EDEM3, a Soluble EDEM Homolog, Enhances Glycoprotein Endoplasmic Reticulum-associated Degradation and Mannose Trimming*

Quality control in the endoplasmic reticulum ensures that only properly folded proteins are retained in the cell through mechanisms that recognize and discard misfolded or unassembled proteins in a process called endoplasmic reticulum-associated degradation (ERAD). We previously cloned EDEM (ER degradation-enhancing α-mannosidase-like protein) and showed that it accelerates ERAD of misfolded glycoproteins. We now cloned mouse EDEM3, a soluble homolog of EDEM. EDEM3 consists of 931 amino acids and has all the signature motifs of Class I α-mannosidases (glycosyl hydrolase family 47) in its N-terminal domain and a protease-associated motif in its C-terminal region. EDEM3 accelerates glycoprotein ERAD in transfected HEK293 cells, as shown by increased degradation of misfolded α1-antitrypsin variant (null (Hong Kong)) and of TCRα. Overexpression of EDEM3 also greatly stimulates mannose trimming not only from misfolded α1-AT null (Hong Kong) but also from total glycoproteins, in contrast to EDEM, which has no apparent α1,2-mannosidase activity. Furthermore, overexpression of the E147Q EDEM3 mutant, which has the mutation in one of the conserved acidic residues essential for enzyme activity of α1,2-mannosidases, abolishes the stimulation of mannose trimming and greatly decreases the stimulation of ERAD by EDEM3. These results show that EDEM3 has α1,2-mannosidase activity in vivo, suggesting that the mechanism whereby EDEM3 accelerates glycoprotein ERAD is different from that of EDEM.

ER 3 quality control is an elaborate mechanism conserved from yeast to mammals, ensuring that newly synthesized proteins in the ER fold and assemble correctly and that only proteins that acquire their correct conformations are sorted further into the secretory pathway (1)(2)(3)(4). During this process, proteins that fail to attain their native conformation due to mutations of the polypeptides or to ER stress conditions adverse for protein folding as well as orphan subunits are degraded in a process known as ER-associated degradation (ERAD) (3,(5)(6)(7). The recognition of misfolded proteins for ERAD is still poorly understood, but there is increasing evidence for a role of mannose trimming in the targeting of glycoproteins for ERAD (8,9). In mammalian cells, overexpression of ER ␣-mannosidase I stimulates ERAD of misfolded glycoproteins (10,11), whereas the ␣1,2mannosidase inhibitors kifunensine and 1-deoxymannojirimycin stabilize misfolded glycoproteins (12)(13)(14)(15)(16). These observations suggested that Man 8 GlcNAc 2 isomer B, the major product of the ER ␣1,2-mannosidase, is a recognition marker for ERAD of glycoproteins, but this view is being challenged, since there is increasing evidence that trimming to smaller oligosaccharides occurs on ERAD substrates (10,(17)(18)(19). We previously cloned mouse EDEM (ER degradation enhancing ␣-mannosidase-like protein) as a cDNA whose expression is up-regulated by ER stress and showed that EDEM accelerates glycoprotein ERAD (20). EDEM is an integral ER membrane protein that has all the signature motifs of Class I ␣1,2-mannosidases (glycosylhydrolase family 47) but no detectable enzyme activity as a processing ␣-mannosidase in vivo or in vitro. Recently, it was found that EDEM extracts terminally misfolded glycoproteins from the calnexin cycle (21,22). In S. cerevisiae, the ER ␣1,2-mannosidase as well as Htm1p/Mnl1p belonging to the same protein family are also involved in ERAD, since disruption of the genes delays the ERAD of glycoproteins (23,24). Although EDEM and Htm1p/ Mnl1p were postulated to be lectins involved in targeting misfolded glycoproteins for ERAD, the precise mechanisms whereby EDEM and Htm1p/Mnl1p recognize and sort misfolded glycoproteins for degradation are still unclear, and their role as lectins has not been established directly. While this manuscript was in preparation, EDEM2 was reported to stimulate ERAD of misfolded glycoproteins without affecting mannose trimming (25,26).
Here, we show that EDEM3 is a soluble EDEM homolog located in the ER of transfected mammalian cells that accelerates ERAD of misfolded glycoproteins through a mechanism likely to be different from that of EDEM or EDEM2, since EDEM3 greatly stimulates mannose trimming in vivo.

EXPERIMENTAL PROCEDURES
Cloning of Mouse EDEM Homolog-Five expressed sequence tag clones were sequenced, and one clone (G431003D06), which was kindly provided by Dr. Y. Hayashizaki (RIKEN, Japan) (27), contained the entire EDEM3 cDNA. Sequencing was performed by the PCR-based dideoxy termination method using BigDye version 3 (ABI) and PCR 9700 (PerkinElmer Life Sciences) and then analyzed with the ABI 3100 capillary sequencer.
Plasmid Construction-The coding region of EDEM3 cDNA was subcloned into pcDNA3.1ϩ by PCR, and the HA tag was introduced prior to the -KDEL ER retrieval signal (EDEM3-HA). The null Hong Kong (NHK)-QQQ mutant was created by replacing Asn residues of NHK glycosylation sites with Gln, using the QuikChange TM site-directed mutagenesis kit (Stratagene). The EDEM3 E147Q mutant was constructed by the same method. FLAG-tagged TCR␣ was kindly provided by Dr. F. Tokunaga (Osaka City University Graduate School of Medicine).
In Vitro Translation and Translocation-EDEM3-HA cDNA was linearized with EcoRI, and transcribed in vitro by T7 RNA polymerase (Promega). The transcript was then translated in vitro in the reticulocyte lysate (Flexi-lysate; Promega) supplemented with [ 35 S]methionine with and without canine pancreas microsomes. The localization of the EDEM3 product was determined by the alkali floatation method (28), monitored by 10% SDS-PAGE. Mouse EDEM tagged with HA was used as a control ER membrane protein (20), and HSP47 was used as a control ER luminal protein (29). FIGURE 1. Similarity between EDEM3, EDEM, and ER ManI. A, amino acid sequence of mouse EDEM3. Region of similarity with Class I ␣1,2-mannosidase family (glycosyl hydrolase family 47), the protease-associated domain, and the signal sequence are shaded in orange, blue, and green, respectively. The KDEL ER retrieval signal is underlined. The nine conserved acidic amino acids are indicated by closed triangles, and putative N-glycosylation sites are shown by dots. B, domain organization of mouse EDEM3, mouse EDEM, and human ER ManI.
Cell Culture and Transfection-HEK293, BALB/c3T3, COS7, and HeLa S3 cells (provided by Japan Health Science Research Resources Bank) were cultured in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum and antibiotics (100 units/ml penicillin G and 100 ng/ml streptomycin) in humidified air containing 5% CO 2 at 37°C. PC12h cells were cultured in Dulbecco's modified Eagle's medium supplemented with 5% fetal bovine serum and 5% horse serum. HepG2 cells stably overexpressing the EDEM3 were established using the pCX4bsr plasmid, which has a retrovirus long terminal repeat (30), and selected in the presence of 30 g/ml blasticidin (Funakoshi, Japan). Plasmids were purified with a Plasmid Maxi Kit (Qiagen) and transfected using the FuGene6 transfection reagent (Roche Applied Science) as described previously (10).
Northern Blotting-Total RNA from cultured cells was extracted using TRIzol TM reagent (Invitrogen). After separation on formaldehyde-denatured gel, RNAs were blotted onto a nylon membrane (Gene-Screen plus; PerkinElmer Life Sciences) and then were hybridized with 32 P-labeled probes for 2-16 h in PerfectHyb TM (TOYOBO, Japan) hybridization solution. An MTN TM blot membrane (Clontech) was used to examine mouse tissue distribution. Probes for EDEM, BiP (immunoglobulin heavy chain-binding protein), ␤-actin, and HSP70 were labeled by the multiprimed labeling method (Roche Applied Science). To detect EDEM3 mRNA, probes were labeled by unidirectional PCR, using the 571-bp fragment of the EDEM3 cDNA near its translational termination site, to increase the specific activity of the probe and to avoid cross-reactivity with EDEM.
Metabolic Labeling, Immunoprecipitation, and SDS-PAGE-Metabolic labeling, cell lysis, immunoprecipitation, and SDS-PAGE were carried out as previously described (10), except that Dulbecco's modified Eagle's medium lacking both methionine and cysteine was used instead of Dulbecco's modified Eagle's medium lacking methionine. Cells were treated with 5 g/ml tunicamycin (Tu) for 6 -7 h. Twenty g of total RNA were analyzed by Northern blotting with probes for EDEM3, EDEM, BiP, and ␤-actin. C, BALB/c 3T3 cells were subjected to ER stress (tunicamycin, 2-deoxyglucose, and A23187) or to cytosolic stress (arsenite and heat shock). The same blot was rehybridized with BiP (induced by ER stress), ␤-actin (loading control), and HSP70 (induced by cytosolic stress).
Antibodies-Antibodies against ␣1-AT were purchased from DAKO (rabbit polyclonal). Polyclonal antibodies against the HA tag were obtained from Santa Cruz Biotechnology, Inc. (Santa Cruz, CA), anti-FLAG M2 antibody (mouse monoclonal) was from Stratagene, and mouse monoclonal antibody against protein-disulfide isomerase was purchased from StressGen.
Reagents-Lactacystin was purchased from Kyowa Medics (Japan), and kifunensine was a generous gift from Fujisawa Pharmaceutical Co. (Osaka, Japan). Endo H and N-glycanase F (PNGase F) were purchased from Roche Applied Science or New England Biolabs.
Immunocytochemistry-COS7 cells were plated on a coverglass placed in a 3.5-cm diameter dish ϳ24 h prior to transfection. Twentyfour h after transfection, the cells were fixed with 4% paraformaldehyde for 15 min at room temperature and incubated with anti-HA tag antibodies (rabbit polyclonal) and anti-protein-disulfide isomerase antibody (mouse monoclonal) for 1 h and then with an Alexa546-labeled anti-rabbit IgG and an Alexa488-labeled anti-mouse IgG antibody for 1 h. Samples were examined by confocal microscopy (LSM 510 META; Zeiss).
Oligosaccharide Analysis of NHK and Total Cellular Glycoproteins-Oligosaccharide analysis of [ 3 H]mannose-labeled NHK by HPLC was performed as described previously (10,31). For oligosaccharide analysis of total glycoproteins in HepG2 and cell lines stably overexpressing EDEM3, N-glycans were released by hydrazinolysis, fluorescently labeled using 2-aminopyridine, and were fractionated by size on an NH 2 -P HPLC column (Shodex Asahipak NH2P50, 0.46 ϫ 15 cm; Showa Denko, Japan) as described previously (32). Briefly, samples were loaded onto the column, and the oligosaccharides were eluted at 0.5 ml/min by a linear gradient of solvent A (acetonitrile) and solvent B (50 mM ammonium acetate, pH 7.0) from 4:1 (v/v) to 3:7 (v/v) for 50 min. Oligosaccharides eluted in each peak were identified by reverse phase HPLC.

Cloning of EDEM3, a Soluble Homolog of EDEM-A transcript named
C1orf22 was recently mapped in a region susceptible to human hereditary prostate cancer (33). The predicted open reading frame of C1orf22 has a sequence similar to Class I ␣1,2-mannosidases (glycosylhydrolase family 47) and shows highest homology to the KIAA0212 gene product (34). Since KIAA0212 is the human ortholog of the mouse EDEM gene (20), we searched the mouse expressed sequence tag data base for the C1orf22 homolog. Sequencing of expressed sequence tag clones revealed a mouse cDNA of 6,349 bases (GenBank TM /EMBL/DDBJ accession number AB188342), encoding a protein of 931 amino acids (Fig. 1A). This open reading frame consists of a region (amino acids 60 -499) similar to Class I ␣1,2-mannosidases (glycosylhydrolase family 47), followed by a C-terminal region containing a protease-associated motif (amino acids 686 -780) (35) that is lacking in EDEM and in all Class I ␣1,2-mannosidases studied so far. It has a putative signal sequence at the N terminus as well as an ER-retrieval signal (-KDEL) at its C terminus.
The mouse EDEM3 cDNA sequence exhibits 90% identity with its human ortholog C1orf22 in the coding region and 76% overall identity. The mouse and human orthologs are 91% identical in amino acid sequence. The mouse EDEM3 has 44 additional amino acids at the N terminus compared with the C1orf22 translation product. The ␣-mannosidase domain shows 44 and 30% amino acid identity between EDEM3 and EDEM and between EDEM3 and ER ␣-mannosidase I (ER ManI), respectively. All nine acidic amino acids that are essential for ␣-1,2 mannosidase activity (36) are conserved between these three proteins (Fig. 1A), although the two Cys residues important for activity of the yeast ER ␣1,2-mannosidase (37) are not conserved in either EDEM3 or EDEM. The protease-associated motif is a consensus sequence found in several proteases (35), the significance of which in EDEM3 is currently unknown.
Expression of EDEM3 and Effects of ER Stress-Northern blotting shows a major EDEM3 transcript of ϳ6.5 kb in all mouse tissues, as expected from the cloned cDNA, with relatively high levels in liver, heart, and kidney ( Fig.  2A), as reported for the human homolog C1orf22 (33). EDEM mRNA is highly expressed in liver and moderately in kidney, whereas the expression is low in heart, brain, and skeletal muscle ( Fig. 2A). We examined whether EDEM3 mRNA expression is regulated by ER stress, as previously shown for EDEM (20,38). The addition of tunicamycin, which induces ER stress by inhibiting N-glycosylation of proteins, causes a mild induction of EDEM3 mRNA, depending on the cell lines (Fig. 2B). The EDEM3 transcript is increased about 1.5-2-fold in BALB/c 3T3, 293, and PC12h cells, whereas no induction is observed in COS7 and in HeLa S3 cells. EDEM3 mRNA is also up-regulated by treating cells with the glucose analogue (2-deoxyglucose) or the calcium ionophore A23187 (ER stress), but the level of EDEM3 mRNA does not change in cells exposed to cytosolic stress that greatly stimulates HSP70 expression (Fig. 2C).
EDEM3 Is Localized in the ER Lumen-To establish whether the hydrophobic region near the N terminus acts as a cleavable signal sequence upon co-translational translocation into the ER or whether it serves as a transmembrane region, we separated integral membrane proteins from soluble proteins by alkali floatation. When EDEM3 RNA is translated in vitro, most of the radioactive EDEM3 is recovered in the soluble fraction (Fig. 3A). We observed a shift in the size by SDS-PAGE of 110 kDa to 120 kDa when EDEM3 was translocated into microsomes (data not shown). Treatment with Endo H or PNGase F shows the removal of high mannose type N-glycans from EDEM3 (Fig. 3B), as predicted from the sequence (Fig. 1B).
Next, we examined the intracellular localization of EDEM3 by transfecting HA-tagged EDEM3 transiently into COS7 cells. Indirect immunofluorescence shows a fine reticular network pattern around the nucleus that colocalizes with the ER-resident protein, proteindisulfide isomerase (Fig. 3C). EDEM3 is not secreted into the medium in pulse-chase experiments of transiently transfected HEK293 cells (data not shown). Thus, we conclude that EDEM3 is an ER luminal protein.
EDEM3 Accelerates Glycoprotein ERAD-We then investigated whether EDEM3 affects glycoprotein ERAD. We used the ␣1-antitrypsin genetic variant NHK as a soluble ERAD substrate (13, 39). Co-trans-fection of EDEM3 enhances the ERAD of NHK (Fig. 4, A and B). Coimmunoprecipitation of EDEM3 with NHK was observed using antibodies to either ␣1-AT or HA-tag (Fig. 4A), indicating that EDEM3 interacts with NHK in the cells. Co-immunoprecipitation was most prominent after 1 and 2 h of chase. At these times, the shift in mobility of NHK indicates additional trimming of the oligosaccharides. This observation suggests that the interaction between EDEM3 and NHK is stronger with increased mannose trimming from NHK (Fig. 4A, compare lanes 5 and 6 with lane 4, and compare lanes 8 and 9 with lane 7). NHK degradation is inhibited by lactacystin in the presence of cotransfected EDEM3, showing that EDEM3 accelerates glycoprotein ERAD by proteasomes (Fig. 4C). The mobility shift of NHK after chase periods was consistently larger in cells co-transfected with EDEM3 than in cells co-transfected with mock vector (Fig. 4A, compare lanes 4 -6  with lanes 1-3). The electrophoretic mobility shift of NHK in EDEM3-overexpressing cells is compared with that of deglycosylated NHK prepared by PNGase F digestion (Fig. 4D). The addition of the ␣1,2-mannosidase inhibitor kifunensine greatly inhibits NHK degradation in cells overexpressing EDEM3 and reduces the mobility of NHK on SDS-PAGE (Fig. 4E). This suggests that the mobility shift of NHK in cells co-transfected with EDEM3 is caused by the mannose trimming from the N-linked oligosaccharides.
Since different mechanisms for ERAD of soluble and transmembrane proteins have been proposed (40 -42), we examined the effect of EDEM3 on a FLAG-tagged TCR␣, a glycosylated transmembrane ERAD substrate (43). Co-expression of EDEM3 enhances the degradation of TCR␣-FLAG (Fig. 5, A and B), which is partly inhibited by kifunensine treatment (Fig.  5A). However, EDEM3 does not affect the degradation of NHK lacking all three N-glycosylation sites (NHK-QQQ), demonstrating its specificity for glycoproteins (Fig. 5, C and D). NHK-QQQ is degraded faster than NHK that bears the three N-glycans, and we have confirmed that NHK-QQQ was also degraded by ERAD (data not shown). These data indicate that the acceleration of glycoprotein ERAD by EDEM3 depends on mannose trimming from the N-glycans.
Effect of EDEM3 on Mannose Trimming from N-Glycans of NHK and of Total Glycoproteins-Since the results in Fig. 4 suggested that overexpression of EDEM3 stimulates mannose trimming from N-glycans on NHK, the oligosaccharides were examined after labeling 293 cells with [ 3 H]mannose. The N-glycans released from NHK by Endo H were analyzed by HPLC. Overexpression of EDEM3 greatly stimulates trimming of N-glycans from NHK to Man 7 GlcNAc 2 and Man 6 GlcNAc 2 (Fig. 6, A and B), but there is relatively little Man 8 GlcNAc 2 found on NHK in cells overexpressing EDEM3 compared with mock-transfected cells. At 0 h of chase, there is only a trace amount of Man 6 GlcNAc 2 and Man 7 GlcNAc 2 on NHK from mock-transfected cells, whereas significant labeled Man 6 GlcNAc 2 and Man 7 GlcNAc 2 are observed in cells transfected with EDEM3. The relative amount of Man 6 GlcNAc 2 and Man 7 GlcNAc 2 increases with time of chase in EDEM3-transfected cells and is greater at 1 h chase than the proportion of Man 6 GlcNAc 2 and Man 7 GlcNAc 2 in mock-transfected cells. Furthermore, at all time points the percentage of labeled Man 6 GlcNAc 2 and Man 7 GlcNAc 2 is much greater in EDEM3-transfected cells than in the control. In contrast, the percentage of radioactivity in Man 8 GlcNAc 2 is always much lower in the presence of EDEM3. This pattern of oligosaccharides indicates that overexpression of EDEM3 stimulates mannose trimming from NHK. To determine whether the increased mannose trimming is due to intrinsic mannosidase activity of EDEM3, the effects of the E147Q mutant on NHK oligosaccharides were studied. Glu 147 is a conserved residue that corresponds to Glu 132 and to Glu 330 in the active site of yeast and human ER ␣1,2-mannosidases, respectively (44,45). It is essential for enzyme activity, since mutation of this residue abolishes ␣1,2-mannosidase activity (36).
The pattern of oligosaccharides on NHK obtained from cells overexpressing E147Q is identical to that of mock-transfected cells. And the relative amounts of Man 6 GlcNAc 2 and Man 7 GlcNAc 2 are the same as in control cells (Fig. 6, A and B).
Stimulation of mannose trimming by EDEM3 was also demonstrated by analyzing N-glycans of total glycoproteins from HepG2 cells stably overexpressing EDEM3 (Fig. 6C). There is a large increase of Man 6 GlcNAc 2 concomitant with a relative decrease of Man 7-8 GlcNAc 2 , compared with the parental HepG2 cells. All of these results demonstrate that EDEM3 has ␣1,2-mannosidase activity in vivo.
Following labeling with [ 35 S]methionine/cysteine, NHK degradation was greatly reduced in 293 cells overexpressing the E147Q mutant com-  pared with wild-type EDEM3 (Fig. 6D), indicating that the mannosidase activity of EDEM3 is important for its effect on ERAD of NHK.

DISCUSSION
The present work shows that EDEM3 is a soluble homolog of EDEM that accelerates ERAD of both soluble NHK and membrane-bound TCR␣ in an N-glycan-dependent manner. The extent of ERAD stimulation on NHK degradation is similar to that previously reported following overexpression of EDEM and of ER ManI (10,11,20). However, the role of EDEM3 in ERAD of NHK is likely to be different from that of EDEM, since EDEM3 overexpression greatly stimulates mannose trimming of N-glycans from NHK, whereas overexpression of EDEM does not (10).
Furthermore, when EDEM3 is overexpressed, the pattern of oligosaccharides released from NHK is very different from that observed on NHK isolated from ER ManI-transfected cells. In cells overexpressing EDEM3, there is extensive trimming of N-glycans to Man 6 -7 GlcNAc 2 with relatively little Man 8 GlcNAc 2 (Fig. 6, A and  B), whereas in cells overexpressing ER ManI, there is increased accumulation of Man 8 GlcNAc 2 and Glc 1 Man 8 GlcNAc 2 concomitant with increased trimming to smaller oligosaccharides (10). Importantly, the effect of EDEM3 on the trimming of N-glycans is abolished by mutating the essential acidic residue Glu 147 to Gln (Fig. 6, A and B, E147Q), and the increased ERAD of NHK due to EDEM3 overexpression is greatly reduced by this mutation (Fig. 6D). This residue is found in the active site of processing Class I ␣-mannosidases by x-ray crystallography (44 -48) and is essential for enzyme activity (36). Although these results strongly indicate that EDEM3 stimulation of NHK ERAD is caused by its ␣1,2-mannosidase activity, an alternative less likely interpretation is that the mutation of Glu 147 to Gln affects the conformation of EDEM3 and thus abolishes the effect on NHK degradation independently of enzyme activity.
Since overexpression of EDEM3 stimulates mannose trimming from total glycoproteins as well as from the misfolded glycoprotein NHK (Fig. 6), EDEM3 is most likely acting as a processing ␣1,2-mannosidase in vivo, accelerating trimming of Man 8 GlcNAc 2 oligosaccharides to Man 6 -7 GlcNAc 2 . Its specificity appears to be different from that of ER ManI that greatly stimulates trimming to Man 8 GlcNAc 2 and Glc 1 Man 8 GlcNAc 2 (10). The present results indicate that Man 8 GlcNAc 2 is not an exclusive targeting signal for ERAD of glycoproteins and that smaller oligosaccharides (Man 5-7 GlcNAc 2 ) attached to misfolded glycoproteins participate in this recognition, in agreement with other studies (10,(17)(18)(19)49).
While this manuscript was in preparation, two groups reported studies on another EDEM homolog, which they named EDEM2 (25,26). Although EDEM2 stimulates glycoprotein ERAD, it has no effect on mannose trimming from misfolded glycoproteins, indicating that its mechanism of action is different from that of EDEM3 described in the present work. In both manuscripts, the existence of EDEM3 is mentioned, but its function is not further analyzed.
The mechanisms involved in ERAD is an area of active investigation at the present time, not only for fundamental cell biology but also for clinical applications, because ERAD is important in the pathogenesis of a large number of genetic diseases caused by protein misfolding. However, the mechanisms whereby the cell recognizes misfolded proteins and targets them to ERAD are not fully understood. Since earlier studies showed that ER ManI and EDEM both stimulate ERAD of glycoproteins, a relatively simple mechanism has been proposed whereby targeting of misfolded glycoproteins depends on Man8B formed by ERManI, which is then recognized by EDEM. However, it is clear from more recent studies and from the work presented here that the targeting for ERAD is far more complicated, since there are two additional EDEM proteins implicated, and trimming of oligosaccharides on misfolded glycoproteins to species smaller than Man8 occurs. Thus, the cloning and characterization of EDEM3 makes a novel contribution to the understanding of the quality control of misfolded proteins.