Tissue-specific Expression and Dimerization of the Endoplasmic Reticulum Oxidoreductase Ero1β*

Endoplasmic reticulum oxidoreductases (Eros) are essential for the formation of disulfide bonds. Understanding disulfide bond catalysis in mammals is important because of the involvement of protein misfolding in conditions such as diabetes, arthritis, cancer, and aging. Mammals express two related Ero proteins, Ero1α and Ero1β. Ero1β is incompletely characterized but is of physiological interest because it is induced by the unfolded protein response. Here, we show that Ero1β can form homodimers and mixed heterodimers with Ero1α, in addition to Ero-PDI dimers. Ero-Ero dimers require the Ero active site, occur in vivo, and can be modeled onto the Ero1p crystal structure. Our data indicate that the Ero1β protein is constitutively strongly expressed in the stomach and the pancreas, but in a cell-specific fashion. In the stomach, selective expression of Ero1β occurs in the enzyme-producing chief cells. In pancreatic islets, Ero1β expression is high, but is inversely correlated with PDI and PDIp levels, demonstrating that cell-specific differences exist in the regulation of oxidative protein folding in vivo.

Protein folding in the ER 5 attracts considerable interest because the failure of a protein to fold can lead to a host of genetic and acquired diseases (1), ranging from cystic fibrosis to ␣1 anti-trypsin deficiency (2). Professional secretory cells in particular must regulate the synthesis of their ER membranes and chaperones to cope with the demands of increased protein production. This is achieved through ER to nucleus signaling pathways, mediated by the trans-membrane associated proteins Ire1␣, PERK, and ATF6 (3). ATF6 and Ire1␣ induce the transcription of XBP1 and the splicing of its mRNA, culminating in the expression of UPR target genes (4). XBP1 is required for B cell maturation into antibody-producing plasma cells (5), and recently, XBP1 and chronic unfolded protein responses have been implicated in obesity and the onset of type 2 diabetes (6), suggesting that targeting physiological unfolded protein responses may have therapeutic value in this disease.
Disulfide bond formation is an essential component of the protein folding process, and disulfide bonds are required for structural stability, enzymatic function, and regulation of protein activity (7). The catalytic events involving the oxidation, reduction, and isomerization of disulfide bonds take place in the ER. During protein oxidation, PDI introduces native disulfide bonds into substrate proteins, and is reoxidized by the Ero proteins (Ero1p in yeast, Ero1␣ and Ero1␤ in humans) (8 -11). In yeast, Pdi1p is capable of both oxidizing and isomerizing disulfide bonds, although the relative importance of each function has been debated (12). In humans, PDI also contributes to collagen biosynthesis as a component of the prolyl-4-hydroxlase complex (13) and can act as a component of the ER degradation machinery, particularly with respect to the unfolding and retro-translocation of toxins (14). Numerous PDI homologues exist in yeast (Mpd1p, Mpd2p, Eps1p, and Eug1p) (15) and in humans (e.g. ERp57, ERp72, P5, PDIR, and PDIp) (16). The redundancy of these proteins has made their precise functions difficult to analyze, but ERp57 at least has a specialized bЈ domain that selectively allows this PDI homologue to interact with the lectin-like ER chaperones calnexin and calreticulin (17).
PDI is the only human protein with an intact WCGHC motif that can be trapped in association with Eros (18,19), although Ero1␣ can form complexes with the unconventional PDI family members ERp44 (20) and PDILT (21). In yeast, Mpd1p and Mpd2p both interact with Ero1p (22). The crystal structure of the Ero1p core revealed that the N-terminal CXXXXC motif of the protein is likely to transfer electrons to the latter two residues of the C-terminal CXXCXXC motif, in close proximity to the isoalloxazine ring of FAD (23). Although co-crystals of Ero1p and Pdi1p are not available, it has been demonstrated biochemically that Pdi1p/PDI binding is disulfide-dependent and requires the N-terminal CXXXXC motif of Ero1p/Ero1␣ (24,25). A supply of reduced glutathione is also required for appropriate oxidative protein folding in the ER (26,27).
In humans, Ero1␣ and Ero1␤ are up-regulated by hypoxia (28) and the UPR (11), respectively, suggesting that the two proteins have undergone functional specialization in response to the different demands of oxygen tension and high throughput protein folding. However, the relationship between Ero mRNA and protein levels, and the relationship between UPR induction of Ero1␤ by a chemical versus a physiological (nutritional) stimulus is not known. Cells in culture generally express low levels of Ero proteins. The assumption is that low expression of an oxidoreductase is enough to drive the oxidation of many substrate proteins by fuelling many PDI molecules. Here we show that this model is too simple and that the pathway of an oxygen electron acceptor-Ero-PDI-substrate chain is incomplete. Our experiments demonstrate that in stomach and pancreatic tissues, Ero1␤ protein expression is much higher than expected for an oxidoreductase catalyst and is restricted to groups of enzyme-producing cells, notably the chief cells in the stomach, and the insulin-and enzyme-producing cells of the pancreas. Constitutively high expression of Ero1␤ is differentially correlated to PDI expression in professional secretory cells in vivo. At a molecular level, both Ero1␣ and Ero1␤ do more than simply engage PDI. Ero1␤ can form alkylation-independent homodimers, which require Cys 396 of the CXXCXXC active site, and Ero1␤ complexes are more abundant in the pancreas than the stomach. Our biochemical analysis is supported by the Ero1p crystal structure, which can be modeled as a symmetrical dimer in which interfacial FAD-FAD contacts are prominent. Our results show that the ER oxidation machinery is more complex than anticipated in mammals, demonstrating how disulfide bond formation could be regulated in vivo through differential complex formation.

EXPERIMENTAL PROCEDURES
Cell Lines, Antibodies, and Tissues-Human cervical carcinoma HeLa cells were maintained in minimal essential medium (Invitrogen), and human fibrosarcoma HT1080 cells were maintained in Dulbecco's modified Eagle's medium (Invitrogen), both supplemented with 8% fetal calf serum (Sigma), 2 mM Glutamax, 100 units/ml penicillin, and 100 g/ml streptomycin (Invitrogen) at 5% CO 2 . The polyclonal antiserum against Ero1␤ was raised against the unique internal Ero1␤ peptide YTGNAEEDADTKTLL (Sigma Genosys). Thirteen of the fifteen residues were conserved between mouse and human Ero1␤. The polyclonal rabbit anti-sera against PDI, Ero1␣ (D5), and ERp57 have been described previously (18). The polyclonal PDIp antiserum was a gift from M. Lan (29). Polyclonal antisera against BiP and insulin (H-86) (both Santa Cruz Biotechnology), the mouse monoclonal antibody HA-7 (Sigma), and the anti-myc antibody 9B11 and the anti-myc polyclonal antibody (both Cell Signaling) were commercially available. Normal goat serum was obtained from DAKO, and normal rabbit serum was obtained from a pre-immune rabbit used to generate the PDI antibody. Murine tissue samples were obtained from male Balb/c or CD1 mice. Anonymous healthy human tissue samples (stomach and pancreas) were obtained embedded in wax from Medical Solutions, Nottingham, UK, with full ethical consent.
Transfections-Transfections with Lipofectamine 2000 (Invitrogen) were performed according to the manufacturer's instructions. Sub-confluent cells in 6-cm dishes were washed twice with phosphate-buffered saline (Invitrogen), and transfected with 1 or 2 g DNA for 6 h in the presence of Opti-Mem serum-free medium (Invitrogen). The medium was replaced after 6 h with complete medium, and the cells were analyzed 24 h post-transfection.
Gel Filtration-Gel filtration was performed using a Superdex 200 Precision Column 3.2/30 (Amersham Biosciences/GE Healthcare). Chicken egg albumin (45 kDa) and bovine serum albumin (66 kDa and 132 kDa) (both Sigma) were used as standards. Proteins and mouse tissue lysates injected in a volume of 140 l were eluted with 50 mM Tris-HCl, 150 mM NaCl, pH 7.0, at a flow rate of 0.5 ml min Ϫ1 , with samples collected every 2 min. Trichloroacetic acid-precipitated fractions were neutralized with Tris and taken up in sample buffer, and equal volumes were analyzed by reducing SDS-PAGE.
Immunohistochemistry-Human tissues were sectioned at 4-m thickness onto Apes coated slides and placed at 60°C for 1 h. The slides were cleared through xylene and 100 -70% alcohol before blocking endogenous peroxidase activity in methanol peroxidase for 10 min. The slides were washed in cold water, and the optimal digestion technique for each tissue type was determined: for stomach tissue, microwave digestion was in 600 ml of citrate buffer (pH 6.0) for 20 min; for pancreatic tissue, pressure cooker digestion was in 3 liters of citrate buffer (pH 6.0) for 2 min. The slides were washed in tap water and then in TBS (pH 7.4) for 5 min. Tissue sections were blocked in 1% normal goat serum (DAKO) for 10 min before incubating with the optimum dilution of primary antibody in TBS (PDI and PDIp, 1:1000; Ero1␤, 1:150; BiP 1:40; and ERp57 and H-86, 1:200) at 4°C overnight. The sections were washed for 5 min in TBS, and 0.01% biotinylated goat anti-mouse/rabbit antibodies (DAKO) diluted in TBS was added for 30 min. The sections were developed using streptavidin AB complex horseradish peroxidase (DAKO) and 3,3-diaminobenzidine. Counterstaining was achieved with hematoxylin.
Western Blotting-Cells or fresh mouse tissues were lysed in lysis buffer (20 mM MES, 30 mM Tris, 100 mM NaCl, pH 7.4) with 1% Triton X-100, supplemented with 10 g/ml each of chymostatin, leupeptin, antipain, and pepstatin A and 20 mM NEM where required. Nuclei were removed by centrifugation at 16,100 ϫ g for 10 min at 4°C. Post-nuclear cell lysates, immunoprecipitates, or comparable amounts of mouse tissue lysates were taken up in sample buffer, boiled, and analyzed by SDS-PAGE. When non-reducing analysis was needed, DTT was left out of the sample buffer. Immunoprecipitations were carried out using 1 l of anti-HA or anti-myc antibodies immobilized on 50 l of a 20% suspension of Protein A-Sepharose beads for 2 h, followed by washing twice with lysis buffer. Proteins were transferred to polyvinylidene difluoride membranes (Millipore) at 150 mA for 2 h or 30 V overnight, and the membranes were blocked in 8% milk/phosphate-buffered saline Tween for 1 h or overnight. The primary antibodies were used at 1:5,000 (␣-myc and ␣-HA), 1:1,000 (␣Ero1␣/D5 and ␣PDI), and 1:50 (␣Ero1␤). After washing six times with phosphate-buffered saline Tween membranes were incubated with corresponding secondary antibodies (DAKO) at 1:3,000 for 1 h, washed extensively, and visualized by ECL (Amersham Biosciences) and exposure to film (Eastman Kodak Co.). Protein markers were from Bio-Rad. Determination of protein concentration of the tissue samples prior to gel loading was carried out by using the Bradford assay (Bio-Rad), with samples equalized to 1.8 g/l. Each blotting experiment was reproduced at least twice.

Ero1␤ Proteins Are Highly Expressed in Stomach and Pancreas-De-
spite much interest in ER oxidoreductases, little is known about the expression of Ero proteins in mammalian tissues. Most work has concentrated on the relationship between Eros and PDI and the mechanism of electron transfer to molecular oxygen. The mammalian cell lines thus far examined have low expression of Eros, making study of endogenous protein behavior in tissue culture difficult. Given that studies at the mRNA level suggest that Ero1␤ transcripts are high in stomach and pancreas (11), we investigated whether Ero1␤ proteins could be detected in these mouse tissues. To do this, we raised a polyclonal antiserum against an internal peptide unique to Ero1␤. To verify that the antiserum recognized Ero1␤, HeLa cells were mock transfected or transfected with Ero1␣-myc, Ero1␤-myc, Ero1␤-HA, or both Ero1␤-HA and Ero1␤-myc. The tagged proteins were only detected in the relevant individual transfectants (Fig. 1, A and B, lanes 1 and 2) and in the co-transfectant (Fig. 1, A and B, lane 3). The Ero1␤ serum specifically detected both tagged Ero1␤ forms (Fig. 1C, lanes 2-4), and no Ero1␤ could be detected in non-transfected HeLa cells (Fig. 1C, lane 5). Ero1␣ was not detected by the Ero1␤ serum, demonstrating that the serum did not cross-react with Ero1␣ (Fig. 1C, lane 1). As a positive control, Ero1␣ could be detected by D5 when the membrane from Fig.  1C was reprobed (without stripping the blot) with the specific polyclonal Ero1␣ serum (Fig. 1D, lane 1). Ero1␤-HA migrated faster than Ero1␤-myc because of size/charge differences in the C-terminal tag. Tagged Ero1␤ proteins can restore viability to yeast Ero1p ts mutants (11) and support oxidative protein folding of model substrates (19), demonstrating that tagging does not affect Ero1␤ function.
Having established the specificity of our Ero1␤ antibody, we then asked whether Ero1␤ expression could be detected in mouse tissues. Samples of mouse tissues were lysed and clarified by centrifugation, and equal amounts of protein were analyzed by Western blotting. Pancreas and stomach (Fig. 1E, lanes 1 and 4) expressed high levels of Ero1␤, whereas liver, small intestine, kidney, and heart did not (Fig. 1E, lanes 2, 3, 5, and 6). Unlike Ero1␤, PDI was expressed in all tissues (Fig. 1F). We conclude that our antibody specifically detected Ero1␤ and that Ero1␤ expression was strikingly high in a restricted range of tissues.
Ero1␤, PDI, ERp57, and BiP Are Highly and Specifically Expressed in the Chief Cells of the Stomach-We further investigated Ero1␤ expression in the digestive system by using immunohistochemistry. Human stomach tissue sections (blocked with normal goat serum) were negative when primary antibody was omitted ( Fig. 2A) and sections exposed to pre-immune rabbit sera instead of primary antibody gave a faint nonspecific staining (Fig. 2B). However, sections probed with the Ero1␤ antiserum showed very strong, specific staining in the basal area of the stomach (Fig. 2C). Higher magnification showed that Ero1␤ had a patchwork distribution, and was present in the enzyme producing chief cells of the stomach (arrow) and absent from the acid producing parietal cells (arrowhead) within the same glands (Fig. 2D).
Next, we asked whether other ER chaperones involved in protein oxidation and the unfolded protein response also showed cell-specific expression in the stomach. PDI (Fig. 2, E and F), ERp57 (Fig. 2, G and H), and BiP (Fig. 2, I and J) were all expressed at a high level in the same cell types. In contrast to Ero1␤, both PDI and BiP showed some additional expression in the gastric epithelium (Fig. 2, E and I). We conclude that enzyme-producing cells in the stomach express high levels of ER chaperones and the oxidoreductase Ero1␤.
Ero1␤ and PDI Are Both Expressed in the Pancreas but Show Differential Expression in Pancreatic Islets-To determine whether high oxidoreductase expression correlates with high expression of ER chaperones in other tissues, we examined human pancreas. This was of particular interest given the fact that the UPR and protein misfolding contribute to type II diabetes (6,30). Tissue sections (blocked with normal goat serum) showed faint background staining when TBS (Fig. 3A) or pre-immune rabbit sera (Fig. 3B) were used instead of specific antibody. Pancreas sections stained for the pancreas-specific PDI, PDIp, showed expression of this enzyme in the acinar cells and weak staining of islets, as expected (Fig. 3, C and D) (29). Pancreatic islets were specif- . D, the membrane from C was reprobed with the ␣Ero1␣ serum D5 without stripping. E, equal amounts of murine tissues were subjected to Western blotting after SDS-PAGE using ␣Ero1␤ and ␣PDI (F). Ero1␤ is specifically detected in transfected cells, stomach, and pancreas.

Ero1␤ Expression and Dimerization
ically detected with the anti-insulin antibody H-86 (Fig. 3, E and F). Sequential sections were used to determine the expression pattern of Ero1␤ (Fig. 3, G and H) and PDI (Fig. 3, I and J). Ero1␤ was expressed in both islets (arrow) and the surrounding enzyme secreting cells, with a stronger expression in islets. In contrast, PDI expression was diminished in islets (arrow) compared with the surrounding acinar cells. We conclude that in the pancreas Ero1␤ is found in both exocrine and endocrine cells, whereas PDI and PDIp are both more strongly expressed in exocrine acinar cells.
Ero1␤ Engages in Disulfide-dependent, Alkylation-independent Interactions in the ER-Having established that the expression patterns of Ero1␤ and PDI were not strictly correlated, we decided to investigate the nature of Ero1␤-PDI complex formation more closely. Ero1␣ interacts with PDI in a disulfide-dependent manner, and Ero-PDI complexes can only be recovered when NEM or another alkylating agent is included in the lysis buffer (18). PDI forms mixed disulfide bonded complexes with Ero proteins that are normally rapidly resolved. NEM blocks the transfer of these intermolecular disulfides by binding to free thiols, thus preventing disulfide bond isomerization and rearrangement. Mutations in the Ero1␣ CXXCXXC active site alter the structural integrity of the protein (18). However, the role of the CXXCXXC motif in Ero1␤ has not been studied. We therefore investigated the properties of Ero1␤, its CXXCXXC motif, and its relationship to PDI by transfecting three Ero1␤ myc-tagged mutants (C390A, C393A, and C396A) into HeLa cells. The transfected cells were lysed either in the presence or absence of the alkylating agent NEM, and the post nuclear supernatants were examined by Western blotting after non-reducing and reducing 8% SDS-PAGE. Running the gels under non-reducing conditions kept disulfide bonds intact and allowed any disulfide-dependent interactions with PDI to be observed.
At steady state, monomeric wild-type Ero1␤ existed as a population of mainly oxidized molecules (OX) of ϳ60 kDa, based on the faster migration of the protein on non-reducing SDS-PAGE (Fig. 4A, lane 1). The exact ratio of oxidized:reduced (OX:R) Ero1␤ varied somewhat between transfections and was also dependent on cell type (not shown and see also Fig. 4C). Ero1␤C390A was in a less compact state (Fig. 4A,  lane 3), consistent with the loss of a long distance disulfide and comparable with previous results for Ero1␣C391A (18). Ero1␤C393A was found as a mixture of oxidized (OX) and reduced (R) forms (Fig. 4A, lane 5), whereas Ero1␤C396A was found in a predominantly reduced state (Fig. 4A, lane 7). Alkylation was required to trap the oxidation states (compare Fig. 4A, lanes 1 and 2), and mock-transfected cells were negative (Fig. 4A, lanes 9 and 10). Reduction of the samples with DTT resulted in the appearance of a single reduced band (Fig. 4B), and treatment of living Ero1␤ transfected HT1080 cells with DTT prior to nonreducing SDS-PAGE (Fig. 4C) demonstrated that the shifts seen in the non-reducing gel are the result of compaction by disulfide bonds and not degradation. These results are similar but not identical to those found with Ero1␣. The most notable difference is that the oxidation states of Ero1␣ are more discrete when analyzed by SDS-PAGE (18).
The same lysates were used to compare the ability of Ero1␤ CXX-CXXC mutants to interact with PDI by taking advantage of the fact that disulfide-dependent Ero-PDI dimers are preserved when trapped by alkylation and separated by non-reducing SDS-PAGE. The transfected HeLa cell lysates were either loaded directly onto non-reducing gels (Fig. 4D), or were first subjected to immunoprecipitation with an antiserum raised against PDI (Fig. 4E). The transferred proteins were then probed with the myc antibody to detect Ero1␤. In the presence of NEM, wild-type Ero1␤ formed a ladder of higher molecular mass bands (Fig.  4D, lane 1), of which a component was an Ero1␤-PDI complex of ϳ130 kDa (Fig. 4E, lane 1). The Ero1␤-PDI dimer was NEM-dependent, as expected (Fig. 4E, lanes 1 and 2). However, in the absence of NEM, disulfide-dependent higher molecular mass complexes still existed (Fig.  4D, lane 2). These must represent Ero1␤ in association with another protein.
The Ero1␤C390A mutant also formed disulfide-dependent complexes (Fig. 4D, lane 3), but hardly interacted with PDI (Fig. 4E, lane 3). Again, Ero1␤C390A complexes were preserved in the absence of an alkylating agent, showing that Ero1␤C390A could associate with partner protein(s) other than PDI (Fig. 4D, lane 4). Ero1␤C393A (Fig. 4, D  and E, lanes 5 and 6) and Ero1␤C396A (Fig. 4, D and E, lanes 7 and 8) both interacted with PDI less strongly than wild type in an NEM-dependent manner. Disulfide-linked complexes were dispersed to a single Ero1␤ band upon reduction with DTT ( Fig. 4B and not shown). Note that the 9E11 ␣-myc monoclonal antibody gave a nonspecific background band at ϳ130 kDa (Fig. 4D, lanes 9 and 11). The nonspecific nature of this band, and the reproducibility of the results, was confirmed using a polyclonal ␣-myc antiserum, which specifically detected Ero1␤containing complexes only (Fig. 4D, lanes 12 and 13, and not shown). Also note the expected cross-reactive PDI antibody in the myc blot of the immunoprecipitations (Fig. 4E, Ab).
To further confirm the distinct nature of the Ero1␤ complexes, we performed an experiment in which PDI was depleted from Ero1␤-transfected cell lysates by consecutive rounds of immunoprecipitation. All detectable PDI was removed by this procedure, whereas calnexin levels remained unaffected (Fig. 4F, compare lane 1 with lanes 3 and 4). Anal- ysis of the PDI-depleted lysates by non-reducing SDS-PAGE and immunoblotting with ␣-myc showed that the PDI-Ero1␤ complex was selectively removed, whereas the lower complex remained (*), along with the Ero1␤ monomers (Fig. 4G). Thus, in cell lysates, the PDI-Ero population was discrete. We conclude that Ero1␤ can form non-PDI redox-sensitive complexes that are not lost by disulfide reshuffling in the absence of alkylating agents.
Ero1␣ and Ero1␤ Form Mixed Disulfide-dependent Heterodimers-We next asked whether the two oxidoreductases, Ero1␣ and Ero1␤, could form mixed complexes at steady state. Thus HeLa cells were transfected with Ero1␣myc, Ero1␤-HA, or Ero1␣-myc and Ero1␤-HA together (Fig. 5, B and C). Cell lysates were analyzed under reducing conditions, and membranes were sequentially probed with the myc antibody (Fig. 5B, lanes 1-4) and the HA-7 antibody (Fig. 5B, lanes 5-8) to confirm the expression of the Ero proteins. Cell lysates were subjected to an HA-7 immunoprecipitation followed by blotting the reducing gels for myc (Fig. 5C). A myc signal could only be observed in the double transfectant (Fig. 5C, lane 3), confirming that Ero1␣ and -␤ interacted. This experiment therefore demonstrates that the two different oxidoreductases can associate in the endoplasmic reticulum.
Ero1␤ Complexes Differ in Abundance in Stomach and Pancreas-To see whether Ero1␤ dimers occurred in vivo, we lysed mouse stomach and mouse pancreas tissues in the absence of NEM, using conditions in which PDIs and Ero1␤ do not interact (see Fig. 4E, lanes 1 and 2). Our Ero1␤ antiserum does not immunoprecipitate, or readily detect Ero1␤ when used for immunoblotting under non-reducing conditions. We therefore subjected tissue lysates to gel filtration and analyzed trichloroacetic acid-precipitated fractions by reducing SDS-PAGE, prior to immunoblotting with the Ero1␤ serum. Ero1␤ from stomach was  1 and 2), Ero1␤C390A (lanes 3 and 4), Ero1␤C393A (lanes 5 and 6), Ero1␤C396A (lanes 7 and 8), and mock transfectants (lanes 9 and 10) lysed in the presence (lanes 1, 3, 5, 7, and 9) or absence (lanes 2, 4, 6, 8, and 10) of NEM, immunoprecipitated with ␣-myc and blotted for PDI. PDI co-immunoprecipitated with wtEro1␤, Ero1␤C393A, and Ero1␤C396A in the presence of NEM. Antibody complexes (Ab) migrate just above the PDI-Ero dimer. F, reducing SDS-PAGE of Ero1␤-myc transfected HeLa cell lysates subjected to sequential immunodepletion of PDI prior to detection of calnexin (Cnx) and PDI. All detectable PDI is specifically removed from the lysate. G, non-reducing SDS-PAGE of Ero1␤-myc transfected HeLa cell lysates subjected to sequential immunodepletion of PDI, prior to detection of Ero1␤-myc with the ␣myc monoclonal antibody. 1st ϭ first immunodepletion, 2nd ϭ second immunodepletion, sup ϭ supernatant, tcl ϭ total cell lysate prior to immunodepletion. The Ero-PDI complex is depleted (arrow), whereas the major Ero complex remains (asterisk). recovered in fractions 10 -13, with a peak in fraction 12 corresponding to the likely size of the glycosylated monomer (Fig. 6A). In contrast, Ero1␤ from pancreas peaked in fractions 9 and 10, corresponding to the expected size of a glycosylated Ero1␤ dimer, with a minor amount eluting in fractions 13 and 14 (Fig. 6B). The elution profiles of Ero1␤ were compared with bovine serum albumin (monomer of 66 kDa and dimer of 132 kDa, Fig. 6C) and chicken egg albumin (monomer of 45 kDa, Fig.  6D). We conclude from this experiment that non-PDI-bound Ero1␤ complexes exist in vivo and that their relative abundance can be different in the stomach and the pancreas.
Our results demonstrate that an Ero1␤ AXXCXXC mutant can interact with itself and with a wild-type Ero1␤ protein. An Ero1␤ CXXCXXA mutant can interact with wild-type Ero1␤ protein but cannot dimerize with the mutant counterpart. These data are consistent with a symmet-rical, redox-dependent Ero dimer in which the Cys 396 residue is required for dimerization.
Modeling the Ero1 Dimer-To identify a molecular basis for the dimer of Ero1␤, and to understand the role of the C396A mutant in disrupting it, we analyzed the recently determined crystal structure of Ero1p (23). Although there is only a single monomer in the asymmetric unit of the two different crystal forms of the enzyme reported (PDB codes 1RP4 and 1RQ1), examination of the crystal contacts revealed a single dimer interaction between monomers that was found in both structures (Fig. 9). This is consistent with the symmetrical dimer suggested by the immunoprecipitation experiments in Figs. 7 and 8. From our model of the dimer, the N termini are brought into close proximity, and therefore the N terminus, which is absent in the construct of the enzyme used for the crystal structure, could also interact and form a part of the dimer interface. The crystal structure was of a truncated Saccharomyces cerevisiae Ero1p protein, which shares ϳ25% sequence identity with full-length human Ero1␤ investigated here. The human enzyme has several amino acid inserts ranging in size from 5 to 25 residues. Analysis of the crystal structure confirms that they are all found at the surface in positions where additional structure can be accommodated. In addition, none of the inserts would disrupt the presumed dimer interface.
The proposed Ero1p dimer buries ϳ1100 Å 2 of predominantly nonpolar solvent-accessible surface area per monomer, and the monomers are arranged so that the active sites are on the same side of the dimer. The dimer brings together the adenosine groups of the two FAD moieties in close apposition; in fact the two hydroxyls of the ribose sugars of the adenosine moiety of FAD form hydrogen bonds to each other (Fig.  9A). The FADs contribute 20% of the dimer interface and therefore may be essential for dimer formation. Interpreting the mutagenesis data in light of the crystal structure, it is notable that the Ero1␤ C396A mutant does not dimerize given the buried location of this residue and its contact with the FAD cofactor (Fig. 9B). It is highly unlikely that this residue is involved in any disulfide to form the dimer but, rather, disrupts the FAD binding so that the FAD-FAD dimer interface is lost.

DISCUSSION
Our experiments show that Ero proteins can homodimerize and heterodimerize, as well as interact with PDI in a redox-dependent manner . Prior to this study, it was known that Ero proteins and PDI were required for disulfide bond catalysis in the ER. However, an Ero1␣ C391A mutant can rescue a temperature-sensitive yeast ero1-1 mutant, despite having little affinity for PDI (10,11,25). Taken together with this observation, the experiments in this report suggest that both dimerization and binding to PDI are required for appropriate maintenance of disulfide bond formation in the ER. The active C390A Ero1␤ mutant binds poorly to PDI but can homodimerize, whereas the inactive C396A Ero1␤ mutant interacts with PDI but cannot homodimerize (Figs. 4 and 8), suggesting that Ero-Ero dimers are important for the functional activity of this protein family.
Our results show that dimerization is partly controlled by the CXX-CXXC motif, although other covalent and non-covalent interactions are likely to be important for dimer establishment. Both the Ero-Ero interactions and the Ero-PDI interactions are stable under non-reducing conditions (in the presence of SDS). However, the strength of these interactions is different, given that the PDI-Ero interaction requires trapping by an alkylating agent, whereas the Ero-Ero interaction does not. Dimers are present both in vivo and in transfected cells, can occur between various tagged and non-tagged forms of Ero1␣ and Ero1␤, and can be retrieved from the yeast Ero1p crystal structure, suggesting that Ero-Ero dimers are conserved throughout evolution and are biologically

Ero1␤ Expression and Dimerization
significant. The fact that mutating Cys 396 gives rise to no dimers also suggests that dimerization is genuine, because nonspecific dimers resulting from misfolding would be expected to increase, not decrease, upon exposure of the free -SH of C393.
Reconstruction of an Ero-Ero dimer from the Ero1p crystal structure shows that the Ero1p, Ero1␣, and Ero1␤ glycosylation sites are all located away from the dimer interface, and would not prevent Ero-Ero interactions (Fig. 9). The likely site of PDI docking, via a flexible Ero hinge, is also positioned away from the dimer interface, suggesting that PDI-Ero heterodimers and Ero-Ero dimers are not necessarily mutually exclusive. Our immunodepletion data (Fig. 4G) suggests that Ero-PDI complexes are independent of Ero-Ero dimers in a Triton X-100 lysate. However, it remains possible that tetrameric Ero-PDI complexes could operate under different situations in vivo. Regulation of a dimer/tetramer pool, for example, could allow for rapid mobilization or storage of disulfide donors in response to the fluctuating secretory requirements of a tissue.
The Ero1p dimer that we have modeled is held together by a hydrogen-bonding network at the dimer interface (Fig. 9). Two hydrogen bonds link the two FAD moieties, which are likely to be important for dimer stability, given that the C396A mutation of Ero1␤ disrupts the dimer. Cys 396 , in concert with Cys 393 , probably passes electrons on to FAD for donation to oxygen. Thus the loss of Cys 396 is likely to lead to loss of dimerization indirectly, by disturbing FAD binding. In the light of our studies, further experiments can now be conducted to determine whether targeting the dimer interface can be used to specifically engineer and manipulate the process of disulfide bond formation. The proximity of the N termini in the model of the truncated Ero1p dimer also suggests that N-terminal cysteines upstream of the CXXXXC redox active site may be close enough to form an intermolecular disulfide bond. This would explain the appearance of dimers under non-reducing conditions and can now be tested using the relevant cysteine mutants.
Prior to this study, it was known that Ero1␤ could be induced by the UPR and that mRNA levels were highest in secretory tissues (11). Nothing was known about the expression or behavior of the Ero1␤ protein in vivo. Our results show that Ero1␤, a UPR-responsive protein, is constitutively yet cell specifically expressed in stomach and pancreas (Figs.  1-3). The balance between monomeric and complex forms of Ero1␤ is also different in the pancreas, where complexes predominate, and in the stomach (Fig. 6). The data imply that Ero1␤ is subject to tissue-and cell-specific regulation, at both the transcriptional and post-translational level. It will be very interesting to identify these regulatory factors and to determine whether Ero1␤ expression levels and/or dimerization depend on the nutritional status of an animal.
In a simple model of Ero-PDI function, a low abundance electron acceptor (Ero) can fuel a large number of catalytic cycles. Our results show that this model is too simple and that cells can accommodate high amounts of Ero1␤ in vivo. This suggests that Ero1␤ is either an inefficient catalyst, that it is performing an additional function, or that it can remain in the ER in an inactive state. The difference between Ero1␤ and PDI expression in pancreatic islets and acinar cells suggests that the expression levels of these two proteins can be independently controlled (Fig. 3). Perhaps PDI is not the only electron donor for Ero1␤ in the pancreas. The PDI homologue PDIp is expressed specifically in this organ, and although PDIp can use hydroxyaryl groups as ligands (32), it could in theory replace PDI function in islets. However, like others (29) we find that PDIp and PDI expression patterns are similar, and that PDIp is expressed exclusively in acinar cells. This suggests that further analysis of the islet-specific physiology of Ero1␤ electron donors may be worthwhile, because tissue-specific differences in the control of disulfide bond formation may provide an opening for therapeutic intervention in pancreatic protein secretion and type 2 diabetes. In this respect, it is interesting to note that QSOX1 is highly expressed in the islets of Langerhans (33). The QSOX (Quescin-sulfhydryl oxidases) family utilize oxygen to generate disulfide bonds with the production of peroxide. However, the true function of QSOX1 in vivo and its relationship to Ero and PDI proteins is not yet known.
Our data raise the unexpected possibility that intermolecular electron transfer between two Ero molecules might occur via FAD (Fig. 9). This may seem unlikely given the weight of evidence for a linear Ero-PDI-substrate oxidation chain (34,35). The Ero-PDI heterodimer can mediate oxidative protein folding in vitro, but it is possible that Ero-Ero dimers contribute to this task in vivo. Many thioredoxin family proteins can dimerize, and mutant bacterial periplasmic DsbC molecules that are unable to do so are converted from isomerases into oxidases (36). Engineered dimerization by domain hybridization also confers novel properties onto redox enzymes (31).
The very high relative expression of the Ero1␤ oxidoreductase in stomach and pancreas raises the question of how the ER protects itself from free radical damage and oxidative stress in chief cells and islets. Perhaps Ero dimerization is related to the need to protect irrelevant proteins from random oxidative damage by "exposed" electrons, with dimerization preventing inappropriate electron flow from an exposed flavin to a bystander protein. Alternatively, the equilibrium between the dimeric and PDI-bound forms of Ero could be used to regulate oxidoreductase activity through occupation or protection of the active site. In this respect, it will be interesting to determine whether overexpression of Ero1␤ mutants, which cannot dimerize, generate more oxidative stress in the ER than the C390A mutant (which is active but can dimerize).
In terms of physiology, we note that, although a chemical UPR is necessary to induce Ero1␤ expression in cell lines, the default state of Ero1␤ in the chief cells and islets of the stomach and pancreas is "on." Specific cells in some organs are therefore pre-primed for oxidative protein folding. Whether this is due to constitutive or partial maintenance of a UPR state requires further investigation, for example by assessing the status of XBP1, ATF6, PERK, and other regulators of the UPR in chief cells and islets. Evidently, some cells can sustain high levels of Ero1␤ expression without incurring sufficient oxidative damage to induce cell death, whereas other neighboring cells remain refractory to Ero1␤ activation. Our studies emphasize the importance of relating work in cell lines to the whole animal, and in this respect an Ero1␤ tissue-specific knock-out mouse may be very informative. Finally, mixed Ero1␣/␤ heterodimers (Fig. 5) add another level to this disulfide bond formation pathway, indicating that Ero1␣␤ dimers may be important in the response to combined UPR and hypoxia. This may be relevant to protein misfolding diseases in the CNS or in tumor development, where hypoxic conditions induce the expression of Ero1␣ (37). It is apparent that the interactions between PDI and Ero proteins are more complex than previously imagined in mammals. Our analysis of the in vivo expression pattern of Ero1␤ and the molecular requirements for homodimer formation significantly develops our cellular and molecular knowledge of an essential biological process, the control of protein folding in the endoplasmic reticulum.