Hairy and enhancer of split 1 (HES1) protects cells from endoplasmic reticulum stress–induced apoptosis through repression of GADD34

Disruption in endoplasmic reticulum (ER) function, termed ER stress, occurs in many diseases, including neurodegenerative disorders, diabetes, and cancer. Cells respond to ER stress with the unfolded protein response (UPR), which triggers a broad transcriptional program to restore and enhance ER function. Here, we found that ER stress up-regulates the mRNA encoding the developmentally regulated transcriptional repressor hairy and enhancer of split 1 (HES1), in a variety cell types. Depletion of HES1 increased cell death in response to ER stress in mouse and human cells, in a manner that depended on the pro-apoptotic gene growth arrest and DNA damage-inducible protein GADD34 (also known as Protein phosphatase 1 regulatory subunit 15A, or MyD116). Furthermore, HES1 bound to the GADD34 promoter, and its depletion led to an up-regulation of GADD34 expression during ER stress. Our results identify HES1 as a repressor of GADD34 expression, and reveal that HES1 contributes to cell fate determination in response to ER stress.

As a central organelle in the protein secretory pathway, the endoplasmic reticulum (ER) 4 is responsible for folding and processing secreted and membrane proteins. Alterations in ER function can arise from various stimuli such as increased secretory protein synthesis during the differentiation of secretory cells, or diseases that disrupt the function of protein folding pathways. These perturbations lead to an imbalance between the load and capacity of the ER, referred to as ER stress. To avoid or counteract the potentially toxic accumulation of misfolded proteins, the ER responds to stress through the unfolded protein response (UPR) (1). This collection of signaling pathways relieves the protein-folding load on the ER by global translational attenuation and mRNA decay (2,3), while simultaneously increasing the ability of the ER to fold proteins by up-regulation of genes encoding ER-specific chaperones (1).
The UPR is initiated by three ER membrane-embedded signaling proteins: PKR-like ER kinase (PERK), inositol-requiring enzyme 1 (IRE1), and activating transcription factor 6 (ATF6). Translational regulation is mediated by PERK, which phosphorylates eukaryotic translation initiation factor 2␣ (eIF2␣) (2). This inhibits general translation but promotes synthesis of proteins such as ATF4, whose mRNAs contain upstream open reading frames (4). ATF4 subsequently activates transcription of ER chaperones and genes involved in amino acid metabolism and antioxidant pathways, which are required for ER quality control (5). A second transducer of the UPR is IRE1, whose endoribonuclease is activated by ER stress. IRE1 mediates the unconventional splicing of the mRNA encoding the transcription factor X-box-binding protein 1 (XBP1) (6), thereby increasing expression of many genes encoding ER chaperones and other proteins that function in the secretory pathway (7). IRE1 also cleaves other mRNAs localized to the ER, leading to their degradation through regulated IRE1-dependent decay (RIDD) (8). ATF6 is activated by proteolysis during ER stress and, along with ATF4 and XBP1, up-regulates genes necessary to reestablish protein homeostasis (9).
In addition to its cytoprotective roles, the UPR induces apoptotic cell death if the stress is not mitigated (10). IRE1 recruits TNF receptor-associated factor 2 (TRAF2), which interacts with apoptotic-signaling kinase-1 (ASK1) (1,11), triggering kinase cascades that promote apoptosis (12,13). The PERK-ATF4 branch up-regulates a pro-apoptotic transcription factor, C/EBP homologous protein (CHOP), causing changes in gene expression that favor apoptosis (14,15). For example, CHOPmediated up-regulation of GADD34 enhances dephosphorylation of eIF2␣, reversing translational attenuation, which further increases the protein-folding burden on the ER (16). Although many studies have uncovered molecules involved in the ER stress-related apoptosis pathway, it is still unclear how the numerous signals from the stressed ER are integrated and how cells decide to activate apoptosis.
Here we report a novel role for hairy and enhancer of split 1 (HES1) in cell fate decisions during ER stress. HES1 is a basic helix-loop-helix transcriptional repressor (17). As a primary target of the Notch signaling pathway, HES1 regulates cell qui- cro ARTICLE escence and proliferation in the development of multiple organs and cell types (18). HES1 mRNA levels have been reported to increase during ER stress in mouse cells (19), although the mechanism of this regulation has not been previously explored. Here we show that HES1 mRNA levels increase in a manner that depends on the ability of PERK to attenuate translation, and affects the survival of cells exposed to chemical inducers of ER stress by repressing GADD34.

Hairy/HES1 mRNA levels increase during ER stress
To measure the response of HES1 to ER stress, we treated Drosophila melanogaster S2, mouse MC3T3-E1, and human HEK293 cells with either dithiothreitol (DTT), a reducing agent that disrupts disulfide bonds, or thapsigargin (Tg), which depletes ER calcium reserves. We then measured relative mRNA levels of fly hairy, or mammalian HES1 by quantitative real-time RT-PCR (qPCR). We found that hairy/HES1 mRNA levels increased during ER stress in all three cell lines (Fig. 1A).

HES1 mRNA up-regulation depends on PERK-mediated translational attenuation
To determine which branch of the UPR signaling network is responsible for the up-regulation of hairy/HES1 mRNA during ER stress, we depleted UPR transducers from S2 cells using RNAi, then compared the mRNA levels of hairy in cells treated with and without DTT for 2.5 h (the time of maximal hairy induction in these cells). Depletion of Perk, but not its downstream target Atf4, resulted in loss of hairy mRNA up-regulation (Fig. 1B). To test whether mammalian PERK is necessary for HES1 mRNA up-regulation, we transfected MC3T3-E1 or Hek293 cells with siRNAs targeting either PERK or a negative control sequence (Neg), and induced ER stress with DTT. Induction of HES1 mRNA was significantly blocked by PERK knockdown (Fig. 1, C-E).
In contrast to Perk knockdown, depletion of Ire1 in MC3T3-E1 cells resulted in increased induction of the Hes1 mRNA during stress (Fig. 1F). We have not explored the reason for this, but speculate that knockdown of Ire1 may exacerbate

HES1 regulates GADD34 during ER stress
ER stress, leading to enhanced PERK signaling and therefore higher Hes1 mRNA levels. Splicing of Xbp1 served as a control for Ire1 knockdown efficiency (Fig. 1G). Depletion of Atf4 did not have significant effects on Hes1 mRNA levels, although it did block induction of its target gene Trib3 as expected (Fig. 1, H and I). These results indicate a conserved effect of PERK, but not ATF4, on HES1 mRNA up-regulation across cells from flies, mice, and humans.
PERK phosphorylates the translation initiation factor eIF2␣, as well as other targets such as NRF2 (20), diacylglycerol (21), and FOXO1 (22). To determine which aspect of PERK function is important for HES1 mRNA regulation, we used integrated stress response inhibitor (ISRIB), a small molecule that blocks translational attenuation upon ER stress by inhibiting the downstream effects of eIF2␣ phosphorylation (23). ISRIB reduced HES1 mRNA levels induced by either DTT or Tg treatment in MC3T3-E1 (Fig. 2, A and B) and Hek293 cells (Fig. 2C), indicating that the translational attenuation mediated by PERK is important for HES1 regulation. Translational attenuation was also sufficient to increase HES1 mRNA levels, as seen when we treated cells with the translation elongation inhibitor cycloheximide (CHX) (Fig. 2D).
HES1 is known to repress its own expression by directly binding to N-box sequences in its promoter (24). The HES1 protein and HES1 mRNA are also highly unstable, and thus their levels would potentially be very sensitive to acute changes in translation and transcription (25). The observation that translation attenuation is both necessary (during ER stress) and sufficient for up-regulation of HES1 mRNA suggested that increased expression of HES1 mRNA may be a direct consequence of the loss of HES1 protein. To address this possibility, we first monitored HES1 protein (Fig. 3, A and B) and mRNA ( Fig. 3C) levels over time in Hek293 cells treated with DTT. HES1 protein levels rapidly declined with DTT, then began to recover after ϳ2 h (Fig. 3B). In contrast, HES1 mRNA levels increased in the presence of DTT until 2-4 h and then returned to normal levels by 6 -8 h (Fig. 3C). We carried out five replicates of these time course experiments. In one of the replicates (circles), the HES1 protein was down-regulated more strongly and for ϳ2 h longer (Fig. 3B); in this same replicate, the HES1 mRNA continued its upward trajectory for 2 h longer before returning to baseline (Fig. 3C). These data are consistent with a negative autoregulation model, where the HES1 protein controls expression of its own mRNA.
We next asked whether HES1 is released from its own promoter during ER stress (Fig. 3, D and E). Chromatin immunoprecipitation (ChIP) analysis revealed robust binding of the endogenous HES1 protein to the HES1 promoter in unstressed Hek293 cells (Fig. 3E). We observed a trend where binding of HES1 was reduced after treatment of the cells with DTT for 4 h (p ϭ 0.058, paired t test, n ϭ 5). Taken together, these results support a model where during ER stress, PERK-mediated trans-  cross-linked, and immunoprecipitated endogenous HES1, using an IgG antibody as a control in parallel. We then measured the relative abundance of the HES1 promoter region (E) in the precipitated chromatin by qPCR with RPL30 exon3 as a negative control (D). The colors of bars representing individual replicate experiments are maintained across panels D and E. *, p value Ͻ0.05, paired two-tailed Student's t test on log 2 -transformed data to test for fold-changes.

HES1 regulates GADD34 during ER stress
lational attenuation of HES1 leads to up-regulation of HES1 mRNA through the loss of negative autoregulation.

Loss of HES1 sensitizes mammalian cells to ER stress
To characterize the role of HES1 during ER stress, we examined the viability of MC3T3-E1 cells transfected with either Neg or Hes1 siRNAs. Hes1 knockdown by itself did not significantly affect MC3T3-E1 cell viability in the absence of ER stressors (Fig. 4, A and B). In contrast, DTT or Tg treatment significantly compromised cell viability in a dose-dependent manner in Hes1-depleted cells as compared with Neg siRNA-transfected cells. To assess whether the reduced viability of Hes1depleted cells during ER stress results from apoptosis, we labeled cells with Annexin V/propidium iodide (PI) and analyzed by flow cytometry. Knockdown of Hes1 increased the percentage of cells that, after DTT treatment, were stained with Annexin V but not with PI, indicative of apoptotic signaling as opposed to necrosis (Fig. 4C). We confirmed this finding in Hek293 cells; depletion of HES1 reduced cell viability upon DTT treatment (Fig. 4D).

HES1 regulates GADD34 during ER stress
Severe ER stress triggers cell death through expression of various pro-apoptotic genes (10). We measured the mRNA levels of several of these genes in the presence and absence of ER stress, in control and Hes1-depleted MC3T3-E1 cells (Fig. 5,  A-E). Interestingly, expression of Gadd34 mRNA was further increased during ER stress in cells depleted of Hes1 (Fig. 5, A  and B), consistent with a role for HES1 in repressing the transcription of Gadd34. This effect was conserved in human cells; HES1-depleted Hek293 cells showed elevated levels of GADD34 mRNA (Fig. 5, F and G) and protein (Fig. 5, I-K). Induction of other ER stress-related pro-apoptotic genes in mouse cells, including Chop, Trib3, and Puma, were not affected by Hes1 knockdown. However, CHOP mRNA levels did increase with HES1 knockdown in Hek293 cells (Fig. 5H). The difference between mouse and human cells in terms of HES1 effects on CHOP expression may be explained by the presence of N-box motifs in the promoter of target genes; human CHOP and both mouse and human GADD34 contain N-box motifs within 2 kb upstream of their transcription start sites, whereas mouse Chop does not.
To test whether HES1 directly regulates GADD34 through repressing its transcription, we used ChIP as in Fig. 3, D and E. Endogenous HES1 protein was enriched at the GADD34 promoter, consistent with its repressive effects on GADD34 mRNA levels (Fig. 6). This ChIP signal did not vary significantly between untreated and DTT-treated cells, although there was a minor trend toward less binding after DTT treatment (p ϭ 0.085, paired t test, n ϭ 5).

Depletion of HES1 induces cell death in a GADD34-dependent manner in cells undergoing ER stress
To test whether increased expression of GADD34 can explain the ER stress-induced cell death associated with loss of HES1, we compared ER stress sensitivity in cells depleted of HES1, GADD34, and both. Co-depletion of HES1 and GADD34 nearly completely rescued the cell death caused by depletion of HES1 alone in both MC3T3 (Fig. 7, A-C) and Hek293 cells (Fig. 7, D-F).
GADD34 is thought to promote cell death by dephosphorylating eIF2␣, leading to the recovery of protein synthesis and potentially exacerbating ER stress if activated prematurely (16,26). To determine whether phosphorylation of eIF2␣ was affected in HES1-depleted cells, we compared Neg and HES1 siRNA-transfected Hek293 cells with or without DTT for 4 h, followed by washing out the DTT and allowing cells to recover. We then measured phospho-eIF2␣ (p-eIF2␣) by Western blotting (Fig. 8, A, B, and D). Total eIF2␣ levels were not affected (Fig. 8C). HES1 knockdown alone led to a small increase in p-eIF2␣ levels (average 2-fold higher in HES1-depleted cells, p value ϭ 0.03, n ϭ 3; Fig. 8B), suggesting that loss of HES1 causes stress at sublethal levels. However, the p-eIF2␣ levels after DTT treatment were slightly lower in the HES1 knockdown cells (average 20% lower in HES1-depleted cells, p value ϭ 0.05, n ϭ 3). Notably, the proportional increase in p-eIF2␣ (in DTTtreated divided by untreated conditions) was lower in the HES1-depleted cells compared with control cells, and this trend continued throughout the recovery period (Fig. 8D). This suggests that when HES1 is depleted, increased GADD34 enhances dephosphorylation of eIF2␣ during ER stress.

Discussion
In this study, we investigated the regulation and function of HES1 in response to ER stress. Although HES1 has not been previously identified as a regulator of ER protein folding or the secretory pathway, its mRNA levels were up-regulated by ER stress, as a result of PERK activation. We propose that translational attenuation by PERK, coupled with the documented instability of HES1 (25), leads to the initial drop in HES1 protein levels during acute ER stress (Fig. 3B). This in turn leads to a relief of transcriptional repression of its own promoter, result-

HES1 regulates GADD34 during ER stress
ing in an increase in mRNA levels (Fig. 3C), followed by a rise in protein levels. Therefore, although HES1 protein levels do not typically increase above baseline levels during ER stress, this feedback regulation likely contributes to the cells' ability to maintain the correct overall abundance of HES1.
The ability to maintain HES1 protein levels during ER stress is important, as its depletion sensitized cells to ER stressinduced apoptosis. This effect appears to be a consequence of the direct regulation of the pro-apoptotic gene GADD34 by HES1: HES1 bound to the GADD34 promoter, HES1 depletion enhanced GADD34 expression in response to ER stress, and depleting cells of GADD34 mitigated the stress-induced apoptosis observed in HES1-depleted cells. HES1 depletion also caused a 2-fold increase in eIF2␣ phosphorylation in the absence of chemical inducers of ER stress, suggesting that loss of HES1 activates a mild stress response. Overall, we propose that lack of HES1 induces apoptosis during ER stress through a combination of causing low levels of stress and enhancing GADD34-mediated dephosphorylation of eIF2␣, thereby reversing the prosurvival effects of PERK.
Expression of GADD34 is induced during ER stress by ATF4 and CHOP (27,28), which are downstream targets of PERK. Because ATF4 -CHOP and HES1 are each regulated by PERK but have opposite effects on GADD34 expression, the balance and timing of these components of the UPR may be important in deciding whether cells live or die during ER stress. Interestingly, in human but not mouse cells, HES1 depletion also led to enhanced up-regulation of CHOP mRNA, suggesting that HES1 represses CHOP transcription (Fig. 5H). This regulation would further strengthen the effects of HES1, by repressing both GADD34 and its upstream activator simultaneously, and suppressing the pro-apoptotic effects of CHOP.   Fig. 3, D and E, and measured the relative abundance of the GADD34 promoter region in the precipitated chromatin by qPCR with RPL30 exon3 as a negative control. Cells were untreated or treated with DTT (2 mM, 4 h) prior to ChIP. *, p value Ͻ0.05, paired two-tailed Student's t test on log 2 -transformed data to test for fold-changes.

HES1 regulates GADD34 during ER stress
Like HES1, the zinc finger transcription factor NMP4 has been shown to repress transcription of GADD34 and sensitize cells to ER stress, although it represses transcription even in the absence of stress (29). HES1, in contrast, did not affect GADD34 expression in the absence of stress (Fig. 5, B and G); HES1 may instead have a specific role in dampening the induction of GADD34 and preventing the premature resumption of protein synthesis.
The regulation of HES1 by the UPR, and its effects on GADD34, may extend beyond the acute induction of ER stress studied here. For example, many components of the UPR have been implicated in cancer development (30). Cancer cells often show elevated levels of HES1 (31,32), which is associated with poor prognosis in colorectal cancer (33,34). It will be interesting to see whether PERK plays a role in regulation of HES1 in cancer, and whether HES1 can enhance tumor survival by repressing the expression of GADD34 and/or CHOP. More broadly, because HES1 is essential for the development of some tissues where the UPR is also important (18,35), HES1 may be targeted by the UPR or may influence UPR pathways in physiological situations.

General information, cell culture, and treatments
We cultured pre-osteoblast mouse MC3T3-E1 cells (American Type Culture Collection) in MEM␣ with nucleosides and no ascorbic acid (Invitrogen), and Hek293 cells in Dulbecco's modified Eagle's medium, both at 37°C and 5% CO 2 . We cultured Drosophila S2 cells (Invitrogen) in Schneider's media at room temperature. All media were supplemented with 10% heat-inactivated fetal bovine serum. We carried out experiments at low passage numbers and did not allow cells to become confluent. To induce ER stress, we added 2 mM DTT (Sigma) or 2 M Tg (Sigma) to cell media. To inhibit the integrated stress response, we added 200 nM ISRIB (kind gift from the Peter Walter lab, University of California at San Francisco) to cells for ϳ5 min before adding ER stressors. For inhibition of translation, we treated cells with 35 M CHX.

RNAi
To deplete S2 cells of individual UPR transducers, we used PCR to amplify regions of cDNAs encoding Xbp1, Perk, and Atf4, using S2 cell cDNA as a template and primers containing T7 RNA polymerase sites on the 5Ј ends. As a control, we amplified a region of the GFP coding sequence. We then used these   HES1 affects eIF2␣ phosphorylation. A, we transfected Hek293 cells with either Neg or HES1 siRNAs, incubated with or without DTT (2 mM, 4 h), washed out the DTT, and monitored phosphorylated eIF2␣ (Ser 51 ) over time by Western blotting. B and C, we quantified p-eIF2␣ or total eIF2␣ levels by Western blotting, from three independent experiments. The bars represent individual replicate experiments, and bars of the same color in both B and C indicate treatments done in parallel. D, we quantified p-eIF2␣ levels from three independent experiments as in A, and normalized to the unstressed cells from each set of siRNA transfections. Markers of the same shape indicate treatments done in parallel. *, p value Ͻ0.05, paired two-tailed Student's t test on log 2 -transformed data to test for fold-changes. Note that in B, differences between untreated and DTT-treated conditions were also statistically significant (p value Ͻ0.05).

HES1 regulates GADD34 during ER stress
PCR products to generate dsRNA by in vitro transcription (Megascript T7 kit, Ambion). We incubated S2 cells with dsRNA in serum-free media for 45 min, replaced the serum, and allowed the cells to recover for 4 -5 days. We then repeated the dsRNA treatment and induced ER stress 1 day following the second dsRNA treatment.
For RNAi in mammalian cells, we followed Invitrogen RNAimax guidelines for transfection of siRNAs. We combined multiple siRNAs (Qiagen) targeting each gene (Table 1). Negative siRNA-transfected cells were included as controls for all experiments. For the double knockdown experiments in Fig. 7, we used the same total amount of siRNAs (900 ng) for each treatment: single knockdowns contained either 675 ng of HES1 or 225 ng of GADD34 siRNA pools, with Neg siRNA pools making up the rest, whereas the double knockdowns contained both 675 ng of HES1 and 225 ng of GADD34 siRNA pools. We subjected cells to inhibitors and/or ER stressors 48 -72 h after transfection, when cells were ϳ70 -80% confluent.

mRNA isolation and analysis
We isolated mRNA using either TRIzol reagent (Invitrogen) or Quick RNA MiniPrep kits (Zymo Research), and synthesized cDNA using 700 ng to 2 g of total RNA as a template, a T18 primer, and Moloney murine leukemia virus reverse transcriptase (New England Biolabs). We measured relative mRNA abundance by qPCR using the Mastercycler ep realplex (Eppendorf) with SYBR Green fluorescent dye. We measured each sample in triplicate and normalized the target mRNA levels to those of ribosomal protein (RPL19) mRNA.
We measured Xbp1 splicing by amplifying cDNA with primers encompassing the Xbp1 splice site and running the products on a 2% agarose gel. We then quantified the relative band intensities for the spliced and unspliced Xbp1 products. All primer sequences are listed in Table 2.

ChIP
We cross-linked proteins and DNA by adding formaldehyde (final concentration of 1%) to Hek293 cells and quenched 10 min later by adding glycine (final concentration 0.125 M). Cross-linked chromatin was sonicated six times at 30% power (9 s on and 1 s off) using a Branson Sonifier 450 with a microtip probe. We removed insoluble debris by centrifugation and incubated the supernatant with antibodies to HES1 (Santa Cruz Biotechnology 166410), or as a control, rabbit IgG (Cell Signaling Technology 2729), for 12 h at 4°C. We then precipitated the immunocomplexes using ChIP-Grade Protein G Magnetic Beads (Cell Signaling Technology 9006), treated with RNase A at 37°C for 30 min, and reversed cross-links by incubating at 65°C for 8 h. We purified the resulting DNA using DNA Clean & Concentrator kits (Zymo Research) and analyzed by qPCR. Serial dilutions of input chromatin were used to generate stan-

HES1 regulates GADD34 during ER stress
dard curves for determining the relative amount of product, which was normalized to the input levels.

Cell viability and apoptosis assays
For viability assays in MC3T3-E1 cells, we removed floating cells by aspiration, trypsinized, and counted live cells using a hemocytometer. For Hek293 cells, which tended to release from the flasks during ER stress, we collected all cells, centrifuged (1000 ϫ g, 5 min) the cells to separate dead cells (in the supernatant), resuspended cell pellets in media, and counted live cells on the hemocytometer. We used the Annexin V-Alexa Fluor 488 apoptosis assay kit (Invitrogen) to determine the percentage of apoptosis as described by the manufacturer's protocol. After staining, cells were analyzed by flow cytometry (BD Accuri C6 Flow Cytometer, BD Biosciences) and BD Accuri C6 Plus (BD Biosciences) software.

Data presentation
We displayed all data for individual experiments. In bar graphs, experimental treatments done in parallel are correlated by color, and when appropriate these colors are consistent across panels (for example, when different mRNAs are measured in the same samples, the replicates are color-coded across panels and noted in the figure legends). In scatterplots, experimental treatments done in parallel are correlated by symbol and connected by lines. Statistical significance for all expression data were determined using paired two-tailed Student's t tests on log 2 -transformed data to test for fold-changes.