PERK is a type I transmembrane protein located in the ER, which senses the accumulation of unfolded proteins in the ER lumen [61–63]. The luminal portion of PERK is involved in sensing unfolded proteins, whereas the cytoplasmic portion contains a kinase domain. In the absence of ER stress, BiP binds to the luminal domain of PERK and keeps it from being activated (Figs 4 and 5A). In response to ER stress, BiP is released from PERK, and PERK is activated through oligomerization and trans-phosphorylation . Activated PERK phosphorylates and inactivates the α-subunit of eukaryotic translational initiation factor 2 (eIF2α), leading to translational attenuation. The phosphorylation of PERK is transient as the protein is dephosphorylated by specific phosphatases such as CReP (constitutive repressor of eIF2α phosphorylation), protein phosphatase 2C-GADD34, and p58IPK. CReP is constitutively expressed, whereas the expression of GADD34 and p58IPK is induced on ER stress by PERK and activating transcription factor (ATF)6 pathways, respectively.
Figure 5. Activation of the PERK pathway. (A) Activation of PERK, IRE1, and ATF6. In the absence of ER stress, BiP prevents PERK, IRE1, and ATF6 from being activated by binding to these sensors. BiP prevents the activation of IRE1 and PERK by keeping them from being oligomerized, whereas BiP inhibits the translocation of ATF6 by masking the Golgi-localization signal (GLS). When BiP is sequestered from sensors by unfolded proteins, these sensor molecules are activated. (B) Regulation of ATF4 expression. In the absence of ER stress, most of the eIF2α is active (not phosphorylated), and translation starts at the small ORFs, leading to the release of ribosomes before they reach the ATF4 ORF. Upon ER stress, most of the eIF2α becomes inactive (phosphorylated), and translation rarely starts at the small ORFs, thus ribosomes can reach the ATF4 ORF and induce translation of ATF4 protein.
Download figure to PowerPoint
Interestingly, translation of the transcription factor ATF4 is up-regulated by eIF2α-mediated translational attenuation. There are several small ORFs in the 5′-UTR of ATF4 mRNA (Fig. 5B). The ribosome first binds to a 5′-cap structure, slides on the ATF4 mRNA, and then starts translation at the small ORFs with unphosphorylated (active) eIF2α. As the ribosome is released from the ATF4 mRNA upon the termination of translation at the stop codon of small ORFs, the ATF4 ORF cannot be translated in the absence of ER stress. In contrast, as phosphorylated (inactive) eIF2α cannot start translation, the probability that the ribosome reaches the ATF4 ORF is increased in the presence of ER stress. Thus, the translation of ATF4 is remarkably enhanced in response to ER stress. The targets of ATF4 include CHOP (C/EBP homology protein), a transcription factor involved in the induction of apoptosis, and proteins involved in amino-acid metabolism such as asparagine synthetase or those involved in resistance to oxidative stress .
eIF2α is also phosphorylated by other kinases, such as dsRNA-dependent protein kinase (PKR), GCN2 (general control of amino-acid synthesis 2) and heme-regulated translational inhibitor. These kinases are activated by viral infections, amino-acid starvation, and heme deficiency, respectively, indicating that translational attenuation and ATF4 induction is induced by not only ER stress but also these physiological situations. Thus, the cellular response mediated by the phosphorylation of eIF2α is called the integrated stress response and is essential for cell survival .
There is another sensor molecule, ATF6, on the ER membrane [67–70]. ATF6 is a type II transmembrane protein, the luminal domain of which is responsible for the sensing of unfolded proteins. The cytoplasmic portion of ATF6 has a DNA-binding domain containing the basic-leucine zipper motif (bZIP) and a transcriptional activation domain. In the absence of ER stress, BiP binds to the luminal domain of ATF6 and hinders the Golgi-localization signal, leading to inhibition of ATF6 translocation (Fig. 5A) [71–75]. In response to the accumulation of unfolded proteins, BiP dissociates from ATF6, and ATF6 is moved to the Golgi apparatus by vesicular transport (Fig. 4). In the Golgi apparatus, ATF6 is sequentially cleaved by a pair of processing proteases called site 1 protease (S1P) and site 2 protease (S2P), and the resultant cytoplasmic portion of ATF6 [pATF6(N)] translocates into the nucleus. In the nucleus, pATF6(N) binds to a cis-acting element, the ER stress response element (ERSE), and activates the transcription of ER chaperone genes such as BiP, GRP94 and calreticulin . The consensus sequence of the ER stress response element is CCAAT-(N9)-CCACG, and ATF6 binds to the CCACG portion, whereas a general transcription factor, NF-Y (nuclear factor Y), binds to the CCAAT portion.
The cleavage of ATF6 is unique, especially as the second cleavage by S2P occurs in the transmembrane region . This process is called regulated intramembrane proteolysis (RIP), which is well conserved from bacteria to mammals (Fig. 6). The most characterized substrate of RIP is sterol response element-binding protein (SREBP) . SREBP is a transcription factor that is located in the ER membrane like ATF6. Upon a deficiency of sterol, SREBP is transported to the Golgi apparatus, cleaved by S1P and S2P, and activates the transcription of genes involved in the biosynthesis of sterol. Thus, the activation of ATF6 and SREBP is mainly regulated at the level of vesicular transport. The regulation of the transport of SREBP has been well characterized, and regulatory components such as the sensor-escort protein SCAP (SREBP cleavage-activating protein) and the anchor protein INSIG (insulin-induced gene 1) have been identified .
Figure 6. Molecules regulated by RIP. RIP is conserved from bacteria to mammals, and is involved in various biological processes. SREBP senses a sterol deficiency and activates the transcription of genes involved in sterol synthesis. Cleavage of APP by RIP results in the production of antibody, which is responsible for the onset of Alzheimer's disease. Notch is a cell surface protein that is cleaved by RIP upon binding Delta, leading to the activation of target genes involved in differentiation. Bacterial RseA protein anchors a transcription factor, σE, to keep it inactive. In response to accumulation of unfolded proteins in the periplasm, RseA is cleaved by RIP, leading to transcriptional activation of periplasmic chaperones.
Download figure to PowerPoint
There are two genes for ATF6, called ATF6α and ATF6β, which have a similar function and are ubiquitously expressed [68,77]. Recently, several bZIP transcription factors located in the ER and regulated by RIP have been reported. cAMP response element-binding protein H (CREBH) is specifically expressed in liver, and processed by S1P and S2P in response to ER stress . CREBH activates the transcription of acute-phase response genes involved in acute inflammatory responses. OASIS (old astrocyte specifically induced substance) is also cleaved by S1P and S2P in response to ER stress in astrocytes and activates the transcription of BiP . A spermatid-specific transcription factor, Tisp40 (transcript induced in spermiogenesis 40), is also severed by S1p and S2P and activates the transcription of EDEM . These tissue-specific ATF6-like molecules may contribute to the ER stress response.
Luman/LZIP/CREB3 can be cut by S1P and S2P and activates the transcription of EDEM through a cis-acting element, unfolded protein response element (UPRE), although ER stress cannot induce Luman RIP [80–82]. CREB4 is transported to the Golgi apparatus in response to ER stress, is cleaved by S1P and S2P, and activates the transcription of BiP, although cleavage is not observed upon ER stress . These ATF6-like molecules, which are insensitive to ER stress, might be activated in situations other than ER stress and activate transcription of ER chaperones.
The third sensor molecule in the ER membrane is IRE1 (inositol requirement 1) [84–86]. The luminal domain of IRE1 is similar to that of PERK and involved in the sensing of unfolded proteins, whereas the cytoplasmic domain contains a kinase domain and an RNase domain. There are two genes for IRE1, IRE1α and IRE1β. Upon ER stress, BiP suppression of IRE1 activation is released, and IRE1 is activated through dimerization and transphosphorylation (Figs 4 and 5A) . Activated IRE1α converts XBP1 (x-box binding protein 1) pre-mRNA into mature mRNA by an unconventional splicing mechanism [69,87]. As the DNA-binding domain and the activation domain are encoded in ORFs in XBP1 pre-mRNA, a protein translated from pre-mRNA [pXBP1(U)] cannot activate transcription. In contrast, a protein translated from mature mRNA [pXBP1(S)] activates the transcription of ERAD component genes such as EDEM, HRD1, Derlin-2, and Derlin-3 through a cis-acting element, unfolded protein response element, as these two ORFs are joined in mature mRNA [37,88,89]. pXBP1(S) also induces the expression of proteins involved in lipid synthesis and ER biogenesis, as well as the expression of ER chaperones such as BiP, p58IPK, ERdj4, PDI-P5 and HEDJ [90,91]. Thus, XBP1 is essential to the function of cells that produce large amounts of secretory proteins such as pancreatic β-cells, hepatocytes, and antibody-producing plasma cells [92–95].
The splicing of XBP1 pre-mRNA by IRE1α is quite different from conventional mRNA splicing (Fig. 7A) . Conventional splicing is catalyzed by the spliceosome, and the consensus sequence at the exon–intron border is GU-AG or AU-AC (Chambon's rule). The splicing reaction is sequential: the 5′ site is cleaved first, then the 3′ site after a lariat structure is formed. In contrast, unconventional splicing of XBP1 pre-mRNA is catalyzed by IRE1α and RNA ligase, and there is a pair of stem–loop structures at the exon–intron border instead of GU-AG or AU-AC. Moreover, the splicing reaction is not sequential but random.
Figure 7. Cytoplasmic splicing. (A) Comparison between nuclear and cytoplasmic splicing. Conventional splicing is catalyzed by the spliceosome in the nucleus, and there is a consensus sequence at the exon–intron boundary such as GU-AG or AU-AC. The splicing reaction is sequential: the 5′ site is cleaved first, the lariat structure is formed, and then the 3′ site is cleaved. In contrast, unconventional splicing is catalyzed by IRE1 and RNA ligase in the cytoplasm, there is a characteristic stem–loop structure at the boundary, and the splicing reaction is random without forming a lariat structure. (B) Biological significance of cytoplasmic splicing. As nuclear splicing cannot splice premRNA exported to the cytoplasm, de novo transcription is required to change the character of the protein encoded in the premRNA. In contrast, as cytoplasmic splicing can splice pre-mRNA that is translated in the cytoplasm, it can rapidly change the character of a protein in response to external or internal stimuli, without de novo transcription.
Download figure to PowerPoint
The most important difference between conventional and unconventional splicing is where the reaction occurs (Fig. 7B). Conventional splicing (nuclear splicing) takes place in the nucleus, whereas unconventional splicing (cytoplasmic splicing) occurs in the cytoplasm. The biological significance of cytoplasmic splicing is that pre-mRNA used for translation in the cytoplasm can be spliced when it is necessary to change the nature of the protein translated from the mRNA, in response to extracellular or intracellular signaling. In contrast, as nuclear splicing cannot splice mRNA exported to the cytoplasm, it is necessary for pre-mRNA to be transcribed de novo and spliced. Thus, cytoplasmic splicing would be a very rapid, versatile, and energy-efficient mechanism with minimal waste as compared with conventional mRNA splicing. Recently, it was found that pXBP1(U) encoded in XBP1 pre-mRNA is a negative feedback regulator of pXBP1(S). Thus, in the case of XBP1, pre-mRNA and mature mRNA encode negative and positive regulators, respectively, and their expression is switched by cytoplasmic splicing in response to the situation in the ER .
IRE1β is specifically expressed in epithelial cells of the gastrointestinal tract, and thought to cleave rRNA to attenuate translation in response to ER stress . When IRE1β–/– mice were exposed to an inducer of inflammatory bowel disease, they actually developed colitis, possibly because of the enhanced ER stress .
Recently, the crystal structure of the luminal domain of IRE1α was solved . The luminal domain is similar in structure to the peptide-binding domain of major histocompatibility complexes, suggesting the interesting possibility that it directly senses ER stress by directly binding unfolded proteins.
The accumulation of unfolded proteins in the ER is toxic to cells. Thus, if the PERK, ATF6, and IRE1 pathways cannot suppress ER stress, an apoptotic pathway is triggered to ensure survival of the organism as a last line of defense. A number of pathways have been reported to be involved in ER stress-induced apoptosis, and the full induction of apoptosis seems to require the concomitant activation of several death pathways, although there remain many arguments over ER stress-induced apoptosis [99–105]. In this section, I will briefly summarize the known death pathways, focusing on recent progress (Fig. 8).
Figure 8. ER stress-induced apoptotic pathways. The subcelluar distribution of factors involved in ER stress-induced cell death is shown. Pro-apoptotic and anti-apoptotic factors are indicated in black and white letters, respectively. Pro-apoptotic Bcl2 proteins positively regulate the IRE1 and caspase pathways, whereas anti-apoptotic Bcl2 proteins negatively regulate the latter. p53 enhances expression of pro-apoptotic Blc2 family proteins such as PUMA and NOXA. The IRE1 pathway activates JNK and SAPK, leading to apoptosis. c-Abl translocates from the ER to the mitochondria in response to ER stress. A fuller explanation is given in the text.
Download figure to PowerPoint
The most characterized pathway is the CHOP pathway. CHOP/GADD153 (growth arrest and DNA damage 153) is a transcription factor, the expression of which is induced by the ATF6 and PERK pathways upon ER stress [70,106,107]. CHOP–/– cells exhibit less programmed cell death when faced with ER stress , suggesting that the CHOP pathway is a major regulator of ER stress-induced apoptosis. As for the target genes of CHOP, CHOP activates the transcription of GADD34, ERO1, DR5 (death receptor 5), and carbonic anhydrase VI, which seem to be responsible for apoptosis. GADD34 associated with protein phosphatase 2C enhances dephosphorylation of eIF2α and promotes ER client protein biosynthesis , whereas ERO1, which encodes an ER oxidase, makes the ER a more hyper-oxidizing environment . DR5, which encodes a cell surface death receptor, may activate caspase cascades . Carbonic anhydrase VI may change the cellular pH, affecting various cellular processes [112,113]. However, the exact signaling mechanism from CHOP to apoptosis is still unclear.
The second apoptotic pathway is the IRE1–TRAF2–ASK1 pathway. The cytoplasmic part of IRE1 binds to an adaptor protein, TRAF2 (tumor necrosis factor receptor-associated factor 2), which couples plasma membrane death receptor to Jun kinase (JNK) and stress-activated protein kinase (SAPK) . IRE1 and TRAF2 form a complex with a mitogen-activated protein kinase kinase kinase, ASK1 (apoptosis signal-regulating kinase 1), and this IRE1–TRAF2–ASK1 complex is responsible for the phosphorylation and activation of JNK . Actually, IRE1–/– cells as well as ASK1–/– cells are impaired in the activation of JNK and apoptosis by ER stress. In contrast, TRAF2–/– cells are more susceptible to apoptosis triggered by ER stress, which might be inconsistent with the above model . TRAF2 also associates with caspase-12 and regulates its activation . IRE1–TRAF2 activates the transcriptional repressor ATF3 as well, leading to the activation of apoptosis . These suggest that the IRE1–TRAF2–ASK1 pathway is a major regulator of ER stress-induced apoptosis. TNFR1 (tumor necrosis factor receptor 1), a receptor for TNF-induced cell death, associates with IRE1α upon ER stress, and the activation of JNK by ER stress is impaired in TNFR1–/– cells. This suggests that TNFR1 mediates the ER stress-induced activation of JNK , possibly forming a complex with IRE1α, TRAF2, and ASK1. The expression of TNFα is up-regulated by the IRE1 pathway during ER stress [120,121], which may contribute to the activation of TNFR1.
Caspases are well-known pro-apoptotic components, and caspases 2, 3, 4, 7, 9 and 12 are reported to be involved in ER stress-induced cell death [122–131]. Caspase-12 is associated with the ER membrane, and activated by ER stress, possibly by calpain . Then caspase-12 activates caspase-9, which in turn activates caspase-3 , leading to cell death. Caspase-12–/– mice are resistant to ER stress-induced apoptosis but sensitive to other death stimuli, suggesting that caspase-12 is a regulator specific to ER stress-induced apoptosis . The activation of caspase-12 by ER stress is observed in models of various diseases such as Alzheimer's disease, polyglutamine disease, ischemia, and viral infection [135–139], suggesting that ER stress-induced apoptosis is closely involved in these diseases (see below). However, the involvement of caspase-12 in apoptosis of human cells is still open to question, as the human caspase-12 gene contains several mutations critical to function . It is possible that an unidentified caspase other than caspase-12 is responsible in human cells.
Bcl2 family proteins are well-known components of the programmed cell death machinery, and some key components are involved in ER stress-induced apoptosis. In general, pro-apoptotic members of the Bcl2 family seem to be recruited to the ER surface and to activate caspase-12, whereas the anti-apoptotic members inhibit this recruitment, although the exact relationship between these factors is still unclear. So I will briefly describe our current understanding of Bcl2 family proteins. The anti-apoptotic factor Bcl2 is down-regulated by the transcription factor CHOP upon ER stress, which leads to enhanced oxidant injury and apoptosis . Overexpression of Bcl2 inhibits the activation of caspase-12 and apoptosis during ER stress . The pro-apoptotic factor, BAD (Bcl2 antagonist of cell death), is dephosphorylated and activated in response to ER stress , whereas other pro-apoptotic factors, Bax (Bcl2-associated X protein) and Bak (Bcl-2 homologous antagonist/killer), are present in the ER membrane as well as the mitochondrial membrane [144,145]. During ER stress, Bax and Bak oligomerize and activate caspase-12. Interestingly, Bax and Bak associate with IRE1α and modulate IRE1α function during ER stress . Bax and Bak are required for most forms of apoptosis . The transcription of PUMA (p53 up-regulated modulator of apoptosis) and NOXA (neutrophil NADPH oxidase factor), pro-apoptotic members of the BH3 (homology domain-3) domain-only family, is up-regulated by p53 during ER stress, and PUMA–/– cells and NOXA–/– cells are resistant to ER stress-induced apoptosis [148,149]. Another pro-apoptotic component, Bim (Bcl2-interacting mediator of cell death), translocates from the dynein-rich compartment to the ER membrane and activates caspase-12 in response to ER stress, whereas an anti-apoptotic factor, Bcl-xL (Bcl-2-like 1), binds to Bim and inhibits its translocation . Bim-knockdown cells are resistant to ER stress. The ER-localized anti-apoptotic factor BI-1 (Bax inhibitor-1) inhibits the activation of Bax during ER stress, and BI-1–/– mice are sensitive to ER stress, whereas mice overexpressing BI-1 are resistant . BIK (Bcl2-interacting killer) is an ER-localized pro-apoptotic component which enhances the recruitment of BAX and BAK to the ER . Bap31 (B cell receptor-associated protein 31) is a pro-apoptotic factor that is cleaved and activated upon ER stress, its cleavage being dependent on calnexin . In calnexin-deficient cells, the cleavage of Bap31 and ER stress-induced apoptosis are inhibited.
The inhibitor of apoptosis (IAP) family has also been reported to be involved in ER stress-induced apoptosis. Human inhibitor of apoptosis 2 (HIAP2) is an IAP that inhibits caspase-3 and caspase-7. Expression of HIAP2 is induced upon ER stress at the level of translation: caspases activated by ER stress cleave eukaryotic initiation factor, p97/DAP5/NAT1, and the cleavage product specifically activates the HIAP2 internal ribosome entry site, leading to enhanced translation of HIAP2 . Transcription of IAP-2 and XIAP (inhibitor of apoptosis, X-linked), two other IAPs, is up-regulated during ER stress, and cells in which these IAPs have been knocked down are sensitive to ER stress-induced apoptosis . Cells overexpressing XIAP or HIAP1 are resistant to ER stress [122,155]. These results suggest involvement of IAP proteins in ER stress-induced apoptosis.
c-Abl (Abelson murine leukemia viral oncogene homolog 1) is a protein tyrosine kinase distributed in the nucleus and cytoplasm, and c-Abl activated by a death signal induces phosphorylation and activation of pro-apoptotic JNK and SAPK. Interestingly, c-Abl is also located on the ER surface, and translocates to the mitochondria upon ER stress, where it induces the release of cytochrome c. c-Abl–/– cells are resistant to ER stress-induced apoptosis. A c-Abl-interacting protein, Aph2 (anterior phalynx defective 2), is also located on the ER and shows pro-apoptotic activity , suggesting that c-Abl forms a distinct pathway leading to ER stress-induced apoptosis.
PKR is a dsRNA-dependent protein kinase which is activated upon viral infection, and its phosphorylation in the nucleus is up-regulated in response to ER stress. Interestingly, PKR-knockdown cells or PKR mutant cells are resistant to ER stress-induced apoptosis, suggesting that PKR is a pro-apoptotic factor during ER stress .
TDAG51 (T-cell death-associated gene 51) is a member of the pleckstrin homology-related domain family, and its transcription is induced by ER stress through the PERK pathway . Overexpression of TDAG51 induces apoptosis, suggesting that TDAG51 is involved in ER stress-induced apoptosis and the development of atherosclerosis (see below).
Nuclear respiratory factor (NRF)1 and NRF2 are transcription factors that regulate the oxidative stress response. NRF2 is distributed in the cytoplasm through its association with the microtubule-associated protein Keap1 (Kelch-like Ech-associated protein 1). Upon ER stress, PERK phosphorylates NRF2 and dissociates it from Keap1, leading to the nuclear recruitment of NRF2 . Remarkably, NRF2–/– cells are sensitive to ER stress-induced apoptosis, whereas NRF1 is located in the ER membrane, and translocates to the nucleus upon ER stress , suggesting that these proteins are involved in ER stress-specific apoptosis.
ATF6 is also involved in the apoptotic process during myogenesis. In differentiating myoblast, the ATF6 pathway is activated, and expression of BiP and CHOP is up-regulated, which may activate caspase-12. Moreover, AEBSF [4-(2-aminoethyl)benzenesulfonyl fluoride hydrochloride], an inhibitor of ATF6 activation, blocks apoptosis, suggesting that the ATF6 pathway contributes to apoptosis during myogenesis .
TRB3 (Tribbles homolog 3) is a human ortholog of Drosophila tribble, and its transcription is induced by ER stress through the PERK–ATF4–CHOP pathway. Interestingly, TRB3-knockdown cells are resistant to ER stress-induced apoptosis, suggesting that TRB3 is a pro-apoptotic factor during ER stress [163,164].
p53 is a transcription factor that induces growth arrest and apoptosis in response to various forms of cellular stress, such as DNA damage. There are several reports suggesting a connection between p53 and ER stress. Upon ER stress, p53 is phosphorylated by glycogen synthase kinase 3B, leading to the distribution of p53 in the cytoplasm and destabilization of p53, and attenuation of p53-dependent apoptosis [165–167]. Interestingly, Scotin and SCN3B (sodium channel subunit beta 3), p53-inducible pro-apoptotic proteins, are located in the ER [168,169]. Moreover, transcription of PUMA and NOXA, which are involved in ER stress-induced apoptosis, is up-regulated by p53 during ER stress , suggesting that the p53 pathway regulates the apoptotic pathway during ER stress.
Other components such as VCP and ALG-2 (apoptosis-linked-gene 2) , AIGP1 , elongation factor-1α, and NRADD (neurotrophin receptor associated death domain)  are also reported to be involved in ER stress-induced apoptosis. Their precise functions and working mechanisms are still to be clarified.