Tumor Necrosis Factor α (TNFα) Induces the Unfolded Protein Response (UPR) in a Reactive Oxygen Species (ROS)-dependent Fashion, and the UPR Counteracts ROS Accumulation by TNFα*

Accumulation of unfolded proteins in the endoplasmic reticulum (ER) causes ER overload, resulting in ER stress. To cope with ER stress, mammalian cells trigger a specific response known as the unfolded protein response (UPR). Although recent studies have indicated cross-talk between ER stress and oxidative stress, the mechanistic link is not fully understood. By using murine fibrosarcoma L929 cells, in which tumor necrosis factor (TNF) α induces accumulation of reactive oxygen species (ROS) and cell death, we show that TNFα induces the UPR in a ROS-dependent fashion. In contrast to TNFα, oxidative stresses by H2O2 or arsenite only induce eukaroytic initiation factor 2α phosphorylation, but not activation of PERK- or IRE1-dependent pathways, indicating the specificity of downstream signaling induced by various oxidative stresses. Conversely, the UPR induced by tunicamycin substantially suppresses TNFα-induced ROS accumulation and cell death by inhibiting reduction of cellular glutathione levels. Collectively, some, but not all, oxidative stresses induce the UPR, and pre-emptive UPR counteracts TNFα-induced ROS accumulation.

Newly synthesized secretory and membrane-associated proteins are correctly folded and assembled in the endoplasmic reticulum (ER). 2 Once ER function is perturbed by various pathological conditions, newly synthesized unfolded proteins accumulate in the ER, resulting in ER stress. To cope with accumulated unfolded ER proteins, mammalian cells trigger a specific response termed the unfolded protein response (UPR) (1)(2)(3). There are three distinct signaling pathways that are trig-gered in response to ER stress, mediated by PERK, ATF6, and IRE1. Under non-pathological conditions, all three components associate with the abundant lumenal chaperon Bip (also known as glucose-regulated protein 78) and this interaction keeps these signaling molecules in an inactive state (4,5). Once unfolded proteins accumulate in the ER, Bip preferentially associates with the unfolded proteins instead of PERK, ATF6, and IRE1, resulting in activation of their downstream signaling molecules. PERK is an ER-resident serine/threonine protein kinase that phosphorylates the ␣ subunit of eukaryotic translation initiation factor 2 (eIF2␣) (6). Phosphorylation of eIF2␣ subsequently inhibits protein synthesis to prevent further influx of nascent proteins into an already saturated ER lumen. Paradoxically, eIF2␣ phosphorylation induces translation of a transcription factor ATF4 and subsequent expression of the ATF4 target genes, GADD34 and CHOP (7). The second signaling pathway is mediated by the basic leucine zipper-type transcription factor, ATF6. ATF6 is synthesized as a type II transmembrane precursor protein with a molecular mass with 90 kDa (p90 ATF6), and anchored to the ER membrane where it is retained by Bip. In response to ER stress, ATF6 is released from Bip and transported to the Golgi complex (8), where ATF6 undergoes sequential cleavages by two proteases, S1P and S2P (9). The processed form of ATF6 (p50ATF6) translocates to the nucleus and binds to the ER stress responsive element (ERSE), and subsequently activates target genes. Many ER chaperons, including Bip, glucose-regulated protein 94, and calreticulin, contain an ERSE in their promoter regions and are induced by ATF6 (10). The third signaling pathway is composed of IRE1 and XBP1. IRE1 is a type I transmembrane protein containing a serine threonine kinase and ribonuclease domains. Under normal conditions, only the unspliced form of XBP1[XBP1(U)] mRNA is translated, but its product is a weak transcriptional activator with a short protein half-life. Upon ER stress, IRE1 is activated and cuts 26 nucleotides out from XBP1(U) mRNA to generate spliced XBP1[XBP1(S)] mRNA, which encodes the more stable and transcriptionally active XBP1(S) protein. XBP1(S) binds to the unfolded protein response element (UPRE) and activates target genes (11)(12)(13).
There are four known kinases, including PERK, PKR, GCN2, and HRI, that phosphorylate eIF2␣ (2,14). PKR is activated by doublestranded RNA and phosphorylates eIF2␣ during viral infection. GCN2 and HRI play the equivalent role in amino acid-starved cells and hemedeprived reticulocytes, respectively. However, the kinase(s) responsible for phosphorylating eIF2␣ in response to oxidative stress remain to be identified (15). Interestingly, a recent study has revealed an intimate correlation between ER stress and accumulation of reactive oxygen spe-* This work was supported in part by grant-in-aid for 21st Century COE Research, grantin-aid for Scientific Research (B) from the Japan Society for the Promotion of Science, grant-in-aid for Scientific Research on Priority Areas (C) from The Ministry of Education, Culture, Sports, Science and Technology, Japan, the Takeda Science Foundation, and by a grant from Human Frontier Science Program. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. 1 To whom correspondence should be addressed. cies (ROS) (16). Whereas tunicamycin, an inhibitor of protein N-linked glycosylation, only weakly induced accumulation of ROS in wild-type cells, the same treatment induced marked accumulation of ROS in cells lacking PERK or ATF4. Furthermore, microarray analysis revealed that the PERK-and ATF4-dependent pathways induce genes associated with amino acid transport and reduced glutathione (GSH) biosynthesis, which might be responsible for elimination of ROS.
Although accumulating evidence has indicated cross-talk between oxidative stress and ER stress response pathways, its detailed molecular mechanism remains to be elucidated. By using murine fibrosarcoma L929 cells, in which TNF␣ induces ROS accumulation, we have investigated whether TNF␣ induces the UPR, and whether preemptive activation of the UPR induced by tunicamycin could affect ROS accumulation induced by TNF␣. Here we show that TNF␣, but not H 2 O 2 or arsenite, induces the UPR in a ROS-dependent fashion, revealing the specificity of downstream signaling components induced by various oxidative stresses. Moreover, the UPR induction by pretreatment with tunicamycin substantially inhibits TNF␣-induced ROS accumulation and cell death. Given that ER stress induces accumulation of ROS in PERK or ATF4 knockout (KO) cells to greater levels than wild-type cells (16), the UPR normally functions to counteract ROS accumulation induced by various stresses, including ER stress itself and TNF␣, to protect cells from cell death.
Western Blot Analysis-L929 cells (5 ϫ 10 5 ) were stimulated with TNF␣ or tunicamycin in the absence or presence of BHA for the indicated time periods, and lysed in a lysis buffer containing 50 mM Tris-HCl (pH 8.0), 150 mM NaCl, 1% Nonidet P-40, 0.5% deoxycholate, 0.1% SDS, 1 mM phenylmethylsulfonyl fluoride, and 1 g/ml aprotinin. After centrifugation, cell lysates were subjected to 10% SDS-PAGE and transferred onto polyvinylidene difluoride membranes (Millipore). The membranes were probed with Abs to phospho-eIF2␣, total eIF2␣, and XBP1. To confirm the equal loading of the lysates, the membranes were reprobed with anti-␣-tubulin Ab. The membranes were developed with Enhanced Chemiluminescence (ECL) Western blotting Detection System Plus (Amersham Biosciences).
To detect phosphorylation of PERK, L929 cells (3.5 ϫ 10 6 ) were unstimulated or stimulated with TNF␣ or tunicamycin in the presence or absence of BHA. The cells were lysed in a lysis buffer containing 20 mM Hepes (pH 7.5), 150 mM NaCl, 1% Triton X-100, 10% glycerol, 1 mM EDTA, 17.5 mM ␤-glycerophosphate, 20 mM NaF, 2 mM sodium orthovanadate, 1 mM phenylmethylsulfonyl fluoride, 4 g/ml aprotinin, and 1 g/ml leupeptin. The lysates were immunoprecipitated with anti-PERK Ab, and the immunoprecipitates were subjected to 6% SDS-PAGE. The membranes were probed with anti-PERK Ab and developed by ECL as described above.
To detect detergent-insoluble Bip, L929 cells (3.5 ϫ 10 6 ) were unstimulated or stimulated with TNF␣ or tunicamycin in the absence or presence of BHA for the indicated time periods. The cells were lysed in a lysis buffer containing 20 mM Hepes (pH 7.4), 120 mM NaCl, 1% Triton X-100, 10% glycerol, 5 mM EDTA, 1 mM phenylmethylsulfonyl fluoride, 1 g/ml aprotinin, and 1 g/ml leupeptin at 4°C for 30 min. The suspensions were collected by centrifugation at 15,000 ϫ g for 5 min, and the supernatants were used as detergent-soluble fractions. The insoluble pellet fractions were washed three times with the lysis buffer, and then solubilized in 1% SDS buffer containing 62.5 mM Tris (pH 6.8), 350 mM 2-mercaptoethanol, and 10% sucrose, and incubated at room temperature for 30 min. After centrifugation at 15,000 ϫ g for 5 min, the supernatants were used as detergent-insoluble fractions. The equal amounts of protein (50 g for the detergent-soluble fractions, and 100 g for detergent-insoluble fractions) were subjected to SDS-PAGE and probed with anti-Bip Ab.
Northern Blot Analysis-Northern blot analysis was performed as described previously (17). L929 cells and wild-type MEFs were stimulated with TNF␣ or tunicamycin for the indicated time periods in the absence or presence of BHA. cDNA probes for GADD34, XBP1, and ␤-actin were prepared by RT-PCR, and 32 P-labeled with Rediprime TM kit (Amersham Biosciences). After hybridization, radioactive signals were analyzed on BAS2500 (Fuji Photo Film) using Image Gauge software (Fuji) and normalized based on the signal of ␤-actin.
Analysis of XBP1 mRNA Splicing by IRE1-RNAs were reverse transcribed using Oligo(dT) primer (Invitrogen), and amplified using sense primer mXBP1.3S (5Ј-AAACAGAGTAGCAGCGCAGACTGC-3Ј) and antisense primer mXBP1.2AS (5Ј-GGATCTCTAAAACTAGAG-GCTTGGTG-3Ј) to generate cDNA product encompassing the IRE1 cleavage sites as previously described (12). The unspliced and spliced mRNAs generate 480-and 454-bp cDNA products, respectively. These fragments were further digested by PstI to check whether a PstI restriction site was lost after IRE1-mediated splicing of mRNA. The cDNA fragments were resolved on 2% agarose gel. cDNA products from the unspliced mRNA yielded two short fragments (289 and 191 bp) after digestion.
Measurement of ROS Accumulation-L929 (4 ϫ 10 5 ) cells were plated in 6-well plates and stimulated with TNF␣, H 2 O 2 , or arsenite for the indicated time periods. After stimulation, the cells were incubated with phenol red-free medium (Opti-MEM) (Invitrogen) containing CM-H 2 DCFDA (1 M) or DHE (1 M) in the dark for 30 min at 37°C. Then, the cells were harvested and analyzed on a flow cytometer (FAC-SCalibur, BD Biosciences). Data were processed by using the CellQuest program (BD Biosciences).
Small Interfering RNAs (siRNAs)-Duplex siRNAs with two nucleotides overhang at the 3Ј-end of the sequence were purchased from Qiagen. The target sequences were as follows: murine PERK, 5Ј-GGUAAUGCGGGAAGUUAAA-3Ј; green fluorescent protein (GFP), 5Ј-GGCUACGUCCAGGAGCGCACC-3Ј. L929 cells (3 ϫ 10 6 ) were transfected with the indicated siRNAs using a Nucleofector according to a manufacturer's instructions (Amaxa). Transfection efficiency was determined by counting GFP-positive cells using flow cytometry after transfection with an expression vector for GFP.
Electron Microscopy-L929 cells (4 ϫ 10 6 ) were unstimulated or stimulated with TNF␣ or tunicamycin in the presence or absence of BHA for 6 h, and then serially fixed with 2% glutaraldehyde in phosphate-buffered saline for 2 h and then with 2% OsO 4 for 2 h before embedding in Epon 812. Thin sections were prepared using a MT-5000 ultramicrotome (Dupont Pharmaceuticals), stained with uranyl acetate followed by lead citrate, and then observed (ϫ5,000 or 15,000) on a JEM1230 electron microscope (JEOL).
Measurement of GSH-Cellular GSH levels were measured by using a colorimetric assay kit (OxisResearch TM ) as previously described (19).
Statistical Analysis-Statistical analysis was performed by Student's t test. p value Ͻ 0.05 was considered to be significant.

TNF␣ Induces the UPR in L929
Cells-Treatment of cells with oxidative agents such as arsenite and heavy metal ions has been shown to induce oxidative stress and eIF2␣ phosphorylation (20). However, it is still controversial whether oxidative stresses induced by various agents elicit the UPR. Given that TNF␣ induces ROS accumulation in murine fibrosarcoma L929 cells (21), we first investigated whether TNF␣ induces eIF2␣ phosphorylation in L929 cells. We stimulated L929 cells with TNF␣ or tunicamycin for 2-10 h and examined eIF2␣ phosphorylation using Ab specific for the phosphorylated form of eIF2␣. Phosphorylation of eIF2␣ was detected at 4 h and persisted up to 10 h after tunicamycin stimulation (Fig. 1A). Interestingly, eIF2␣ phosphorylation was also induced at 2 h and persisted up to 10 h after TNF␣ stimulation. Because eIF2␣ phosphorylation subsequently induces ATF4 and GADD34 (2), we next examined the induction of GADD34 mRNA after TNF␣ stimulation by Northern blot analysis. TNF␣, as well as tunicamycin, induced GADD34 mRNA in L929 cells (Fig. 1B). Notably, eIF2␣ phosphorylation and the induction of GADD34 mRNAs after TNF␣ treatment were faster than those after tunicamycin treatment (Fig. 1, A  and B). These results indicate that TNF␣ induces a pathway initiated by phosphorylation of eIF2␣ and followed by the up-regulation of GADD34.
In addition to eIF2␣ phosphorylation, ER stress activates the ATF6dependent pathway that subsequently induces the expression of many genes containing the ERSE in the promoter regions, including Bip, calnexin, calreticulin, and XBP1 (10,22). Upon ER stress, ATF6 is converted from a 90-kDa protein (p90ATF6) to a 50-kDa protein (p50ATF6), resulting in nuclear translocation. Thus, we examined whether TNF␣ induces p50ATF6 by Western blotting using anti-ATF6 Ab (22). Although the induction levels of p50ATF6 by TNF␣ were weaker than tunicamycin, TNF␣ substantially induced p50ATF6 with a peak at 4 h after stimulation (Fig. 1A). We then examined whether TNF␣ induces the expression of XBP1 mRNA by Northern blot analysis. As shown in Fig. 1B, TNF␣ as well as tunicamycin markedly increased the expression of XBP1 mRNA, indicating that TNF␣ also activates the ATF6-dependent pathway.
ER stress also activates the IRE1-dependent pathway (11)(12)(13). Activated IRE1 cuts out 26 nucleotides of the unspliced XBP1[XBP1(U)] mRNA to generate the spliced XBP1[XBP1(S)] mRNA, which encodes transcriptionally active XBP1(S) with a molecular mass of 50 kDa. The RT-PCR products produced from the spliced and unspliced XBP1 mRNA can be easily detected following PstI digestion of PCR products generated using primers flanking the XBP1 intron, because the unspliced product contains a PstI site that is lost in the spliced product (12). As shown in Fig. 1C, PstI-digested fragments of the RT-PCR products were detected prior to stimulation, and these fragments decreased at 4 h and were no longer present at 6 h after tunicamycin stimulation. Residual undigested fragments prior to stimulation may reflect the basal level of IRE1 activation as reported previously (23). TNF␣ treatment also resulted in accumulation of XBP1(S) mRNA as seen by the presence of PstI-resistant RT-PCR products beginning at 2 h and continuing through 6 h of stimulation (Fig. 1C). We also examined whether TNF␣ FIGURE 1. TNF␣ induces the UPR in L929 cells. A, Western blot analysis of eIF2␣ phosphorylation, p50ATF6, and XBP1 induction. L929 cells were stimulated with TNF␣ (10 ng/ml) or tunicamycin (Tm) (2 g/ml) for the indicated time periods. Whole cell lysates were probed with Ab specific for a phosphorylated form of eIF2␣, ATF6, or XBP1. To verify the equal loading, the membranes were reprobed with anti-total eIF2␣ or anti-tubulin Ab. The asterisk indicates cross-reacted bands recognized by anti-ATF6 Ab. B, Northern blot analysis of GADD34 and XBP1 mRNAs. L929 cells were stimulated with TNF␣ or tunicamycin as above. Then, total RNAs were extracted and subjected to Northern blot analysis. The membranes were hybridized with 32 P-labeled GADD34 or XBP1 cDNA probes, and subsequently rehybridized with the ␤-actin cDNA. C, RT-PCR analysis of XBP1(S) and XBP1(U) mRNA. Diagrams of the RT-PCR products of XBP1(S) and XBP1(U) mRNAs. The dotted line indicates splicing of the PstI containing intron. Total RNAs from TNF␣-or tunicamycin (Tm)-stimulated L929 cells were subjected to RT-PCR as described under "Experimental Procedures." After digestion with PstI, the PCR products were subjected to 2% agarose gel electrophoresis. The PCR products of XBP1(S) mRNA remained intact (454 bp), whereas the PCR products of XBP1(U) mRNA were cut into two fragments of 289 and 191 bp, as indicated by the arrows.
induces XBP1(S) protein expression by Western blotting using anti-XBP1 Ab. Consistent with the RT-PCR analysis, the kinetics of TNF␣induced expression of XBP1(S) were faster than those of tunicamycin induction (Fig. 1A). Combined together, these data indicate that TNF␣ treatment induces the full UPR in L929 cells.
The TNF␣-induced UPR Is ROS-dependent-To investigate the contribution of ROS to TNF␣-induced UPR in L929 cells, we examined the inhibitory effect of the antioxidant, BHA, on the TNF␣-induced UPR. We previously demonstrated that BHA has a stronger antioxidant activity than N-acetyl-L-cysteine (19,24). Indeed, BHA completely inhibited TNF␣-induced ROS accumulation in L929 cells ( Fig. 2A). We then examined the effect of BHA on TNF␣-induced UPR in L929 cells. As shown in Fig. 2B, BHA significantly inhibited eIF2␣ phosphorylation and the up-regulation of XBP1(S). Moreover, BHA almost completely inhibited induction of GADD34 and XBP1 mRNAs (Fig. 2C). We next tested the inhibitory effect of BHA on the splicing of XBP1 mRNA. As shown in Fig. 2D, BHA significantly inhibited TNF␣-induced disappearance of the unspliced XBP1 mRNA. These results indicate that TNF␣-induced UPR in L929 cells is mediated by ROS.
In contrast to TNF␣-induced UPR, the tunicamycin-induced UPR has both ROS-dependent and -independent components. Tunicamycin-induced eIF2␣ phosphorylation and induction of GADD34 mRNA were not significantly inhibited by BHA (Fig. 2, B and C). However, BHA weakly and transiently inhibited induction of XBP1 mRNA and protein expression, but not splicing of XBP1 mRNA (Fig. 2, B-D). These results suggest that antioxidants have little effect on preventing PERK and IRE1 activation, and a modest effect on inhibiting ATF6-dependent up-regulation of XBP1 mRNA.
We have previously shown that TNF␣ does not induce ROS accumulation in wild-type MEFs (19). Then, we therefore examined whether TNF␣ induces the UPR in wild-type MEFs. Whereas tunicamycin induced phosphorylation of eIF2␣ (Fig. 2G), the expression of GADD34 and XBP1 mRNAs (Fig. 2E), splicing of XBP1 mRNA (Fig. 2F), XBP1(S) (Fig. 2G), and TNF␣ did not up-regulate any of these molecules in wild-type MEFs. These results substantiate that ROS play an essential role in inducing the UPR by TNF␣.
TNF␣ Activates Promoters Containing the UPRE and ERSE in a ROSdependent Fashion-Taken that TNF␣ induces the UPR in L929 cells in a ROS-dependent fashion, we next examined whether TNF␣ activates the promoter activities of ER stress-responsive genes. Overexpression of truncated XBP1(S) or p50ATF6 activates promoters containing an ER stress responsive element termed the UPRE, which was originally identified as a binding sequence of ATF6 in vitro (18). To examine whether TNF␣ up-regulates transcriptional activity of a promoter containing the UPRE, we transiently transfected L929 cells with the pGL3-5 ϫ ATF6 plasmid containing 5 copies of the UPRE, and stimulated the cells with TNF␣ or tunicamycin. Consistent with the induction of XBP1(S) by TNF␣ (Fig. 1A), TNF␣ increased the reporter activity to 2.5-fold compared with unstimulated cells (Fig. 3A). Importantly, BHA treatment completely inhibited TNF␣-induced 5 ϫ ATF6 reporter gene  Fig. 1. C, Northern blot analysis of GADD34 and XBP1 mRNAs. L929 cells were stimulated with TNF␣ or tunicamycin (Tm) in the absence or presence of BHA as above. Total RNAs were subjected to Northern blot analysis as described in the legend to Fig. 1B. D, RT-PCR analysis of XBP1 mRNA splicing. L929 cells were stimulated with TNF␣ or tunicamycin in the absence or presence of BHA as above. Total RNAs were subjected to RT-PCR analysis as described in the legend to Fig. 1D. Bands undigested or digested with PstI are indicated by the arrows. E, Northern blot analysis of GADD34 and XBP1 mRNAs. Wild-type MEFs were stimulated with TNF␣ or tunicamycin as above. Total RNAs were subjected to Northern blot analysis, as described in the legend to Fig. 1B. F, RT-PCR analysis of XBP1 mRNA splicing. Wild-type MEFs were stimulated with TNF␣ or tunicamycin as above. Total RNAs were subjected to RT-PCR analysis as described in the legend to Fig. 1D. G, Western blot analysis of XBP1 induction in wild-type MEFs. Wild-type MEFs were stimulated with TNF␣ or tunicamycin as above, and whole cell lysates were probed with the indicated Abs as described in the legend to Fig. 1. activity, indicating that ROS were essential for the UPRE-dependent transcriptional activation by TNF␣. Interestingly, tunicamycin-induced 5 ϫ ATF6 reporter gene activity was also partially inhibited by BHA (Fig.  3B), which is consistent with its partial inhibitory effect on the induction of XBP1(S) (Fig. 2B). We next examined whether TNF␣ activates the 5 ϫ ATF6 reporter gene activity in other types of cells. Consistent with the finding that TNF␣ does not induce ROS accumulation in wild-type MEFs (19) or HeLa cells (data not shown), TNF␣ did not induce the 5 ϫ ATF6 reporter gene activation in either wild-type MEFs (Fig. 3C) or HeLa cells (Fig. 3D). Furthermore, BHA treatment only marginally inhibited tunicamycin-induced 5 ϫ ATF6 reporter gene activation in wild-type MEFs (Fig. 3C) and HeLa cells (Fig. 3D), indicating that the contribution of ROS to tunicamycin-induced UPR activation is variable among cell types.
In addition to the UPRE, expression of ER stress responsive genes, such as Bip and calreticulin is regulated by another element, termed the ERSE. Thus, we next examined whether TNF␣ also activates the promoter activity of Bip (10). To this end, we transiently transfected L929 cells with pGL3-Bip (Ϫ132/ϩ7) containing three ERSE sites and stimulated the cells with TNF␣ or tunicamycin. TNF␣ weakly, but substantially activated the Bip reporter gene activity, and this activation was inhibited by BHA (Fig. 3E). Similarly, tunicamycin-induced Bip reporter gene activity was also partially inhibited by BHA (Fig. 3E). In contrast, TNF␣ did not activate the Bip reporter gene activity and BHA did not inhibit tunicamycin-induced Bip reporter gene activity in HeLa cells (Fig. 3F). Collectively, these results reveal an intimate correlation between TNF␣-induced ROS accumulation and activation of UPREand ERSE-dependent reporter gene expression.
TNF␣ and Tunicamycin Induce Dilatation of the ER-Given that TNF␣ induced the UPR in a ROS-dependent fashion, it was intriguing to examine whether TNF␣ induces morphological changes of the ER in L929 cells using electron microscopy. We treated L929 cells with TNF␣ or tunicamycin in the absence or presence of BHA for 6 h. In unstimulated cells, endoplasmic reticulum was recognized as long and thin cisternae surrounded by ribosomes observed as electron-dense dots (Fig. 4, B and C). TNF␣ induced prominent vacuolization in the cytoplasm without affecting nuclear morphology, consistent with necrotic cell death (Fig. 4, D-I). Taken that the dilated lumen was surrounded by electron-dense dots, the dilated organelles appeared to be the ER (Fig. 4,  F and I). Similarly, tunicamycin also induced a modest dilatation of the ER (Fig. 4, M-O). Importantly, BHA treatment significantly inhibited dilatation of the ER in TNF␣-and tunicamycin-treated cells (Fig. 4, J-L and P-R), indicating that ROS contribute to both TNF␣-and tunicamycin-induced dilatation of the ER in L929 cells.
PERK Is an eIF2␣ Kinase That Responds to Oxidative Stress Induced by TNF␣-Although oxidative stresses induce eIF2␣ phosphorylation, the kinase(s) that phosphorylate eIF2␣ remain to be identified (15). Because PERK has been implicated in UPR-mediated eIF2␣ phosphorylation (6), we examined whether PERK is activated by TNF␣ in a ROS-dependent fashion in L929 cells. We previously showed that phosphorylated PERK (activated PERK) migrates more slowly compared with non-phosphorylated PERK in SDS-PAGE, thereby phosphorylated PERK can be discriminated from non-phosphorylated PERK by Western blotting (6). As shown in Fig. 5A, phosphorylation of PERK was progressively induced at 4 -8 h after tunicamycin stimulation. Upon TNF␣ stimulation, phosphorylation of PERK was similarly induced at 2 h, peaked at 4 h, and then gradually decreased (Fig. 5A). Importantly, BHA treatment substantially inhibited TNF␣-, but not tunicamycininduced phosphorylation of PERK (Fig. 5B). These results suggest that TNF␣ induces a ROS-dependent ER stress that activates PERK.
To investigate whether PERK is indispensable for TNF␣-induced phosphorylation of eIF2␣, we knocked down expression of PERK by using siRNA against PERK. As shown in Fig. 5C, transfection efficiency was almost 90% based on the percentages of GFP-positive cells after transfection with a GFP expression vector. Under this condition, transfection of PERK siRNA, but not control GFP siRNA significantly reduced endogenous PERK protein levels (Fig. 5D). We then tested whether knockdown of PERK abolishes eIF2␣ phosphorylation induced by TNF␣. As expected, tunicamycin-induced eIF2␣ phosphorylation was significantly reduced in PERK knockdown cells (Fig. 5E). However, TNF␣-induced eIF2␣ phosphorylation was not impaired. This indicates that a kinase other than PERK is also involved in TNF␣-induced eIF2␣ phosphorylation.
TNF␣ Induces an Increase in Detergent-insoluble Bip in a ROS-dependent Fashion-A recent study showed that Bip accumulates in the detergent-insoluble complex in tunicamycin-treated cells and this accumulation is enhanced under oxidized cellular conditions (25). This suggests that oxidized cellular conditions enhance accumulation of Bip in  OCTOBER 7, 2005 • VOLUME 280 • NUMBER 40 JOURNAL OF BIOLOGICAL CHEMISTRY 33921 the detergent-insoluble fractions, resulting in the release of Bip from PERK, ATF6, and IRE1, and finally elicit the UPR. To test this possibility, we examined the detergent-soluble and -insoluble Bip before and after TNF␣ or tunicamycin stimulation. As shown in Fig. 5F, the expression levels of detergent-soluble Bip were not significantly altered before and after stimulation (up to 12 h after tunicamycin stimulation). Whereas detergent-insoluble Bip was barely detected before stimulation, detergent-insoluble Bip significantly increased at 4 h after TNF␣ stimulation. Importantly, this increase was inhibited in the presence of BHA. Moreover, this kinetics was well correlated with those of induction of XBP1(s) in protein levels and GADD34 in mRNA levels (Fig. 1, A and B). Whereas tunicamycin similarly induced an increase in detergent-insoluble Bip, the kinetics were delayed compared with TNF␣ stimulation, and the inhibitory effect of BHA was only transiently observed at 6 h after stimulation.

H 2 O 2 and Arsenite Induce eIF2␣
Phosphorylation, but Do Not Activate PERK-or IRE1-dependent Pathways-Given that TNF␣ induces eIF2␣ phosphorylation and also the UPR in a ROS-dependent fashion in L929 cells, we examined whether the ROS-producing stresses H 2 O 2 and arsenite similarly induce the UPR in L929 cells. Consistent with a previous study (20), H 2 O 2 and arsenite induced eIF2␣ phosphorylation (Fig. 6, A and B). However, these agents did not induce up-regulation of XBP1(S) protein, total XBP1 mRNA level, or XBP1 mRNA splicing (Fig.  6, A-D). Furthermore, H 2 O 2 and arsenite did not induce phosphorylation of PERK (Fig. 6E). Together, these results indicate that the downstream signaling pathways activated by H 2 O 2 and arsenite appear to be different from those induced by TNF␣. One possible scenario to explain these differences is that different oxidative stresses might induce accumulation of different types of ROS. To address this possibility, we used two ROS-sensitive fluorescent dyes, CM-H 2 DCFDA and

FIGURE 5. TNF␣ induces phosphorylation of PERK and accumulation of Bip in the detergent-insoluble fractions in a ROS-dependent fashion.
A, L929 cells were stimulated with TNF␣ (10 ng/ml) or tunicamycin (Tm) (2 g/ml) for the indicated time periods. The lysates were immunoprecipitated with anti-PERK Ab, and then the immunoprecipitates were subjected to Western blotting with anti-PERK Ab. Phosphorylated and non-phosphorylated PERK are indicated by arrows. B, L929 cells were stimulated with TNF␣ (10 ng/ml) for 4 h or tunicamycin (Tm) (2 g/ml) for 8 h in the absence or presence of BHA (100 M). Phosphorylated and non-phosphorylated PERK were detected as above. C, L929 cells were transfected with 2.5 g of an expression vector for GFP. After 48 h, the cells were analyzed by flow cytometry. Thin and bold lines indicate non-transfected and GFP-transfected cells, respectively. D, L929 cells were transfected with PERK siRNA or control GFP siRNA. Expression of endogenous PERK protein was examined by Western blotting with anti-PERK Ab. To verify the equal loading of the lysates, the membrane was reblotted with anti-tubulin Ab. E, knockdown of PERK does not inhibit TNF␣-induced phosphorylation of eIF2␣. Control or PERK knockdown cells were stimulated with TNF␣ (10 ng/ml) or tunicamycin (Tm) (2 g/ml) for the indicated time periods, the whole cell lysates were probed with the indicated Abs, as described in the legend to Fig. 1A. F, L929 cells were stimulated with TNF␣ (10 ng/ml) or tunicamycin (Tm) (2 g/ml) for the indicated periods in the absence or presence of BHA (100 M). The detergent-soluble and -insoluble fractions were prepared as described under "Experimental Procedures," and subjected to Western blotting using anti-Bip Ab. DHE, which are mainly oxidized to increase fluorescence by H 2 O 2 and superoxide anions, respectively. We stimulated L929 cells with TNF␣, H 2 O 2 , or arsenite, and analyzed the labeling with CM-H 2 DCFDA or DHE by flow cytometry. As shown in Fig. 6F, TNF␣, H 2 O 2 , and arsenite induced a substantial increase in fluorescent intensity of CM-H 2 DCFDA. The increase of the fluorescent intensity of CM-H 2 DCFDA induced by TNF␣ was significantly higher that those induced by H 2 O 2 or arsenite. Notably, H 2 O 2 and arsenite, but not TNF␣, induced a substantial increase in fluorescent intensity of DHE (Fig. 6G). These results demonstrate both qualitative and quantitative differences of accumulated ROS induced by TNF␣, H 2 O 2 , and arsenite, which might be responsible for the differential responses to these agents in L929 cells.
UPR Activation Inhibits TNF␣-induced ROS Accumulation and Cell Death-To investigate the mechanism by which TNF␣ induces ROS accumulation in L929 cells, we examined cellular levels of the major antioxidants, GSH, before and after TNF␣ stimulation in wild-type MEFs and L929 cells. Whereas GSH levels were not significantly altered before and after TNF␣ stimulation in wild-type MEFs as we previously reported (19), TNF␣ induced substantial decreases in GSH levels in L929 cells (Fig. 7A). We previously showed that the UPR up-regulates genes associated with amino acid transport and GSH biosynthesis, which may be responsible for elimination of ROS under normal conditions (16). Thus, we next examined whether pretreatment of L929 cells with tunicamycin counteracts the TNF␣-induced decrease in GSH levels. As expected, pretreatment of tunicamycin for 12 h almost completely inhibited the TNF␣-induced decrease in GSH. These results prompted us to investigate whether tunicamycin pretreatment inhibits TNF␣-induced ROS accumulation in L929 cells. After pretreatment of L929 cells with tunicamycin, the cells were then stimulated with TNF␣ for 4 -12 h and ROS accumulation was analyzed by flow cytometry. As shown in Fig. 7B, accumulation of ROS was progressively observed at 4 -12 h after TNF␣ stimulation in untreated L929 cells. Interestingly, pretreatment with tunicamycin substantially inhibited TNF␣-induced ROS accumulation. Furthermore, tunicamycin pretreatment significantly increased the viability of L929 cells after TNF␣ stimulation compared with untreated cells (Fig. 7C). Collectively, these results indicate that the tunicamycin-induced UPR inhibits TNF␣-induced ROS accumulation and cell death.
We previously showed that TNF␣ induces accumulation of ROS in NF-B activation-deficient cells, such as RelA (a major component of the NF-B complex), KO, or DKO MEFs (19). Although this NF-B function is presumably mediated by up-regulating antioxidant enzymes (26 -28), the detailed molecular mechanism is not fully understood. Given that pretreatment with tunicamycin inhibited TNF␣-induced accumulation of ROS in L929 cells (Fig. 7B), it is interesting to test whether preemptive UPR also suppresses TNF␣-induced ROS accumulation in RelA KO and DKO MEFs. Interestingly, pretreatment with tunicamycin substantially inhibited TNF␣-induced ROS accumulation (Fig. 7D), and cell death also in these cells (Fig. 7, E and F).

DISCUSSION
In the present study, we have demonstrated that TNF␣ induces the UPR, including PERK-mediated eIF2␣ phosphorylation, and ATF6-and IRE1-mediated induction of XBP1(S), in a ROS-dependent fashion. In contrast to TNF␣, oxidative stresses by H 2 O 2 and arsenite induced only eIF2␣ phosphorylation, but not activation of PERK-or IRE1-dependent pathways. Collectively, these results indicate that different oxidative stresses specify activation of different downstream signaling pathways. Furthermore, pretreatment with tunicamycin significantly inhibited TNF␣-induced ROS accumulation and cell death in L929 cells, RelA KO, and DKO MEFs, indicating that gene(s) induced by the UPR also prevent ROS accumulation by TNF␣.
Although oxidative stresses induced by arsenite and heavy metal ions have been shown to induce eIF2␣ phosphorylation (20), it remained to be solved which of these oxidative stresses may induce the UPR. Moreover, the molecular link between ER stress and oxidative stress is largely unknown. We have shown that TNF␣ activates PERK-and IRE1-dependent pathways in a ROS-dependent fashion in L929 cells (Figs. 2 and 5). Activation of the three known signaling components of the UPR, including PERK, ATF6, and IRE1, depends on their dissociation of each molecule from the abundant ER chaperone Bip. It is likely that ROS may induce accumulation of malfolded proteins that subsequently form the complex with Bip, resulting in translocation of the Bip-containing complex in the detergent-insoluble fractions (Fig. 5F). This translocation of Bip induces the dissociation of Bip from the PERK, ATF6, and IRE1, and ultimately elicits the UPR.
It is notable that H 2 O 2 or arsenite did not induce similar responses despite inducing eIF2␣ phosphorylation. Together, these results indicate that qualitative difference exists in the signaling cascades induced by various oxidative stresses. Given that H 2 O 2 and arsenite, but not TNF␣, induced accumulation of superoxide anions (Fig. 6G), accumulated superoxide anions might directly activate another eIF2␣ kinase other than PERK. Alternatively, endogenously and exogenously produced ROS might be qualitatively different. For example, ROS generated in the cells may come from mitochondria, ER, or peroxisome and specifically damage proteins in those subcellular compartments. However, there is currently no reliable way to determine the subcellular location of ROS generation using oxidation-sensitive dyes. Thus it is crucial to develop such detection systems for better understanding of the signaling specificity induced by various oxidative stresses.
Although TNF␣ induced phosphorylation of PERK in a ROSdependent fashion (Fig. 5, A and B), knockdown of endogenous PERK using siRNA did not inhibit TNF␣-induced eIF2␣ phosphorylation (Fig.  5E). This indicates that a kinase other than PERK also phosphorylates eIF2␣ in response to TNF␣-induced oxidative stress. Interestingly, a previous study showed that PKR responds to TNF␣ and phosphorylates eIF2␣ in NIH3T3 cells (29). Therefore, PERK and PKR may act redundantly as eIF2␣ kinases in response to TNF␣-induced oxidative stress. Nevertheless, it is still possible that a kinase other than PERK or PKR also phosphorylates eIF2␣. A further study will be required to address this issue.
Our present electron microscopic analysis has shown that TNF␣ induces dilatation of the ER in L929 cells in a ROS-dependent fashion (Fig. 4, D-L). These results are reminiscent of a previous study in which ER dilatation was observed in pancreatic ␤ cells in PERK knock-out mice (30). Similarly, tunicamycin induced mild dilatation of the ER (Fig.  4, M-O). In contrast to TNF␣, which induced marked accumulation of ROS in L929 cells ( Fig. 2A), tunicamycin (2 g/ml) did not induce apparent accumulation of ROS (data not shown). However, a higher dose of tunicamycin (10 g/ml) did induce accumulation of ROS at a detectable level (data not shown). These results suggest that the UPR normally functions to prevent ROS accumulation, however, the antioxidant activity induced by the UPR is overwhelmed upon severe ER stress, resulting in ROS accumulation. These data suggest that a small amount of ROS under detection levels may contribute to tunicamycininduced dilatation of ER in L929 cells.
We previously showed that TNF␣ induces a reduction of cellular GSH levels, which might be responsible for accumulation of ROS, in RelA KO and DKO MEFs (19). A similar reduction of GSH levels was observed in L929 cells upon TNF␣ stimulation (Fig. 7A). Given that TNF␣-induced NF-B activation is impaired in RelA KO and DKO MEFs, these results indicate that some NF-B-dependent pathways FIGURE 7. Pretreatment with tunicamycin inhibits TNF␣-induced ROS accumulation and cell death. A, cellular levels of GSH in wild-type MEFs and L929 cells. Wild-type MEFs (WT) and L929 cells were untreated (Ϫ) or pretreated with tunicamycin (Tm) (2 g/ml) for 12 h, and then the cells were unstimulated (open columns) or stimulated with TNF␣ (10 ng/ml) for 4 h (closed columns). Cellular levels of GSH were measured as described under "Experimental Procedures," and are presented as % reduction in stimulated cells compared with unstimulated cells. Results are presented as mean Ϯ S.E. of triplicate samples and represent two independent experiments with similar results. *, p Ͻ 0.05 compared with unstimulated cells. B, flow cytometric analysis of ROS accumulation. L929 cells were untreated (left panels) or pretreated (right panels) with tunicamycin (Tm) (2 g/ml) for 12 h. After washing to remove tunicamycin, cells were unstimulated (thin line) or stimulated (bold line) with TNF␣ (10 ng/ml) for the indicated time periods. Accumulation of ROS was analyzed as described in the legend to Fig. 2A. The ratios of the mean fluorescent intensity of stimulated cells to that of unstimulated cells are indicated at the upper right corner in each panel. C, cell viability after TNF␣ stimulation. L929 cells were untreated or pretreated with tunicamycin (Tm) (2 g/ml) for 12 h, and then stimulated with the indicated doses of TNF␣ for 16 h. Cell viability was determined by WST assay. Results are presented as mean Ϯ S.E. of triplicate samples and represent three independent experiments with similar results. *, p Ͻ 0.05 compared with untreated cells. D, flow cytometric analysis of ROS accumulation in RelA KO and DKO MEFs. RelA KO and DKO MEFs were untreated (left panels) or pretreated (right panels) with tunicamycin (Tm) (2 g/ml) for 4 h. Then, cells were unstimulated (thin line) or stimulated (bold line) with TNF␣ (10 ng/ml) for 4 h. Accumulation of ROS was analyzed as described in the legend to Fig.  2A. E and F, cell viability after TNF␣ stimulation. RelA KO and DKO MEFs were untreated or pretreated with tunicamycin (Tm) (2 g/ml) for 4 h, and then stimulated with the indicated doses of TNF␣ for 16 h. Cell viability was analyzed as described in the legend to Fig.  7C. *, p Ͻ 0.05 compared with unstimulated cells. prevent ROS accumulation. Consistent with this notion, we recently showed that TNF␣ induces various antioxidants in an NF-B-dependent fashion (31). On the other hand, we showed that the PERK-and ATF4-dependent pathways induce several genes associated with amino acid transport and glutathione biosynthesis, which are responsible for elimination of ROS (16). Intriguingly, pretreatment with tunicamycin substantially inhibited ROS accumulation induced by TNF␣ in L929 cells, RelA KO, and DKO MEFs (Fig. 7, B and D). Consequently, ROSmediated cell death was significantly inhibited by tunicamycin pretreatment in these cells (Fig. 7, C, E, and F). Collectively, these results suggest that the NF-Band UPR-dependent pathways have a similar function to suppress ROS accumulation. It remains to be determined whether the target genes are overlapped between these two pathways. Identification of the molecule(s) induced by NF-B or the UPR that inhibit ROS accumulation is crucial for understanding and manipulating the pathological conditions, in which ROS are critically involved.