Distinct Regions within the Erlins Are Required for Oligomerization and Association with High Molecular Weight Complexes*

The group of stomatin/prohibitin/flotillin/HflK/C (SPFH) domain-containing proteins comprise members of diverse subcellular localization and function. Association with detergent-resistant membranes (DRMs) and the propensity to form oligomers are two common properties of SPFH domain proteins and likely important for the function of these proteins. Our laboratory recently discovered two novel members of this protein group, which, based on their endoplasmic reticulum (ER) localization and association with DRMs, were named ER lipid raft-associated protein (erlin)-1 and -2. Here we characterized erlin oligomerization and identified domains within the erlins responsible for oligomerization and DRM association. Using co-immunoprecipitation and sucrose density gradient centrifugation approaches on endogenous and ectopically expressed erlin proteins, we found that they formed homo- and hetero-oligomers and were part of large multimeric complexes. These properties were independent of their DRM association. By analyzing truncation and point mutants of erlin-2 we discovered that interaction between erlin monomers (oligomerization) and association with high molecular weight complexes require distinct regions within the protein. Although oligomerization and DRM association were mediated by a region immediately downstream of the SPFH domain (residues 228–300), integration into high molecular weight complexes was absolutely dependent on a phenylalanine residue C-terminal of this region (Phe-305), which lies within a short stretch of hydrophobic residues. Our data demonstrate that lower order oligomerization and incorporation into multimeric complexes are two separate biochemical properties of the erlins, because they are mediated by distinct regions.

Proteins containing the stomatin/prohibitin/flotillin/HflK/C (SPFH) 4 domain comprise a large protein group with members in many different species from prokaryotes to eukaryotes (1). Mammalian members include the flotillins, the prohibitins (PHBs), podocin, stomatin, and stomatin-like proteins (2,3). The SPFH domain, also known as PHB or HflK/C domain, was originally considered to be evolutionarily conserved, but it was recently proposed that it emerged independently in different proteins by convergent evolution (4). The role of this domain mostly remains elusive, although the SPFH domain of podocin was found to be involved in cholesterol binding (5). Members of the SPFH domain-containing proteins are attached to membranes of different subcellular compartments, including mitochondria (prohibitin), trans-Golgi network (flotillins), endosomes, and the plasma membrane (flotillin and stomatin) (2,3). In conjunction with this diverse targeting these proteins carry out diverse functions, like regulation of membrane channels, endocytosis, and coupling of plasma membrane signaling complexes to the cytoskeleton (2). All mammalian SPFH domaincontaining proteins associate with detergent-resistant membranes (DRMs), and thus appear to occupy lipid raft-like domains of different cellular membranes (2,3).
Using an antibody screen against isolated lipid raft proteins from human myelomonocytic cells, our laboratory recently identified two novel members of this protein group named erlin (endoplasmic reticulum lipid raft protein)-1 and -2 (previously known as KE04p/SPFH1 and C8orf2/SPFH2, respectively) (6). The initial study characterized them as components of lipid raft-like domains in the endoplasmic reticulum (ER) membrane (6). Erlin-2 was independently identified as a component of endothelial caveolae (7) and as an interacting partner of activated inositol 1,4,5-trisphosphate receptors (7,8). The function of the erlins is still relatively unexplored, but the latter study suggests that erlin-2 acts as a substrate recognition factor during ER-associated protein degradation of activated inositol 1,4,5-trisphosphate receptors and other substrates (8).
Both erlins are closely related as they share ϳ80% identity at the amino acid level, whereas the extreme N and C termini show the lowest conservation (2,9,10). The erlins are evolu-tionarily conserved with homologous proteins being found in Caenorhabditis elegans and Arabidopsis thaliana (8 -10). The extreme N terminus of erlin-1 and -2 contains a transmembrane domain, which anchors and targets these proteins to the ER membrane (6). Erlin-2 was shown to be a type II ER membrane protein, with only the first three N-terminal amino acids on the cytosolic side of the ER membrane, and most of the protein, including central and C-terminal regions, being found in the ER lumen (8). Erlin-1 likely has the same membrane topology, due to its high similarity to erlin-2.
In addition to DRM association, another common characteristic of SPFH domain-containing proteins is the propensity to form oligomers, although the size of these complexes varies somewhat between members. Stomatin and podocin both form large homo-oligomers of ϳ9 -12 and ϳ20 -50 monomers, respectively (5,11), whereas prohibitin-1 and -2 form multimeric complexes of ϳ1.2 MDa, which were shown to have a ringlike structure (12). Flotillin-1 and -2 on the other hand form homo-and heterotetramers (13), which likely assemble to larger clusters with a diameter of ϳ100 nm (14,15). Oligomerization of all SPFH domain-containing proteins investigated thus far depends on regions close to their C terminus. Domains of predicted coiled-coil secondary structure are required for oligomerization of the prohibitins and the flotillins (12,13). Although the oligomerization domains of most SPFH domaincontaining proteins have only been defined in relatively broad terms, the domain necessary for oligomerization and DRM association of stomatin was mapped to a 9-amino acid stretch downstream of the SPFH domain (16).
Because oligomerization and DRM association are common features of SPFH domain-containing proteins (2), they are likely essential for their function. Domains or motifs, which mediate these properties, may be functionally important. In the first part of this study, we characterized complex formation of endogenous and ectopically expressed erlin proteins. Consequently our second aim was to identify domains within the erlins mediating oligomerization as well as DRM association, because these two properties seem to be mediated by the same general regions in other SPFH domain-containing proteins. Here we report the identification of two distinct erlin regions, which mediated specific biochemical properties of these proteins. Although one region was found to be required for lower order oligomerization and DRM association, another one was responsible for incorporation into multimeric complexes.
Bioinformatics-Conserved domains in erlin-1 and -2 were determined as follows: The PHB domain (smart00244) was identified by the Simple Modular Architecture Research Tool. The SPFH domain corresponds to the Band_7/SPFH domain (pfam01145), and the HflK/C domain corresponds to the conserved domains Band_7/HflK (cd0304) and Band_7/HflC (cd0305), which were determined based on homologies to the bacterial membrane protease subunits HflK and HflC. All are available at the conserved domain data base at NCBI.
Plasmid Constructs-HA-and FLAG-tagged full-length (FL) erlin-1, FL erlin-2FLAG, and HA-tagged erlin-2 truncation mutants were generated by PCR using erlin-1 and -2 cDNAs described in a previous study (6) as templates. FL erlin-2HA has been described previously (6). All HA-tagged constructs were cloned into the pLPCX, and all FLAG-tagged constructs were cloned into the pLNCX2 retroviral vector (Clontech). In all cases constructs were inserted into vectors using XhoI and ClaI restriction sites. Erlin-2 (E2) F305A HA and E2 IPMFM3 A HA were generated from the erlin-2HA/pLPCX plasmid using the QuikChange site-directed mutagenesis kit (Stratagene) with minor modifications.
Cell Culture and Generation of Stable Cell Lines-HeLa, NIH-3T3, and Bosc23 cells were maintained in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin. NIH-3T3 cell lines stably expressing erlin constructs were generated by retroviral infection as described previously with some modifications (17). Briefly, retrovirus-containing supernatants were produced by transiently transfecting Bosc23 cells with retroviral vectors described above using Fugene6 (Roche Applied Science) according to the manufacturer's instructions. For infection, retroviral supernatant diluted 1:1 in growth media containing 8 g/ml Polybrene was added to cells. Cells infected with HAtagged erlin constructs in pLPCX were selected for positive clones using 2 g/ml puromycin. To generate NIH-3T3 cells co-expressing HA-tagged and FLAG-tagged erlin constructs, cell lines expressing erlin-HA constructs were re-infected with erlin-FLAG constructs in pLNCX2 and selected for positive clones by adding 2 mg/ml G418 (Invitrogen) to puromycin containing growth medium. Frozen pellets of two 150-mm tissue culture dishes of nearly confluent cells were used for immunoprecipitations, sucrose density gradient centrifugations, DRM isolations, and methyl-␤-cyclodextrin (M-␤-CD) treatment of membrane fractions. For harvesting, cells were washed briefly in phosphate-buffered saline (PBS), detached with a cell scraper and pelleted by centrifugation at 1000 ϫ g for 5 min. Cell pellets were stored at Ϫ80°C until use.
Immunoprecipitation-For immunoprecipitations frozen cell pellets were lysed in lysis buffer (1% Triton X-100, 1% sodium deoxycholate, 0.1% SDS, 10% glycerol, 150 mM NaCl, 10 mM Tris-HCl, pH 8.0) containing protease inhibitors (1 mM phenylmethylsulfonyl fluoride, 10 g/ml each of aprotinin and leupeptin) and homogenized with 20 strokes in a Dounce homogenizer. Lysates were cleared by centrifugation at 16,000 ϫ g for 10 min at 4°C, and 0.5-1 mg of protein of the cell lysate was used per immunoprecipitation reaction. After preclearing lysates by incubation with protein G-Sepharose for 45 min, samples were incubated with either ␣-HA, mouse ␣-FLAG, normal IgG, erlin antiserum, or pre-immune serum for 1 h, followed by protein G-Sepharose for an additional hour. Immunoprecipitates were washed in lysis buffer, resuspended in Laemmli sample buffer, and analyzed by Western blotting. Rabbit ␣-FLAG was used as primary antibody for all ␣-FLAG Western blots of immunoprecipitation samples. For all other Western blots, the identical antibodies used for immunoprecipitations were also employed as primary antibodies in combination with light chain-specific secondary antibodies.
siRNA Transfection-Predesigned siRNA duplexes specific for human erlin mRNAs were purchased from Qiagen. We obtained two erlin-1-specific siRNAs with catalogue numbers SI00731402 (Hs_SPFH1_2) and SI00731409 (Hs_SPFH1_3), which we herein refer to as siRNAs E1_1 and E1_2, respectively. Catalogue numbers of the two erlin-2-specific siRNAs purchased are SI00731437 (Hs_SPFH2_3) and SI03167234 (Hs_SPFH2_6), herein referred to as E2_1and E2_2. "AllStars negative control siRNA" (Qiagen) was used as the negative control. siRNA transfection of HeLa cells was performed using HiPerFect Transfection Reagent (Qiagen). Cells were transfected in 6-well plates (2.3 ϫ 10 6 cells/well) according to the manufacturer's protocol using 10 nM siRNA and 12 l of HiPer-Fect for each well. 72 h after transfection cells were lysed in radioimmune precipitation assay buffer, and erlin protein levels were analyzed by Western blot.
Sucrose Density Gradient Centrifugation-Sucrose gradient centrifugation was performed as described previously (18) with minor modifications. Briefly, frozen cell pellets were lysed in 1% Triton X-100 in MES-buffered saline (MBS, 25 mM MES, 150 mM NaCl, pH 6.5, and protease inhibitors) and homogenized with 25 strokes in Dounce homogenizer. Lysates were cleared by centrifugation as described above, and 1 ml of lysate was layered on top of an 11-ml 40 -5% continuous sucrose gradient (containing 1% Triton X-100). Gradients were spun at 170,000 ϫ g for 17 h at 4°C in an SW41 rotor (Beckman). Gel filtration molecular weight standards (Amersham Biosciences) were resuspended in 1% Triton/MBS and run in parallel. Twelve 1-ml fractions were collected from each gradient, and equal volumes of each fraction were mixed with Laemmli sample buffer. Protein samples were resolved by SDS-PAGE and analyzed by Coomassie staining using GelCode Blue Stain (Pierce) or Western blotting.
M-␤-CD Treatment-Treatment of crude membrane fractions with 50 mM M-␤-CD was performed as previously reported (6). Subsequently, DRM isolation, immunoprecipitation, and sucrose gradient centrifugation was performed with M-␤-CD-treated membranes and controls, which were treated with PBS only.
Isolation of DRM Fractions-DRM fractions were prepared as described previously (17). Briefly, frozen cell pellets were lysed in 1% Triton X-100/MBS with protease inhibitors, and lysate was adjusted to 40% sucrose in a volume of 2 ml. Lysate was overlaid with a 10-ml continuous 30 -5% sucrose gradient and spun at 170,000 ϫ g for 17 h in SW41 rotor. After centrifugation the top ϳ3.5 ml of the gradient was discarded, and the next 3.5-4 ml containing the opaque band of detergent-resistant material was diluted with 8 ml of MBS and subjected to another ultracentrifuge spin of 1 h to pellet the DRM fraction. The soluble fraction in the bottom 2 ml of the gradient was trichloroacetic acid-precipitated, both DRM and soluble fraction were resuspended in a 1:1 mixture of 2ϫ Laemmli sample buffer and 1 M Tris (pH 11), and equal parts of each fraction were analyzed by Western blotting. When DRM isolation was performed on M-␤-CD-treated membranes the gradient was fractionated into six 2-ml fractions and further subfractionated into insoluble and soluble material as described previously (6). Equal portions of insoluble material from all fractions (1-6) and soluble material from fraction 6 were analyzed on Western blots.
Microscopy-For immunofluorescent staining, NIH-3T3 cells were seeded onto acid-washed coverslips in Dulbecco's modified Eagle's medium containing 10% charcoal-stripped fetal bovine serum 1 day prior to fixation. Transient transfection with E2 N300 HA was performed on cells seeded onto coverslips using Fugene6 (Roche Applied Science) according to the manufacturer's instructions, and cells were fixed 24 h post transfection. Cells were fixed with 4% paraformaldehyde in PBS for 10 min at room temperature and permeabilized for 5 min with 0.2% Triton X-100 in PBS. Primary and secondary antibodies were diluted in 1% bovine serum albumin in PBS, and cells were incubated with each antibody for 1 h at room temperature. After fixation and each antibody incubation, cells were washed trice in PBS. Endogenous erlins in 3T3 pLPCX cells were stained with PAN-erlin antibody and donkey ␣-rabbit AlexaFluor 488. HA-tagged erlin-2 constructs were stained with 3F10 antibody, cells were co-stained with rabbit ␣-calnexin as ER marker, and goat ␣-rat AlexaFluor 488 and donkey ␣-rabbit AlexaFluor 568 were used as secondary antibodies (all fluorescently labeled secondary antibodies from Molecular Probes). Coverslips were mounted using Fluorescence Mounting Medium (Dako). Images were acquired as Z-stacks using an LSM 510 Meta confocal on an Axiovert 200M microscope with a 63ϫ/1.4 Plan Apochromat objective (all from Carl Zeiss) and are presented as projections of three focal planes, which were generated using LSM Image Browser (Carl Zeiss). For presentation purposes, levels, contrast, and brightness were adjusted across the entire image using Adobe Photoshop.

Erlin-1 and -2 Form Homo-and Hetero-oligomers-
We previously reported that, in addition to the 43-kDa band corresponding to the monomeric form of ectopically expressed HAtagged erlin-2, a ϳ80-kDa band, which might represent a dimeric version of the protein, can be detected in DRM fractions isolated from NIH-3T3 murine fibroblasts (6). Similar higher molecular weight bands were also found on erlin-1 Western blots (Data not shown). This finding and the fact that oligomerization appears to be a common feature of SPFH domain-containing proteins (2), led us to investigate whether the erlins form homo-and hetero-oligomers.
We used co-immunoprecipitation experiments to test whether erlin monomers could interact with each other, which is the prerequisite for oligomerization. To examine homo-oligomerization, we stably co-expressed HA-and FLAG-tagged versions of either human erlin-1 or -2 and performed immunoprecipitations with HA-and FLAG-specific antibodies (Fig.  1A). We found that erlin-1HA precipitated with erlin-1FLAG, and vice versa, and the same was observed for HA-and FLAGtagged erlin-2 (Fig. 1A). This showed that erlin-1 and -2 monomers have the ability to homo-oligomerize. Next, we examined interactions between erlin-1 and erlin-2 by immunoprecipitations on NIH-3T3 cells co-expressing HA-tagged erlin-1 and FLAG-tagged erlin-2, or vice versa, and found an interaction between ectopically expressed erlin-1 and erlin-2 (Fig. 1B). To confirm this finding in an endogenous setting, immunoprecipitations on endogenous erlin proteins were performed in HeLa cells, a human cervical carcinoma cell line (Fig. 1C). We used polyclonal antibodies specific for either erlin-1 or -2, which were raised against the extreme N terminus of erlin-1 or erlin-2, respectively, a region with very little conservation between erlin-1 and -2 (supplemental Fig. S1; see also Fig. 5A). These antibodies were specific for human erlin proteins, because they recognized endogenous protein in HeLa cells but not in vector control murine NIH-3T3 cells (supplemental Fig. S1). Consistent with our data on overexpressed proteins, we observed an interaction of endogenous erlin-1 and -2 (Fig. 1C). The interaction between erlin monomers appears to be very stable, because all immunoprecipitation experiments described here (Fig. 1, A-C) were carried out in a lysis buffer, containing substantial amounts of various detergents (see "Experimental Procedures"). Based on these data we concluded that erlin-1 and -2 form homo-and hetero-oligomers.
Protein Levels of Erlin-1 and -2 Are Independent from Each Other-Earlier studies on the prohibitins and the flotillins suggest that hetero-oligomerization is required for protein stability (2,12,13,19). PHB-1 and -2 are both dependent on each other for protein stability, because loss of one PHB protein leads to the loss of the partner protein (12,19). siRNA-mediated knockdown of flotillin-2 causes proteasomal degradation of flotillin-1, whereas protein levels of flotillin-2 are not dependent on its partner protein (13). We therefore investigated whether protein levels of erlin-1 and -2 are also dependent on each other. To this end, HeLa cells were transfected with siRNA duplexes against either erlin-1 or -2 (Fig. 1D). For targeting of each erlin mRNA, two siRNA sequences were used in separation, to control for off-target effects of siRNAs. Although siRNAs against Immunoprecipitations were performed with HA and FLAG tag specific antibodies and normal rat and mouse (ms) IgGs, respectively, were used as negative controls. Lysate: 10% of input. B, co-immunoprecipitation of HA-tagged erlin-1 with FLAG-tagged erlin-2 (left panel) or vice versa (right panel) indicates hetero-oligomerization of ectopically expressed erlin-1 and -2. Lysate: 10% of input. C, co-immunoprecipitation of endogenous erlin-1 and -2 was performed in HeLa cells using rabbit polyclonal antibodies against either erlin-1 (␣-E1) or erlin-2 (␣-E2). Pre-immune sera (PI) from corresponding rabbits were used as controls. Lysate: 15% of input. D, HeLa cells were transfected with siRNA duplexes against either erlin-1 (E1_1 and E1_2) or erlin-2 (E2_1 and E2_2), or with a control siRNA duplex. Knockdown of erlin proteins was assessed by Western blotting using a panel of erlin antibodies.
erlin-1 and -2 dramatically reduced protein levels of erlin-1 and -2, respectively, they did not have an obvious effect on levels of the partner protein (Fig. 1D). Lysates of siRNA-transfected HeLa cells were probed with a previously described erlin-1 antibody (7D3) (6), the newly generated polyclonal antibodies described above, which were specific for either erlin-1 or erlin-2, and a polyclonal antibody, which detected both erlin proteins (PAN-erlin antibody) (Fig. 1D). The PAN-erlin antibody was raised against the C-terminal half of erlin-1, which for the most part is conserved between erlin-1 and -2 (supplemental Fig. S1). The antibody detected a double band on Western blots, whereas the lower band corresponded to erlin-1 and the upper band to erlin-2, which was corroborated by the siRNA experiment ( Fig. 1D and supplemental Fig. S1).
In contrast to the erlins, siRNAmediated knockdown of PHB1 or PHB2 reduces protein levels of PHB1 and PHB2 by the same degree, whereas only reducing PHB1 or PHB2, respectively, at the mRNA level (19). A similarly dramatic effect on reduction of flotillin-1 levels after siRNA-mediated knockdown of flotillin-2 has also been reported (13). However, our data suggested that protein levels of erlin-1 and -2 are not dependent on each other.
The Erlins Associate with High Molecular Weight Complexes-Because our initial experiments showed that the erlins have the ability to form homo-and hetero-oligomers, we wanted to characterize these oligomeric complexes in more detail. Total lysates of HeLa cells or NIH-3T3 cells, expressing empty vector (pLPCX), erlin-1HA, or erlin-2HA, were subjected to sucrose density gradient centrifugation and gradient fractions were analyzed by SDS-PAGE and Western blotting (Fig. 2). Stained SDS-PAGE gels of gradient fractions demonstrated that this technique resulted in efficient and reproducible separation of cellular proteins, with the bulk of the proteins being concentrated in the low density fractions 2-6. Thus most proteins appeared to be part of low molecular weight protein complexes. HeLa and vector control NIH-3T3 cells showed a similar distribution of cellular proteins across sucrose gradient fractions (Fig. 2A). Molecular weight standards were also run on the gradient to confirm separation of size markers (Fig. 2B, bottom). Interestingly, both erlin-1 and -2 were enriched in high density fractions 7-12, with endogenous proteins in HeLa and NIH-3T3 cells and ectopically expressed HA-tagged proteins all showing similar distributions (Fig. 2B). Based on the migration pattern of molecular weight standards within the gradient, we estimated that most of the erlin protein was associated with high molecular weight complexes of Ն600 kDa. In NIH-3T3 cells stably expressing erlin-2HA, a smaller pool of ectopically expressed erlin-2HA was also found in lower density fractions 3-5, which might represent monomeric and dimeric versions of erlin-2. To FIGURE 2. Sucrose density gradient centrifugation shows association of erlins with high molecular weight complexes. Sucrose density gradient centrifugation was performed on total lysates from HeLa cells or 3T3 cells stably expressing pLPCX vector only, E1HA, or E2HA. Cells were lysed in 1% Triton X-100 and fractionated by 40 -5% sucrose gradient centrifugation. A, gels with gradient fractions from HeLa and 3T3 pLPCX cells were stained to visualize distribution of cellular proteins within the gradient. B, Western blot analyses of fractions: Distribution of endogenous erlin proteins within gradient was determined by probing Western blots with antibodies specific for either erlin-1 or erlin-2 (HeLa) or with PAN-erlin antibody, which detects both erlins (3T3). Ectopically expressed HA-tagged erlin-1 and -2 were detected with an HA-specific antibody. Western blots were also probed for calnexin, which localizes to the ER, but is not enriched in high density fractions. Distribution of molecular weight standards is indicated at the bottom. 440-and 669-kDa standards were not separated by this procedure and therefore concentrated in identical fractions.
confirm that enrichment in high density fractions is not just a general feature of ER membrane proteins, we also examined the distribution of the ER-membrane-associated chaperone calnexin across density gradient fractions. In contrast to the erlins, calnexin, an ϳ80-kDa protein, was almost exclusively found in low density fractions of both HeLa and NIH-3T3 cells (Fig. 2B), indicating that it primarily exists as a monomer, which is consistent with previous reports (11).

Interactions between Erlin Monomers and Association of Erlins with High Molecular Weight Complexes Are Not Dependent on Their Association with Lipid Rafts-
The erlins associate with lipid rafts, which was inferred by their presence in DRMs (6). Hence, we wanted to test whether the ability of erlin monomers to interact with each other and to associate with high molecular weight complexes requires lipid raft association. DRM association of erlin proteins can be disrupted by treating purified membrane fractions with the cholesterol sequestering agent M-␤-CD (6). Treatment of HeLa cell membranes with 50 mM M-␤-CD prevented erlin-1 and -2 from associating with floating, Triton X-100-insoluble fractions (Fig. 3A, insoluble  fractions 2 and 3), indicating that erlin lipid raft association was efficiently disrupted. A similar effect was observed with the commonly used lipid raft marker protein flotillin-1 (Fig. 3A). M-␤-CD-treated HeLa cell membranes were subjected to immunoprecipitation and sucrose gradient centrifugation, but M-␤-CD neither affected interaction between erlin-1 and -2, nor their association with high molecular weight complexes (Fig. 3, B and C). Thus, these biochemical properties of the erlins seemed to be independent from lipid raft association.

Interactions between Monomers, Incorporation into High Molecular Weight Complexes, and Lipid Raft Association of Erlin-2 Depend on Regions between the SPFH Domain and the
Non-conserved C Terminus-Oligomerization and DRM association of the erlins are properties shared with other SPFH domain-containing proteins and are likely required for their function. Consequently, domains or motifs within the erlins mediating these properties could be functionally important. The next objective of this study was therefore to identify erlin regions responsible for interactions between erlin monomers and association with high molecular weight complexes and DRMs. We focused on erlin-2 for this part of the study, because erlin-1 and -2 are very similar at the amino acid level (see Fig.  5A), and functional domains are likely conserved between the two proteins. Because oligomerization and DRM association domains of other SPFH domain-containing proteins are localized to regions close to their C terminus (2), we assumed a similar scenario for the erlins. For the initial characterization we generated three different C-terminal erlin-2 truncations (Fig. 4A). The shortest truncation mutant encompassed amino acids 1-188, therefore called E2 N188, and consisted of the N terminus with the transmembrane domain and the complete PHB domain (Fig. 4A). The PHB domain is essentially the same as the SPFH domain, but different domain recognition algorithms were used to define the domains, resulting in slightly different boundaries (2). The PHB domain is therefore 23 residues shorter than the SPFH domain (Fig. 4A). The next erlin-2 truncation mutant was named E2 N227 and was composed of the N terminus and the complete HflK/C domain, which again is another term to describe the SPFH domain, but it covers a slightly larger part of the protein. The longest truncation mutant generated, E2 N305, only lacked the extreme C terminus, which shows very little conservation between erlin-1 and -2 (Fig. 4A, also see Fig. 5A).
To examine whether these erlin-2 truncation mutants are able to interact with FL proteins, we stably co-expressed HAtagged versions of the mutants together with a FLAG-tagged version of FL erlin-2 in NIH-3T3 cells and carried out HA-and FLAG-specific immunoprecipitations with these cell lines (Fig.  4B). Although E2 N305HA, like FL erlin-2HA, precipitated with FL erlin-2FLAG, this was not the case for E2 N188 and N227 (Fig. 4B). This showed that E2 N188 and N227 lacked the domain required for interaction between erlin monomers, which is on the other hand included in the N305 mutant. Next we performed sucrose gradient centrifugation on NIH-3T3 cells stably expressing HA-tagged erlin-2 truncations alone, to test their association with high molecular weight complexes (Fig. 4C). Consistent with co-immunoprecipitation results, only E2 N305 exhibited the same distribution across gradient fractions as FL erlin-2, with most of the protein being enriched in high density fractions 7-12, and a smaller pool in low density , and sucrose gradient centrifugation (C). A, pretreated membrane fractions were lysed in 1% Triton X-100, lysate was adjusted to 40% sucrose and overlaid with a 30 -5% continuous sucrose gradient. After ultracentrifugation six fractions were collected from gradient, which were further separated into soluble and insoluble subfractions. Equal parts of insoluble fractions 1-6 and soluble protein from fraction 6 were analyzed by Western blotting using the PAN-erlin antibody and a flotillin-1-specific antibody as lipid raft marker. B, immunoprecipitation of endogenous erlins was performed as in Fig. 1D. Lysate: 20% of input. C, sucrose gradient centrifugation was performed as in Fig. 2, and Western blot was probed with PAN-erlin antibody.
fractions 2-5 (Fig. 4C, also compare with Fig. 2B). In contrast, E2 N188 and N227 accumulated in low density fractions, demonstrating that they do not contain the erlin-2 region necessary for incorporation into high molecular weight complexes (Fig.  4C). Lastly DRM association of these erlin-2 truncations was examined. To asses DRM association equal parts of soluble and DRM (raft) fractions isolated from NIH-3T3 cells expressing HA-tagged erlin-2 truncations were analyzed by Western blot-ting. Although the FL and the E2 N305 constructs were found in both soluble and raft fraction, E2 N188 and N227 were detected in the soluble fraction only (Fig. 4D). Taken together, these data showed that the region of erlin-2 located downstream of the SPFH domain and upstream of the non-conserved C terminus (residues 228 -305) was required for interaction between monomers, high molecular weight complex formation, and lipid raft association of erlin-2.

. A C-terminal region of Erlin-2 between the SPFH domain and the non-conserved C terminus is required for its oligomerization and association with high molecular weight complexes and DRMs.
A, schematic of full-length and truncated erlin-2 constructs used for initial characterization in this study: Transmembrane (TM) domain is shown in gray, prohibitin (PHB) domain is shown in red, red and orange boxes combined represent the SPFH domain, and red, orange, and yellow boxes combined represent HflK/C domain. The blue box indicates a C-terminal region, which shows very little conservation between erlin-1 and -2. HA and FLAG tags (green) were added to the C terminus of the erlins. Erlin-2 truncations lack various parts of the C terminus and were named N188, N227, and N305, because they consist of the N-terminal 188, 227, and 305 amino acids, respectively. B, full-length (FL) FLAG-tagged erlin-2 and HA-tagged truncated (or FL) erlin-2 were stably co-expressed in NIH-3T3 cells, and immunoprecipitations with HA-and FLAG-specific antibodies were performed. Western blots on the left show total lysate, the middle and right panels show immunoprecipitations HA-and FLAG-specific antibodies, respectively. Lanes in the middle and right panels marked with "ϩ" show IPs with HA-or FLAG-specific antibodies, whereas lanes marked with "-" show control IPs with normal rat or mouse IgGs, respectively. Lysate: 20% of input. C, sucrose gradient centrifugation was performed on 3T3 cells stably expressing truncated or FL erlin-2HA, as described for Fig. 3. D, lipid raft isolation was performed on 3T3 cells stably expressing truncated or FL erlin-2HA. Equal portions of soluble (s) and raft (r) fractions were analyzed by Western blotting to determine lipid raft association of erlin-2 truncation mutants. Flotillin-1 was used as a marker for lipid raft fractions.
To ensure that the phenotypes of erlin-2 truncations mutants were not just caused by incorrect subcellular targeting, we confirmed their ER localization by immunofluorescence microscopy (supplemental Fig. S2). We did not expect C-terminal mutations to interfere with ER localization, because the N-terminal transmembrane domain has been shown to be suf- FIGURE 5. Residue Phe-305 within a short hydrophobic stretch is dispensable for oligomerization and lipid raft association but is required for high molecular weight complex association of erlin-2. A, ClustalW alignment of amino acid sequences of erlin-1 and -2 (modified from reference 6); the small gray box highlights residues 301-310, which are magnified in the large gray box. Hydrophobic residues between 301 and 306 are marked as bold letters, and underlined residues are included in erlin-2 N305 truncation mutant. The black box contains a list of additional erlin-2 mutants. B, testing new erlin-2HA mutants for oligomerization using the immunoprecipitation approach described for Fig. 4B. On ␣-HA, a Western blot shows total lysate (top left), and 10ϫ the amount of protein was loaded onto the gel for IPMFM3 A compared with N300 and F305A. Longer exposure times are shown for IP blots of IPMFM3 A mutant, due to low expression levels of this mutant in that specific cell line. Lysate: 10% of input. C, testing lipid raft association of new erlin-2HA mutants using same approach as in Fig. 4D. D, association of new erlin-2HA mutants was examined using sucrose gradient centrifugation, as described for Fig. 4C. Western blot of FL E2HA gradient fractions was included for comparison. ficient for correct ER targeting (6). Endogenous erlins in vector control NIH-3T3 cells (supplemental Fig. S2A) and ectopically expressed erlin-2HA (supplemental Fig. S2B, first row) revealed a reticular cytoplasmic and perinuclear staining pattern, indicative of ER localization, which is consistent with previous reports (6 -8). The erlin-2 truncation mutants also showed ER staining, which greatly overlapped with staining for the ER protein calnexin (supplemental Fig. S2B).
Erlin-2 Residue Phe-305 Is Essential for Association with High Molecular Weight Complexes, but Not for Interaction between Erlin Monomers or DRM Association-Our next goal was to further define the erlin-2 region essential for monomer interaction and association with high molecular weight complexes and DRMs. A recently published study on stomatin identified a short, largely hydrophobic stretch of 9 amino acids close to the C terminus, which is necessary for oligomerization as well as DRM association of stomatin. This stretch contains a phenylalanine, which is absolutely crucial for both oligomerization and DRM association of stomatin (16). E2 N305 was the only truncation mutant showing oligomerization and DRM and multimeric complex association. Interestingly, residue 305 is a phenylalanine, and three out of four residues upstream and one residue downstream of Phe-305 are hydrophobic amino acids between, which was conserved in both erlin proteins (Fig. 5A). We speculated that the short hydrophobic stretch around erlin residue Phe-305 was the equivalent to the hydrophobic DRM association and oligomerization domain of stomatin (16). To test this, three additional erlin-2 mutants were generated. We made another erlin-2 truncation mutant E2 N300, which lacks the complete hydrophobic region; a mutant in which all hydrophobic residues between positions 301 and 306 were replaced by alanines, called E2 IPMFM3 A; and an alanine substitution mutant of residue Phe-305 only, named E2 F305A (Fig. 5A).
However, when these new erlin-2 mutants were tested in co-immunoprecipitation and DRM association experiments described above, no impairment of their ability to interact with FL erlin-2 or to associate with DRMs could be detected (Fig. 5, B and C). Based on these results it appeared that the hydrophobic stretch between residues 301 and 306 was not involved in DRM association and oligomerization. Finally, we examined high molecular weight complex formation of these erlin-2 mutants by sucrose gradient centrifugation. Strikingly, none of the mutant proteins were enriched in high density fractions as observed for the FL or N305 construct of erlin-2 (Fig. 5D, also compare with Fig. 4C). Instead, they accumulated in low density fractions, suggesting, that despite their ability to interact with FL erlin-2, they were unable to associate with high molecular weight complexes (Fig. 5D). E2 F305A exhibited the same phenotype as mutants lacking the entire hydrophobic region (E2 N300 and IPMFM3 A), demonstrating that mutation of a single residue, E2 F305, was sufficient to prevent erlin-2 from incorporating into high molecular weight complexes. The three new erlin-2 mutants were concentrated in the same low density fractions (fractions 2 to 5 and 6) as E2 N188 and N227 (Fig. 4, C  and D). Although the new mutants were still able to interact with FL erlin-2, and thus likely formed at least dimers, E2 N188 and N227 were unable to oligomerize and probably exist as monomers only. It therefore appears that erlin monomers and dimers could not be efficiently separated by sucrose gradient centrifugation due to the limited resolution of the technique. These data also demonstrated that the ability to interact with other erlin monomers and to associate with high molecular weight complexes are two separate properties of the erlins and are mediated by distinct regions within the protein.
The new erlin-2 mutants were appropriately targeted to the ER, as determined by immunofluorescence. However, distribution of the E2 N300 mutant differed from the rest of the mutants. Although it showed ER staining, large cytoplasmic aggregates and vesicle-like structures were also observed (supplemental Fig. S2, C, first row, and D). These structures were only detected when E2 N300 was expressed at high levels, whereas they were not present in cells expressing low levels of E2 N300 (supplemental Fig. S2D). These structures appeared to be derived from ER, because most of them co-stained for ER chaperone calnexin, and potentially represented aggregates of misfolded proteins (supplemental Fig. S2, C, first row, and D). Thus E2 N300 might be misfolded, because it is lacking C-terminal residues required for proper folding of this part of the protein. Five additional amino acids seemed to be sufficient for correct folding, because the E2 N305 mutant did not form these aggregate-like structures. The phenotype of E2 N300 was not caused by its inability to associate with high molecular weight complexes, because it was not observed in the E2 F305A and IPMFM3 A (supplemental Fig. S2C), which otherwise exhibited the same biochemical properties as E2 N300. Nevertheless, protein stability of E2 N300 did not seem to be severely compromised, because it could be stably expressed at high levels in NIH-3T3 cells, which also did not show any obvious defects in cell viability (data not shown).

DISCUSSION
The goal of the present study was to characterize oligomer formation of the erlins and to identify erlin regions that mediate oligomerization as well as DRM association. We showed here that erlin-1 and -2 have the ability to interact with themselves and with each other, which suggests they form homo-and hetero-oligomers. Furthermore, we demonstrated that the erlins are part of high molecular weight complexes. Whereas interaction between erlin monomers, here referred to as oligomerization, and DRM association required a region of erlin-2, which lies downstream of the SPFH domain and upstream of the nonconserved C terminus (residues 228 -300), incorporation into multimeric complexes was absolutely dependent on residue Phe-305. Oligomerization and high molecular weight complex association therefore appear to be two distinct properties of the erlins, because they are mediated by different regions.
Residue Phe-305 of erlin-2 lies within a short stretch of hydrophobic amino acids, and replacing it by an alanine was sufficient to prevent erlin-2 from associating with high molecular weight complexes. Mutation of this residue might prevent incorporation into multimeric complexes by reducing the hydrophobicity of this stretch or altering its secondary structure, or both. Because domains involved in protein-protein interactions are often hydrophobic (20), it is reasonable to assume that Phe-305 promotes erlin multimerization due to its hydrophobic character and that the surrounding hydrophobic amino acids are also involved in this process. Similarly, oligomerization of stomatin, another SPFH domain-containing protein, depends on a short hydrophobic stretch close to its C terminus (16). However, the hydrophobic residues at positions 301-306 in erlin-2 are not entirely equivalent to the hydrophobic domain in stomatin. The nine-residue hydrophobic domain of stomatin is not only required for oligomerization, but also for DRM association (16), whereas erlin DRM association and lower order oligomerization are mediated by a separate region of erlin-2. Because 81% of residues between positions 228 and 300 are identical between erlin-1 and -2, and the hydrophobic stretch between residues 301 and 306 is also conserved in erlin-1, these regions are likely also required for DRM association, oligomerization, and high molecular weight complex association of erlin-1.
The function of the SPFH domain still remains elusive, but it has been previously proposed to facilitate oligomerization and DRM association, because these are characteristics shared by most SPFH domain-containing proteins (3,21). Our study showed that the erlin SPFH domain alone is not sufficient to induce these biochemical properties. Likewise, oligomerization of the flotillins and stomatin requires domains outside of their SPFH domain, which are unique for the each SPFH domain protein (13,16,22). Nevertheless the SPFH domain might contribute to DRM association by binding to cholesterol, which has been demonstrated for the SPFH domain of podocin and its C. elegans homologue MEC-2 (5), and potentially other membrane lipids.
Based on our results, one could envision the formation of multimeric erlin complexes as a two-step process. The first step would be the formation of lower order oligomers, potentially dimers, which is mediated by the region between erlin residues 228 and 300. Bands of ϳ80 kDa observed on erlin Western blots likely represent erlin dimers (6), indicating that dimers are relatively stable intermediates in the process of higher molecular weight complex formation. Oligomeric complexes of only one or both types of erlins are conceivable, because we have found homo-and heterotypic interactions between erlin-1 and -2 and erlin protein levels did not seem to be dependent on the partner protein. This is in contrast to PHB-1 and -2, which both depend on their partner for protein stability and can therefore only form hetero-oligomers (12,19). Erlin-2 mutants, which contain residues 1-300, but not Phe-305 (E2 N300, F305A, and IPMFM3 A), would only undergo the first step of multimeric complex formation. In a second step, lower order erlin oligomers would assemble into high molecular mass complexes of Ն600 kDa. These complexes might either consist of erlin proteins only, or include other ER membrane proteins. Such a twostep process has been demonstrated for the formation of ringshaped multimeric complexes by yeast PHBs in the inner mitochondrial membrane (12). Upon import into the mitochondria, PHB-1 and -2 form hetero-oligomeric intermediates of ϳ120 kDa, which then aggregate into ring-like structures of 1.2 MDa. Similar to the erlins, each step requires distinct regions within the PHBs (12). Hence erlin multimerization at the ER membrane might proceed in a similar fashion, especially because the erlins have a membrane topology similar to PHB-1, and, of all SPFH domain-containing proteins, the prohibitins and the erlins are most similar to each other (6).
In this study we also explored the relationship between DRM association of the erlins and their oligo-and multimerization. We found that disrupting erlin DRM association by removing cholesterol from cellular membranes after hypotonic lysis affected neither the interactions between erlin-1 and -2 nor their association with high molecular weight complexes. DRM association therefore does not seem to be required for stabilization of existing erlin oligomers and high molecular weight complexes. Due to their intracellular localization to ER membranes, DRM association of the erlins cannot be disrupted in live cells (6). This prevented us from examining whether DRM association is needed for initial assembly of erlin oligomers and multimers in live cells. However, multimerization of podocin and its C. elegans homologue MEC-2 is independent of cholesterol binding (5), suggesting that DRM association is generally not required for multimerization of SPFH domain-containing proteins. This is in contrast to multimers of the caveolins, which require cholesterol for their stability (23).
We observed DRM association of erlin-2 mutants, which were unable to associate with high molecular weight complexes, indicating that multimerization is not a prerequisite for lipid raft association of the erlins. As has been shown for podocin and MEC-2 (5), the erlins might be able to bind cholesterol and potentially other ER membrane lipids and thereby nucleate specific lipid molecules around them. Binding of specific lipids could be sufficient to induce detergent insolubility in the erlins. Erlin monomers and lower order oligomers, each bound to a certain number of cholesterol/lipid molecules, might assemble into multimeric complexes and thus form lipid raft-like microdomains in the ER membrane. According to this model, the erlins would provide a protein scaffold for the formation of specialized microdomains in the ER membrane, where they create a lipid microenvironment distinct from the remainder of the ER membrane. The flotillins have been proposed to form such "scaffolding microdomains" at the plasma membrane, which could provide platforms for flotillin-dependent processes, which include certain plasma membrane receptor signaling pathways (21) and a novel type of endocytosis (24). Similarly, "erlin microdomains" in the ER membrane might facilitate certain ER-associated processes by clustering proteins involved. ER-associated protein degradation of several protein substrates has been shown to require erlin-2 (8) and could therefore take place, at least partially, within the proposed "erlin microdomains," but other ER-associated processes might also be facilitated by the erlins.
Our data suggest that protein levels of erlin-1 and -2 are not dependent on each other, and it is therefore physically possible for each erlin protein to function in isolation from its partner. The strong conservation between the two proteins could hint toward redundant functions. However, siRNA-mediated knockdown of erlin-2 only is sufficient to inhibit ER-associated protein degradation of activated inositol 1,4,5-trisphosphate receptors and other substrates, which shows that erlin-1 cannot functionally substitute for erlin-2 (8). Instead these data suggest two other possibilities: 1) hetero-oligomers of both erlins might be required for their function or 2) the function of erlin-1 is at least partially different from