (Evans et al, 2021 Preprint), to model the three-dimensional interaction between TMUB1-L, RNF170, and a dimer of ERLIN1 and ERLIN2 (Fig 3C). The best model predicted that the conserved N-terminal motif of RNF170 and TMUB1 interacts with the SPFH domain of each ERLIN subunit at different interfaces. The first interaction occurs with a β-sheet in the ERLINs. Hydrogen bonds are predicted between Tyr51 of ERLIN1 and Gly21 of RNF170, whereas Phe49 and Leu51 in ERLIN2 contact Gly6 and Gly10 of TMUB1, respectively. On the second interface, Glu5 of TMUB1-L or the corresponding Glu20 of RNF170 binds to the adjacent ERLIN subunit with hydrogen bonds predicted with Arg38 and Thr44 in ERLIN1 and Arg36 and Thr42 in ERLIN2 (Fig 3D).

A Gly50Val substitution in ERLIN1 has been implicated in autosomal recessive SPG62 (Novarino et al, 2014), whereas His50Tyr (ClinVar ID: 947803) and Arg36Lys (ClinVar ID: 1720746) in ERLIN2 have been reported as a variant of unknown significance in individuals with spastic paraplegia. We explored whether these variants alter the interaction of the ERLINs with TMUB1-L and RNF170, by modelling the interaction after mutating the corresponding residues in ERLIN2 (Fig 3E). The size of the ERLIN2 β-sheet between Gly48 and His50 is shortened by introducing the mutations, and the hydrogen bonds predicted between TMUB1-L and the ERLIN subunits at the two interaction interfaces are lost (Fig 3E). The expression of ERLIN2-FLAG mutant constructs carrying these variants individually or in combination showed that the Gly48 is required for the stability of the protein (Fig 3F). Similarly, the simultaneous mutation of the two residues predicted to form hydrogen bonds with both TMUB1-L and RNF170 also induces protein degradation (Fig 3F). These data support the hypothesis that these variants are pathogenic and that the interaction of ERLINs with TMUB1-L and RNF170 plays a role in stabilizing the ERLIN complex.

Loss of ERLINs perturbs pathways linked to cell adhesion and vesicular trafficking

To uncover pathways regulated by the ERLIN complex, we first used unbiased omics-based approaches. Transcriptomics analysis of WT and DKO cells confirmed activation of the SREBP pathway upon ERLIN depletion (Huber et al, 2013); however, this was milder than previously reported (Fig 4A and Table S3). Over-representation analysis of the DKO transcriptome showed a significant down-regulation of GO terms related not only to the ER, but also to focal adhesion, cell adhesion, and cadherin binding (Fig 4B). We used label-free quantitative MS to reveal changes at the proteome level in DKO cells. We only considered proteins measured by at least two peptides, showing significant changes in abundance between DKO and WT, and substantially rescued by re-expressing the ERLINs. Of 2,742 proteins measured in the post-nuclear supernatant, only 34 fulfilled these criteria (Fig 4C and Table S4). Many of these proteins functionally belonged to pathways related to sterol and lipid metabolism and cell adhesion (Fig 4C). Most of these changes were mirrored by transcriptional alterations (Fig 4C). ER-resident enzymes involved in neutral lipid and cholesterol metabolism (ACSL1, HMGCR, and SQLE) and regulated by SREBPs were increased. Furthermore, we found that β-catenin (CTNNB1) was increased in DKO cells and rescued in the DKO^+E1/E2 cells. Alteration of the β-catenin pathway was further confirmed via Western blot (Fig S3A and B) and was paralleled by a migration defect in a scratch assay, in which DKO cells took more time to fill a gap on a plate compared with WT and DKO^+E1/E2 cells (Fig S3C and D).

We reasoned that the lack of the scaffolding function of the ERLIN complex in the ER may not necessarily affect overall protein abundance but may change their dynamic subcellular distribution by impairing recruitment to cholesterol-enriched nanodomains. We therefore combined the flotation gradient used to separate DRMs with stable isotope labelling by amino acids in cell culture (SILAC) (Ong et al, 2002) and analysed protein abundance in five collected fractions with different densities. The major advantage of this approach is allowing us to combine WT, DKO, and DKO^+E1/E2 cell lysates before loading them on the flotation column, thus reducing the variability of fraction collection among samples (Fig 5A). Proteins associated with DRMs are recovered in low-density fractions; however, they can shift from one fraction to another upon altered cellular lipid composition or signalling stimulation (Simons & Toomre, 2000; Foster et al, 2003). This approach confirmed that in WT cells, ERLIN1, ERLIN2, and other DRM-associated proteins, such as FLOT1 and FLOT2, are enriched in fraction II, as previously detected by Western blot (Figs S4 and 1D). The overall distribution of the log₂ normalized protein ratio was similar between all comparisons in fractions II to V (Fig 5B). Fraction I contained significantly less proteins than the other fractions, which accumulated in the DKO compared with the other genotypes (Fig 5B). We used unsupervised clustering to identify proteins that were significantly changed in the three comparisons (KO/WT, DKO^+E1/E2/DKO, and DKO^+E1/E2/WT). We identified cluster 170 in fraction I and cluster 155 in fraction II as containing proteins that were up-regulated in the DKO and rescued in the DKO^+E1/E2 cells (Fig 5C, D, F, and G and Table S5). Pathway analysis showed that these clusters were enriched in proteins belonging to adherens and anchoring junctions, to the plasma membrane, and to membrane-bound vesicles (Fig 5E and H). Surprisingly, cluster 577 in fraction IV contained proteins belonging to the same GO terms, showing, however, an opposite trend compared with the previous two clusters. Thus, proteomics of DKO cells suggests perturbations of proteins that follow the secretory route.

Loss of the ERLIN complex leads to a collapse of ER tubules and Golgi fragmentation

Based on the previous results, we hypothesized that trafficking along the secretory compartment may be impaired in DKO cells. As a first step, we used antibodies against reticulon 4 (RTN4), which is enriched in the tubular ER, to assess whether the general morphology of the ER was affected in DKO cells. Although the total levels of RTN4 did not change in cells of different genotypes (Table S4), immunofluorescence analysis showed altered ER morphology in DKO cells, which were intriguingly depleted of ER peripheral tubules in comparison with both WT and DKO^+E1/E2 cells (Fig 6A and B). We also noticed that the cell area labelled by RTN4 staining in DKO cells appeared smaller 24 h after plating (Fig S5A), although the cell size was unchanged when assessed in cytofluorometry (Fig S5B). Therefore, the ER phenotype may be influenced by a defect of the cells in spreading on the plate, consistent with the cell adhesion defect.

To assess trafficking along the secretory system, we employed the retention using selective hooks (RUSH) system (Boncompain et al, 2012). This system comprises a hook fused to streptavidin, which is bound to the donor compartment (the ER), and a reporter protein fused to a streptavidin-binding domain. The addition of biotin to the medium synchronously competes with the streptavidin-binding domain, releasing the reporter protein from the hook and allowing its trafficking along the secretory pathway. We employed the vesicular stomatitis virus G (VSVG) glycoprotein expressed with a fluorescent EGFP tag as a reporter and monitored in live imaging its trafficking after biotin addition in WT, DKO, and DKO^+E1/E2 cells (Fig S5C and Video 1, Video 2, and Video 3). In WT and DKO^+E1/E2 cells, VSVG-GFP translocated from the ER to the Golgi compartment within ∼10 min, followed by trafficking to the exocytic compartment. In ERLIN-deficient cells, VSVG-GFP moved from the ER to a fragmented vesicular compartment within 10 min after adding biotin, and then remained largely confined within these structures. When we analysed colocalization of the VSVG signal with the Golgi marker GM130 in cells fixed at different time points after biotin addition, we confirmed that in DKO cells, the VSVG-EGFP remained elevated in the Golgi after 60 min from biotin addition, in contrast to WT and DKO^+E1/E2 cells (Fig 6C and D). Furthermore, the loss of ERLINs induced Golgi fragmentation (Fig 6E and F), a phenotype that was rescued in DKO^+E1/E2 cells (Fig 6E–G). We then examined whether Golgi proteins changed their abundance in the different fractions analysed by SILAC among genotypes (Table S5), as a further indication of the organelle dispersal in DKO cells. ARF1 was affected in clusters 170, 155, and 577 in fractions I, II, and IV, respectively. Additional Golgi proteins, including GORASP2 (GRASP55), were changed in abundance in cluster 577 of fraction IV (Table S5). The steady-state levels of GM130 (GOLGA2) were unaffected in the post-nuclear supernatant proteomics but were perturbed in the fraction III analysed by SILAC (Tables S4 and S5). Thus, the lack of the ERLIN complex does not impair ER-to-Golgi trafficking but induces morphological defects of the Golgi apparatus and affects post-Golgi trafficking. Notably, we observed dispersion of the Golgi apparatus also upon TMUB1-L depletion (Fig S5D and E).

Excessive cholesterol esterification in the absence of ERLINs causes Golgi fragmentation

The Golgi apparatus is a central sorting station not only for proteins, but also for lipids that traffic along the secretory pathway. Although the presence of lipid rafts in the ER is still debated, it is accepted that lipid nanodomains enriched with cholesterol and sphingolipids are present along the secretory pathway, with increasing abundance from the ER to the plasma membrane. These domains are implicated in the intracellular trafficking of proteins, such as GPI-linked receptors, integrins, and other adhesion molecules. By binding cholesterol at the ER, complexes of ERLINs may control the distribution of cholesterol in cellular membranes. We measured levels of ceramides and sphingomyelin in the DRMs of WT, DKO, and DKO^+E1/E2 cells using targeted lipidomics. We found a higher quantity of species of both lipids containing fatty acids with chain lengths between C18 and C24 in the DKO cells, which were restored to WT levels in the DKO^+E1/E2 cells (Fig 7A and B and Table S6). The protocol used for the preparation of DRMs prevented measuring cholesterol levels in these fractions. We therefore assessed the content of free cholesterol and cholesterol esters (CE) in the total cell lysates. Unesterified cholesterol was not affected upon ERLIN depletion (Fig 7C). However, total CE and especially the abundant CE 18:1 species were significantly increased when ERLINs were depleted and recovered to WT levels in DKO^+E1/E2 cells (Fig 7D and E and Table S6). Avasimibe, an inhibitor of SOAT1 (sterol O-acyltransferase 1), the enzyme that catalyses the formation of fatty acid–cholesterol esters, strongly suppressed levels of CE18:1 in WT cells, confirming that this species is mainly synthesized endogenously (Fig 7E). Levels of triacylglycerides (TAGs) also increased in the DKO compared with both WT and DKO^+E1/E2 cells (Fig 7F and Table S6). Consistent with the accumulation of neutral lipids, larger LDs labelled by Bodipy 493/503 accumulated in DKO cells, a phenotype that was rescued upon re-expression of ERLINs (Fig 7G).

Esterification of excess cholesterol at the ER and its accumulation in LDs are observed when cholesterol transport from the ER to the Golgi is impaired (Mesmin et al, 2017). Cholesterol trafficking between these two organelles mostly occurs via non-vesicular transport, at sites of close apposition between the membranes of the organelles (Ikonen & Zhou, 2021). This transfer is mediated by oxysterol-binding proteins (OSBPs), which bind cholesterol at the ER and release it at the Golgi in exchange for a molecule of phosphatidylinositol-4-phosphate (Ikonen & Zhou, 2021). The FFAT (two phenylalanines—FF—in an acidic tract) domain of OSBP binds the protein VAPA (VAMP-associated protein A) at the ER, whereas the pleckstrin homology (PH) domain binds ARF1 at the Golgi membrane. Notably, ARF1 was identified in the ERLIN2 interactome (Fig 1E), suggesting a close apposition between the ERLIN complex and the Golgi. Cholesterol balance in the Golgi is essential for protein transport, and both cholesterol depletion and cholesterol loading block trafficking of secretory proteins from the trans-Golgi network (Stüven et al, 2003). In addition, lack of cholesterol in the Golgi may lead to vesiculation (Stüven et al, 2003). DKO cells showed a modest but significant activation of SREBPs, indicating that they evade feedback inhibition by cholesterol synthesis. Although the destabilization of INSIG1 upon ERLIN knockdown could explain this result (Huber et al, 2013), an alternative explanation could be that increased cholesterol esterification keeps levels of cholesterol in the ER low, impairs transport to the Golgi, and prevents feedback inhibition of SREBPs. To test this hypothesis, we treated WT and DKO cells with avasimibe. Remarkably, inhibition of SOAT1 activity not only reduced the size of the LDs in DKO cells (Fig 8A and C), but also rescued the Golgi fragmentation (Fig 8B and D), supporting our hypothesis. SOAT1 inhibition suppressed the transcriptional up-regulation of 3-hydroxy-3-methylglutaryl-CoA synthase (HMGCS), a SREBP target gene in both WT and DKO cells, blunting the difference between genotypes (Fig 8E). We propose a model in which the ERLIN complex, by binding cholesterol, restricts its esterification and thus promotes its transport to the Golgi. The loss of ERLIN1/2 scaffolds on the one hand depletes the ER of cholesterol, activating the SREBP pathway, and on the other hand leads to Golgi vesiculation and defects in post-Golgi trafficking.

Cell culture

HeLa cells were cultured at 37°C and 5% CO₂ in DMEM containing 4.5g/litre of glucose (# 11960044; Gibco), 2 mM glutamine (# 25030024; Gibco), 100 U/ml penicillin, 100 μg/ml streptomycin (# 15140122; Gibco), and 10% FetalClone III (#SH30109.03; GE Healthcare Hyclone). For SILAC labelling, cells were grown in arginine- and lysine-free DMEM with 4.5g/litre glucose (# 28001300; Silantes), containing dialysed FCS (#26400044; Gibco), penicillin–streptomycin–glutamine (# 10378016; Gibco), and different isotopes of arginine and lysine. We used medium isotopes (Arg6, Lys4) (# 201203902 and 211103913; Silantes) for the WT, light (Arg0, Lys0) (# 201003902 and 211003902; Silantes) for the DKO, and heavy (Arg10, Lys8) (# 201603902 and 211603902; Silantes) for the DKO^+E1/E2. Cells were regularly tested to be mycoplasma-free.

Knock-out of ERLIN1 and ERLIN2 in HeLa cells via CRISPR/Cas9

ERLIN1 and ERLIN2 were knocked out using a CRISPR/Cas double-nickase approach as described previously (Ran et al, 2013). CRISPR guide RNAs (gRNAs) were designed to target exon 1 of ERLlN1 (guide sequence sense strand 5′-GCCATCTGGCTGTGTACTAC-3′, guide sequence antisense strand 5′-TGTGGATGGAGGCGTAGAGC-3′) and exon 2 of ERLIN2 (guide sequence sense strand 5′-GTTGGGAGCAGTTGTGGCTG-3′, guide sequence antisense strand 5′-AGGATAAAGGCTCACTGATG-3′), respectively, and confirmed with CRISPOR (http://crispor.tefor.net). Oligonucleotides were purchased from Sigma-Aldrich and cloned into the pX335-U6-Chimeric_BB-CBh-hSpCas9n(D10A) vector, a gift from Feng Zhang (plasmid # 42335; Addgene) (Cong et al, 2013). Constructs were transfected into HeLa cells on five consecutive days, and clones were obtained via serial dilution. Positive clones were identified by Western blot analysis using anti-ERLIN1 and anti-ERLIN2 antibodies and confirmed by subcloning targeted gene regions into pcDNA3 (amplification of exon 1 of ERLIN1 with 5′-TTAGGATCCGCGAGAGAGCGCCAAGTTTC-3′ and 5′-AATGCGGCCGCCAACATCTGGAGGGTGCT-3′ and amplification of exon 2 of ERLIN2 with 5′-TTAGGATCCGAAAGGGGAACGTTGACTGA-3′ and 5′-AATGCGGCCGCCTTTCCCGTGGTTACTGACA-3′) followed by sequencing.

Generation of retroviruses and transduction

Coding regions of ERLIN1 and ERLIN2 (amplified with 5′-GGGGACAAGTTTGTACAAAAAAGCAGGCTAGAATGAATATGACTCAAGCCCG-3′ and 5′-GGGGACCACTTTGTACAAGAAAGCTGGGTCTCAACCTGTGCTCTCTTTGTTTT-3′; 5′-GGGGACAAGTTTGTACAAAAAAGCAGGCTACCATGGCTCAGTTGGGAGCAGTT-3′ and 5′-GGGGACCACTTTGTACAAGAAAGCTGGGTCTCAATTCTCCTTAGTGGCCGTCT-3′, respectively) were cloned into the pDONR221 and inserted into the pBABE-puro vector using the gateway recombinant cloning (Morgenstern & Land, 1990). The pBABE-puro vector was a gift from Hartmut Land & Jay Morgenstern & Bob Weinberg (plasmid # 1764; http://n2t.net/addgene:1764; RRID:Addgene_1764; Addgene). Respective constructs were transfected into Phoenix-AMPHO packaging cells for 24 h. The harvested virus-containing medium was filtered through a 0.45-μm filter. For infection, HeLa cells at 70% confluency were incubated for 24 h in a virus-containing medium and 4 μg/ml polybrene. Cells with positive genome integration were selected by incubation of infected cells in DMEM containing 5 μg/ml puromycin for 48 h. The expression of the gene of interest was confirmed by Western blot analysis using ERLIN1 and ERLIN2 antibodies, respectively.

TMUB1-L KO generation

The start codon of the TMUB1 gene was mutagenized with an adenine base editor to convert the ATG into GTG (Gaudelli et al, 2017). The gRNA was designed using the rgenome webtool (http://www.rgenome.net/), and primers were purchased from Sigma-Aldrich and cloned into gRNA Cloning Vector, a gift from George Church (plasmid # 41824; Addgene) (Mali et al, 2013) after digestion with AflII and through the Gibson assembly of the PCR product of the primer pair (guide sequence sense strand 5′-TTTCTTGGCTTTATATATCTTGTGGAAAGGACGAAACACCGCGCCATGACCCTGATTGAAG-3′, guide sequence antisense strand 5′-GACTAGCCTTATTTTAACTTGCTATTTCTAGCTCTAAAACCTTCAATCAGGGTCATGGCGC-3′). The gRNA-containing vector was cotransfected with the pCMV_ABEmax_P2A_GFP, a gift from David Liu (plasmid # 112101; Addgene) (Koblan et al, 2018), and after 24 h sorted using INFLUX Cell-Sorter (BD Biosciences). Two 96-well plates were prepared, and the successful mutagenesis was tested via Western blot analysis.

ERLIN2 overexpression constructs

pERLIN2-3xFLAG, pERLIN2-3xFLAG (R36K), pERLIN2-3xFLAG (G48V), pERLIN2-3xFLAG (H50Y), pERLIN2-3xFLAG (R36K-H50Y), and pERLIN2-3xFLAG (R36K-G48V-H50Y) were cloned in the p3XFLAG-CMV-14. All the mutations were confirmed via Sanger sequencing. For cloning, ERLIN2 was amplified with 5′-TTAGAATTCATGGCTCAGTTGGGAGCAGTT-3′ and 5′-AATGGATCCATTCTCCTTAGTGGCCGTCTC-3′ from cDNA and the product was cloned into p3XFLAG-CMV-14 using EcoRI and BamHI. The R36K variant was generated with Q5 Site-Directed Mutagenesis Kit (# E0554S; NEB) from the pERLIN2-3XFLAG using the manufacturer’s protocol (primer sequence: Fwd 5′-GTATATTACAAAGGCGGTGCC-3′ and Rev 5′-CCCAATATGTCCCTCTTCTATC-3′). All the other variants were generated using the Gibson assembly (HiFi DNA Assembly Master Mix, # E2621S; NEBuilder) from the pERLIN2-3XFLAG using the manufacturer’s protocol. The pERLIN2-3XFLAG was amplified using four primers, two of which binding the backbone common for all the mutations (Fwd 5′-TAATGAGTGAGCTAACTCACATTAATTGCGTTGCG-3′ and Rev 5′-GTGAGTTAGCTCACTCATTAGGCACCCC-3′), and other two primers specific for the desired mutation (G48V: 5′-CAGCGGCCCTGTTTTCCATCTCATGCTCC-3′ and 5′-GATGGAAAACAGGGCCGCTGGTCGAAGTC-3′; H50Y: 5′-CCCTGGTTTCTATCTCATGCTCCCTTTCATC-3′ and 5′-GCATGAGATAGAAACCAGGGCCGCTGGTC-3′; G48V-H50Y: 5′-CAGCGGCCCTGTTTTCTATCTCATGCTCCCTTTC-3′ and 5′-GATAGAAAACAGGGCCGCTGGTCGAAGTC-3′).

TMUB1 siRNA

For TMUB1 down-regulation, stealth siRNA from Thermo Fisher Scientific was used at 100 nM final concentration. After 2 d from the first transfection, performed using Lipofectamine 2000, a second transfection was performed and the cells were analysed after 2 d. Shorter protocols failed to down-regulate TMUB1 in our system. Three siRNAs were used: 5′-ACCUCCCUCCCAACUGCGUUCUCCA-3′, 5′-GGGAACAGCAGGUGCGACUCAUCUA-3′, and 5′-GCCAUGGCAGCUACCGACAGCAUGA-3′.

Flow cytometry

Cell size was compared using BD LSRFortessa Analyzer. 24 h after seeding, cells were detached with trypsin, resuspended in fresh DMEM, and washed twice with PBS. Pellets were resuspended in PBS and analysed. Cell debris and cell doublets were excluded from the analysis by gating. For the TMUB1-L KO sorting, an untransfected control not expressing the GFP was used to discriminate the GFP-positive events.

Scratch assay

2 × 10⁶ cells were seeded on a six-well plate and grown confluent over 3 d. A wound was introduced using a 1-ml pipette tip before the plate was washed with PBS and the fresh culture medium was added. Wound size was subsequently imaged with an Echo Revolve microscope. Plates were washed with PBS, and fresh culture medium was added, before imaging every 24 h. The area of wound size was quantified manually using Fiji/ImageJ 2.9.0 (Schindelin et al, 2012).

Immunofluorescence staining

Cells were washed once with PBS 1X (# 18912014; Thermo Fisher Scientific) and fixed at RT for 15 min in 4% PFA, pH 7.4 (# P6148-500G; Sigma-Aldrich). After washing with PBS, cells were incubated for 10 min at RT in PBS containing 0.2% Triton X-100 (# 93418; Fluka) for permeabilization. Subsequently, cells were washed again and treated with 10% goat serum in PBS for 10 min at RT. Depending on the primary antibody, cells were incubated for 2 h with the antibody diluted in 1% goat serum in PBS. After three washing steps with PBS for 5 min, cells were incubated for 1 h in a secondary antibody diluted in 1% goat serum in PBS. Cells were washed again three times for 5 min with PBS, whereas DAPI (# D9542-1MG; Sigma-Aldrich) was added at a final concentration of 1.25 μM during the second wash. Samples were mounted with FluorSave reagent (# 345789; Millipore). The antibodies used in this study are the following: anti-ERLIN2 (1:1,000, 2959; Cell Signaling), anti-TMUB1 (1:1,000, ab180586; Abcam), anti-GM130 (1:1,000, 610822; BD Biosciences), Alexa Fluor 488 anti-rabbit IgG (1:1,000, # A-11008; Thermo Fisher Scientific), and Alexa Fluor 594 anti-mouse IgG (1:1,000, # A-11012; Thermo Fisher Scientific). For LD staining, no permeabilization was performed and Bodipy 493/503 (D3922; Thermo Fisher Scientific) was applied just after fixation with PFA 4% for 30 min at a final concentration of 5 μM and washed five times for 10 min in PBS.

Transfection and live-cell imaging

Transfection of cells was performed using Lipofectamine 2000 (# 11668019; Invitrogen), 24 h after seeding. Before the transfection, the medium of the cells was changed to OptiMEM. For the transfection, 2 μl of Lipofectamine was diluted in 250 μl of OptiMEM (# 11058021; Gibco) and kept for 5 min at RT. Afterward, the Lipofectamine:OptiMEM mix was added to 250 μl of OptiMEM containing 1 μg of DNA. After 15 min, the DNA:Lipofectamine complex was added dropwise to the cells, and after 6-h incubation, the medium was exchanged by DMEM. Live imaging was performed 24 h after the transfection and at 37°C with 5% CO₂ to resemble the incubator conditions. The RUSH system was used to analyse ER-to-Golgi trafficking similarly as described before (Boncompain et al, 2012). Cells were transfected 24 h before imaging with Str-Ii_VSVG-SBP-EGFP, a gift from Franck Perez (plasmid # 65300; http://n2t.net/addgene:65300; RRID:Addgene_65300; Addgene). Live imaging was performed with a spinning disc confocal microscope (UltraVIEW VoX; PerkinElmer) for 45 min, and one frame was acquired every 30 s using Volocity software. Imaging was performed using a 60x oil objective (NA 1.49), and acquisition of the video was initiated directly after the biotin was added to the medium.

Image analysis

Image analysis was performed using Fiji/ImageJ 2.9.0 (Schindelin et al, 2012). ER morphology was quantified as previously described (Hu et al, 2009). By visual inspection, cells were classified into three categories (normal, intermediate, or collapsed) according to the ER shape using RTN4 immunostaining. The RTN4 area was calculated using the same staining after a Gaussian blur (Sigma = 20) filter was applied. The area was later obtained by thresholding with fixed values for each biological replicate. For the quantification of the Golgi apparatus and LDs, after manual segmentation of the cells, the Golgi apparatus and LDs were automatically segmented using GM130 immunostaining and Bodipy staining, respectively. Segmentation was performed using a fixed threshold for all genotypes in each biological replicate. The objects derived by thresholding were counted per cell using the “Analyze particles” function available in Fiji. For VSVG-GFP/GM130 colocalization, after subtraction of the background calculated from an area outside the cells, the cells were manually segmented, and for each cell, both the total VSVG-GFP fluorescence intensity and the fluorescence intensity within a mask obtained from GM130 signal thresholding were calculated. The value used is the ratio between the fluorescence intensity inside the Golgi and the total fluorescence intensity for each cell. The scale bar is equal to 10 μm.

RNA isolation, RNA-seq, and qRT-PCR analysis

RNA of cells was isolated with TRIzol reagent (# 15596026; Invitrogen), according to the manufacturer’s protocol (Life Technologies). Reverse transcription was performed using SuperScript First-Strand Synthesis System (# 11904018; Thermo Fisher Scientific). Real-time qRT-PCR was performed in a thermocycler (Applied Biosystems) with SYBR Green Master Mix (# 4309155; Thermo Fisher Scientific) and respective primer pairs (ERLIN1 sense primer 5′-GAAAGCTCACTCCCCTCTAAG-3′ and antisense primer 5′-TGTTCCCACTTAACCCCTTG-3′; ERLIN2 sense primer 5′-ACGCTTCAAGAGGTCTACATTG-3′ and antisense primer 5′-ATTGCCTCTGGTATGTTGGG-3′; HMGCS sense primer 5′-CATTAGACCGCTGCTATTCT-3′ and antisense primer 5′-AGCCAAAATCATTCAAGGTA-3′). GAPDH was used as a reference gene in all analyses (GAPDH sense primer 5′-AATCCCATCACCATCTTCCA-3′ and antisense primer 5′-TGGACTCCACGACGTACTCA-3′), and the fold enrichment was calculated using the formula 2(−ΔΔCt). For RNA-seq analysis, the RNA was submitted to the Cologne Center for Genomics. Libraries were prepared using the Illumina stranded TruSeq RNA sample preparation kit. ERCC RNA Spike-In Mix (Thermo Fisher Scientific) was added to the samples before library preparation. Library preparation started with 1 μg total RNA. After poly-A selection (using poly-T oligo-attached magnetic beads), mRNA was purified and fragmented using divalent cations under elevated temperature. The RNA fragments underwent reverse transcription using random primers. This was followed by second-strand cDNA synthesis with DNA polymerase I and RNase H. After end repair and A-tailing, indexing adapters were ligated. The products were then purified and amplified (15 PCR cycles) to create the final cDNA libraries. After library validation and quantification (Agilent TapeStation), equimolar amounts of the library were pooled. The pool was quantified using the Peqlab KAPA Library Quantification Kit and Applied Biosystems 7900HT Sequence Detection System. The pool was sequenced on an Illumina NovaSeq 6000 sequencing instrument with a PE100 protocol.

Analysis of RNA-seq

RNA-seq was performed with a directional protocol. Quality control, trimming, and alignment were performed using the nf-core (Ewels et al, 2020) RNA-seq pipeline (v3.0) (10.5281/ZENODO.1400710). Details of the software and dependencies for this pipeline can be found at the referenced DOI for the pipeline and at https://github.com/nf-core/rnaseq/blob/master/CITATIONS.md. The reference genome sequence and transcript annotation used were Homo sapiens genome GRCh38 from Ensembl version 103. Differential expression analysis was performed in R version 4.1.2 (2021-11-01) (R core Team, 2021) with DESeq2 v1.34.0 (Love et al, 2014) to make pairwise comparisons between groups. Log fold change shrinkage estimation was performed with ashr (Stephens, 2016). Only genes with a minimum coverage of 10 reads in all samples from each pairwise comparison were considered as candidates to be differentially expressed. Genes were considered differentially expressed if they showed a |log₂(fold change)| > 1 and were below an FDR of 0.05. Genes with a minimum coverage of 10 reads in all samples from each pairwise comparison were included in functional enrichment analyses and considered as the “gene universe” for over-representation–based analyses. Functional enrichment analysis was performed with clusterProfiler v4.2.0 (Yu et al, 2012).

Protein extraction and immunoblotting

Cells were harvested by scraping, and the pellet was lysed in an equal volume of RIPA buffer (50 mM Tris–HCl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.25% sodium deoxycholate, 1 mM sodium orthovanadate) containing 1X Protease Inhibitor Cocktail (# P2714-1BTL; Sigma-Aldrich) for 30 min on ice. Samples were centrifuged for 30 min at 20,000g at 4°C, and protein concentrations of the supernatants were measured with a standard Bradford assay (Bio-Rad Laboratories). 25 or 50 μg of proteins was diluted in Laemmli buffer 3X (188 mM Tris–HCl, 6% SDS, 12% glycerol, 0.05% bromophenol blue, 10% β-mercaptoethanol) and then analysed by SDS–PAGE followed by immunoblot analysis. Proteins were blotted on a polyvinylidene difluoride (PVDF) membrane (#10600023; Amersham) of pore size 0.45 μm using wet chambers and blotting buffer (25 mM Tris, 192 mM glycine, 20% methanol) at 4°C for either 1.5 h at 300 mA or for 16 h at 30 V. After blotting, proteins were stained for 10 min in Ponceau S. Membranes were blocked for 30 min in 5% skim milk (T145.3; Carl Roth) in TBST (20 mM Tris–HCl, pH 7.4, 150 mM NaCl, 0.1% Tween-20) at RT. Primary antibodies were diluted in the same blocking solution and incubated overnight at 4°C or for 2 h at RT. Respective HRP-linked secondary antibodies were used at 1:10,000 dilution and incubated for 1 h at RT. After incubation with both primary and secondary antibodies, membranes were washed three times in TBST. HRP was detected using ECL Western blotting detection reagents (# RPN2106; VWR). Chemiluminescence was detected using X-ray films (# 4005194; Bema) or the Vilber Fusion Solo S. Intensities of bands of interest were quantified using ImageJ. The antibodies used in this study are the following: anti-ERLIN1 (1:1,000, HPA011252; Sigma-Aldrich), anti-ERLIN2 (1:1,000, 2959; Cell Signaling), anti-TMUB1 (1:1,000, ab180586; Abcam), anti-TMUB2 (1:1,000, 28044-1-AP; Proteintech), anti-GAPDH (1:2,000, #MAB374; Millipore), anti-flotillin-1 (1:1,000, #610821; BD Transduction Laboratories), anti-KIDINS220 (1:1,000, PA5-22116; Thermo Fisher Scientific), anti-RTN4 (1:1,000, NB100-56681; Novus Biologicals), anti-VDAC2 (1:1,000, 9412S; Cell Signaling), anti-TMEM259 (1:1,000, HPA042669; Atlas Antibodies), anti-PMP70 (1:1,000, ab85550; Abcam), anti-FLAG (1:5,000, # F7425; Sigma-Aldrich), HRP-linked anti-mouse IgG (1:10,000, #A9044; Sigma-Aldrich), HRP-linked anti-rabbit IgG (1:10,000, #A0545; Sigma-Aldrich).

Immunoprecipitation

1.5 × 10⁷ cells were seeded on a 15-cm dish 1 d before the lysis. The next day, cells were washed once with PBS at RT, harvested with a cell scraper, pelleted at 60g for 3 min, and resuspended in 400 μl IP buffer (Tris–HCl, pH 7.4, 50 mM, KCl 50 mM, Triton X-100 0.5%) containing 1X Protease Inhibitor Cocktail (# P2714-1BTL; Sigma-Aldrich). Then, cells were passed six times through a 29 Gauge needle, incubated for 30 min on ice, and centrifuged at 20,000g for 30 min at 4°C. Protein concentration was determined via the Bradford assay, 500 μg of proteins was transferred in a new microcentrifuge tube, and the volume was adjusted to 250 μl using IP buffer. Next, 0.5 μg of the respective antibody was added to the proteins and incubated overnight at 4°C on a rotor. An anti-RFP antibody (600-401-379S; Rockland) was used as a control antibody for the TMUB1 IP. Afterward, 20 μl of Dynabeads Protein G (# 10003D; Thermo Fisher Scientific) was equilibrated three times with 1 ml of ice-cold IP buffer using a magnetic rack, added to the protein:antibody lysate, and incubated for 3 h at 4°C. After incubation, the magnetic rack was used to separate the beads from the flow-through and beads were washed three times with 1 ml of IP buffer. Depending on the followed application, elution was performed either in 30 μl of Laemmli buffer 3X, for Western blot analysis, or in SP3 buffer (5% SDS in 1X PBS) for proteomics analysis. After the addition of the respective elution buffer, beads were vortexed for 1 min and incubated at 95°C for 5 min. The eluate was separated from the beads using the magnetic rack.

DRM isolation

DRMs were isolated according to the protocol from George and colleagues (George et al, 2010) with the following changes. Lipid–protein crosslinking step was not performed, and the Triton X-100 concentration of the raft isolation buffer contained 5% Triton X-100, 150 mM NaCl, 5 mM EDTA, 50 mM Tris–HCl, pH 7.4. Per sample, 1.5 × 10⁷ cells were seeded on 15-cm plates, after 2 d washed twice with cold PBS + 5 mM EDTA, harvested using a cell scraper, and centrifuged for 10 min at 11,200g and at 4°C. Cell pellets were kept overnight at −80°C. Samples were washed once with TBS and lysed in raft isolation buffer. After lysis, cell lysates were centrifuged for 15 min at 112g, 4°C, and protein concentrations of the supernatants were determined via the Bradford assay. 1 mg of proteins was diluted in 500 μl raft isolation buffer and subsequently mixed with 1 ml 60% OptiPrep (# 1114542; Axis-Shield) solution to obtain a final concentration of 40% iodixanol. Homogenates were transferred into ultra-clear tubes (# 344062; Beckman and Coulter) and overlaid with a step gradient of 30% iodixanol solution and 5% iodixanol solution to a final volume of 4 ml. The step gradients were centrifuged for 5 h at 132,000g at 4°C (SW60# 335649; Beckman and Coulter). Subsequently, five fractions were obtained by taking 800 μl from the top to bottom.

Proteomics analysis of IPs, post-nuclear lysate, and SILAC-labelled DRMs

Lipidomics

AlphaFold-Multimer prediction of the ERLIN1, ERLIN2, TMUB1, and RNF170 complex

Multimer predictions of the ERLIN1-ERLIN2-TMUB1-RNF170 complex were performed using Cologne High Efficient Operating Platform for Science (CHEOPS) provided by the Regional Computing Center of Cologne (RRZK) using 4 CPU cores of a node with 2x Intel(R) Xeon(R) Gold 6248 CPU @ 2.50 GHz, 60 GB RAM, and Nvidia Tesla V100-SXM2 GPU with 32 GB VRAM running the AlphaFold module 2.2.0 (Evans et al, 2021 Preprint; Jumper et al, 2021), with 5 seeds per model generating in total 25 models. Peptide sequences of ERLIN1 (sp|O75477|ERLN1_HUMAN Erlin-1), ERLIN2 (sp|O94905|ERLN2_HUMAN Erlin-2), TMUB1 (sp|Q9BVT8|TMUB1_HUMAN transmembrane and ubiquitin-like domain–containing protein 1), RNF170 (sp|Q96K19|RN170_HUMAN E3 ubiquitin–protein ligase RNF170), VCP (sp| P55072|TERA_HUMAN Transitional endoplasmic reticulum ATPase), and FAF2 (sp|Q96CS3|FAF2_HUMAN FAS-associated factor 2) were extracted from Uniprot.org. The top-ranked model, with the highest confidence, of the Multimer prediction was used for further analysis. AlphaFold models were visualized using the latest version of ChimeraX 1.5, and hydrogen bonds were predicted using the ChimeraX function “H-Bonds.”

Statistical analysis

A one-way ANOVA followed by post hoc Tukey’s test was used to compare multiple groups. The statistical analysis of the MS analysis is described in detail in the “Data processing” section of the proteomics section. The number of biological replicates, tests used, and P-values are specified in the figure legends.