|Home | About | Journals | Submit | Contact Us | Français|
The unfolded protein response (UPR) controls the levels of molecular chaperones and enzymes involved in protein folding in the endoplasmic reticulum (ER). We recently isolated ATF6 as a candidate for mammalian UPR-specific transcription factor. We report here that ATF6 constitutively expressed as a 90-kDa protein (p90ATF6) is directly converted to a 50-kDa protein (p50ATF6) in ER-stressed cells. Furthermore, we showed that the most important consequence of this conversion was altered subcellular localization; p90ATF6 is embedded in the ER, whereas p50ATF6 is a nuclear protein. p90ATF6 is a type II transmembrane glycoprotein with a hydrophobic stretch in the middle of the molecule. Thus, the N-terminal half containing a basic leucine zipper motif is oriented facing the cytoplasm. Full-length ATF6 as well as its C-terminal deletion mutant carrying the transmembrane domain is localized in the ER when transfected. In contrast, mutant ATF6 representing the cytoplasmic region translocates into the nucleus and activates transcription of the endogenous GRP78/BiP gene. We propose that ER stress-induced proteolysis of membrane-bound p90ATF6 releases soluble p50ATF6, leading to induced transcription in the nucleus. Unlike yeast UPR, mammalian UPR appears to use a system similar to that reported for cholesterol homeostasis.
The endoplasmic reticulum (ER) provides the optimal environment for proper folding and assembly of newly synthesized proteins destined for the secretory pathway (Gething and Sambrook, 1992 ; Helenius et al., 1992 ; Huppa and Ploegh, 1998 ). Eukaryotic cells from yeast to human have developed specific and tight communication systems between the ER and cell nucleus to gear the folding capacity in the ER to the requirements within the ER. Thus, accumulation of unfolded proteins under so-called “ER stress” conditions leads to induction of transcription of genes encoding molecular chaperones and folding enzymes localized in the ER (Lee, 1987 ; Kozutsumi et al., 1988 ; Normington et al., 1989 ; Rose et al., 1989 ). This transcriptional induction process coupled with intracellular signaling from the ER to the nucleus is called the unfolded protein response (UPR) (McMillan et al., 1994 ; Shamu et al., 1994 ).
The most proximal event in the UPR is the sensing step of ER stress, which is considered to be mediated by Ire1p/Ern1p, a transmembrane protein kinase in the ER originally isolated by genetic screening in Saccharomyces cerevisiae (Cox et al., 1993 ; Mori et al., 1993 ). Although the detailed mechanism is unknown, Ire1p is activated via oligomerization and autophosphorylation upon accumulation of unfolded proteins in the ER (Shamu and Walter, 1996 ; Welihinda and Kaufman, 1996 ). Ire1p is likely to be negatively regulated by Ptc2p, a serine/threonine-specific protein phosphatase (Welihinda et al., 1998 ). Recent cloning of the mammalian homologue of yeast Ire1p revealed the presence of at least two different molecular species in mammalian cells (Tirasophon et al., 1998 ; Wang et al., 1998 ). Overexpression of either type of mammalian Ire1p constitutively activated transcription of ER chaperone genes as in the case of yeast Ire1p, suggesting similarity in the sensing system between yeast and mammals. Interestingly, mammalian ER contains one more transmembrane protein kinase called PEK/PERK, the lumenal domain of which shows significant homology to that of Ire1p (Shi et al., 1998 ; Harding et al., 1999 ). PEK/PERK is capable of phosphorylating the α subunit of eukaryotic translation initiation factor 2 and thus appears to be responsible for ER stress-induced translational attenuation (Prostko et al., 1992 ; reviewed by Kaufman, 1999 ).
The most distal step in the UPR is the transcription step that depends on the specific interaction of a trans-acting factor with a cis-acting element present in the promoter regions of UPR target genes. In S. cerevisiae, the cis-acting unfolded protein response element (UPRE) was originally identified as a sequence necessary and sufficient for induction of the KAR2 gene encoding the yeast homologue of a major ER chaperone, GRP78/BiP (Mori et al., 1992 ; Kohno et al., 1993 ). We recently showed that KAR2 UPRE contains an E box-like partially palindromic sequence separated by a spacer of one nucleotide (CAGCGTG) that is essential for its function (Mori et al., 1996 ). Subsequent characterization of UPRE sequences responsible for induction of other ER stress-responsive genes led us to propose that this unique feature of UPRE explains why only a specific set of ER proteins are induced in yeast UPR (Mori et al., 1998 ).
In contrast to yeast UPR, multiple cis-acting elements appeared to be involved in the mammalian UPR (Wooden et al., 1991 ). Recently, however, we and others independently found that a unique sequence with a consensus of CCAAT-N9-CCACG is commonly present in promoter regions of almost all ER stress-responsive mammalian genes (Yoshida et al., 1998 ; Roy and Lee, 1999 ). This novel cis-acting element was designated the ER stress response element (ERSE). By extensive mutational analysis, we demonstrated that ERSE is necessary and sufficient for induction of at least three major ER chaperones (GRP78, GRP94, and calreticulin) (Yoshida et al., 1998 ). The CCAAT portion of ERSE was confirmed to be a binding site for the general transcription factor NF-Y/CBF as shown previously (Roy and Lee, 1995 ; Roy et al., 1996 ). Thus, we proposed that the CCACG portion of ERSE exactly 9 bp downstream of the CCAAT portion provides specificity of the mammalian UPR (Yoshida et al., 1998 ). The CCACG may correspond to a half site of the palindromic sequence found in the yeast UPRE.
In yeast, Hac1p/Ern4p was identified as the UPR-specific transcription factor (Cox and Walter, 1996 ; Mori et al., 1996 ; Nikawa et al., 1996 ). Hac1p contains a basic leucine zipper (bZIP) motif and specifically binds to the E box-like palindromic sequence in UPRE. Yeast cells lacking Hac1p show the same phenotype as those lacking Ire1p; cells are unable to induce transcription of ER chaperone genes in response to ER stress. Subsequent analysis conducted in Walter’s and our laboratories revealed an unexpected connection between the most proximal event in the ER and the most distal event in the nucleus, namely the unconventional splicing system for HAC1 mRNA. HAC1 mRNA is constitutively expressed, but its translation is tightly blocked because of the presence of an intron of 252 nucleotides within the Hac1p-coding region. The HAC1 intron is specifically removed by a splicing event that is activated by signaling from the ER. Spliced mRNA is translated, and the Hac1p thus synthesized activates transcription of UPR target genes through the UPRE (Cox and Walter, 1996 ; Chapman and Walter, 1997 ; Kawahara et al., 1997 , 1998 ). Enzymes responsible for this unconventional mRNA splicing were identified by Walter and colleagues. The splicing reaction is initiated by Ire1p, a sensor molecule of ER stress by itself; both 5′- and 3′ splice sites are cleaved by the action of the C-terminal endonuclease domain of Ire1p, which is postulated to be activated upon ER stress (Sidrauski and Walter, 1997 ). The splicing reaction is completed by Rlg1p (Sidrauski et al., 1996 ), a tRNA ligase located in the nucleus (Clark and Abelson, 1987 ). This unique system allows yeast cells to produce transcription factor Hac1p only when they sense ER stress and require increased amounts of ER chaperones to cope with unfolded proteins accumulated in the ER (reviewed by Sidrauski et al., 1998 ).
Recently, we screened for putative human ERSE-binding proteins and obtained the bZIP protein ATF6 as a candidate (Yoshida et al., 1998 ). The amino acid sequence of the ATF6 basic region shows significant similarity with that of yeast Hac1p. In contrast to Hac1p, however, ATF6 is constitutively expressed as a 90-kDa protein, and ATF6 mRNA is not spliced in response to ER stress (Zhu et al., 1997 ; Yoshida et al., 1998 ). Here, we analyzed the mechanism of activation of ATF6 by ER stress and revealed an important step that connects the event in the ER with that in the nucleus.
Recombinant DNA techniques were performed according to standard procedures (Sambrook et al., 1989 ). The ATF6 expression plasmid pCGN-ATF6 (Zhu et al., 1997 ) was kindly provided by Dr. R. Prywes (Columbia University, New York, NY). This plasmid contains the full-length ATF6 cDNA at the XbaI site of the mammalian expression vector pCGN and thus expresses ATF6 protein tagged with the influenza virus hemagglutinin (HA) epitope at the N terminus under the control of the cytomegalovirus promoter [referred to here as pCGN-ATF6 (670)]. pCGN-ATF6 (402), pCGN-ATF6 (373), and pCGN-ATF6 (366) encoding truncated forms of ATF6 were constructed by PCR-mediated amplification of the region corresponding to amino acids 1–402, 1–373, and 1–366 of ATF6, respectively, together with a stop codon (TAG) followed by insertion of the resultant fragments into the XbaI site of pCGN after their sequences had been confirmed.
HeLa cells were grown in Dulbecco’s modified Eagle’s medium supplemented with 10% fetal calf serum, 2 mM glutamine, 100 U/ml penicillin, and 100 μg/ml streptomycin sulfate. Cells were maintained in a 5% CO2 incubator at 37°C. Transfection was carried out by the standard calcium phosphate method (Sambrook et al., 1989 ) as described previously (Yoshida et al., 1998 ) with some modifications. Briefly, HeLa cells plated at ~10% confluency 1 d before transfection were incubated with calcium phosphate–DNA complex for 16 h at 37°C. After washing with PBS, cells were cultured in fresh medium for 24 h before extraction for immunoblotting or fixation for immunofluorescence analysis unless otherwise specified. Transfection efficiency depended on the amount of DNA used: usually ~5% with 1 μg of DNA and ~50% with 10 μg of DNA. To induce UPR, cells were treated with 2 μg/ml tunicamycin (Sigma, St. Louis, MO), 300 nM thapsigargin (Sigma), or 1 mM dithiothreitol for various periods as indicated.
Rabbit anti-B03N antiserum raised against the N-terminal region of ATF6 (amino acids 6–307) fused to Escherichia coli maltose-binding protein (Yoshida et al., 1998 ) was purified by removing materials bound to CH-Sepharose 4B (Amersham Pharmacia Biotech, Buckinghamshire, UK) on which soluble proteins extracted from E. coli cells producing large amounts of maltose-binding protein from the plasmid pMAL-c2 (New England BioLabs, Beverly, MA) had been immobilized. The flow-through fraction highly specific to ATF6 (see Figure Figure1)1) was used as anti-ATF6 antibody.
Immunoblotting analysis was carried out according to the standard procedure (Sambrook et al., 1989 ) as described previously (Yoshida et al., 1998 ) using an enhanced chemiluminescence Western blotting detection system kit (Amersham Pharmacia Biotech). Mouse anti-KDEL monoclonal antibody (clone 10C3), mouse anti-HSP70 monoclonal antibody (clone C92F3A-5), and rabbit antisera against the N or C terminus of calnexin were obtained from StressGen Biotechnologies (Victoria, British Columbia, Canada). Rabbit anti-HA epitope polyclonal antibody (Y-11) and goat anti-lamin B polyclonal antibody (M-20) were obtained from Santa Cruz Biotechnology (Santa Cruz, CA).
HeLa cells cultured in 60-mm dishes until ~80% confluency were incubated for 20 min in methionine- and cysteine-free Dulbecco’s modified Eagle’s medium supplemented with 2 mM glutamine, antibiotics, and 10% dialyzed fetal calf serum. Cells were then pulse labeled for 30 min using 0.5 mCi (18.5 MBq)/plate EXPRE35S35S protein labeling mix (DuPont, Wilmington, DE) dissolved in 1 ml of the above medium. To chase 35S-labeled proteins, cells were washed with fresh complete medium three times and incubated in fresh complete medium for various periods. After removal of medium, cells were lysed by incubation for 10 min on ice in 0.1 ml of lysis buffer (1× Tris-buffered saline containing 1% SDS, 1 mM PMSF, and 5 μg/ml leupeptin). The lysates were boiled for 5 min and, after addition of 0.4 ml of dilution buffer (1× Tris-buffered saline containing 2% Triton X-100), clarified by centrifugation at 15,000 × g for 10 min. The supernatant was incubated with 50 μl of normal rabbit serum (Sigma) for 3 h at 4°C, and the mixture was rotated overnight at 4°C after addition of 50 μl of 50% protein A-Sepharose 4 Fast Flow (Amersham Pharmacia Biotech). The mixture was clarified by brief centrifugation, and 10 μl of anti-ATF6 antibody were added to the resulting supernatant. After standing for 3 h at 4°C, 50 μl of 50% protein A-Sepharose 4 Fast Flow were added, and the mixture was rotated for 2 h at 4°C. The resin was then washed as described by Franzusoff et al. (1991) , and the immunoprecipitates were subjected to SDS-PAGE.
HeLa cells were grown directly on slide glasses in 90-mm dishes. Cells were fixed with PLP solution (2% p-formaldehyde, 10 mM sodium metaperiodate, and 75 mM l-lysine in PBS) for 10 min at room temperature, permeabilized with ice-cold acetone for 10 min, and then reacted with various primary antibodies for 1 h at 37°C. Rabbit anti-ATF6 antibody was diluted by 50-fold with 1% BSA-TBST solution (10 mM Tris-HCl, pH 8.0, 150 mM NaCl, 0.05% Tween 20, and 1% BSA) before use. Rabbit anti-HA epitope antibody and mouse anti-KDEL antibody were diluted with 1% BSA-TBST solution to concentrations of 4 and 20 μg/ml, respectively. Primary antibodies were visualized by incubation for 1 h at 37°C with either 10 μg/ml FITC-conjugated goat anti-rabbit immunoglobulin G antibody (Vector Laboratories, Burlingame, CA) or 50 μg/ml rhodamine-conjugated goat anti-mouse immunoglobulin G antibody (ICN Pharmaceuticals, Aurora, OH). The cells were counterstained with 0.3 μg/ml DAPI in PBS-glycerol (1:9). Photographs were taken using a 465- to 495-nm excitation filter and a 515- to 555-nm emission filter for fluorescein isothiocyanate and a 525- to 540-nm excitation filter and a 605- to 655-nm emission filter for rhodamine to avoid mutual interference.
We and others have shown that ATF6 is constitutively expressed in HeLa cells as a 90-kDa protein (p90ATF6) (Zhu et al., 1997 ; Yoshida et al., 1998 ). Interestingly, however, we found that the p90ATF6 level decreased, and instead a new band of 50 kDa (p50ATF6) appeared in ER-stressed cells before the induction of GRP78, a major target protein of the mammalian UPR (Yoshida et al., 1998 ). As the anti-B03N antiserum used in the previous study reacted with multiple proteins present in HeLa cell extracts, we purified it as described in MATERIALS AND METHODS. The purified antibody recognized p90ATF6 almost exclusively on immunoblotting of extracts of unstressed HeLa cells (Figure (Figure1A,1A, lane 1). We thus used this as anti-ATF6 antibody, which should recognize the N-terminal region of ATF6 (amino acids 6–307, indicated by the thick line in Figure Figure4B).4B).
We reexamined the time course of the conversion of p90ATF6 to p50ATF6 using the purified antibody in HeLa cells treated with thapsigargin, which causes ER stress by inhibiting the ER Ca2+-ATPase (Li et al., 1993 ). Consistent with our previous report (Yoshida et al., 1998 ), p90ATF6 level decreased, and p50ATF6 appeared from 1 h after thapsigargin treatment, whereas an increase in the level of GRP78 was detected from 4 h (Figure (Figure1A).1A). To examine whether p90ATF6 was directly converted to p50ATF6 upon ER stress, we performed pulse–chase experiments (Figure (Figure1B).1B). ATF6 was immunoprecipitated from HeLa cells pulse labeled for 30 min with [35S]methionine and [35S]cysteine and then chased for various periods. In the absence of thapsigargin in the chase medium, p90ATF6 level decreased with a half-life of ~2 h, and no band corresponding to p50ATF6 was detected (Figure (Figure1B,1B, lanes 1–5). In the presence of thapsigargin, p50ATF6 appeared after the chase (Figure (Figure1B,1B, lanes 6–10), suggesting that ATF6 is synthesized as a precursor protein (p90ATF6) in unstressed cells and processed into a mature form (p50ATF6) in response to ER stress.
To further confirm whether ER stress did not induce de novo synthesis of p50ATF6, we examined the effects of blocking new protein synthesis on the conversion of p90ATF6 to p50ATF6. It was recently shown that PERK, a transmembrane protein kinase involved in ER stress-induced translational attenuation, was activated by thapsigargin treatment of COS-1 cells even in the absence of protein synthesis (Harding et al., 1999 ). In marked contrast, p50ATF6 was not produced when HeLa cells (Figure (Figure2,2, lanes 1–3) or COS-1 cells (our unpublished observation) were pretreated for 30 min with cycloheximide, an inhibitor of protein synthesis, before thapsigargin treatment. The reason for the difference is currently unknown. Importantly, however, we found that the conversion of p90ATF6 to p50ATF6 could occur in the absence of protein synthesis under a certain condition such as dithiothreitol treatment, which disrupts disulfide bonding and thus malfolds proteins in the ER directly (Figure (Figure2,2, lanes 4–6). All of the above results strongly indicate that the appearance of p50ATF6 in ER-stressed cells is due to proteolytic cleavage of preexisting p90ATF6.
We also reexamined the time course of the processing of ATF6 in HeLa cells treated with tunicamycin, which causes ER stress by inhibiting N-glycosylation of newly synthesized proteins (Lee, 1987 ; Kozutsumi et al., 1988 ). Consistent with our previous report (Yoshida et al., 1998 ), p90ATF6 level decreased, and p50ATF6 appeared from 2 h after tunicamycin treatment, whereas an increase in GRP78 level was detected from 6 h (Figure (Figure3).3). Although the processing of ATF6 was significantly slower in tunicamycin-treated cells than that in thapsigargin-treated cells, appearance of p50ATF6 accompanied the increase in level of GRP78 mRNA (Figure (Figure3),3), suggesting that the proteolytic conversion of p90ATF6 to p50ATF6 is a key regulatory step in mammalian UPR.
The experiments shown in Figures Figures11–3 revealed the production of a novel band migrating slightly faster than p90ATF6 in addition to p50ATF6 in tunicamycin-treated cells but not in thapsigargin- or dithiothreitol-treated cells, suggesting that p90ATF6 might be glycosylated and thus associated with the ER. In fact, when unstressed HeLa cells were examined by indirect immunofluorescence analysis (Figure (Figure4A),4A), fine reticular structures surrounding the nucleus were observed using purified anti-ATF6 antibody, and the staining pattern was very similar to that obtained with anti-KDEL antibody, which recognizes two major ER chaperones (GRP78 and GRP94) in HeLa cells. Furthermore, hydropathy analysis identified a hydrophobic stretch of 21 amino acids (amino acids 378–398) near the center of the ATF6 molecule, which is long enough to span the membrane once (Figure (Figure4B).4B). These results suggested that p90ATF6 is a transmembrane protein in the ER.
We conducted fractionation experiments using untreated HeLa cells and those treated with tunicamycin for 4 h as starting materials to monitor recovery of p90ATF6 and p50ATF6, respectively (Figure (Figure5).5). After the first low-speed centrifugation of the homogenate, p50ATF6 was recovered exclusively in the nuclear pellet similarly to the nuclear protein lamin B (Figure (Figure5,5, lane 4), whereas the majority of p90ATF6 remained in the supernatant (Figure (Figure5,5, lane 5). After the second high-speed centrifugation, all p90ATF6 was recovered in the membrane fraction (Figure (Figure5,5, lane 9). The distribution pattern of p90ATF6 was almost identical to that of calnexin, a transmembrane-type chaperone localized in the ER (Wada et al., 1991 ), but completely different from those of lamin B (Moir et al., 1995 ) and the cytosolic protein HSP70 (Welch and Suhan, 1986 ), revealing that p90ATF6 is indeed associated with membranes. It should be noted that p90ATF6 remaining in tunicamycin-treated cells as well as a protein band migrating slightly faster than p90ATF6 (lane 2) were distributed in each fraction similarly to p90ATF6 in untreated cells.
To determine whether p90ATF6 is a peripheral or integral membrane protein, differential solubilization experiments were carried out (Figure (Figure6A).6A). The supernatant fraction obtained after low-speed centrifugation of untreated HeLa cell homogenate was treated with various reagents and then centrifuged at high speed. Neither p90ATF6 nor calnexin was released from membranes by 0.5 M NaCl or 0.1 M Na2CO3, pH 11, treatment, which should extract peripheral membrane proteins. In contrast, both proteins were released into the supernatant by detergents such as 1% SDS and 1% sodium deoxycholate, whereas only calnexin was extracted by 1% Triton X-100; Triton X-100 may cause aggregation of p90ATF6 as in the case of yeast Ire1p (Mori et al., 1993 ). Thus, p90ATF6 is an integral membrane protein.
Next, the orientation of p90ATF6 in membranes was examined by trypsin treatment (Figure (Figure6B).6B). Calnexin, a type I transmembrane protein, was used as a control and digestion of calnexin was monitored using two different antibodies specific to either the N-terminal region present in the ER lumen (Calnexin-N) or the C-terminal region located in the cytoplasm (Calnexin-C). At the concentration of trypsin that decreased the amount of full-length calnexin, anti-Calnexin-N antibody (Figure (Figure6B,6B, lanes 7 and 8) but not anti-Calnexin-C antibody (Figure (Figure6B,6B, lanes 11 and 12) detected a fragment the size of which matched that of the N-terminal region of calnexin, indicating that proteins or segments in the lumen were protected from digestion by trypsin as expected. Under these conditions, p90ATF6 disappeared at the lowest concentration of trypsin used, and no fragments of ~50 kDa were detected by anti-ATF6 antibody raised against the N-terminal region of ATF6 (Figure (Figure6B,6B, lane 2). These results strongly indicated that the N-terminal domain of ATF6 containing the bZIP region is oriented facing the cytoplasm.
Examination of the amino acid sequence of the lumenal domain of ATF6 indicated the presence of three potential glycosylation sites as schematically presented in Figure Figure4B.4B. To determine whether p90ATF6 is actually glycosylated, unstressed HeLa cell extracts were denatured and treated with endoglycosidase H followed by immunoblotting analysis (Figure (Figure6C).6C). p90ATF6 treated with endoglycosidase H (Figure (Figure6C,6C, lane 3) migrated faster than untreated p90ATF6 (Figure (Figure6C,6C, lane 2) but migrated at the same position as in vitro–translated full-length ATF6 (Figure (Figure6C,6C, lane 1), whereas the mobility of calnexin was unaffected by endoglycosidase H treatment (Figure (Figure6C,6C, lanes 4 and 5); calnexin contains no potential glycosylation sites. The sensitivity of the carbohydrate moiety to digestion with endoglycosidase H implied that p90ATF6 stays in the ER and is not subjected to retrograde transport from the Golgi apparatus.
We further determined the localization of p90ATF6 after transfection. pCGN-ATF6 (670) expressed full-length ATF6 tagged with the HA epitope at the N terminus (see MATERIALS AND METHODS). Immunofluorescence analysis of transfected HeLa cells using anti-HA epitope antibody revealed that expression of ATF6 (670) was restricted to perinuclear structures, which were also stained with anti-KDEL antibody (Figure (Figure6D,6D, compare a with b and c), demonstrating that p90ATF6 was localized in the ER even when overexpressed. Taken together, we concluded that p90ATF6 is a type II transmembrane glycoprotein localized in the ER.
In contrast to the tight association of p90ATF6 with ER membranes, p50ATF6 detected in ER-stressed cells was recovered exclusively in the nuclear pellet fraction (Figure (Figure5,5, lane 4). As a small portion of p90ATF6 was also recovered in the same fraction (Figure (Figure5,5, lane 3), we further extracted the nuclear pellet with high salt buffer (Figure (Figure7A).7A). As a result, more than half of p50ATF6 was released into the supernatant (Figure (Figure7A,7A, compare lane 2 with lane 4), whereas all p90ATF6 remained in the pellet (Figure (Figure7A,7A, compare lane 1 with lane 3). Furthermore, when HeLa cells were extracted by repeated freezing and thawing (Figure (Figure7B),7B), all p50ATF6 was recovered in the supernatant (Figure (Figure7B,7B, compare lane 2 with lane 4), whereas all p90ATF6 was recovered in the pellet (Figure (Figure7B,7B, compare lane 1 with lane 3). From these results, we concluded that p50ATF6 is a soluble nuclear protein in marked contrast to p90ATF6.
Nonetheless, we could not determine the localization of endogenous p50ATF6 by immunofluorescence analysis. We reproducibly observed decreased staining of the ER with anti-ATF6 antibody 4 h after tunicamycin treatment but hardly detected increased staining of the nucleus, probably because the small amounts of p50ATF6 produced diffused out in the nucleus, the volume of which was much larger than that of the ER (see Figure Figure4A).4A). Because of this technical problem, it was also not possible to visualize the change in localization of ATF6 from the ER to the nucleus after tunicamycin or thapsigargin treatment of HeLa cells transfected with pCGN-ATF6 (670) and thus overproducing full-length ATF6. It should be noted that in thapsigargin-treated cells the amounts of p90ATF6 were much greater than those of p50ATF6 at all time points (Figure (Figure1A).1A). Similarly, the total amounts of p90ATF6 plus unglycosylated p90ATF6 were much greater than those of p50ATF6 in tunicamycin-treated cells at all time points (Figure (Figure3).3). To circumvent this difficulty, we overexpressed various ATF6 mutants and examined their localization in transiently transfected HeLa cells (Figure (Figure8).8).
The above results indicated that ER stress triggers processing of ATF6, which not only decreases its molecular mass from 90 to 50 kDa but also alters its subcellular localization. We hypothesized that such dual outcomes could result from proteolysis of ATF6 in the C-terminal region; p50ATF6 reacts with anti-ATF6 antibody raised against the N-terminal region of ATF6. We therefore constructed various C-terminal deletion mutants and examined the localization of expressed proteins (see schematic structures depicted in Figure Figure8A).8A). The boundaries of three functional domains should be noted; the basic region (R308-Y330), the leucine zipper (L334-X6-A-X6-L-X6-L-X6-V-X6-L369), and the transmembrane domain (V378-L398). Thus, ATF6 (402) lacked the majority of the C-terminal lumenal domain but retained the transmembrane domain, whereas two mutants, ATF6 (373) and ATF6 (366), lacked both lumenal and transmembrane domains. ATF6 (373) contained the entire bZIP region. ATF6 (366) contained the entire basic region and majority of the leucine zipper region. All mutant proteins were tagged with the HA epitope at the N terminus (Figure (Figure88A).
Before localization analysis, we examined by immunoblotting whether these mutant forms of ATF6 were actually expressed in transfected HeLa cells (Figure (Figure8B).8B). Simultaneously, this analysis provided useful information on the cleavage site in p90ATF6. The doublet protein bands detected at ~50 kDa in cells transfected with pCGN-ATF6 (670) (Figure (Figure8B,8B, lane 2) served as a molecular size marker for p50ATF6; these proteins were considered to represent p50ATF6, which was produced by constitutively activated proteolysis of the HA-tagged ATF6 (670) overexpressed in transfected cells (see DISCUSSION for explanation). The faster-migrating band might have lost the HA tag. Comparison of the mobilities of various C-terminal deletion mutants on SDS-PAGE revealed that the size of p50ATF6 was close to those of ATF6 (373) and ATF6 (366) (Figure (Figure8B,8B, lanes 4 and 5, respectively), suggesting that p90ATF6 may be cleaved between the bZIP and transmembrane domains to produce p50ATF6 when the cellular UPR was activated. We thus used ATF6 (373) and ATF6 (366) as representatives of p50ATF6 in the following experiments.
Immunofluorescence analysis showed that ATF6 (402) was localized in the ER (Figure (Figure8C,8C, compare a with b and c), indicating that the C-terminal lumenal domain is dispensable for the association with ER membranes. In marked contrast, ATF6 mutants lacking both lumenal and transmembrane domains entered the nucleus; the nucleus was clearly stained with anti-HA epitope antibody in cells overexpressing ATF6 (373) and ATF6 (366) (Figure (Figure8C,8C, compare d and g with e, f, h, and i). These results demonstrated that the localization of ATF6 is determined by the presence or absence of the transmembrane domain in the molecule and strongly suggested that the ATF6 molecule released from ER membranes (p50ATF6) can translocate into the nucleus.
We finally examined the effects of overexpression of ATF6 mutants on the levels of endogenous UPR targets at both mRNA and protein levels (Figure (Figure9).9). The level of GRP78 mRNA was slightly enhanced by overexpression of full-length ATF6, ATF6 (670), compared with the vector control (Figure (Figure9,9, compare lane 2 with lane 1), although ATF6 (670) was primarily localized in the ER (Figure (Figure6D;6D; see DISCUSSION for explanation). Most importantly, the levels of both GRP78 mRNA and GRP78 were markedly enhanced by overexpression of ATF6 (373) and ATF6 (366) (Figure (Figure9,9, lanes 3 and 4), both of which were localized in the nucleus (Figure (Figure8C).8C). We concluded that these two ATF6 mutants representing p50ATF6 directly activated transcription of the GRP78 genes in the nucleus, leading to enhanced levels of GRP78 in the ER.
Eukaryotic cells possess multiple intracellular signaling pathways from the ER to the nucleus, each modulating gene expression in response to changes in or surrounding the ER (reviewed by Pahl and Baeuerle, 1997 ). One of these, the UPR, deals with homeostasis of the folding capacity in the ER. Without this system, both yeast cells (Cox et al., 1993 ; Mori et al., 1993 , 1996 ; Nikawa et al., 1996 ) and mammalian cells (Li and Lee, 1991 ; Little and Lee, 1995 ; Morris et al., 1997 ; Liu et al., 1998 ) are unable to survive under ER stress conditions that cause continuous accumulation of unfolded proteins in the ER.
In this study, we analyzed the mechanism of activation of ATF6 by ER stress, which we recently isolated as a candidate for a mammalian UPR-specific transcription factor. We showed that constitutively expressed 90-kDa protein (p90ATF6) is directly converted to a 50-kDa protein (p50ATF6) specifically in ER-stressed cells before the induction of GRP78, a major target protein of the mammalian UPR (Figures (Figures11–3). The results presented here revealed the most important consequence of this conversion, that is, alteration of subcellular localization (see our current model depicted in Figure Figure10).10). Fractionation (Figure (Figure5),5), differential solubilization (Figure (Figure6A),6A), topological studies (Figure (Figure6B),6B), glycosylation analysis (Figure (Figure6C),6C), and immunoflurescence analysis (Figures (Figures4A4A and and6D)6D) clearly demonstrated that p90ATF6 is a type II transmembrane glycoprotein embedded in the ER membrane. In contrast, p50ATF6 is a soluble nuclear protein (Figures (Figures55 and and7).7). When overexpressed, ATF6 mutants representing p50ATF6 were accumulated in the nucleus (Figure (Figure8)8) and constitutively activated transcription of the GRP78 gene, leading to enhanced levels of GRP78 in the ER (Figure (Figure9).9). We thus propose that p50ATF6 represents an active and mature form of the mammalian transcription factor ATF6, and its production is mediated by ER stress-induced proteolysis of p90ATF6, which is synthesized as a precursor protein embedded in the ER membrane. Upon ER stress, p50ATF6 is released from the ER membrane, allowing it to enter the nucleus. In the nucleus, p50ATF6 containing a bZIP domain activates transcription of ER chaperone genes such as GRP78 through ERSE in collaboration with the general transcription factor NF-Y; we recently found that ATF6 recognizes ERSE and directly interacts with NF-Y (our unpublished observation).
Interestingly, p90ATF6 appears to be turned over fairly quickly; its half-life within the cell is ~2 h (Figure (Figure1).1). This rapid turnover rate allowed the cells to restore p90ATF6 at 16 h after thapsigargin treatment (Figure (Figure1A).1A). Similarly, p90ATF6 was restored at 16 h after tunicamycin treatment, although it was unglycosylated (Figure (Figure3).3). Therefore, p90ATF6 itself might serve as a sensor molecule of ER stress; under the conditions that cause accumulation of unfolded proteins in the ER, p90ATF6 would not be able to fold properly, and this may somehow activate proteolytic processing of p90ATF6, resulting in production of p50ATF6. In this connection, it is noteworthy that overexpression of full-length ATF6, ATF6 (670), constitutively activated transcription of the GRP78 gene, albeit only slightly (Figure (Figure9).9). We reasoned that overproduction of ATF6 (670), a transmembrane protein in the ER, is sensed as ER stress by the cell probably because normal levels of ER chaperones are insufficient for proper folding of exogenous proteins expressed at high levels. As a result, portions of endogenous p90ATF6 and exogenous ATF6 (670) are subjected to proteolytic processing constitutively, resulting in enhanced transcription of ER chaperone genes by constitutively produced p50ATF6 through ERSE in the nucleus. Indeed, p50ATF6-like doublet protein bands were detected in extracts of cells transfected with pCGN-ATF6 (670) (Figure (Figure8B,8B, lane 2). This also explains why various promoters of ER chaperone genes were constitutively activated when full-length ATF6 was overexpressed in HeLa cells (Yoshida et al., 1998 ).
We showed here that the mammalian UPR uses a system very similar to that previously identified for cholesterol homeostasis, another well-investigated intracellular signaling pathway from the ER to the nucleus (Yokoyama et al., 1993 ; Wang et al., 1994 ; Sakai et al., 1996 , reviewed by Brown and Goldstein, 1997 ). Cholesterol metabolism is primarily regulated at the level of transcription, and sterol regulatory element binding proteins (SREBPs) mediate transcriptional activation of genes involved in cholesterol biosynthesis as well as receptor-mediated endocytosis of cholesterol-containing lipoproteins from plasma. Under normal conditions, SREBPs are bound to membranes of the ER and nuclear envelope through two hydrophobic stretches separated by a spacer of 31 amino acids and located around the center of the molecule. Depletion of sterols from culture medium activates proteolytic cleavage of SREBPs at the two sites in a sequential manner, allowing entrance of the N-terminal fragment into the nucleus. This released N-terminal fragment contains all of the functional domains necessary for active transcription factors such as DNA-binding (basic helix–loop–helix), dimerization (leucine zipper), and transactivation (acid blob) domains and thus enhances transcription of target genes in the nucleus.
In contrast to SREBPs, the precise cleavage site of ATF6 is as yet unknown. The region from amino acid 378 to 398 functions as a transmembrane domain because it is the only hydrophobic segment found in ATF6 (Figure (Figure4B).4B). The calculated molecular weight of the N-terminal region (1–377) is 41,161 and is thus significantly smaller than the size of p50ATF6 estimated from the mobility on SDS-PAGE. However, yeast Hac1p behaves as a 41-kDa protein on SDS-PAGE despite its calculated molecular weight of 26,903, presumably because of high contents of basic amino acids (Kawahara et al., 1997 ). Similarly, full-length ATF6 behaves as a 90-kDa protein on SDS-PAGE despite its calculated molecular weight of 74,597 (Zhu et al., 1997 ; Yoshida et al., 1998 ). Comparison of the mobilities of various C-terminal deletion mutants with that of p50ATF6 on SDS-PAGE (Figure (Figure8B)8B) suggested that ATF6 is cleaved between the bZIP and transmembrane domains, although we cannot exclude the possibility that ATF6 is cleaved within the transmembrane domain, similarly to cleavage at site 2 in SREBP2 (Sakai et al., 1996 ), or cleaved around the boundary between the cytoplasmic and transmembrane domains, similarly to Notch-1 (Chan and Jan, 1998 ; Schroeter et al., 1998 ). We are currently carrying out mutational analysis to determine the precise cleavage site. Such information will be useful for identification of the protease(s) responsible for proteolysis of ATF6.
We have identified a key regulatory step that connects events in the ER with those in the nucleus in mammalian UPR, and our observations have raised an important question: i.e., whether ER stress-induced proteolysis of ATF6 is regulated by mammalian Ire1p, a putative sensor molecule of ER stress identified recently (Tirasophon et al., 1998 ; Wang et al., 1998 ). The detection of endonuclease activity in the C-terminal tail region of human Ire1p suggests the presence of a mammalian mRNA splicing system similar to that for yeast HAC1 mRNA (Tirasophon et al., 1998 ), but ATF6 mRNA is not spliced (Yoshida et al., 1998 ). This raises the question of whether there are any substrates of such a mammalian splicing system, and if so, what is the consequence of such mRNA splicing. In yeast, transcriptional induction of ER chaperones is coupled to phospholipid biosynthesis; yeast cells lacking either Ire1p or Hac1p require exogenous inositol for growth (Nikawa and Yamashita, 1992 ; Cox et al., 1993 ; Mori et al., 1993 , 1996 ; Nikawa et al., 1996 ; Sidrauski et al., 1996 ). It has been proposed that yeast UPR coordinates the synthesis of ER chaperones and ER membranes via the Ire1p–Hac1p pathway (Cox et al., 1997 ). This provides a basis for one intriguing speculation that in mammalian cells ER membrane biosynthesis may be controlled differently from transcriptional regulation of ER chaperone genes. Thus, the mammalian mRNA splicing system may be specialized to adjust the production of phospholipids according to the requirements within the ER. On the other hand, mammalian Ire1p may regulate the synthesis of ER chaperones by phosphorylating a putative protease(s), which activates ATF6. Answers to these important questions will further extend our understanding of the molecular mechanism of the mammalian UPR.
We thank Masako Nakayama, Mayumi Ueda, Hideaki Kanazawa, Rika Takahashi, Seiji Takahara, and Tomoko Yoshifusa for technical assistance.