The Nrf3 Transcription Factor Is a Membrane-bound Glycoprotein Targeted to the Endoplasmic Reticulum through Its N-terminal Homology Box 1 Sequence*

Transcription factor Nrf3 (NF-E2 p45-related factor 3) is targeted to the endoplasmic reticulum (ER). Mouse Nrf3 is subject to proteolysis, Asn glycosylation, and deglycosylation reactions. It is synthesized as a ∼96-kDa protein that is subsequently converted into isoforms of ∼90, 80, and 70 kDa. In the ER, the ∼90-kDa glycoprotein is predominant and gives rise to ∼80- and ∼70-kDa isoforms. The ∼90- and ∼80-kDa polypeptides were observed in the nuclear envelope, whereas the ∼70-kDa isoform was detected primarily in the nucleoplasm. Our experiments showed the N-terminal homology box 1 (NHB1, residues 12-31) is part of a tripartite signal peptide sequence, comprising n, h, and c regions. The h region (residues 12-23) was demonstrated to target Nrf3 to the ER and is necessary for its Asn glycosylation. The n region (residues 1-11) controlled the abundance of the ∼90-kDa glycoprotein. The c region (residues 24-39) was found to contain a signal peptidase cleavage site that is responsible for production of the ∼90-kDa mature Nrf3 glycoprotein from a ∼96-kDa precursor. We have found that Nrf3 is activated by the ER stressors tunicamycin and brefeldin A, and that NHB1 is required for this response. Amino acids between the c region and NHB2 (residues 76-100) controlled the proteolytic processing of mouse Nrf3 into cleavage products of ∼80-kDa (glycated) and ∼70-kDa (non-glycated); by contrast, human Nrf3 lacked a signal peptidase cleavage site between its c region and NHB2. Lastly, data are presented suggesting that the NHB2 sequence in mouse Nrf3 may regulate the topology of the transcription factor within the ER membrane.

part of human Nrf3, between residues 298 and 399, contributes to transactivation. This region is herein included in the part of the protein designated TAD ( Fig. 1 and supplemental Fig. S1), and sequence comparisons have revealed that it shares homology with three domains in Nrf1, namely, Neh5L, NST, and AD2. Lastly, Nrf3 contains a sequence that resembles the Neh6 domain in Nrf2 between its TAD and CNC domains.
Nrf3 can be divided into seven domains (i.e. NTD, PEST, TAD, Neh6L, CNC, bZIP, and Neh3L). As mentioned above, Nrf1 also contains NTD and TAD (i.e. Neh5L, NST, and AD2) regions that resemble those in Nrf3. In the case of Nrf1, its NTD is responsible for targeting and anchoring the factor to the endoplasmic reticulum (ER) through a non-cleavable signal sequence in a Leu-rich peptide, called N-terminal homology box 1 (NHB1), located between residues 11 and 30 (11). Within the NTD, Nrf1 also contains a Leu/Val-rich peptide, called NHB2, between residues 82 and 106, but its function is not known. Upon insertion into the ER, Nrf1 is subject to Asn glycosylation through its NST domain, and this post-synthetic modification probably contributes to its retention in the organelle.
The NTD of Nrf3 contains both NHB1 (residues 12-31) and NHB2 (residues 76 -100) sequences. Following identification of NHB1 in Nrf1 and Nrf3 (11,12), the human Nrf3 protein was shown to be localized and glycosylated in the ER (13). It has not however been established whether Nrf3 is targeted to the ER through its NHB1 sequence and whether it is activated by ER stress. Importantly, the NHB1 in Nrf1 is part of a tripartite signal peptide, comprising an n region, h region, and c region. The n and c regions surrounding NHB1 are not highly conserved between Nrf1 and Nrf3. Although the signal peptide in Nrf1 is not cleaved by a signal peptidase (SPase), it is not known if the putative signal peptide in Nrf3 also fails to be cleaved from the bZIP protein. To identify the mechanism by which Nrf3 is directed to the ER, and to determine how it is processed before being transported to the nucleus, we have examined whether: (i) NHB1 and its adjacent residues conform to a signal peptide sequence that targets Nrf3 to the ER; (ii) this signal sequence contains a potential cleavage site for SPase; (iii) Nrf3 is integrated into the ER membrane and the nuclear envelope; and (iv) Nrf3 transactivation activity is increased by ER stress.

EXPERIMENTAL PROCEDURES
Chemicals, Enzymes, and Other Reagents-These were all of the highest quality and were readily available commercially. The ER extraction kit and all chemicals were purchased from Sigma-Aldrich. Endoglycosidase H (Endo H), peptide:N-glycosidase F (PNGase F), and proteinase K (PK) were obtained from New England Biolabs. Rabbit polyclonal antibodies against calreticulin (CRT), calnexin, and GFP were bought from Calbiochem (San Diego, CA) and Abcom PLC (Cambridge, UK), respectively. Mouse monoclonal antibody against the V5 epitope was from Invitrogen, whereas those against Sec61␣, Lamin A/C, and retinoblastoma 1 gene protein were supplied by Upstate (Dundee, Scotland, UK). Goat polyclonal antibodies against Nrf3 (N14, V15, and L16) and small Maf (sMaf) proteins were bought from Santa Cruz Biotechnology (Santa Cruz, CA). The siRNA oligonucleotides against human Nrf3 were purchased from Ambion, Inc. (Austin, TX).
Expression Plasmids-The cDNA encoding mouse Nrf3, previously cloned into the pEF-BOS vector (1), was amplified in 50 l of reaction buffer containing 1 unit of KOD Hot Start DNA polymerase (Merck Chemicals Ltd., Nottingham, UK), 0.5 mM MgSO 4 , 2% (v/v) DMSO, and 12.5 pmol of each of the forward and reverse primers. The first-strand cDNA encoding human Nrf3 was synthesized in the AccuScript TM High Fidelity RT-PCR system that contained 1 g of total RNA extracted from human choriocarcinoma JAR cells. The double-stranded cDNA sequence, generated by the above PCR reaction, was cloned into pcDNA3.1/V5His B (Invitrogen) following KpnI/ XhoI digestion. From this expression plasmid, various Nrf3 mutants were created by site-directed mutagenesis performed in the KOD Hot Start DNA PCRs containing the appropriate pairs of sense and antisense primers (14).
Nucleotide sequences encoding the NTD and N-terminal 66 amino acids (N66) of mouse Nrf3, as well as the N-terminal 126 residues of human Nrf3 (called hN126), along with several mutants, were generated by PCR using suitable primers. These products were ligated to the 5Ј-end of the cDNA for either Nrf2 or GFP in which their translation initiation sites had been mutated, to ensure that neither free Nrf2 nor free GFP were produced. Therefore, the cDNA for the fused protein was inserted into pcDNA3.1/V5His B. The fidelity of all cDNA products was confirmed by sequencing. All oligonucleotide primers in this study were synthesized by MWG Biotech (Ebersberg, Germany) and are listed in supplemental Table S1.
previously (12). The significance of differences in the transactivation activity of wild-type Nrf3 and mutant forms of Nrf3 was determined using the Student's t test.
siRNA Knockdown, Immunocytochemistry, and Confocal Microscopy-COS-1 and JAR cell lines were transfected with an expression construct for Nrf3, or siRNAs targeted against Nrf3. Approximately 24 h following transfection, the cells were examined by immunocytochemistry and confocal microscopy (11,12).
Subcellular Fractionations-The intact ER, microsome-containing membrane, nuclear, and cytosolic fractions were prepared as described previously (11,16,17). The salt-extracted nuclear fraction was prepared according to an established method (18,19). Briefly, purified nuclei were incubated with an extraction buffer (10 mM Hepes-KOH, pH 7.4, 0.42 M NaCl, 2.5% (v/v) glycerol, 1.5 mM MgCl 2 , 0.5 mM sodium EDTA, 0.5 mM EGTA, 1 mM dithiothreitol, 1 mM phenylmethylsulfonyl fluoride (PMSF), and 1% complete protease inhibitor mixture). After being gently rotated for 60 min at 4°C, the mixture was centrifuged at 14,000 ϫ g for 5 min at 4°C. The resulting supernatant was saved and is referred to as "salt-extracted nuclei." The nuclear envelope membrane fraction was obtained by the high-ionic-strength extraction method (20). In brief, purified nuclei were suspended at a concentration of 5 mg DNA/ml in the SMT buffer (250 mM sucrose, 5 mM MgCl 2 , 50 mM Tris-HCl, pH 7.4, supplemented with 1 mM PMSF, and 1% complete protease inhibitor mixture), before they were gently homogenized by passing the mixture through a 25-gauge needle with 10 strokes. The nuclear homogenate (1 ml) was incubated with both DNase I and RNase (at a concentration of 250 g of protein/ml) for 1 h at 4°C before being sedimented by centrifugation at 1000 ϫ g for 10 min at 4°C. The resulting pellet was resuspended in an equal volume of MT buffer (0.2 mM MgCl 2 , 10 mM Tris-HCl, pH 7.4, containing 1 mM PMSF and 1% complete protease inhibitor mixture), and was subsequently mixed with 4 volumes of high-NaCl buffer (2 M NaCl, 0.2 mM MgCl 2 , 10 mM Tris-HCl, pH 7.4, supplemented with 1% (v/v) 2-mercaptoethanol, 1 mM PMSF, and 1% complete protease inhibitor mixture). After stirring for 30 min, the nuclear envelope membranes were pelleted by centrifugation at 1600 ϫ g and 4°C for 30 min.
Membrane Proteinase Protection Reactions-Briefly, either intact ER-rich or microsome-containing fractions were purified and resuspended in 100 l of 1ϫ isotonic buffer (10 mM Hepes, pH 7.8, containing 250 mM sucrose, 1 mM EGTA, 1 mM EDTA, and 25 mM KCl). An aliquot (30 g of protein) of the membrane-containing preparation was incubated for 30 min on ice with proteinase K at a final concentration of 50 g protein/ml. The reactions were terminated by heating at 90°C for 10 min following the addition of 1 mM PMSF.
Deglycosylation Reactions, Co-immunoprecipitation, and Western Blotting-These were performed as reported elsewhere (11,21). During Western blotting, some antibody-blotted nitrocellulose membranes were washed for 30 min with stripping buffer (7 M guanidine hydrochloride, 50 mM glycine, 0.05 mM EDTA, 0.1 M KCl, and 20 mM 2-mercaptoethanol at pH 10.8) before being re-probed with an additional primary antibody (22). The intensity of some Western blots was calculated using the Java-based image-processing program (ImageJ) developed at the National Institutes of Health.

RESULTS
Nrf3 Transcriptionally Activates ARE-driven Gene Expression-Electrophoretic mobility shift assays have demonstrated that Nrf3 can bind ARE sequences as a heterodimer with sMaf proteins (1,2,23). There are however conflicting reports about whether Nrf3 activates or suppresses ARE-driven gene expression. We found that Nrf3 activated a P TK nqo1-ARE-Luc reporter gene about 1.7-fold ( Fig. 2A). By comparison, a parallel experiment using similar amounts of expression vector revealed that Nrf1 transactivated the same reporter gene ϳ4-fold, whereas Nrf2 increased the reporter activity 18-fold.
Based on the mass of DNA transfected, Nrf3 demonstrated less transactivation activity than either Nrf1 or Nrf2. To determine whether differences in protein levels might account for the low activity of Nrf3, Western blotting was undertaken. It has been shown previously that post-synthetic modification of Nrf3 yields three electrophoretic bands, called A, B, and C, which in LDS-NuPAGE we estimate to have molecular masses of ϳ90, ϳ80, and ϳ70 kDa (Fig. 2B). Although Nrf3 exhibited less activity than either Nrf1 or Nrf2, both its ϳ90-kDa A and ϳ80-kDa B forms (which seemed to be present in similar amounts) appeared to be more abundant than either Nrf1 or Nrf2. The ϳ70-kDa Nrf3 C polypeptide was not always clearly discernible, probably because it is less abundant than the larger isoforms. Approximately 36 h after transfection, the cells were harvested, and luciferase reporter assays were performed as described in the text. Luciferase activity was normalized for transfection efficiency, and data are presented as a -fold change (mean Ϯ S.D.) from three independent experiments, each performed in triplicate. Statistical significance was determined using the Student's t test and is shown as a p value. B, total cell lysates expressing Nrf1, Nrf2, or Nrf3 proteins were resolved using 7% LDS-NuPAGE in the Tris acetate buffer system. These proteins were visualized by Western blotting using an antibody against the V5 epitope. The amount of protein applied to each electrophoresis sample well was adjusted to ensure equal loading of ␤-gal activity. The antibody-blotted nitrocellulose membrane was exposed to x-ray film for 10 s (upper panel) or 30 s (lower panel). The arrows indicate the mobilities of the Nrf3 A, B, and C bands. The relative intensities of these three bands were calculated using the ImageJ software program and are shown as an A:B:C ratio.
The NHB1 Sequence Targets Nrf3 to the ER-The low activity of Nrf3, like Nrf1, may be due to the fact that it contains an ER-targeting sequence in its N-terminal domain. Bioinformatic examination of human and mouse Nrf3 revealed that NHB1 and its adjacent residues conform to a classic tripartite ERtargeting signal peptide sequence. In the murine protein the n, h, and c regions are located between residues 1-11, 12-23, and 24 -39, respectively (Fig. 3A). Residues between amino acids 7 and 27, which encompasses the h region, are predicted to fold into a hydrophobic ␣-helix that spans the membrane (Fig. 3B). Whether these sequences contribute to the ER targeting of Nrf3 was examined by confocal microscopy. As shown in Fig. 3C, ectopic wild-type Nrf3 gave predominantly ER staining in 65% of COS-1 cells examined. A significant increase in ER staining was seen in 90% of cells that had been transfected with an expression construct for either the Nrf3 ⌬2-11 or Nrf3 ⌬24 -39 mutants, in which the n region and the c region were deleted, respectively. By contrast, a marked increase in nuclear staining was observed upon transfection with either the Nrf3 ⌬12-23 or Nrf3 ⌬12-30 mutants, in which the h region and the essential NHB1 sequences were removed. Extranuclear staining was not observed in the case of Nrf3 ⌬2-39 as it was located exclusively in the nucleus (Fig. 3C). We next examined whether the different subcellular distributions of wild-type murine Nrf3 and its mutants influence transactivation of the P TK nqo1-ARE-Luc reporter gene. Fig. 3D shows that the activity of Nrf3 ⌬2-11 and Nrf3 ⌬24 -39 was significantly lower than the wildtype protein. Conversely, the Nrf3 ⌬12-23 , Nrf3 ⌬12-30 , and Nrf3 ⌬2-39 mutants exhibited a modestly increased transcriptional activity when compared with the wild-type protein. To examine whether differences in transactivation of ARE-driven transcription affected by these various Nrf3 mutants might be due to their relative abundance or post-translational modification, Western blotting experiments were performed. By comparison with wild-type protein, ectopic Nrf3 ⌬2-11 was expressed in COS-1 cells primarily as a ϳ90-kDa A polypeptide with a relative reduction in the amount of ϳ80-kDa B polypeptide (Fig. 3E). Conversely, the ϳ90-kDa protein was not detected upon transfection with an expression construct for the Nrf3 ⌬12-23 , Nrf3 ⌬12-30 , or Nrf3 ⌬2-39 mutants. Production of the ϳ90-kDa protein was not altered in the case of the Nrf3 ⌬24 -39 mutant, whereas the amount of the ϳ80-kDa form was significantly reduced. These data suggest that the h region within NHB1 contributes to the negative regulation of Nrf3 by its NTD, presumably through association with the ER.
Endogenous Nrf3 Isoforms Are Located in the ER and the Nucleus-To confirm that endogenous Nrf3 is also localized in the ER, we performed immunocytochemistry using antibodies against an N-terminal region (N14) or an internal region (L16) of the Nrf3 protein. Confocal microscopy showed that human Nrf3 is located primarily in the extranuclear compartment of JAR cells, yielding a staining pattern similar to that obtained for Nrf3 Is a Membrane-bound Glycoprotein calreticulin (Fig. 4A); this is consistent with the hypothesis that endogenous Nrf3 is associated with the ER. To support the immunocytochemistry data, knockdown experiments were performed. ER staining for Nrf3 was markedly diminished following transfection of JAR cells with two separate siRNA targeted against mRNA for the bZIP protein (Fig. 4A, right two lines). Western blotting using two different Nrf3 antibodies revealed that three immunoreactive polypeptides of ϳ50, 80, and 85 kDa were recovered in the ER and nuclear factions (Fig.  4B). The level of all three polypeptides was significantly reduced following knockdown of Nrf3 mRNA using specific siRNA species.
Generation of the ϳ90-kDa Nrf3 Glycoprotein Requires the h Region-Differences in the molecular mass of Nrf3 isoforms may arise from ER-directed post-translational modification, such as Asn glycosylation. To test this possibility, we subjected wild-type mouse Nrf3 and its mutant proteins to enzymatic deglycosylation. As shown in Fig. 5A, while ectopic wild-type Nrf3 in COS-1 cells migrated during NuPAGE as two protein bands with molecular masses of ϳ80 kDa and ϳ90 kDa, digestion of the cell lysates with PNGase F resulted in a complete loss of the larger of the two proteins. This loss of the ϳ90-kDa band was accompanied by a modest increase in the amount of the ϳ80-kDa protein and a slight increase in the amount of the ϳ70-kDa protein. Two similarly migrating proteins were resolved by NuPAGE from lysates prepared from COS-1 cells expressing the Nrf3 ⌬2-11 or Nrf3 ⌬24 -39 mutants, although in both cases the ϳ90-kDa protein was predominant (Fig. 5A). Again, the largest ϳ90-kDa band from Nrf3 ⌬2-11 or Nrf3 ⌬24 -39 disappeared following digestion with PNGase F and was replaced with substantially increased amounts of the ϳ80-kDa protein, along with a modest increase in the ϳ70-kDa band (Fig. 5A). A similar experiment using COS-1 cell lysates expressing Nrf3 ⌬12-23 showed that neither of the ϳ70and ϳ80-kDa bands changed following digestion with PNGase F. Furthermore, when forms of Nrf3 lacking the entire n, h, and c regions were examined, the mutant Nrf3 ⌬2-39 or Nrf3 ⌬2-75 proteins each migrated as a single band of between ϳ70 and ϳ80 kDa. Digestion with PNGase F caused a slight increase in the electrophoretic mobility of the single Nrf3 ⌬2-39 band, whereas the mobility of the single Nrf3 ⌬2-75 band did not change.
Western blotting of subcellular fractions from COS-1 cells expressing ectopic wild-type mouse Nrf3 showed that the ϳ90-kDa protein was almost exclusively recovered in the intact ERrich fraction, although a small amount of the ϳ80-kDa isoform was also detected in this subcellular fraction (Fig. 5B). In vitro digestion with Endo H of Nrf3 in the ER resulted in its apparent molecular mass being reduced from ϳ90 to ϳ80 kDa (Fig. 5B). In addition, both the ϳ70and ϳ80-kDa polypeptides were also detected in the microsome-rich membrane fraction, but no change in their molecular masses was observed following digestion with Endo H (Fig. 5B).
Taken together, the results in Fig. 5 suggest that the ϳ90-kDa Nrf3 A isoform represents a glycosylated protein, and its generation in the ER is dependent on the h region of the N-terminal signal sequence. The ϳ80-kDa Nrf3 B isoform may represent a glycosylated cleaved protein or a deglycosylated non-cleaved protein. The ϳ70-kDa Nrf3 C isoform may be a non-glycosylated, cleaved polypeptide. Examination of human Nrf3 protein showed that upon its expression in COS-1 cells two major isoforms of ϳ95 and ϳ80 kDa were observed, the former of which is a glycoprotein (Fig. 5C).
Nrf3 Can Heterodimerize with sMaf Proteins-To examine whether Nrf3 isoforms form heterodimers with sMaf proteins, we performed co-immunoprecipitation by incubating COS-1 lysates that had been co-transfected with expression constructs for both sMaf and Nrf3 with specific antibodies against these two bZIP proteins. Western blotting using antibodies against a V5 epitope revealed that a major immunoreactive Nrf3 isoform of ϳ80 kDa was immunoprecipitated by a sMaf antibody (Fig.  5D). Conversely, in a separate experiment, sMafK-V5 protein was precipitated by Nrf3 antibodies. Further, endogenous sMaf proteins from RL34 and JAR cell lysates were precipitated by Nrf3 antibodies (lower two panels); it was noted that the ϳ90-kDa Nrf3 protein was not observed in these pulldown experiments, possibly because it may be deglycated to become the ϳ80-kDa isoform by a glycosidase present in the immunoprecipitation buffer. Antibodies against the N-terminal region of Nrf3 (N14) precipitated neither the ϳ70-kDa protein nor an additional ϳ45-kDa Nrf3 isoform, but they were pulled down by antibodies against sMaf (Fig. 5D, upper two panels). These findings indicate that while cleavage of Nrf3 results in loss of FIGURE 3. Targeting of mouse Nrf3 to the ER by its N-terminal signal peptide. A, diagrammatic representation of sequences deleted from the N terminus of mouse Nrf3. Residues placed on a dark background represent the NHB1 sequence. Within the putative signal peptide sequence, its N-terminal extension region, hydrophobic core region, and C-terminal polar region are abbreviated as n, h, and c regions. B, residues 7-27 were wheeled into a hydrophobic ␣-helix structure. Aromatic and hydrophobic amino acids are placed on a dark background, a basic arginine residue is on a blue background, a polar glutamine residue is on a light red background, two nucleophilic residues serine and threonine are on a green background, and four small glycine residues are shown on a gray background. C, COS-1 cells were transfected with 1.3 g of an expression construct for V5-tagged wild-type Nrf3, or 1.3 g of any one of five different expression constructs for mutant forms of Nrf3 lacking portions of the N-terminal signal peptide, or 1.3 g of an empty pcDNA3.1/V5His B vector. These expression vectors were each co-transfected with 0.7 g of an ER/DsRed expression construct. Approximately 24 h after transfection, the subcellular location of the proteins was examined by immunocytochemistry followed by confocal imaging. Fluorescein isothiocyanate-labeled second antibody was used to locate V5-tagged proteins. Nuclear DNA was stained by 4Ј,6-diamidino-2-phenylindole (DAPI). The merge signal represents the results obtained when the three images were superimposed. The corresponding quantitative data shown here were calculated by determining the percentage of cells in which the extranuclear stain (i.e. cytoplasmic plus ER) was greater than or equal to the nuclear stain, as opposed to the percentage of cells in which the extranuclear stain was less than the nuclear stain. Bar ϭ 20 m. D, COS-1 cells were transfected with 1.2 g of each of the expression constructs for wild-type or mutant Nrf3, or an empty pcDNA3.1/V5His B vector, together with 0.6 g of P TK nqo1-ARE-Luc and 0.2 g of a ␤-gal reporter plasmid. Approximately 36 h after transfection, the cells were harvested, and luciferase reporter assays were performed as described in the text. Luciferase activity was normalized for transfection efficiency, and data are presented as a -fold change (mean Ϯ S.D.) from three independent experiments, each performed in triplicate. Significant differences were determined using the Student's t test and are shown as p values. E, total cell lysates were resolved using a 7% LDS-NuPAGE Tris acetate system. The V5-tagged proteins were visualized by Western blotting. The positions of the A, B, and C Nrf3 bands are indicated by arrows. The relative abundance of the three bands in both wild-type Nrf3 and the Nrf3 ⌬24 -39 mutant is shown.
portions of its NTD, the processed protein is still capable of forming a heterodimer with sMaf.
The NHB1 Sequence of Nrf3 Directs an NTD/Nrf2 Fusion Protein to the ER Where It Is Asn-glycosylated-To confirm the functional significance of the NHB1-associated signal sequence within Nrf3, we created N66/Nrf2 and NTD/Nrf2 fusion proteins through attachment of residues 1-66 or 1-126 of the mouse bZIP factor to the N terminus of mouse Nrf2. Because the part of human Nrf3 that is equivalent to the c region of mouse Nrf3 contains significant amino acid differences, we also fused the NTD of the human bZIP factor to the N terminus of mouse Nrf2, giving hN126/Nrf2, as an additional control. Following transfection into COS-1 cells, ectopic wild-type Nrf2 was localized primarily in the nucleus (Fig. 6A). By contrast, both N66/Nrf2 and NTD/ Nrf2 gave a predominantly ER stain in ϳ85% of COS-1 cells, whereas the remaining cells expressing these fusion proteins gave primarily a nuclear stain. Similar experiments using the hN126/Nrf2 chimeric fusion protein, showed it to be localized almost exclusively in the ER of all cells examined.
Deletion of the major portion of NHB1 from NTD/Nrf2, giving NTD ⌬12-30 /Nrf2, resulted in the ectopic mouse fusion protein becoming located exclusively in the nucleus of almost all cells. Similarly, a construct in which the NHB2 of mouse Nrf3 was deleted, NTD ⌬76 -100 /Nrf2, was also localized preferentially in the nucleus of ϳ55% cells, although in this case the mutant fusion protein was observed in the ER of the remaining cells. These results suggest that the NHB1 of Nrf3 is required to target an NTD/Nrf2 fusion protein to the ER and that this process may also be modulated by NHB2.
To test whether the targeting of Nrf2 to different subcellular locations through Nrf3 sequences influences its activity, we performed luciferase reporter assays. Fig. 6B shows that transient transfection of COS-1 cells with wildtype mouse Nrf2 activated AREdriven gene expression ϳ30-fold, whereas NTD/Nrf2 and N66/Nrf2 activated ARE-driven transcription by only ϳ18and ϳ25-fold, respectively (left panel). By contrast, transfection with either NTD ⌬12-30 /Nrf2 or NTD ⌬76 -100 /Nrf2 activated transcription to a similar extent as did wild-type Nrf2 (right panel). These results suggest that the targeting of CNC bZIP factors to the ER via NHB1 and NHB2 results in a partial inhibition of their activity.
Western blotting showed that cell lysates expressing NTD/ Nrf2 produced a major ϳ95-kDa band and a relatively weaker band of ϳ85-kDa, but upon digestion with PNGase F their elec- . Endogenous Nrf3 is present in the ER and nucleus and its abundance can be reduced by siRNA knockdown. A, human choriocarcinoma JAR cells were transfected using 1 ml of serum-free Opti-MEM1 medium containing 100 nM of either human Nrf3-targeted or scrambled siRNA with 10 l of Lipofectamine 2000. Approximately 24 h after transfection, the cells were fixed, permeabilized, and then subjected to immunocytochemistry with primary antibodies against human Nrf3 (N14 and L16) and calreticulin (CRT). The confocal images were visualized with a green-fluorescent Alexa-Fluor 488 (fluorescein isothiocyanate (FITC))-labeled secondary antibody against rabbit IgG and a red-fluorescent Alexa-Fluor 594 (RFP)-labeled secondary antibody against sheep IgG. Nuclear DNA was stained by 4Ј,6-diamidino-2-phenylindole (DAPI). The merge signal represents the results obtained when the three images were superimposed. Bar ϭ 20 m. B, JAR cells were transfected with either human Nrf3-targeted or scrambled siRNA, and then the cells were subject to subcellular fractionation followed by electrophoresis and immunoblotting with antibodies against Nrf3. The hNrf3-V5 indicates that the human C-terminally V5-tagged Nrf3 protein was loaded in the first lane of the NuPAGE gel. The samples N and T indicate nuclear fraction and total cell lysates, respectively. Blotting with antibodies against lamin A/C and a membrane-bound protein calnexin (CNX) was undertaken to confirm the efficacy of subcellular fractionation. The numbers 1 and 2 indicate two different siRNA oligonucleotides against human Nrf3.

Nrf3 Is a Membrane-bound Glycoprotein
trophoretic mobilities increased (Fig. 6C). This suggests that both the ϳ85and 95-kDa NTD/Nrf2 proteins are glycosylated in the ER. Furthermore, the ϳ85-kDa NTD/Nrf2 protein may arise through cleavage of a portion of its NTD in the ER, because wild-type Nrf2 protein gave a single band of ϳ81-kDa and was unaffected by the deglycosylation reaction. Similarly, cell lysates expressing N66/Nrf2 also yielded two polypeptides of ϳ81 and ϳ85 kDa, and following digestion with PNGase F, the major ϳ85-kDa form disappeared and was replaced with a polypeptide band of ϳ81-kDa (Fig. 6C, left panel). This suggests that the ϳ85-kDa form is a glycosylated N66/Nrf2 fusion protein, whereas the less abundant ϳ81-kDa protein is a non-glycated or deglycated form that may be produced through a mechanism in which the N66 moiety is proteolytically cleaved in the ER. By contrast, lysates from cells expressing the hN126/Nrf2 chimeric fusion protein produced a single band of ϳ95 kDa with a very low activity, and its molecular mass did not appear to change following digestion with PNGase F (Figs. 6B,  6D, and 5C). This suggests that a proteolytic cleavage event is unlikely to occur within the N-terminal 126 residues of human Nrf3.
The Signal Peptide c Region and Its Flanking Residues 67-78 Control Proteolytic Processing of Mouse Nrf3-In the experiments described above, we found that N66/Nrf2 and NTD/ Nrf2, both of which were tagged C-terminally with a peptide containing the V5 epitope, were each represented by two polypeptides of molecular masses that were equal to, or greater than, wild-type Nrf2. Because the two electrophoretic bands persisted after PNGase F digestion, it appeared likely that proteolytic cleavage of residues within the NTD of mouse Nrf3 accounted for the appearance of the two protein species because Nrf2 migrated during electrophoresis as a single band. We postulated that proteolysis might occur at a site that lies close to the C terminus of the N66 peptide.
Alanine and glycine are frequently found at the Ϫ1 position of an SPase cleavage site, whereas the Ϫ3 position is often occupied by alanine, serine, or valine (24). We therefore considered whether the consensus AXA site at Ala 39 or Ala 55 might be recognized by an SPase (Fig. 7A). To test this hypothesis, we mutated the potential 37 AAA 39 consensus site into VVV or NNN (to create Nrf3 VVV or Nrf3 NNN , respectively). We also mutated the other 53 ASA 55 site into VLV (to yield Nrf3 VLV ). As shown in Fig. 7B, wild-type Nrf3 migrated during NuPAGE in Tris acetate buffer primarily as the A and B isoforms. However, cell lysates expressing the Nrf3 VVV mutant were resolved in NuPAGE gels into the A, B, and C isoforms at a ratio of 71:14:11 (left panel). By comparison with the wild-type mouse Nrf3, recovery of the ϳ80-kDa B protein was significantly diminished in the Nrf3 VVV mutant, suggesting that the 37 AAA 39 site is recognized by an SPase. This interpretation is also supported by additional comparison of the Nrf3 VVV mutant and the wild-type Nrf3 that were resolved by ϳ4 -12% A, total lysates of COS-1 cells expressing wild-type Nrf3 and its mutants each were either incubated (ϩ), or were not incubated (Ϫ), for 1 h with 500 units of PNGase F in a G7 reaction buffer. The products were examined, following electrophoresis in 4 -12% LDS/NuPAGE containing Tris-Bis buffer, by Western blotting using antibodies against the V5 epitope. The antibody-blotted nitrocellulose membrane was exposed to x-ray film for the indicated times. B, intact ER-rich and microsome-containing membrane (M) fractions were purified from COS-1 cells expressing Nrf3, and after washing the fractions were either incubated (ϩ), or were not incubated (Ϫ), with 500 units of Endo H in a G5 reaction buffer. Following 1 h of digestion, Nrf3 was analyzed by the 4 -12% NuPAGE Bis-Tris buffer system followed by Western blotting. The blot was reprobed with an antibody against CRT (calreticulin), to confirm integrity of the ER membrane. The sample T in the first lane of the blot indicates total cell lysate. C, total lysates of COS-1 cells expressing wild-type mouse and human Nrf3, and their N-terminal domains fused to the N terminus of mouse Nrf2 each either were incubated (ϩ) or were not incubated (Ϫ) with PNGase F, before the products were resolved using either 4 -12% LDS/NuPAGE in the Tris-Bis buffer system (lower panel) or 7% LDS/NuPAGE in the Tris acetate buffer system (upper panel), and subsequently immunoblotted with an antibody against the V5 epitope. D, total lysates of COS-1 cells that had been co-transfected with a MafK expression construct, along with an expression construct for mouse or human Nrf3 (upper panel) were subjected to immunoprecipitation (IP) followed by Western blotting (WB) with antibodies against the V5 epitope and sMaf protein.
Lysates prepared from RL34 and JAR cells were also subject to IP (lower two panels). The Input sample indicates a total lysate co-expressing Nrf3 and sMafK that had been incubated with the IP buffer alone.
NuPAGE in a Tris-Bis buffer system (Fig. 7B, right panel). Further comparisons of the Nrf3 VVV and Nrf3 VVV/VLV mutants revealed that generation of the ϳ80-kDa type B isoform from the latter protein was diminished, but not completely abolished. By contrast, the Nrf3 NNN mutant protein still produced substantial amounts of the ϳ80-kDa B protein (Fig. 7B, right panel), suggesting that it contains an additional SPase-recognized cleavage site. Further examination showed that the Nrf3 VLV mutant did not influence production of the ϳ80-kDa B protein (Fig. 7B, right panel), suggesting that the 53 ASA 55 site is probably not recognized by an SPase.
To abolish its cleavage more completely, we deleted both the 37 AAA 39 SPase consensus site and residues 67-78, which contains two potential Site-1 protease recognition sites, to create Nrf3 ⌬37-39ϩ⌬67-78 . As expected, the cleaved ϳ80-kDa Nrf3 isoform was not observed upon transfection of COS-1 cells with a construct for the double deletion mutant (Fig. 7C). Formation of this cleavage product was also blocked by a combined mutant Nrf3 VVV/VLV/PxxV (in which PxxV indicates a mutant of RXXL 70 ). It was however observed following transfection with constructs for either Nrf3 ⌬37-39ϩ⌬67-70 or Nrf3 ⌬67-70 , indicating that there are two Site-1 protease cleavage sites between residues 67 and 78.
As shown in Fig. 7C (left panel), a mouse Nrf3 precursor protein of estimated molecular mass of ϳ96-kDa (called AЈ) was generated upon ectopic expression of the Nrf3 ⌬37-39 mutant. This AЈ form could be converted by PNGase F digestion into a ϳ90-kDa form (Fig. 7C). The ϳ96and ϳ90-kDa proteins were detected at relatively low levels following expression of the Nrf3 ⌬37-39ϩ⌬67-78 mutant, but the larger ϳ96-kDa protein was not observed from the Nrf3 ⌬37-39ϩ⌬67-70 mutant. These results demonstrate that the SPaserecognized 37 AAA 39 cleavage site is responsible for proteolytic processing of the ϳ96-kDa mouse Nrf3 precursor into a mature glycoprotein of ϳ90 FIGURE 6. The NHB1 within NTD redirects an Nrf2 fusion protein to the ER where it is Asn-glycosylated. A, the upper panel shows alignment of the N-terminal amino acids (residues 1-126) in mouse Nrf3 (mNrf3) and human Nrf3 (hNrf3), which were linked to the N terminus of mouse Nrf2. Identical residues are placed on a dark background, whereas residues possessing similar physicochemical properties are placed on a gray background. The positions of NHB1 and NHB2 are indicated. The lower panel shows the confocal images obtained from immunocytochemistry of COS-1 cells that were transfected with an expression construct for wild-type Nrf2, N66/Nrf2, NTD/Nrf2, mutants of NTD/Nrf2, hN126/Nrf2, or hNrf3. Each of these constructs was co-transfected with an ER/DsRed expression construct. Approximately 24 h after transfection, the subcellular location of proteins was examined by immunocytochemistry followed by confocal imaging. B, COS-1 cells were cotransfected with each of the indicated expression constructs along with the P SV40 GSTA2-6ϫARE-Luc and pRL-TK reporter plasmids. Luciferase activity was determined 36 h later. Statistical significance was examined using the Student's t test and is indicated as a p value. C, total cell lysates expressing the indicated protein were either incubated (ϩ), or were not incubated (Ϫ), for 1 h with 500 units of PNGase F, before being resolved using 4 -12% LDS/NuPAGE with Tris-Bis buffer and immunoblotted with antibodies against the V5 epitope.

Nrf3 Is a Membrane-bound Glycoprotein
kDa. Furthermore, the stability of the two proteins is probably influenced by residues 67-70.
Orientation of Nrf3 Protein within the ER Membrane Is Modulated by Residues between NHB1 and NHB2-Following expression of NTD/GFP and mouse Nrf3 in COS-1 cells, a difference was observed in the relative abundance of the cleaved ϳ33-kDa NTD/GFP band (supplemental Fig. S2B) and the ϳ80-kDa Nrf3 band (Fig. 7), when compared with the total ectopic proteins. To test the hypothesis that this was due to the position of the SPase-cleavable peptide bond in the c region being influenced by flanking sequences, we attached the NTD of mouse Nrf3, and also various NTD mutants, to the N terminus of Nrf2. We speculated that if it were translocated into the ER lumen it might become glycosylated through Asn 159 , Asn 331 , and Asn 340 in Nrf2. As anticipated, NTD/ Nrf2 was targeted to the ER where it was glycosylated (Figs. 8, A and  B). Targeting of the fusion protein to the ER was increased in cells expressing either the NTD VVV / Nrf2 or NTD VVV/VLV /Nrf2 mutants but no change was observed in the case of the NTD VLV /Nrf2 mutant (Fig. 8A). Deglycosylation reactions revealed that NTD VLV /Nrf2, like NTD/Nrf2, was Asn-glycosylated, whereas neither NTD VVV /Nrf2 nor NTD VVV/VLV /Nrf2 were glycated (Fig. 8B). To help explain these findings, we proposed that NTD/Nrf2 and NTD VVV /Nrf2 adopt different conformations within the ER membrane (Fig. 8C). The NTD VVV /Nrf2 fusion protein was less active than NTD/Nrf2 at transactivating P SV40 GSTA2-6ϫARE-Luc (Fig. 8D), suggesting that NTD VVV /Nrf2 may be improperly folded within the ER membrane.
Nrf3 Activity Is Modulated by Its NHB2 Sequence-As described above, attachment of the NTD of mouse Nrf3 to the N terminus of Nrf2 diminished its activity, whereas negative regulation by the NTD was completely abolished by deletion of NHB2 in the NTD ⌬76 -100 /Nrf2 mutant (Fig. 6B). These results suggest that Nrf3 may also be controlled by NHB2. Residues within and around NHB2 are predicted to adopt basic hydrophobic ␣-helix and ␤-sheet secondary structures, which could either lie in the plane of the membrane or span the membrane (Fig.  9A). To test whether the amphipathic NHB2 sequence controls subcellular distribution of Nrf3 and its transactivation activity, we created a construct encoding mouse Nrf3 ⌬76 -100 . Confocal microscopy showed that, although this mutant protein localized to the ER, it gave a stronger nuclear stain than the wild-type bZIP factor (Fig. 9B). Nrf3 ⌬2-75 , in which both the NHB1 signal sequence and the linker region between the NHB1 and NHB2 were omitted, was localized predominantly in the nucleus, presumably because it lacks the ER signal peptide.
Western blotting of cell lysates expressing the murine Nrf3 ⌬76 -100 mutant revealed that it, like wild-type Nrf3, also yielded three ϳ70-, ϳ80-, and ϳ88-kDa isoforms (Fig. 9C). Comparison with wild-type Nrf3 indicated that the level of the ϳ80-kDa Nrf3 B isoform was significantly reduced by deletion of the NHB2 sequence. This suggests that NHB2 is possibly required to position the signal peptide c region correctly in the FIGURE 7. The signal c region and its flanking residues between the NHB1 and NHB2 control proteolytic processing of mouse Nrf3. A, a sequence alignment of residues 1-101 from mouse Nrf3 with residues 1-115 of human Nrf3 and residues 1-107 of mouse Nrf1 is presented. The classic tripartite signal peptide comprises n, h, and c regions, and the sequences encompassed by these regions are indicated. The c region in mouse Nrf3 is overlined, whereas the c region in human Nrf3 is underlined. Within the c region, putative SPase-recognition sites are overlined and indicated by vertical solid-headed arrows, one of which consists of tri-alanine between position 37 and 39 that is absent from the human Nrf3. Additional cleavage sites for Site-1 protease were predicted in close proximity to the NHB2 sequence and are indicated by vertical empty-headed arrows. These putative cleavage sites were deleted or mutated. B, COS-1 cells were co-transfected with each of expression constructs for Nrf3 and the indicated mutants, along with a ␤-gal plasmid, and allowed to recover for 36 h. Total cell lysates were either incubated (ϩ), or were not incubated (Ϫ), for 1 h with 500 units of PNGase F, and resolved using LDS/NuPAGE that contained either 7% polyacrylamide in a Tris acetate buffer (left panel) or 4 -12% gradient polyacrylamide in a Tris-Bis buffer (right panel). Immunoblotting was performed with antibodies against the V5 epitope. The amount of protein added to each electrophoresis sample well was adjusted to ensure equal loading of ␤-gal activity. The arrows indicate the positions of A, B, and C Nrf3 bands, the intensity of which in some lanes was calculated and is shown as a ratio of A:B:C. C, COS-1 cells were transfected with each of expression constructs for Nrf3 and the indicated mutants. Total cell lysates were either incubated (ϩ), or were not incubated (Ϫ), with PNGase F, and then resolved using the 4 -12% gradient LDS/NuPAGE Tris-Bis buffer system, before being immunoblotted with a V5 antibody.
ER membrane and allow cleavage to occur. In addition to these three forms of Nrf3 ⌬76 -100 , two faster migrating bands were resolved that were estimated to have molecular masses of ϳ54 and ϳ50 kDa, called Nrf3 bands D and E, respectively (Fig. 9C). The D and E bands were identical with those of Nrf3 ⌬1-172 , indicating that they may have arisen through either translation from two internal start codons at Met 173 and Met 211 or proteolytic cleavage occurring within a central region.
We examined the effect that deleting NHB2 had on the activity of mouse Nrf3. Transactivation of the P TK nqo1-ARE-Luc reporter gene was significantly reduced from ϳ1.7-fold by wild-type Nrf3 to ϳ1.2-fold activation by the Nrf3 ⌬76 -100 mutant (Fig. 9D). This loss of activity may be due to a failure to synthesize the ϳ80-kDa Nrf3 isoform and a relative increase in the amount of the ϳ88-kDa glycoprotein. Both the Nrf3 ⌬2-75 and Nrf3 ⌬2-100 mutants increased AREdriven gene activity to ϳ2.5and ϳ3.0-fold, respectively (Fig. 9D); this was accompanied by the appearance of ectopic polypeptides of between ϳ70 and ϳ80 kDa (Fig.  9C). Taken together, these results suggest that the ϳ70and ϳ80-kDa Nrf3 polypeptides are both active, but the former is unstable.
Deletion of the NHB2 Sequence Results in the Retention of Nrf3 within the ER Lumen-To address the question of why mouse Nrf3 ⌬76 -100 is less active than wildtype Nrf3 (Fig. 9D), whereas the NTD ⌬76 -100 /Nrf2 fusion protein has the same activity as wild-type Nrf2 (Fig. 6B, right panel), we examined whether Nrf3 ⌬76 -100 and NTD ⌬76 -100 /Nrf2 differ in their subcellular location and/or membrane topology. Fig. 10A shows that a small fraction of the ϳ88-kDa Nrf3 ⌬76 -100 protein was recovered in the nucleus, but the majority was recovered in the ER. The ϳ88-kDa Nrf3 ⌬76 -100 protein appeared to be retained in the ER lumen; this conclusion was drawn because, like the luminal protein calreticulin, the ϳ88-kDa Nrf3 ⌬76 -100 protein was protected by the membrane against digestion with PK, but following disruption of the membrane by addition of 1% Triton X-100 it was highly susceptible to digestion (Fig. 10C). FIGURE 8. Membrane topology of an NTD/Nrf2 fusion protein is modulated by residues between the NHB1 and NHB2. A, each of the expression constructs for NTD/Nrf2 and its mutants, along with an ER/DsRed expression construct, was co-transfected into COS-1 cells and allowed to recover for ϳ24 h. Subcellular location of proteins was examined by immunocytochemistry followed by confocal imaging. B, total cell lysates expressing NTD/Nrf2 and the indicated mutants were resolved using 4 -12% gradient LDS/NuPAGE in either a Tris-Bis buffer system (upper panel) or in a reducing buffer (lower panel), before being immunoblotted with antibodies against the V5 epitope. The amount of protein added to each electrophoresis sample well was adjusted to ensure equal loading of ␤-gal activity. C, the schematic shows two proposed topologies of NTD/Nrf2 and NTD VVV /Nrf2. In the context of the former protein, the wild-type c region is predicted to fold as a ␤-sheet, whereas in the latter protein the mutant c region could fold as a hydrophobic ␣-helix. The scissors indicate a potential cleavage site. D, COS-1 cells were cotransfected with each of the indicated expression constructs along with the P SV40 GSTA2-6ϫARE-Luc and pRL-TK plasmids. Approximately 36 h later, luciferase activity was measured and normalized for transfection efficiency. The data are presented as a -fold change (mean Ϯ S.D.) from three independent experiments, each performed in triplicate. Significant differences were determined using the Student's t test and are shown as p values.

Nrf3 Is a Membrane-bound Glycoprotein
By contrast with the Nrf3 ⌬76 -100 mutant, an NTD ⌬76 -100 / Nrf2 fusion protein of ϳ92 kDa was predominantly recovered in both the nuclear and ER fractions (Fig. 10A, right panel). Although proteinase reactions revealed that the ϳ92-kDa protein was located in the lumen of the ER (Fig. 10C), NTD ⌬76 -100 / Nrf2 also yielded an additional cleaved polypeptide of ϳ85-kDa, with a similar abundance to wild-type Nrf2 (Fig. 10B). Subcellular fractionation showed that the ϳ85-kDa fusion protein was also localized in the nucleus (Fig. 10A, right panel).
Nrf3 Is a Membrane-associated Protein That Locates to the Nuclear Envelope-To explain why Nrf3 is less active than Nrf1, we examined whether wild-type Nrf3 is a luminal ER protein or a membrane-bound protein in the ER by performing protease protection reactions. Fig.  10D shows that following digestion of the ER fraction with PK, most of the ϳ90-kDa Nrf3 A isoform disappeared, to be replaced by a relatively weaker multiple-polypeptide ladder. This digested protein ladder comprised two major bands of ϳ70 and ϳ85 kDa and four additional bands of between ϳ12 and ϳ32 kDa. However, following digestion of the microsome-rich membrane fraction that primarily contained the ϳ80-kDa Nrf3 form, no polypeptide bands were detected indicating that it was not protected from proteolysis (Fig. 10D). In the same experiment, the ER-membrane protein Sec61␣ was completely digested by PK, whereas no change in the abundance of the luminal protein calreticulin was detected (Fig. 10D, lower panels). These results suggest that Nrf3 is translocated into the ER lumen and then inserted into the membrane through its hydrophobic and/or amphipathic regions. Once this bZIP protein is integrated into the membrane, we envisage it would either be released from the ER or it would be transported into the nuclear envelope. Consistent with the second of these proposals, both the ϳ80and ϳ90-kDa Nrf3 proteins were recovered primarily in the nuclear envelope membrane fraction (Fig. 10E). By contrast, the salt-extracted soluble nuclear fraction was found to contain relatively small amounts of the ϳ70and ϳ80-kDa Nrf3 forms, and none of the ϳ90-kDa protein.
The NHB1 Sequence Is Required to Enable Nrf3 Activity to Be Increased by ER Stressors-The biological significance of Nrf3 being targeted to the ER is unclear. We therefore tested whether it responds to ER stress. Fig. 11A shows that wildtype mouse Nrf3 was significantly activated by the ER stressors tunicamycin (TU) and brefeldin A (BFA), but not by thapsigargin (TG). Further, the increase in transactivation activity of Nrf3 produced by TU and BFA was prevented by ALLN (N-acetyl-L-leucyl-L-leucyl-L-norleucinal) (Fig. 11B), an inhibitor of proteolysis catalyzed by proteasomes (25). By contrast, ALLN markedly increased transactivation activity of the Nrf3 ⌬12-30 mutant lacking NHB1, but this factor was not stimulated by TU, TG, and BFA (Fig. 11A). Moreover, both Nrf3 and Nrf3 ⌬12-30 were significantly activated by co- FIGURE 9. The NHB2 sequence controls mouse Nrf3 activity. A, residues within and around the NHB2 are predicted to fold into an N-terminal amphipathic ␣-helix and a C-terminal amphipathic ␤-sheet, which were modeled using the MacPyMol software. The secondary structure appears to be modulated by two PXXP motifs. Aromatic and hydrophobic amino acids are shown in white, with the exception of four prolines that are shown in yellow. Acidic, basic, and neutral amino acids are illustrated in red, blue, and green colors, respectively. B, COS-1 cells were co-transfected with an expression construct for mouse Nrf3 or its mutants, along with an ER/DsRed expression construct. Approximately 24 h later, the cells were subjected to immunocytochemistry followed by confocal imaging. C, total cell lysates expressing Nrf3 or various deletion mutants were resolved using a 7% LDS/NuPAGE Tris acetate buffer system. Proteins were visualized by immunoblotting with a V5 antibody. The amount of protein added to each electrophoresis sample well was adjusted to ensure equal loading of ␤-gal activity. The arrows indicate five discrete forms of Nrf3 A to E, which have molecular masses between ϳ50 and ϳ90 kDa. The intensity of A, B, and C forms that are present in the lanes for Nrf3 and Nrf3 ⌬76 -100 was calculated and is shown as a ratio of A:B:C. D, COS-1 cells were cotransfected with each of the indicated expression constructs along with the P TK nqo1-ARE-Luc and ␤-gal plasmids. Luciferase reporter activity was measured 36 h later. ARE-driven transactivation activity was normalized for transfection efficiency, and data are presented as a -fold change (mean Ϯ S.D.) from three independent experiments, each performed in triplicate.

Nrf3 Is a Membrane-bound Glycoprotein
treatment with ALLN and tert-butyl hydroquinone (tBHQ), but they were also slightly stimulated by tBHQ alone. These results indicate that the NHB1 sequence in Nrf3 is required in order for this bZIP factor to be activated by TU or BFA.
Western blotting showed that the ϳ90-kDa Nrf3 protein was not synthesized following treatment with TU or BFA, but in both cases a relatively weaker band of ϳ85-kDa was detected (Fig. 11, C and D, and see supplemental Fig. S3). This suggests that the ϳ90-kDa Nrf3 protein is produced by Asn glycosylation in both the ER and Golgi apparatus. In addition, TU treatment also increased expression of the ϳ80-kDa Nrf3 isoform. By contrast, neither the ϳ80-kDa nor the ϳ90-kDa Nrf3 proteins were markedly affected by treatment with TG or tBHQ (Fig. 11, C and D). Further, treatment with ALLN increased the abundance of both the ϳ70-kDa and the ϳ80-kDa polypeptides in COS-1 cells expressing Nrf3 ⌬12-30 protein (Fig. 11E). Taken together, these findings indicate that both the ϳ70-kDa and ϳ80-kDa Nrf3 polypeptides are active.
Nrf3 Does Not Competitively Inhibit Nrf2 Activity-We cotransfected COS-1 cells with expression constructs for Nrf2 and Nrf3, together with the P SV40 GSTA2-6ϫARE-Luc reporter plasmid, to test whether the two bZIP factors compete for ARE enhancers. This reporter gene was activated by forced overexpression of mouse Nrf3 in a dose-dependent fashion (Fig. 12A). Under normal and sulforaphane-stimulated conditions, overexpression of mouse and human Nrf3 did not competitively inhibit Nrf2 activity (Fig. 12B), with the exception that human Nrf3 modestly inhibited Nrf2-mediated transactivation only in tBHQ-treated cells (p Ͻ 0.10). We noted that mouse Nrf3 augmented the transactivation activity of Nrf2 in the presence of either tBHQ or sulforaphane, but the increase in activity was not observed in the case of human Nrf3 (Fig. 12B).
Nrf3 Is Negatively Regulated by Its PEST Sequence-The mouse Nrf3 ⌬1-172 mutant, lacking its PEST sequence and NTD, augmented the P TK nqo1-ARE-Luc gene activity from ϳ1.7to ϳ3.8-fold (Fig. 9D). This was accompanied by an increase in the levels of ectopic proteins of ϳ50 and ϳ54 kDa in the nucleus (Fig. 9, B and C). This finding suggests that the PEST sequence, along with the NTD, negatively regulates Nrf3 activity. To test this idea, we deleted the PEST sequence from mouse and human Nrf3 to create mNrf3 ⌬137-152 and hNrf3 ⌬158 -172 . Fig. 12C1 reveals that by comparison with the wild-type proteins, both mNrf3 ⌬137-152 and hNrf3 ⌬158 -172 significantly increased transactivation of the P SV40 GSTA2-6ϫARE-Luc reporter gene from ϳ3.8-fold to between 6-and 7-fold. Immunoblotting showed that expression of two major polypeptides of between ϳ80 and ϳ90 kDa increased in COS-1 cells expressing mNrf3 ⌬137-152 and hNrf3 ⌬158 -172 (Fig. 12C2). Similar results were obtained from mNrf3 ⌬101-136 , lacking 36 amino acids between the PEST and the NHB2 sequences (Fig.  12C). These findings indicate that the PEST sequence and its N-terminal flanking residues contribute to negative regulation of Nrf3 probably by directing its proteolysis.

DISCUSSION
In the present report, we have demonstrated that Nrf3 is targeted to the ER through an NHB1 signal sequence, and that this motif is required to enable the activity of Nrf3 to be increased by the ER stressors TU and BFA. . Wild-type mouse Nrf3 is a membrane-associated protein that is localized in the ER and the nuclear envelope, while an Nrf3 mutant lacking NHB2 is retained in the ER lumen. A, COS-1 cells expressing the Nrf3 ⌬76 -100 mutant or the NTD ⌬76 -100 /Nrf2 fusion protein were subjected to subcellular fractionation. C-terminally V5-tagged protein in the different fractions was resolved using 4 -12% NuPAGE Tris-Bis and visualized by Western blotting. B, total lysates of COS-1 cells expressing wild-type Nrf2 and NTD ⌬76 -100 /Nrf2 fusion protein were subjected to electrophoretic resolution followed by immunoblotting with antibodies against the V5 epitope. The amount of protein added to each electrophoresis sample well was adjusted to ensure equal loading of ␤-gal activity. C, the ER fractions were purified from COS-1 cells expressing Nrf3 ⌬76 -100 and NTD ⌬76 -100 /Nrf2 proteins, and after washing the fractions were resuspended in an isotonic extraction buffer. These membrane samples were incubated with either proteinase K (ϩ) at a final concentration of 100 g/ml or in buffer lacking PK (Ϫ), plus 1% Triton X-100 (ϩ), or not (Ϫ), for the indicated time from 5 to 30 min. After the reactions were stopped, C-terminally V5-tagged proteins were analyzed by Western blotting. The blot was reprobed with an antibody against CRT, to confirm integrity of the ER membrane. D, the intact ER and microsome-containing membrane (M) fractions were purified from COS-1 cells expressing wild-type Nrf3. These membrane samples were incubated for 30 min with either PK (ϩ) or without PK (Ϫ). After the reactions were stopped, C-terminally V5-tagged Nrf3 protein was visualized by Western blotting. The nitrocellulose membrane was reprobed with an antibody against CRT and Sec61␣. E, COS-1 cells expressing wild-type Nrf3 were subjected to subcellular fractionation. V5-tagged Nrf3 polypeptides were resolved by electrophoresis and immunoblotted with V5 antibody. The samples were as follows: cytosol (C) fraction from the 100,000 ϫ g supernatant; microsome-containing (M) membrane fraction; nuclear envelope (NE) isolated from the purified whole nuclei fraction; salt-extracted nuclear fraction (SN); total cell lysate (T). Immunoblotting with antibodies against actin, CRT, lamin A/C, and retinoblastoma 1 (Rb1) was undertaken to confirm the efficacy of subcellular fractionation.

Nrf3 Is a Membrane-bound Glycoprotein
Nrf3 Is a Transcriptional Activator That Is Negatively Regulated by Its NTD-Our data show that mouse Nrf3 can transactivate an ARE-driven reporter gene in COS-1 cells, although it is less active than either Nrf1 or Nrf2. This is consistent with previous results obtained with a reporter construct based around ARE-containing sequences from the chicken ␤-globin gene (1). We have also presented evidence that the activity of Nrf3 is modestly increased by tBHQ and sulforaphane. Although Nrf3 is significantly less active than Nrf2, we found it does not blunt induction of ARE-driven gene expression mediated by the latter factor. These results appear to contradict the report of Sankaranarayanan and Jaiswal (23) that mouse Nrf3 negatively regulates expression of a reporter gene driven by an ARE from human NQO1. It should however be noted that the human NQO1-ARE contains an embedded AP-1 sequence (5Ј-TGACTCAGC-3Ј, the AP-1 motif is underlined), whereas our P TK nqo1-ARE-Luc does not. Thus, the apparent discrepancy is likely to be due to the absence of an AP-1 motif within the ARE reporter construct we employed. We propose that, in the case of an ARE containing an AP-1 binding site, Nrf3 may outcompete Nrf2 through it being recruited to the AP-1 sequence as a heterodimer with c-Fos or Fra1, thereby causing negative regulation of the reporter gene. This does not, however, occur in the case of an ARE that lacks an AP-1 binding site.
Others have shown previously that human Nrf3 contains a transcription activation domain between residues 171 and 399 but that its activity is inhibited by the first 88 N-terminal residues (2). We have similarly found that attachment of residues 1-126 from human Nrf3 to the N terminus of Nrf2 reduced the activity of the fusion protein to only ϳ15% of that of wild-type Nrf2, indicating that NTD serves as a negative regulation domain. By contrast with the NTD of human Nrf3, the NTD of mouse Nrf3 was less effective at repressing Nrf2 when an equivalent fusion protein was created. The greater ability of the NTD from human Nrf3 to negatively regulate Nrf2, when compared with its murine counterpart, could be due to the absence of a proteolytic cleavage site in the hN126 sequence of the former bZIP protein.
Nrf3 Is Targeted to the ER through an NHB1 Signal Peptide-Our present data have shown that Nrf3, like Nrf1, is targeted to the ER through its NHB1-associated signal peptide (aa 1-39). Bioinformatics revealed that the signal peptide sequence in mouse Nrf3 comprises n, h, and c regions.
The n region (aa 1-11) in Nrf3 may regulate its topology within the membrane. According to the positive-inside rule (26), the n region that adjoins the membrane-spanning ␣-helix should reside on the cytoplasmic side of the ER. Deletion of this region in Nrf3 ⌬2-11 caused an increase in the abundance of the ϳ90-kDa Nrf3 glycoprotein in the ER. This FIGURE 11. The NHB1 is required to enable activation of mouse Nrf3 by certain ER stressors. A and B, COS-1 cells were cotransfected with an expression construct for wild-type mouse Nrf3 and its mutant Nrf3 ⌬12-30 lacking the NHB1 sequence, along with the P SV40 GSTA2-6ϫARE-Luc and pRL-TK reporter plasmids. Following transfection, the cells were treated for 24 h with DMSO (0.1%), TU (1 g/ml), TG (1 M), BFA (1 M), ALLN (5 g/ml), or tBHQ (50 M) either individually or in combination. Subsequently, cell lysates were prepared and luciferase reporter assays were performed. ARE-driven gene activity was normalized for transfection efficiency, and data are presented as a -fold change (mean Ϯ S.D.) as described above. Significant differences were calculated using the Student's t test. C-E, COS-1 cells were cotransfected with an expression construct for Nrf3 or Nrf3 ⌬12-30 along with the ␤-gal plasmid, and then treated with the chemicals as described above. Subsequently, total lysates (C and E) of the cells, as well as its nuclear fraction (D), were analyzed by Western blotting with a V5 antibody. The amount of total lysate added to each electrophoresis sample well was adjusted to ensure equal loading of ␤-gal activity. Re-blotting with antibodies against actin and lamin A/C was used as a loading control.

Nrf3 Is a Membrane-bound Glycoprotein
amphipathic regions. The first transmembrane region (called TM1) is likely folded by residues 7-25 and is orientated in an N cyt /C lum fashion (Fig. 8C). However, if the TM1-adjoining signal peptide is cleaved by an SPase, the remaining portion of Nrf3 might either be released from the membrane to the cytoplasm or may translocate into the lumen of the ER.
Our observation that membrane proteinase protection yielded two bands of ϳ70-kDa and ϳ85-kDa suggests that an additional region within the NTD of Nrf3 may be exposed to the cytoplasmic side of the ER. The region in question may comprise residues within and around NHB2. This suggestion is supported by our finding that, upon deletion of the NHB2 sequence, a ϳ88-kDa protein is recovered from the proteinase protection assay.
During the proteinase protection experiments we observed a C-terminally V5-tagged Nrf3 peptide of ϳ12 kDa. The recovery of this peptide suggests that the bZIP protein contains a C-terminal transmembrane (called TMc) region with an N cyt /C lum orientation (supplemental Fig. S4). The TMc region may be formed by residues 625-643. We also predict that, once residues 361-379 insert into the membrane lipids, they may fold into a transmembrane amphipathic ␣-helix (called TMi), which could either lie on the plane of the membrane or span the lipid bilayer (Fig. 13).
Concluding Comments-In this study we have shown that both mouse and human Nrf3 are targeted to the ER through the NHB1 sequence within their respective N-terminal domains. Mouse Nrf3 is subject to N-terminal proteolytic cleavage, whereas the human protein is not. Both proteins are glycosylated in the ER and are also located in the nuclear envelope. Interestingly, although Nrf1 is also targeted to the ER and located in the nuclear envelope (32), our data suggest that Nrf3 is integrated into membranes in a more complete fashion. Proteinase protection assays have shown that the topology of Nrf3 within the membrane is modulated by NHB2 within the NTD.