A technique for delineating the unfolding requirements for substrate entry into retrotranslocons during endoplasmic reticulum–associated degradation

The endoplasmic reticulum–associated degradation (ERAD) pathway mediates the endoplasmic reticulum–to–cytosol retrotranslocation of defective proteins through protein complexes called retrotranslocons. Defective proteins usually have complex conformations and topologies, and it is unclear how ERAD can thread these conformationally diverse protein substrates through the retrotranslocons. Here, we investigated the substrate conformation flexibility necessary for transport via retrotranslocons on the ERAD-L, ERAD-M, and HIV-encoded protein Vpu-hijacked ERAD branches. To this end, we appended various ERAD substrates with specific domains whose conformations were tunable in flexibility or tightness by binding to appropriate ligands. With this technique, we could define the capacity of specific retrotranslocons in disentangling very tight, less tight but well-folded, and unstructured conformations. The Hrd1 complex, the retrotranslocon on the ERAD-L branch, permitted the passage of substrates with a proteinase K–resistant tight conformation, whereas the E3 ligase gp78-mediated ERAD-M allowed passage only of nearly completely disordered but not well-folded substrates and thus may have the least unfoldase activity. Vpu-mediated ERAD, containing a potential retrotranslocon, could unfold well-folded substrates for successful retrotranslocation. However, substrate retrotranslocation in Vpu-mediated ERAD was blocked by enhanced conformational tightness of the substrate. On the basis of these findings, we propose a mechanism underlying polypeptide movement through the endoplasmic reticulum membrane. We anticipate that our biochemical system paves the way for identifying the factors necessary for the retrotranslocation of membrane proteins.

Nearly one-third of human proteins are secretory and membrane proteins that are synthesized at the endoplasmic reticu-lum (ER). 3 When misfolded or failed to be assembled into a complex, these proteins are retained within the ER and ultimately eliminated by a process termed ER-associated degradation (ERAD). ERAD is typically categorized into ERAD-L, ERAD-M, and ERAD-C branches for their preference of substrates with lesions from ER luminal, transmembrane, and cytosolic domains, respectively (1, 2). These branches are known to use different sets of factors to promote substrate destruction (3).
The mechanism of ERAD involves the retrotranslocation of substrates from the ER to the cytosol, where they are ubiquitinated, extracted from membrane by AAA ϩ ATPase p97, and degraded by 26S proteasome (4,5). Because the degradation machinery is mainly localized in the cytoplasm, all ERAD substrates must traverse the ER membrane to be relocated to the cytosol for their destruction. Therefore, a proteinaceous channel termed retrotranslocon should reside in the ER membrane to mediate retrotranslocation of soluble and membrane protein substrates into the cytosol (6). Many candidates were proposed to be the channels, including the transmembrane E3 ligase Hrd1. Recent studies have shown that misfolded CPY* can be retrotranslocated in a reconstituted liposome containing yeast Hrd1p (7,8). Cryo-EM structure revealed that the Hrd1p dimer shows a central cavity, likely the conduit for passing polypeptide (9). However, it is unclear whether Hrd1 is able to retrotranslocate the misfolded protein bearing a folded domain because these studies were performed using misfolded proteins. In mammalian cells, ubiquitin ligases Hrd1 and gp78 both show sequence homology to yeast Hrd1p within the transmembrane region. Hrd1 is strictly required for ERAD-L, whereas gp78 is mainly responsible for ERAD-M, especially for polytopic membrane proteins (10). TRC8, a transmembrane E3 ligase required for cytomegalovirus-encoded protein US2-mediated degradation of MHC-I heavy chain (HC), also shows similar conserved characteristics to yeast Hrd1p (11). There are suggestions that some other (up to ϳ20) ER-resident ubiquitin ligases in mammals could serve as retrotranslocons for poly-peptide movement through the membrane of respective subset of ERAD substrates on ERAD-L, -M, and -C branches of pathway (9,12).
Because folding is an error-prone process, most, if not all, newly synthesized secretory and membrane proteins have the potential to become substrates of ERAD. Therefore, ERAD must recognize and ultimately degrade a highly diverse range of substrates: not only accept individual proteins with different sizes but also handle potential diverse conformations when a protein is misfolded. Adding to the complexity is the fact that many substrates are modified with oligosaccharides in the ER lumen. It is largely unclear how limited number of potential retrotranslocons can handle substrates with potentially unlimited conformations. Specifically, the requirements on the size or conformation flexibility of substrates for threading through the pores of retrotranslocons in ERAD-L, -M, and -C pathways are largely unknown.
A well-characterized strategy to study protein translocation takes advantage of a conformation-tunable cytosolic enzyme, dihydrofolate reductase (DHFR). When bound to folate analogs such as methotrexate (MTX) or the membrane-permeable derivative trimetrexate (TMX) (13), DHFR adopts a very tight conformation that becomes resistant to proteinase K digestion, which generates a 21-kDa proteinase K-resistant fragment. Therefore, ligand-stabilized DHFR can be employed as a testing substrate to evaluate the pore size or the unfolding activity of translocation machineries. Studies using DHFR fusion proteins showed that preunfolded conformation is necessary for protein import into the ER, mitochondria, and lysosome (14 -16), indicating that related translocation machineries lack the capability to unfold DHFR in tight conformation and only allow the passage of at least partially unfolded peptides. However, the tightly folded DHFR-MTX complex can be imported across protein translocons present in the outer and inner chloroplast envelope membranes because of the large functional pore size with a diameter of 25.6 Å (17,18).
The destabilizing domains (DDs) derived from FKBP12 is another example whose conformation can be tuned by ligand binding (19,20). Binding to its ligand Shield-1 stabilizes the structure of several FKBP12 mutants (i.e. the DDs), which would otherwise show highly flexible conformation (20). Although these nearly unstructured FKBP12 mutants can serve as a degron for the ubiquitin-proteasome system (UPS) when present in the cytoplasm, they are very stable when confined within ER lumen (21). Thus, similar to DHFR/ligands, DD/Shield-1 pair could be used to study the conformation requirements for the entry into the pores of retrotranslocation machineries.
To get a comprehensive understanding as to how defective ER proteins with distinct conformations and topologies, such as soluble protein and single-spanning and multispanning membrane proteins, are retrotranslocated during ERAD, we harnessed both DHFR/TMX and DD/Shield-1 pairs to investigate the substrate unfolding requirements for threading through the retrotranslocons on ERAD-L, ERAD-M, and HIV-encoded protein Vpu-mediated ERAD pathways (22). This technique allows us to define the capacity of specific retrotranslocons in disentangling very tight, less tight but well-folded, and unstruc-tured conformations. Our results demonstrated that three retrotranslocon complexes showed distinct protein unfolding requirements for moving substrates across respective pores. Thus, different strategies may be used to accommodate or unfold substrates for entry into respective channels.

An experimental system to report unfolding requirement in cultured cells
DHFR, when fused in-frame to the MHC-I heavy chain (DHFR-HC), can fold properly in the lumen of the ER and bind TMX, a membrane-permeable folate analog (23). We adopted a similar strategy to append DHFR to the ER luminal region of a serial of ERAD substrates that are degraded through ERAD-L or Vpu-hijacked ERAD pathways (Fig. 1A). These included null Hong Kong ␣1-antitrypsin (NHK) and a truncated version of CD4 (mini-CD4, mCD4), which is degraded from the ER by Vpu-hijacked ERAD (24,25). DHFR-fused GFP u containing a C-terminal CL1 degron was used as a control substrate in the cytoplasm (26).
We then confirmed that the TMX was able to bind the DHFR moiety of substrates in cultured cells using the proteinase K digestion assay (23) (Fig. 1B). Treating total cell lysates with proteinase K results in the generation of a ϳ21-25-kDa DHFR fragment, indicating that the DHFR portions of these substrates bind the TMX and show tightly folded conformation ( Fig. 1, C-E). No postlysis binding of TMX to DHFR was observed in control experiments (Fig. 1, B-E, sample 5). This result is consistent with a previous report (23) indicating that the TMX is able to form complex with the DHFR moiety fused with various substrates localized in cytoplasm as well as ER lumen. Quantification of these results indicates that at least 90% of DHFRfused substrates were bound with TMX in cultured cells (Fig.  1F). Thus, TMX binding to DHFR significantly improved substrate solubility in buffers containing mild detergents (Fig. 1G).

UPS is able to eliminate DHFR-GFP u bound with folate analogs
We made HEK293T cells stably expressing DHFR-GFP u and asked whether DHFR-GFP u could be degraded by UPS in the presence of folate analogs. Significant accumulation of DHFR-GFP u in cultured cells was observed when the cells were treated with proteasome or p97 inhibitors (MG132 and CB5083) (27,28), whereas the addition of TMX or MTX had little effect ( Fig.  2A). In addition, the translational arrest by cycloheximide resulted in degradation of DHFR-GFP u that could be impeded by the treatment of either MG132 or CB5083 but not folate analogs (Fig. 2B). These results revealed that DHFR bound with folate analogs, although resistant to proteinase K, could be degraded by protease activities of 20S proteasome and suggested that the AAA-ATPase activity in the 19S proteasome might disentangle the tightly folded conformation of DHFR charged with folate analogs (29).
It has been reported that in the presence of MTX, CL1 degron-appended DHFR (DHFR u ) is not able to be degraded by purified proteasome or in crude rabbit reticulocyte lysate (30,31) because of the lack of a flexible region of peptide chain required for a "tight binding" to proteasome (32). Here, the additional GFP moiety besides DHFR could provide sufficient Unfolding requirements for substrates retrotranslocation flexible regions for the recognition of DHFR portion by the 19S proteasome. Thus, we concluded that once retrotranslocated from the ER, DHFR-linked ERAD substrates can be degraded even in the presence of folate analogs, unless the proteasome activity is blocked.

The ERAD-L pathway is compatible for the retrotranslocation of substrates with a folded domain
To test whether TMX-occupied DHFR could traverse the ER membrane through ERAD-L pathway, we compare the turnover rate of DHFR-NHK in the absence or the presence of TMX. DHFR-NHK were rapidly degraded after blocking ribosomal translation by CHX (Fig. 2, C and D). However, TMX did not slow down degradation speed of substrates from the ER. In contrast, MG132 and CB5083 significantly extended the halflives of substrates, suggesting that the degradation of these ERAD-L substrates depends on both p97 ATPase and proteasome. Notably, TMX somehow enhanced turnover rate of glycosylated DHFR-NHK, whereas cotreatment with MG132 and CHX led to the accumulation of a deglycosylated DHFR-NHK, which represents substrates retrotranslocated because of the removal of glycan by cytosolic N-glycanase (33) (Fig. 2C). These results demonstrated that TMX-charged DHFR is able to be retrotranslocated across ER membrane via the ERAD-L pathway and ultimately degraded by the proteasome in the cytosol.
We then asked whether retrotranslocated DHFR in cytoplasm maintained in the native-like conformation. We used a deubiquitinase from the Ebola virus (EBV-DUB), which could inhibit proteasomal degradation while permitting efficient retrotranslocation when expressed in cultured cells (34). The expression of EBV-DUB led to stabilization of the retrotranslocated DHFR-NHK in the cytoplasm, as shown by the appearance of the deglycosylated band (Fig. 2E, lanes 3, 5, 7, and 9). Proteinase K digestion for DHFR-NHK and DHFR-GFP u showed that MTX only bound with DHFR localized within the cytosol but not that of ER lumen ( Fig. 2E; compare the proteinase K-resistant DHFR in lanes 2 and 4). However, the proteinase K-resistant DHFR band became evident concomitant with Figure 1. TMX binds to the DHFR moiety of various substrates within the ER lumen. A, schematic representation of ERAD substrates fused with DHFR used in this study. NHK is an ERAD-L substrate. Newly synthesized CD4 can be regarded as an ERAD-C substrate in the presence of HIV-encoded protein Vpu. Cytosolic UPS substrate DHFR-GFP u is used as a control. B, schematic representation of the protocol used to detect TMX-bound DHFR. Cells expressing DHFR-fused substrates (DHFR-sub.) were incubated with TMX or DMSO for 4 h, lysed, and treated with the proteinase K (PK) as indicated by the protocol. The indicated samples were analyzed by Western blotting. C-E, binding analyses of TMX to the DHFR moieties of DHFR-GFP u (C), DHFR-NHK (D), and DHFR-mCD4 (E) by immunoblotting (IB) with anti-DHFR antibody. Note that endogenous DHFR (Endo.) is detected, whereas the migration of proteinase K-resistant DHFR (PK res.) is slightly slower than that of endogenous DHFR on the gel. The experiments were performed three times independently with similar results. F, quantify the binding efficiency of TMX to DHFR portion in various ERAD substrates in cells. The bands in samples 1 and 2 in C-E were quantified by densitometry. A percentage of the DHFR that binds to TMX was calculated by dividing the intensity of the 21-kDa proteinase K-resistant fragment in sample 2 by the intensity of the nondigested protein in sample 1. The error bars stand for S.E. from three independent experiments. G, solubility analysis of various substrates in the presence or absence of TMX. Cells expressing the indicated substrates were treated with or without TMX and then lysed with buffers containing sodium deoxycholate (radioimmune precipitation assay buffer (RIPA)), Triton X-100 (TX100), or SDS. After centrifugation to remove insoluble proteins, clarified lysates were subjected to immunoblotting analysis for levels of indicated proteins. Endogenous BAG6 was blotted as a loading control. The asterisk in the top panel indicates nonspecific bands detected by the ␣-GFP antibody. Each substrate was quantified by densitometry and normalized to BAG6 levels.

Unfolding requirements for substrates retrotranslocation
the appearance of retrotanslocated and deglycosylated DHFR-NHK (Fig. 2E, lanes 5 and 6). This result indicates that the retrotranslocated DHFR when in binding with MTX may still maintain the native-like conformation.
Our results on the retrotranslocation of DHFR-NHK are consistent with that of DHFR-HC in the presence of US2, a cytomegalovirus-encoded adaptor protein for mediating MHC-I heavy chain ubiquitination and retrotranslocation (23). The degradation of DHFR-NHK strictly depends on the Hrd1 E3 ligase activity (Fig. S1, A and B) (35), a highly potential retrotranslocon in mammalian cells, because its yeast counterpart, Hrd1p, has been shown to be a retrotranslocation channel (7-9). Because the misfolded NHK functions as a degradation signal, the retrotranslocation should be initiated on at least par-tially unfolded structure of NHK moiety. The capability of TMX-charged DHFR to transverse the membrane suggests that once the Hrd1 retrotranslocon is opened for initiating the passage of the leading NHK moiety, the trailing 21-kDa TMX-bound DHFR is able to enter the channel and finally cross the membrane. However, according to the cryo-EM structure, yeast Hrd1p shows a narrow central channel that is closed at the luminal side (9). It is possible that the opened Hrd1 channel and/or Hrd1-associated factors such as Derlins could be dynamic to accommodate 21-kDa DHFR for successful retrotranslocation of ERAD-L substrates. After retrotranslocation, p97 might unfold and extract DHFR from the rigid DHFR-TMX complex, thread the flexible DHFR peptide through the p97's central pore, and release it into the cytosol (36,37). The DHFR polypeptide could be partially renatured to The error bars stand for the S.E. from four independent experiments. B, cells stably expressing DHFR-GFP u were treated with CHX plus DMSO (as a control) or indicated chemical compounds. At the indicated time points, cell lysates were analyzed by immunoblotting for levels of DHFR-GFP u and BAG6 (loading control). This experiment was performed three times independently with similar results. The asterisks denote nonspecific bands reacting with ␣-GFP antibody. Bottom panel, quantification of DHFR-GFP u in the presence of indicated compounds. DHFR-GFP u level in cells without treatment was set as 100% at 0 h. The error bars are the S.E. from three independent experiments. C, 293T cells expressing DHFR-NHK were incubated with CHX plus TMX, MG132, CB5083, or DMSO where indicated. At indicated time points, the cell lysates were analyzed by immunoblotting for the levels of indicated proteins. Arrowhead in a long exposure (exp.) image indicates deglycosylated DHFR-NHK, a hallmark of retrotranslocation. D, quantification of DHFR-NHK in the presence of indicated compounds shown in C. The error bars are the S.E. from three independent experiments. E, analysis of binding of MTX to retrotranslocated DHFR-NHK in the cytoplasm. The cells were transfected with DHFR-GFP u or DHFR-NHK. Where indicated, deubiquitinase EBV-DUB (or empty vector as a control) was cotransfected with DHFR-NHK. These proteins were expressed for 24 h in cells, and then either MTX or TMX was added to incubate for 4 h before harvesting using a buffer containing 1% Triton X-100. The clarified detergent lysates were then either left untreated or treated with proteinase K, followed by immunoblot analysis with ␣-DHFR antibody. Black arrowheads denote deglycosylated retrotranslocated substrate, whereas the red arrowhead represents proteinase K-resistant (PK resis.) DHFR bound with MTX. Deglyc., deglycosylated; Glyc., glycosylated; no transf., no transfection; endo., endogenous DHFR.

Unfolding requirements for substrates retrotranslocation
restore its native-like conformation before the degradation by proteasome.

Unfolding of luminal domain is a prerequisite for Vpu-mediated CD4 degradation
To investigate how ER luminal domain would affect retrotranslocation of an ERAD-C substrate, we utilized the Vpumediated ERAD system (22). When inserted into the ER membrane of host cells, Vpu is phosphorylated and associates with E3 ligase SCF ␤TrCP to form a tightly bound complex anchored on ER membrane (Fig. S2A). The Vpu-ligase complex mediates degradation of many newly synthesized membrane proteins, including BST-2, PSGL-1, and CD4, by an ERAD-like pathway (Fig. S2B) (38 -41).
Vpu-ligase recognizes CD4 primarily via CD4 cytosolic tail and catalyzes polyubiquitination on the lysine residues in the cytosolic tail (25,42,43). Thus, CD4 can be considered as an ERAD-C substrate. In accordance with this, we observed that Vpu induces highly efficient degradation of various luminal domain mutated CD4, including DHFR-mCD4 in which DHFR was substituted for D1-D3 domains of CD4 ( Fig. 3A and Fig. S2, C-F). Notably, CD4 degradation is independent of Hrd1 and gp78, as evidenced by efficient retrotranslocation and degradation of CD4 variants in Hrd1 and gp78 double KO cells (Fig. S2,  B, E, and G).
We then asked how conformational change would affect substrate retrotranslocation and degradation. To address this, we enhanced conformational rigidity of DHFR portion of DHFR- A, schematic diagram of factors involved in Vpu-mediated ERAD of DHFR-mCD4, which represents the DHFR fused mini-CD4 (mCD4). B, a CHX chase experiment to analyze DHFR-mCD4 degradation. Cells cotransfected with plasmids encoding Vpu and DHFR-mCD4 were treated with TMX, MG132, or CB5083 for 1 h before coincubation with CHX for indicated times. The cell lysates were analyzed by immunoblotting (IB) for the levels of DHFR-mCD4, Vpu, and BAG6 (loading control). Upon MG132 treatment, faster migration species were observed and indicated as deglycosylated substrate. Note that Vpu is phosphorylated (Vpu-P), evidenced by its slower migration on the gel. The result is the representative of three independent experiments. C, quantification of results in B. The error bars are the S.E. from three independent experiments. D, DHFR-mCD4 retrotranslocation in cells. DHFR-mCD4 and WT Vpu (WT) or Vpu-S52N/S56N phosphorylation mutant (SN) were expressed in 293T cells for 24 h. The cells were then treated with MG132 (4 h), CB5083 (4 h), or TMX (5 h). For TMX and MG132 cotreatment, the cells were pretreated with TMX for 1 h before coincubation with additional MG132 for 4 h. Cell lysates were analyzed by immunoblotting for levels of indicated proteins. Retrotranslocated DHFR-mCD4 appeared as a deglycosylated species upon MG132 treatment. This experiment was performed three times independently with similar results. E, analysis of topology of DHFR-mCD4 after TMX treatment. The cells expressing Vpu and DHFR-mCD4 were treated with TMX for 4 h before harvesting and microsome preparation. Crude microsomes were resuspended in a buffer with or without Triton X-100. They were then either left untreated or treated with proteinase K (PK) before denaturation with boiling SDS sample buffer. Samples were analyzed by immunoblotting with ␣-DHFR antibody. F, DHFR-mCD4 was coexpressed with HA-tagged WT or SN Vpu, treated with TMX for 4 h, and analyzed by immunoblotting directly (top) or after IP with ␣-HA antibodies (bottom). G, DHFR-CD4 ubiquitination in cells. DHFR-mCD4 (FLAG tag) and HAubiquitin were cotransfected with or without Vpu, treated with indicated combinations of chemicals for 4 h, subjected to denaturing IP with ␣-FLAG, and analyzed by immunoblotting for HA-ubiquitin and DHFR-mCD4. The numbers of HA-ubiquitins attached on FLAG-tagged DHFR-mCD4 are indicated. This experiment was performed twice independently with similar results. deglyc., deglycosylated; glyc., glycosylated; P.F., protected fragment.

Unfolding requirements for substrates retrotranslocation
mCD4 by charging it with TMX. The CHX chase analysis revealed a rapid degradation of DHFR-mCD4, whereas treatment with the p97 inhibitor CB5083 blocked substrate turnover from the ER membrane and that of MG132 stabilized deglycosylated retrotranslocated intermediate within the cytoplasm (Fig. 3, B and C). We then found that addition of TMX, similar to that of CB5083, significantly slowed down the turnover rate of glycosylated DHFR-mCD4, resulting in stabilization of the glycosylated form of DHFR-mCD4 ( Fig. 3D and Fig. S2, F and  H). These results implied that binding of TMX with DHFR inhibited the retrotranslocation of the DHFR moiety. Furthermore, cells treated with both TMX and MG132 showed a significantly decreased amount of retrotranslocated substrate compared with MG132 treatment alone (Fig. 3D, compare  lanes 3 and 6), strongly suggesting that TMX, similar to that of CB5083, can block the retrotranslocation of CD4 (Fig. S2H,  compare lanes 3 and 4). Consistent with this, membrane protection experiment showed that the DHFR portion indeed bound tightly to TMX and resided within the ER lumen (Fig.  3E).
Additional coimmunoprecipitation result showed that DHFR-mCD4 maintains interaction with Vpu and its phosphorylation mutant Vpu-SN in the presence of TMX (Fig. 3F). Accordingly, TMX treatment allowed Vpu-mediated ubiquitination of DHFR-mCD4 (Fig. 3G). These results clearly showed that TMX did not impair substrate interaction with E3 or substrate ubiquitination. Thus, it was the retrotranslocation process that was blocked by the rigid structure of DHFR formed upon binding to TMX.
Our results demonstrate the conformation-dependent retrotranslocation and suggest that Vpu-mediated retrotranslocation route is able to fully unfold the ligand-free DHFR but not the TMX-occupied DHFR. Although both NHK and CD4 degradation depend on the activity of p97 ATPase, which is the only known source of mechanical force for extracting substrate from the ER, the two substrates show totally different behaviors during retrotranslocation, suggesting that different substrates use distinct retrotranslocons and/or strategies for moving substrates across membrane (44). This was supported by the fact that NHK degradation is strictly dependent on Hrd1, a proteinconducting channel in mammalian cells, whereas Vpu-mediated CD4 degradation remains efficient in Hrd1/gp78 DKO cells (Figs. S1A and S2B).

TMX entraps the retrotranslocating intermediate of DHFR-mCD4 in the ER membrane
Because TMX only binds to the DHFR portion fused with mCD4, we reasoned that the D4 domain could be allowed for retrotranslocation in the presence of TMX (Fig. 4A). To detect the retrotranslocated species, a BioTag (45) was inserted into a loop within the D4 domain ( Fig. 4A and Fig. S3A). We predicted that expression of a cytosolic BirA (BirA cyto ) would lead to biotinylation of the BioTag transported into cytosolic compartment (Fig. 4A).
Indeed, we observed biotinylated DHFR-Bio-mCD4 in the crude cell lysate when cells coexpressing BirA cyto , Vpu, and DHFR-mCD4 were treated by TMX (Fig. 4B). The biotinylated species could be captured by denaturing immunopre-cipitation for substrate, suggesting that the biotin was indeed covalently conjugated onto the DHFR-mCD4 per se (Fig. 4B). The appearance of biotinylated substrate was strictly dependent on TMX treatment and expression of BirA cyto , suggesting that the BioTag on the luminal D4 domain was transported across the ER membrane and reached the cytoplasm. In addition, protease digestion of microsomes containing biotinylated substrates showed that membrane-protected fragment was not modified with biotin, confirming that the biotinylation only occurred in the cytoplasm where BirA cyto was expressed (Fig. 4C).
Previous work showed that inhibition of proteasome by MG132 stabilized retrotranslocated CD4 (46,47), which migrated faster on SDS-PAGE because of glycan removal by cytosolic N-glycanase (Figs. 3B and 3D). As expected, the retrotranslocated deglycosylated substrate was biotinylated by the cytosolic BirA when cells were treated with MG132 (Fig. 4D).
To test whether a BioTag before DHFR can be biotinylated, we inserted the BioTag into DHFR-mCD4 between the FLAG tag and DHFR to make Bio-DHFR-mCD4 (Fig. 4E). We found that the BioTag cannot be modified with biotin in the presence of TMX (Fig. 4F), suggesting that the BioTag and the TMXoccupied DHFR domain may not be retrotranslocated out of the membrane. In contrast, inhibition of the proteasome by MG132 leads to accumulation of biotinylated Bio-DHFR-mCD4, which is also deglycosylated, suggesting that retrotranslocated substrate species is stabilized in the cytosol and thus modified with biotin by the BirA cyto .
We noticed that, in the presence of TMX, the majority of biotinylated substrate fully retained the oligosaccharide chain (Fig. 4, D and G). These results suggested that although the BioTag adjacent to the TMD has been exported out of the membrane, the majority of the glycan moiety has not yet reached the cytosol. It was probably still retained within the ER lumen as depicted in the Fig. 4A or inserted into the lipid bilayer. The width between both phosphate heads of a typical lipid bilayer is ϳ5 nm, whereas the distance from last amino acid of DHFR to the ␣-carbon of the Asn-325 may be not sufficient to span the membrane (Fig. S3A). Alternatively, it is also possible that the oligosaccharide chain does reach the cytosol but not far enough from the membrane surface to be accessed by the N-glycanase, which is known to associate with p97 complex (48). Nevertheless, our results provide a snapshot catching the moment that DHFR-mCD4 was trapped in a retrotranslocating channel in the presence of TMX.
To determine whether the partially retrotranslocated species were integral within membrane, we isolated crude microsomes from TMX-treated cells and performed the alkaline wash assay. The membrane-associated p97 and luminal chaperone calreticulin were removed from microsomes by the alkaline treatment, whereas the biotinylated substrate, similar to Calnexin, still remained associated with the membrane (Fig. 4H). Thus, the biotinylated and glycosylated DHFR-mCD4 is likely spanning the membrane lipid bilayer with the BioTag and the glycan moieties divided by the membrane, representing a bona fide intermediate undergoing retrotranslocation. The residues in the D4 domain are rather hydrophilic and could not favor the hydrophobic environment within lipid bilayer (Fig. S3B). Thus, 293T cells coexpressing Vpu and FLAG-tagged DHFR-Bio-mCD4 with or without BirA cyto were either left untreated or treated with TMX before harvesting. The cell lysates were analyzed by immunoblotting directly or after denaturing IP with ␣-FLAG antibody. The asterisk indicates nonspecific bands that are used as a loading control. This experiment was performed three times independently with similar results. C, analysis of biotinylation analysis of status of full-length (FL) DHFR-Bio-mCD4 and the membrane-protected fragment (PF). Crude microsomes were isolated from 293T cells coexpressing Vpu, DHFR-Bio-mCD4, and BirA cyto with TMX treatment for 4 h and were subjected to digestion with proteinase K (PK) in the absence or presence of detergent (Deter.). The products were immunoprecipitated using antibodies against N-terminal FLAG tag and analyzed by blot with ␣-FLAG antibody and HRP-conjugated avidin. The asterisks indicate nonspecific bands. D, cells coexpressing Vpu, DHFR-Bio-mCD4, and BirA cyto were treated with TMX, MG132, or DMSO (control). Cell lysates were analyzed as in B. The asterisk indicates nonspecific bands that are used as a loading control. This experiment was performed three times independently with similar results. E, diagram of potential forms of Bio-DHFR-mCD4 in which the BioTag is inserted between FLAG tag and the DHFR moiety of the substrate. F, analysis of biotinylation of Bio-DHFR-mCD4 with indicated treatment. Cell lysates were analyzed by Western blotting. The experiment was repeated for two times with the same result. G, analysis of glycosylation status of biotinylated DHFR-Bio-mCD4. Crude microsomes prepared in C were solubilized with detergent and subjected to digestion with endoglycosidase H (EndoH) or peptide:N-glycosidase F (PNGaseF). The resulting materials were analyzed by Western blotting with indicated reagents. H, alkaline wash of microsome containing biotinylated DHFR-mCD4. Microsomes prepared in C were resuspended in low salt buffer, 0.5 M NaCl, 0.1 M NaHCO 3 (pH 11), or 1% SDS. The mixtures were subjected to ultracentrifugation, and the supernatant (S) and pellet (P) fractions were analyzed by blot with indicated antibodies or HRP-conjugated avidin. The total lysate from the cells coexpressing Vpu, DHFR-Bio-mCD4, and BirA cyto and treated with MG132 was loaded in the last lane to show the position of deglycosylated DHFR-Bio-mCD4 on the gel. This experiment was performed three times independently with similar results. deglyc., deglycosylated; glyc., glycosylated.

Unfolding requirements for substrates retrotranslocation
we propose that the intermediate may be integrated into the ER membrane by forming a complex with retrotranslocation machineries. In fact, TMX induced massive accumulation of ubiquitinated DHFR-mCD4 on the ER membrane (Fig. 3G, lane  3), which is probably due to the blockade of potential retrotranslocon by the stalled translocating intermediate. The results in Figs. 2-4 revealed that the mechanisms of the Hrd1and Vpu-mediated retrotranslocation routes are distinct, because TMX-occupied DHFR is able to move through the former but not the latter.

Well-folded structure within ER lumen impairs multispanning membrane protein retrotranslocation and degradation
To test how the folding status of ER luminal region affects retrotranslocation and degradation of multispanning membrane proteins, we used a truncated TAP2 (mini-TAP2, mTAP2) that contains four N-terminal transmembrane segments from polytopic membrane protein TAP2 (Fig. 5A) (49). Thus, mTAP2 has an aberrant transmembrane region and can be considered as an ERAD-M substrate. When expressed in cells, its turnover is predominantly dependent on gp78 E3 ligase (Fig. S4A). The DHFR and a destabilizing domain (DD) of FKBP12 mutant V4A/I7V/F36V (20) were added to the N termini to make DHFR-mTAP2 and DD-mTAP2, respectively (Fig. 5A).
To our surprise, the expression of DHFR-mTAP2 led to massive accumulation of ubiquitinated species with a much slower turnover rate than that of mTAP2 (Fig. 5, B and C). This result clearly demonstrates that the well-folded DHFR domain appended to the luminal region of mTAP2 impairs retrotranslocation and degradation.
The FKBP12-derived DD is intrinsically flexible, but binding to its ligand Shield-1 induces formation of a compact and at least partially folded structure (20). A control experiment showed that the DD with the C-terminal ER-retaining motif KDEL (DD-KDEL) was stable when expressed in ER (Fig. 5D), consistent with previous report that targeting DD to the ER lumen leads to accumulation even in the absence of Shield-1 (21). These results indicated that ER luminal DD itself may not act as a degron. We then found by CHX chase that DD-mTAP2 was rapidly degraded when expressed in WT 293T cells but significantly stabilized in gp78 KO cells (Fig. 5E). Incubation with either MG132, CB5083, or Shield-1 stabilized DD-mTAP2 as well (Fig. 5, F and G, and  Fig. S4B). Incubation with Shield-1 did not change the ubiquitination level of DD-mTAP2 (Fig. 5H), suggesting that Shield-1 may impair postubiquitination steps of ERAD such as retrotranslocation. Consistently, treatment with CB5083 or Shield-1 inhibited MG132-induced deglycosylation of DD-mTAP2, a hallmark of retrotranslocation (Fig. 5I). Therefore, Shield-1 slowed down substrate turnover mostly by inducing formation of a folded structure in DD, which in turn impairs mTAP2 retrotranslocation. This result implicates that unfolding of the luminal part on multispanning membrane protein might be necessary for the retrotranslocation.

Discussion
Although many factors required for ERAD have been identified, substrate configuration before and during movement through the pore of a retrotranslocon remains largely obscure. In the current study, we systematically measured the capacity of three ERAD pathways in unfolding substrates for successful ER-to-cytosol transport. The most interesting finding lies in gradual decreases in the capacities of ERAD-L, Vpu-hijacked ERAD-C, and ERAD-M in unfolding of respective substrates for retrotranslocation (Fig. 6).
The Hrd1-mediated retrotranslocation route in the ERAD-L allows the passage of 21 kDa TMX-charged DHFR. Similar results have been described for protein import through membrane of chloroplast envelope (17) and US2-mediated MHC-I heavy chain retrotranslocation, for which TRC8 E3 ligase was shown to be required (11,23). These results suggested that the pore size of related translocation machineries (e.g. Hrd1 and Derlins) was large enough to accommodate the DHFR in complex with stabilizing ligand. However, cryo-EM structure showed a narrow pore within Hrd1 transmembrane region (9), suggesting that substrate unfolding might be required for threading; this led us to propose that Hrd1, TRC8, and/or its cofactors in ERAD-L pathway might have strong unfoldase activity to disentangle the TMX-charged DHFR to thread the polypeptide into the channel. It is conceivable that the unfolding steps require energy and the Derlins and associated p97 ATPase could be involved. The mechanism on the substrate unfolding before entering the channel requires further investigations.
Contrary to ERAD-L substrates in which retrotranslocation initiates before ubiquitination, ERAD-M and -C substrates, such as mTAP2 and CD4 in this study, are believed to be ubiquitinated on cytosolic lysine residues before being extracted from the membrane. Retrotranslocation factors in the Vpuhijacked pathway, however, could unfold well-folded domains such as D1-D4 domains of CD4, superfolder GFP, and the ligand-free DHFR, but not the TMX-occupied DHFR. In fact, TMX-stabilized DHFR inhibits retrotranslocation probably by docking on the entrance of potential retrotranslocon and causing steric hindrance, as evidenced by the retrotranslocating intermediate integrated into lipid bilayer (Fig. 4H). We surmise that the peptide sequence of the intermediate trapped within lipid bilayer might bind to the potential retrotranslocon stoichiometrically. This potential channel is not Hrd1 or gp78 because they are not required for Vpu-mediated CD4 retrotranslocation and degradation (Fig. S2, B, E, and G).
From current and earlier studies (23), we conclude that there may be three types of retrotranslocation channels for ERAD substrates with distinct properties (Fig. 6). The first type includes Hrd1-Derlins complex and probably TRC8 that function in ERAD-L and mediate retrotranslocation of NHK and MHC-I heavy chain, respectively (11). The channels could either be very dynamic to accommodate a 21-kDa globular protein or have strong unfoldase activity capable to disentangle a tightly folded structure like TMX-bound DHFR. The second type of channel, hijacked by Vpu, permits retrotranslocation of well-folded structure, such as superfolder GFP, DHFR, and Unfolding requirements for substrates retrotranslocation luminal domain of CD4, but does not allow structurally tight DHFR to enter. The threshold of substrate size for entering the channel is currently unknown. The third type of channel, specifically targeting multispanning membrane proteins, is very stringent in terms of substrate flexibility, because even well-folded domains, such as the DHFR and the DD bound with Shield-1, inhibit retrotranslocation. Only disordered peptides such as ligand-free DD are allowed to cross-membrane. These results indicate that at least partial unfolding of luminal domain is required for entering and traversing the last two types of

Unfolding requirements for substrates retrotranslocation
channels. To what extent the luminal part should be unfolded is currently unknown, and fully understanding the question awaits the advent of more quantitative method to precisely measure the extent of protein folding and maturation within ER lumen. Nevertheless, we speculated that an unfoldase might be necessary for ERAD-M substrate retrotranslocation, not only for CD4, but also for polytopic membrane proteins such as TAP2. Prominent candidates include Torsin/Lull1 that is an ER-localized AAAϩ ATPase complex and has been shown to be involved in CFTR degradation (50). Whether and how Torsins and Torsin-interacting proteins are involved in ERAD-M of other substrates requires future studies.
For polytopic membrane protein like mTAP2, our results suggest that multiple transmembrane helices are likely segregated into individual segments or subgroups and then threaded through the channel in a sequential order. It is very unlikely that the channel allows en bloc extraction of a whole bundle of transmembrane helices, because only unstructured DD enters the retrotranslocation route. The FKBP12-derived DD used here contains 107 amino acids, which might show a small globular shape when fixed by Shield-1 according to the structure of the FKBP12-rapamycin complex (51).
In summary, our results provide insights into various routes for substrate retrotranslocation and point out that substrates with diverse sizes or conformational properties may require different types of channel and modes of actions for successful ER-to-cytosol transport. Our biochemical system to capture the retrotranslocating intermediate paves the way to purify and identify the retrotranslocation channel and other factors in the HIV-mediated ERAD pathway.

Plasmid, antibody, and other materials
Constructs encoding Vpu, CD4, BirA cyto , and EBV-DUB have been described (34,45,46). The coding sequence for DHFR and mTAP (amino acids 1-187 of TAP2) was amplified from pCB268 (Addgene plasmid 53363, a gift from Dr. Alexander Varshavsky) and pCDNA5/FRT/TO-TAP2-HA (a gift from Dr. Mals Mariappan), respectively. DNA sequence encoding FKBP12 was amplified from RT-PCR products of total RNA extracted from 293T cells and cloned into pCDNA5/FRT/TO vector carrying a C-terminal FLAG tag. Coding regions of null Hong Kong ␣1-antitrypsin (NHK) were amplified from pCDNA-A1AT-NHK-Venus (a gift from Dr. Mals Mariappan) and cloned into pCDNA5/FRT/TO vector, with a C-terminal FLAG tag. Destabilizing domain (20) and superfolder GFP (52) were made by site-directed mutagenesis using FKBP12-FLAG, NHK-FLAG, and EGFP sequences as templates, respectively. For chimeric constructs shown in Figs. 1A and 5A, Gibson  A strong unfoldase activity might disentangle tightly folded structure in substrates (e.g. TMX-occupied DHFR). Alternatively, the pore of retrotranslocation complex could be large enough to accommodate TMX-bound DHFR. Vpu-mediated protein degradation exploits a currently unidentified retrotranslocation channel (Vpu-hijacked ERAD-C). For moving substrate through membrane, the channel and/or cofactors are able to unfold the DHFR fused to the substrate but not the TMX-bound DHFR. Luminal region of polytopic membrane proteins must be in a flexible conformation to be threaded through the potential channel; the channel or channel-associated cofactors may not disentangle well-folded domains such as DHFR or Shield-1-occupied destabilizing domain (ERAD-M).

Unfolding requirements for substrates retrotranslocation
assembly was used to fuse linearized pCDNA5/FRT/TO, Prolactin signal sequence, DHFR, or DD, and corresponding DNA sequences were amplified as described above. The Prolactin signal peptide at the N terminus directs these molecules to the ER. Plasmid expressing HA-ubiquitin was kindly provided by Dr. Yanfen Liu. All constructs were verified by DNA sequencing.

Cell culture and transfection
293T and HEK293T Flp-In (Invitrogen) cell lines were cultured in high-glucose Dulbecco's modified Eagle's medium (Thermo Fisher Scientific) containing 10% FBS (Hyclone, GE Healthcare) at 5% CO 2 . Transfections with plasmids were performed with Lipofectamine 2000 according to the manufacturer's suggestions. HEK293T Flp-In cell line stably expressing DHFR-GFP u were made by cotransfection of pCDNA/FRT/TO-DHFR-GFP u and pOG44 vector (Invitrogen), followed by screening using hygromycin (100 g/ml) until colonies appeared.

In vivo TMX binding
To measure the extent to which ligands (TMX or MTX) bind to DHFR appended to various protein substrates in cultured cells, we used a reported method with modifications ( Fig. 1B) (23). Typically, cells expressing substrates were treated with 100 M ligand for 4 h and then washed with cold PBS twice before harvesting. The cells were lysed with buffer containing 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 0.5% Triton X-100. The lysate was clarified by centrifugation, and the supernatant was added by either proteinase K or water (as a control) and incubated for 1 h on ice. To measure maximum binding, clarified cell lysate was incubated with 100 M TMX 10 min before digestion with the proteinase K. The reactions were terminated by 2 mM phenylmethylsulfonyl fluoride and subsequently added into boiling sample buffer with continued incubation at 95°C for 3 min. To measure postlysis binding of DHFR with TMX, cells expressing empty vector were treated with TMX, whereas those expressing substrates were treated with DMSO. The cells were then washed with cold PBS for three times. Equal amounts of cells from two treatments were combined, lysed, and clarified as mentioned. The resulting supernatant was treated directly with either 0.5 mg/ml proteinase K or water (control), as described above.

Cycloheximide chase
To measure substrate turnover rate, we performed translational shut-off assay by treating cell with CHX. Except DHFR-GFP u that was stably expressed in HEK293 Flp-In cell line, most degradation substrates were expressed by transfection in 293T cell for 24 h, unless otherwise indicated. The cells were then incubated with 100 g/ml CHX for various times as shown in the figures. At each time point after adding CHX, the cells were directly harvested with 2ϫ SDS-PAGE sample buffer and boiled before loading. To measure how small molecule inhibitors impinge upon protein degradation, the cells were preincubated with individual compound for initial 1 h, unless otherwise indicated, followed by CHX and compound cotreatment for indicated times. DMSO was used as a control.

Immunoprecipitation
Native immunoprecipitations in Fig. 3F were performed as described (46) with some modifications. Basically, 24 -48 h after transfection, the cells expressing the indicated proteins were harvested and incubated in 1 ml of IP buffer containing 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.5% Triton X-100, and protease inhibitor mixture. The lysate was then centrifuged at 21,000 ϫ g for 20 min. The resulting supernatant was incubated with 10 l of agarose conjugated with anti-HA antibody for 2 h. The beads were washed three times with IP buffer and eluted with 50 l of 2ϫ SDS-PAGE sample buffer.
For analysis of biotinylated and ubiquitinated products in Figs. 3-5, the cells were lysed in denaturing buffer. Briefly, cell expressing the indicated constructs for 24 -48 h were washed with cold PBS; lysed by addition of 100 l of lysis buffer containing 100 mM Tris-HCl, pH 8.0, and 1% SDS; and immediately denatured by boiling. The resulting lysate was diluted with 1 ml of IP buffer containing 50 mM Tris-HCl, pH 7.5, 150 mM NaCl, and 1% Triton X-100. The diluted lysate was spun down to remove aggregates, and the supernatant was subjected to IP as described above.

Microsome preparation and alkaline treatment
Crude microsome was isolated from cultured cells as described (46). Briefly, cells expressing BirA cyto , Vpu, and DHFR-Bio-mCD4 for 24 h were washed once while on a 10-cm dish and then collected by pipetting in cold PBS. Cell pellet was resuspended in 1 ml of hypotonic buffer (10 mM HEPES, pH 7.4, 250 mM sucrose, 2 mM MgCl 2 , plus 1ϫ protease inhibitor mixture) and homogenized by repeated passage through a 23-gauge needle. Nuclei and unbroken cells were pelleted by centrifugation at 2000 ϫ g for 20 min. The supernatant was then collected and centrifuged again to remove nuclei thoroughly. The resulting supernatant was centrifuged at 100,000 ϫ g for 60 min at 4°C. The pellet was resuspended and gently homogenized in the hypotonic buffer, 0.5 M NaCl, 0.1 M NaHCO 3 , pH 11, or 1% SDS, as shown in Fig. 4H. The mixtures were kept on ice for 20 min and then spun again at 200,000 ϫ g for 60 min. The resulting supernatant and pellet were diluted and solubilized in 1.5ϫ SDS-PAGE sample buffer, respectively.

Membrane protection experiment
Cells coexpressing Vpu and DHFR-mCD4 (Fig. 3E) or Vpu, DHFR-Bio-mCD4 and BirA Cyto (Fig. 4C) was treated with TMX for 4 h before harvesting. Crude microsome was isolated as described above. The crude microsomes were homogenized in a hypotonic buffer containing 10 mM HEPES, pH 7.4, 2 mM MgCl 2 , and 250 mM sucrose. The resuspended microsomes were then digested with 0.5 mg/ml proteinase K in the absence or presence of detergent. The digesting reaction was stopped by phenylmethylsulfonyl fluoride and subsequently added into boiling denaturing buffer containing 1% SDS. The samples were either directly diluted with SDS-PAGE sample buffer for loading (Fig. 3E) or subjected to denaturing IP (Fig. 4C) as described above.

Deglycosylation assay
Samples were digested with endoglycosidase H or peptide:Nglycosidase F (NEB) at 37°C for 2 h according to the manufacturer's instructions.

Construction of knockout cell line using CRISPR/Cas9 technology
Hrd1 KO and gp78 KO cells were made as described with slight modification (54). The targeting sequence to the first coding exon of Hrd1 (3Ј-GTGCCGTCACTACTACCGC-5Ј) and gp78 (3Ј-CCGCTCGGCCTGGTCGATTG-5Ј) were cloned into the vector pSpCas9(BB)-2A-Puro (55), which expresses both gRNA and Cas9 nuclease simultaneously. 293T cells were transfected with the plasmid with Lipofectamine 2000 and cultured for 24 h for expression of Cas9 and gRNA, which was then selected for by treating cell with puromycin for 3 days. The cells were then cultured in nonselecting medium for additional 3 days, plated at 0.33 cells/well in 96-well plates, and expanded for 3 weeks. Individual clones were examined for Hrd1 or gp78 knockout by Western blotting to test expression of endogenous Hrd1 and gp78. Hrd1/gp78 DKO (DKO) cells were made by transfecting plasmid expressing gRNA targeting Hrd1 exon1 into gp78 KO cells. The rest of the steps are the same as described above.