Redox controls UPR to control redox

A fundamental property of Metazoa is their ability to send, receive and decode signals between neighboring tissues and the extracellular environment. As a prerequisite for such communication, eukaryotic cells developed a specialized endomembrane system known as the secretory pathway, in which peptide hormones, membrane receptors and other secretory cargo proteins are synthesized, folded and eventually released to the extracellular space. The endoplasmic reticulum (ER), where translation and folding of extracellular proteins occurs, is the first component of the secretory pathway (Gidalevitz et al., 2013). When ER homeostasis is compromised, a set of ‘ER stress’ signals is transmitted to the cytoplasm and nucleus. This ER-centric system of signal transduction is defined as the unfolded protein response (UPR), the primary function of which is to re-establish ER homeostasis (Hetz, 2012). If this fails, however, the very same signaling network is capable of sending death messages, leading to programmed cell death (Tabas and Ron, 2011).

The UPR in higher eukaryotes relies on three major signaling cascades originating from the ER membrane UPR sensors inositol-requiring protein-1 (IRE1α or IRE1β, herein referred to as IRE1; also known as ERN1), protein kinase PKR-like ER kinase (PERK; also known as EIF2AK3), and activating transcription factor-6α (ATF6). IRE1 is a ribonuclease that catalyzes non-canonical splicing of X-box binding protein 1 (XBP1) mRNA, a step that is required for the expression of this master transcription factor of UPR target genes (Calfon et al., 2002; Lee et al., 2002; Yoshida et al., 2001). The ribonuclease activity is also required for regulated IRE1-dependent mRNA decay (RIDD) (Hollien et al., 2009; Maurel et al., 2014). In addition, activated IRE1 can trigger the apoptosis signal-regulating kinase-1 (ASK1; also known as MAP3K5) to JUN N-terminal kinase (JNK) pathway (Urano et al., 2000). PERK is a kinase, the main target of which upon ER stress is the α subunit of the eukaryotic translation initiation factor-2 (eIF2α; also known as EIF2S1) (Harding et al., 2000b). Phosphorylated eIF2α, in turn, causes general attenuation of protein synthesis while promoting the expression of the UPR transcription factor ATF4 (Harding et al., 2000a). ATF6 is itself a transcriptional activator of UPR target genes. Normally confined to the ER membrane, it is activated upon ER stress and migrates to the Golgi where its cytosolic DNA-binding domain is freed through intra-membrane proteolysis (Haze et al., 1999).

A biochemical process that is crucial for ER protein homeostasis is the formation of disulfide bridges, which is often referred to as oxidative protein folding. Disulfide bonds are introduced into nascent proteins by ER-resident protein disulfide isomerases (PDIs) (Appenzeller-Herzog and Ellgaard, 2008), resulting in reduction of the latter. A number of pathways exist for the re-oxidation of PDIs, among these the best conserved – catalyzed by endoplasmic oxidoreductin 1 (Ero1α and Ero1β in human; encoded by ERO1L and ERO1LB, respectively) – exploit the oxidizing power of molecular oxygen (Box 1) (Bulleid and Ellgaard, 2011). At the same time, disulfide-reducing pathways are essential for the proof reading of incorrectly introduced disulfide bonds and for efficient degradation of ER proteins (Kojer and Riemer, 2014). As in probably every cell compartment (Schafer and Buettner, 2001), millimolar concentrations of the redox couple glutathione (GSH)–glutathione disulfide (Montero et al., 2013) are important for disulfide reduction and the maintenance of redox homeostasis in the ER (Appenzeller-Herzog, 2011). Thus, a healthy ER environment is characterized by a well-tuned curtailing of both oxidizing and reducing power.

In this Commentary, we summarize the consequences of redox disturbance for ER function. ER dysfunctions can cause cellular redox changes, which in turn feedback onto the ER to influence cell survival and death. These interrelations place redox disturbances at the interface of ER stress and disease.

The early finding that glucose-regulated UPR target genes are induced by treatment with reducing agents, which antagonize the formation of disulfide bonds, was made before the actual discovery of the UPR (Kim and Lee, 1987). Nowadays, disulfide-reducing agents such as dithiothreitol (DTT) are widely being used as rapid inducers of ER stress (DuRose et al., 2006), and it has been shown that even an excess of the endogenous ER reductant GSH can activate the UPR by compromising oxidative protein folding (Kumar et al., 2011).

However, there is accumulating evidence that ER stress signaling is elicited also in response to oxidative triggers. Treatments that cause intracellular release of reactive oxygen species (ROS; Box 2) lead to pro-apoptotic ER stress (Dejeans et al., 2010; Inoue and Suzuki-Karasaki, 2013; Santos et al., 2009), a phenomenon that is being exploited in photodynamic cancer therapies (Garg and Agostinis, 2014). Like ROS generation, nitrosative stress is also linked to UPR-mediated cytotoxicity through the inactivation of active-site cysteine residues in PDIs through nitrosylation (Uehara et al., 2006).

In addition to these non-ER challenges, oxidative protein folding in the ER is assumed to be a source of ER over-oxidation and stress (Han et al., 2013; Marciniak et al., 2004; Tu and Weissman, 2004). Expression of a hyperactive Ero1α mutant causes detectable concentrations of its reaction by-product hydrogen peroxide (H₂O₂; Box 2) in the ER, along with mild UPR activation, which is exacerbated by knockdown of the ER-resident and UPR-inducible glutathione peroxidase 8 (GPX8) (Ramming et al., 2014). Accordingly, Ero1α-derived H₂O₂, rather than disulfide overproduction, appears to affect ER homeostasis. Similarly, overexpression of Ero1β in pancreatic β-cells causes ER stress (Awazawa et al., 2014). Whether H₂O₂ that is derived from endogenous Ero1 is sufficient to trigger ER stress is questionable (Appenzeller-Herzog, 2011; Ramming et al., 2014). Interestingly, peroxide sensing or quenching by GPX7 (a GPX8-related ER peroxidase, also known as NPGPx) (Ramming and Appenzeller-Herzog, 2013) promotes cell survival by stimulating the ER chaperone BiP (also known as GRP78 and HSPA5) (Wei et al., 2012).

Although the mechanism by which reductive challenge of the ER causes stress is mainly based on compromised disulfide bond formation and consequent protein misfolding, it is less clear how hyper-oxidizing conditions affect the ER. As discussed below, we propose that – besides sulfoxidation, nitrosylation or glutathionylation of key cysteine residues in ER lumenal proteins (e.g. in PDIs) – specific oxidation of ER membrane proteins has profound effects on Ca²⁺ homeostasis and UPR signaling.

Ca²⁺ influx into the ER is governed by ATP-dependent Ca²⁺ import pumps that belong to the sarcoendoplasmic reticulum Ca²⁺ transport ATPase (SERCA) family, whereas Ca²⁺ efflux from the ER is controlled by excitable Ca²⁺ release channels from the inositol 1,4,5-trisphosphate (IP₃) receptor (IP₃R) and, mainly in muscle and brain tissue, the ryanodine receptor (RyR) families, as well as a heterogeneous collection of Ca²⁺ leak pores (Berridge et al., 2003). Interestingly, different UPR-triggering stimuli, such as unsaturated fatty acids, DTT, cytokines or expression of misfolded ER proteins, result in loss of Ca²⁺ from the ER (Hara et al., 2014). This can occur owing to inhibition of SERCA, the opening of release pores, or both. Lower SERCA activity, e.g. following disturbed lipid metabolism, glucose starvation or inflammatory mediators, is a well-known phenomenon during ER-stress-dependent toxicity (Cardozo et al., 2005; Fu et al., 2011; Hara et al., 2014; Moore et al., 2011). Likewise, altered IP₃R activity and increases in passive Ca²⁺ leak from the ER are hallmarks of ER-stressed cells (Kiviluoto et al., 2013) (see also below).

ER Ca²⁺ pumps and IP₃R or RyR channels are subject to redox regulation, which we propose is an important component of ROS-driven ER stress. Sulfoxidation of the conserved SERCA cysteine residue 674 in response to ROS inhibits pump activity, whereas glutathionylation at Cys674 by nitric oxide activates SERCA (Raturi et al., 2014; Tong et al., 2010). In the large IP₃R and RyR channel proteins, complex, yet incompletely defined redox mechanisms occur that involve several cysteine residues (Raturi et al., 2014). Nevertheless, ROS-mediated sensitization to submaximal doses of IP₃ agonist and cytosolic Ca²⁺, which enhances ER Ca²⁺ release, is well documented (Bansaghi et al., 2014). Taken together, ROS can cause the depletion of Ca²⁺ from the ER, which disables Ca²⁺-dependent ER chaperones, such as calnexin and calreticulin, and results in ER stress (Hetz, 2012).

Another connection between the redox state and Ca²⁺ that has potential implications for ER stress physiology has been demonstrated by the finding that short-term depletion of Ca²⁺ from the ER, by either inhibition of SERCA or activation of IP₃R, leads to an immediate drop in the reduction potential of the ER (i.e. rendering the ER lumen more reducing) (Avezov et al., 2013; Birk et al., 2013; Enyedi et al., 2010). Although the mechanistic basis for this drop remains to be explored, it is probable that lowering ER Ca²⁺ levels during chronic ER stress (Cardozo et al., 2005; Fu et al., 2011; Hara et al., 2014; Moore et al., 2011) mediates a hypo-oxidizing ER environment. With regards to the reductive activation of ATF6 (see below), ER hypo-oxidation, in turn, will stimulate stress signals.

The UPR sensors ATF6, IRE1 and PERK are activated in response to ER stress, which is mainly mediated through the lumenal domains of the sensors. BiP, which binds to all three sensors in their inactive state, is titrated off when the unfolded protein load rises in the ER, and UPR sensors can also directly sense unfolded proteins, leading to their activation (Walter and Ron, 2011). Two studies have recently described redox regulation of the lumenal domains of the ER sensors that is mediated by dedicated PDIs (Eletto et al., 2014; Higa et al., 2014). This regulation impacts the onset, amplitude and duration of signaling by each of the sensors, thus providing a mechanistically novel concept of sensor-specific UPR activation.

In the unstressed ER, ATF6 exists in the form of disulfide-bonded monomers, dimers or oligomers, which rely on cysteine residues 467 and 618 in the ER lumenal domain (Nadanaka et al., 2007) (Fig. 1A). When oxidized, the sensor remains in the ER, but in response to ER stress, the disulfide bonds in ATF6 are cleaved, and monomeric reduced ATF6 is released from the organelle (Fig. 1B). Not surprisingly, this occurs more rapidly in cells that are treated with DTT compared with cells that are treated with other pharmacological UPR triggers (DuRose et al., 2006; Higa et al., 2014; Nadanaka et al., 2007). In cells that have been depleted of PDIA5 (also known as PDIR), but not of other PDIs, oligomer dissociation, ER-to-Golgi migration and induction of ATF6 target genes are all inhibited, strongly suggesting that PDIA5 cleaves the disulfide bonds in ATF6 (Higa et al., 2014). ER-stress-induced dissociation of BiP from ATF6, however, occurs independently of PDIA5 (Higa et al., 2014) and is still a prerequisite for ER-to-Golgi transport of monomeric ATF6 in which cysteine residues 467 and 618 are mutated (Nadanaka et al., 2007). Thus, PDIA5-mediated reduction of ATF6 is necessary but not sufficient for Golgi targeting, which in addition requires the dissociation of BiP.

Unlike in the case of ATF6, the thiol–disulfide configuration of the three lumenal cysteine residues in IRE1 or the four lumenal cysteine residues in PERK (Fig. 1A) in their inactive states is not clear. However, similar to ATF6, the activities of IRE1 and PERK are controlled by thiol–disulfide exchange in their lumenal domains. Here, PDIA6 (also known as P5) is the specific catalyst, as indicated by the trapping of sensor–PDIA6 interchain disulfide bonds (Eletto et al., 2014). Depletion of PDIA6 increases the duration and amplitude of IRE1 or PERK signaling, causes profound hypersensitivity of cells to ER-stress-inducing agents and mediates a developmental block in Caenorhabditis elegans (Eletto et al., 2014). These observations indicate that PDIA6 terminates sensor activation and prevents exaggerated UPR signaling.

Although the lumenal cysteine residues in IRE1 are dispensable for signal initiation, they can engage in Cys148–Cys148 and Cys332–Cys332 interchain disulfide bonds that result in IRE1 dimers and higher order oligomers (Eletto et al., 2014; Liu et al., 2003). Formation of these interchain disulfide bonds is promoted by ER stress or by IRE1 overexpression (Eletto et al., 2014), which leads to spontaneous IRE1 clustering and activation (Han et al., 2013; Li et al., 2010b). ER-stress-induced IRE1 clustering could then promote the rearrangement of pre-existing intramolecular disulfide bonds (Fig. 1C). Alternatively, interchain disulfide bonds in IRE1 could be formed in response to an oxidative trigger such as ROS, which is consistent with the selective activation of IRE1 by the mitochondrial poison antimycin A (Mori et al., 2013). Irrespective of the precise mechanism, PDIA6 is required to revert this conversion and to inactivate IRE1 through nucleophilic attack, mainly at residue Cys148 (Eletto et al., 2014) (Fig. 1D). Accordingly, Cys148–Cys148 might act as a recognition motif for PDIA6, which in turn limits the signaling and resets the original inactive state of IRE1. To date, the roles of lumenal cysteine residues in PERK are poorly characterized.

Thus, the enzymatic activities of both PDIA5 and PDIA6 specifically act on the stress sensors in response to ER stress to reduce the disulfide bonds in the sensors. The molecular basis for the intriguing selectivity of the PDI–sensor pairs remains to be determined.

The discovery of the PDIA6–IRE1, PDIA6–PERK and PDIA5–ATF6 axes suggests that redox changes not only impact on ER homeostasis (as discussed above) but also on UPR signaling itself (Fig. 2). Accordingly, by promoting oxidation of UPR sensors or of PDIs, ROS signals could potentially inhibit ATF6 signaling while, at the same time, enhancing the duration of IRE1 and PERK signaling. In addition, it is conceivable that physiological triggers of mild ER hypo-oxidation selectively activate the ATF6 arm of the UPR. On a more general note, we propose that the different redox sensitivities of UPR sensors contribute to sensor-selective UPR settings and sensor-specific (in)activation kinetics.

Stress sensor regulation mediated by PDIs probably influences cell physiology. Upregulation of both PDIA5 and PDIA6 is essential for chemoresistant cancer cell growth. Specifically, resistance of leukemic cells to the tyrosine kinase inhibitor Imatinib depends on PDIA5 and pro-survival ATF6 signaling (Higa et al., 2014), and non-small lung cancer cells require robust overexpression of PDIA6 (and PDIA4, also known as ERp72) to withstand treatment with cytotoxic Cisplatin (Tufo et al., 2014). It is attractive to speculate that these pro-survival roles of PDIA5 and PDIA6 in cancer are mainly due to redox regulation of UPR sensors, i.e. promoting pro-survival ATF6 signaling and inhibiting sustained, pro-apoptotic IRE1 or PERK signaling. Indeed, data for PDIA5- or PDIA6-mediated oxidative protein folding is scarce. It has been shown that, in vitro, PDIA5 catalyzes the folding of α₁-antitrypsin (Horibe et al., 2004) and associates with the chaperone calreticulin, which might hint to a role in glycoprotein folding (Vinaik et al., 2013). PDIA6 has been implicated in the folding of BiP substrates (Jessop et al., 2009), although its knockdown has no detectable consequences for oxidative protein folding in hepatocytes or pancreatic β-cells (Eletto et al., 2014; Rutkevich et al., 2010).

One remarkable difference between PDIA5 and PDIA6 is their copy number, with PDIA6 being at least tenfold more abundant, depending upon the cell type (Gidalevitz et al., 2013). PDIA6-mediated control of IRE1 and PERK signaling might be governed by low affinity interactions between the sensor and the enzyme, which could be one reason why PDIA6 is so abundant. Moreover, although PDIA5 also binds to BiP in detectable amounts (Jansen et al., 2012), the non-covalent complex between PDIA6 and BiP appears to be particularly prominent, as shown by several laboratories (Eletto et al., 2014; Jessop et al., 2009; Meunier et al., 2002). Thus, just like the UPR sensors themselves, PDIA6 might be sequestered (or recruited to the vicinity of BiP targets for the catalysis of oxidative folding) under basal conditions through interacting with BiP (Fig. 1C,D). However, a relatively lower amount of PDIA5 appears to suffice to guarantee efficient ATF6 activation. It is possible that PDIA5 is a better reductase than PDIA6, which displays more oxidizing power than other PDIs (Araki et al., 2013).

Collectively, we propose that the pro-survival effects of PDIA5 and PDIA6 can be attributed to their functions as reductants of disulfide bonds in UPR sensors. Genetic or oxidative interference with these functions can contribute to the execution of cell death.

Although ROS trigger ER stress, they can also arise in response to ER stress in a cell-type-dependent manner (Appenzeller-Herzog, 2011) (Fig. 2). Yeast cells that are defective in clearing misfolded ER proteins, or PERK-knockout mouse fibroblasts that are subjected to ER stress, show prominent ROS levels and reduced viability (Harding et al., 2003; Haynes et al., 2004). In both model systems, ROS generation ceases upon inactivation of the mitochondrial electron transport chain, implying that ER–mitochondrial crosstalk plays a vital part in the redox stress response (see next section). Similarly, overexpression of misfolding-prone coagulation factor VIII induces severe ER stress and intracellular ROS, which can be rescued with the addition of a quencher of free radicals (Malhotra et al., 2008).

There is converging evidence that cellular hyper-oxidation during irremediable ER stress is, at least in part, orchestrated by the transcription factor C/EBP homologous protein (CHOP; also known as DDIT3) (Marciniak et al., 2004). At the transcriptional level, CHOP is targeted by ATF4, and ATF4 and CHOP co-regulate a panel of genes that are involved in protein synthesis upon prolonged stress (Han et al., 2013). Thus, phosphorylated-eIF2α-mediated translation attenuation that has been induced during the adaptive phase of UPR downstream of PERK is reverted, which ultimately increases ROS load and leads to cell death. In vivo, a diet containing antioxidants protects pancreatic β-cells from UPR- and protein synthesis-driven demise, suggesting a causal role of ER-stress-induced ROS in apoptosis progression (Back et al., 2009).

One CHOP target gene that contributes to ROS-dependent apoptosis in mammalian cells is ERO1L (Han et al., 2013; Li et al., 2009; Marciniak et al., 2004). Furthermore, nematode ERO-1 contributes to animal-wide ROS accumulation in response to ER stress and to decreased life span of ER-stressed animals (Harding et al., 2003; Marciniak et al., 2004). As Ero1-family enzymes are direct producers of ROS (Box 1) (Bulleid and Ellgaard, 2011), an apparently obvious explanation for these findings in mammalian cells and nematodes is that increased Ero1 activity following ER stress causes diffusion of Ero1-derived ROS to the cytoplasm. Testing this model in mammalian tissue culture, however, failed to support this notion and instead identified GPX8 as an ER-stress-regulated ER peroxidase that is crucial for the control of Ero1α-derived ROS and cell survival (Ramming et al., 2014). At second glance, it appears intuitive that ROS as a regular by-product of oxidative protein folding require tight control. Endogenous sources of ROS, which act in cellular signaling, such as NADPH oxidases (NOXs), are typically inducible (Gough and Cotter, 2011), but disulfide production by Ero1 is a constitutive housekeeping process.

In addition to its role in disulfide bond formation, Ero1α is a positive regulator of IP₃R. Accordingly, IP₃R-dependent Ca²⁺ flux to the cytosol and mitochondria is stimulated by induction of Ero1α (Anelli et al., 2012; Li et al., 2009). Increased cytosolic Ca²⁺ levels feed into the JNK pathway through Ca²⁺/calmodulin-dependent protein kinase II (CaMKII) and thereby lead to augmented ROS production through NOX2 (also known as CYBB) or NOX4, which in turn leads to oxidative stress and positive-feedback regulation of CHOP through double-stranded RNA-dependent protein kinase (PKR; also known as EIF2AK2) (Li et al., 2010a; Pedruzzi et al., 2004) (Figs 2 and 3). Mitochondrial Ca²⁺ elevation, however, increases the generation and release of mitochondrial ROS by various mechanisms, including opening of the mitochondrial permeability transition pore (MPTP) in the inner mitochondrial membrane (Figueira et al., 2013) (see below).

Mitochondrial outer membrane permeabilization (MOMP) is the focal point of the intrinsic apoptosis pathway that is executed by mitochondria-derived endonucleases, caspase-activating factors such as cytochrome c, and ROS (Figueira et al., 2013; Galluzzi et al., 2012). Two pathways mediate MOMP. First, the pro-apoptotic proteins BAX and BAK can assemble into oligomeric pore complexes in the outer mitochondrial membrane. Formation of these complexes is tightly controlled by protein–protein interactions with other factors, including the pro-MOMP BH3-only proteins and anti-MOMP BCL2 (Czabotar et al., 2014). Second, prolonged MPTP opening, for example in response to Ca²⁺ uptake, induces swelling of the mitochondrial matrix, ultimately leading to MOMP (Rizzuto et al., 2012). As discussed below, both of these pathways can be triggered through chronic irremediable ER stress (Fig. 3).

ER stress signals facilitate BAX and BAK oligomerization by inducing the expression of BH3-only proteins, including BIM, NOXA, PUMA, DR5 and CRK (Austgen et al., 2011; Cunha et al., 2012; Li et al., 2006; Puthalakath et al., 2007; Wali et al., 2014) and by repressing BCL2 (McCullough et al., 2001). Furthermore, the BH3-only protein BID is truncated or activated in response to ER stress in a pathway downstream of caspase-8 (Namba et al., 2013) or caspase-2 (Upton et al., 2008) – although the role of caspase-2 is debated (Sandow et al., 2014). The caspase-8 pathway is stimulated by the ER stress effector cell-death-inducing p53 target 1 (CDIP1), which forms a caspase-activating complex with the ER membrane protein BAP31 (also known as BCAP31) (Namba et al., 2013) and, presumably, the mitochondrial outer membrane protein FIS1 (Iwasawa et al., 2011). In addition, the CDIP1–BAP31 complex sequesters BCL2 (Namba et al., 2013). It is worth mentioning that BAX, BAK and some BH3-only proteins also reside on the ER membrane (in addition to mitochondria) where they are necessary for full and sustained activation of IRE1 through direct physical interactions (Hetz et al., 2006; Rodriguez et al., 2012), although the implication of these interactions at mitochondria is currently unclear.

ER–mitochondrial crosstalk is orchestrated at sites of close membrane apposition, termed mitochondria-associated ER membranes (MAMs) (van Vliet et al., 2014). So far, the only established signaling entity that is funneled from the ER to mitochondria at MAMs is Ca²⁺. ER-to-mitochondria Ca²⁺ flux through a gateway comprising the IP₃R, voltage-dependent anion-selective channel protein 1 (VDAC1) and the mitochondrial Ca²⁺ uniporter (MCU) serves a dual role: at physiological amplitude, it stimulates mitochondrial ATP production, whereas at an excessive pathological amplitude, it initiates MOMP (Rizzuto et al., 2012). Several ER stress mechanisms influence mitochondrial function by boosting Ca²⁺ release from the ER. Firstly, a UPR-induced SERCA1 splice variant augments both resting and stimulated Ca²⁺ transmission to mitochondria (Chami et al., 2008). Secondly, CHOP-dependent upregulation of Ero1α potentiates IP₃-induced Ca²⁺ release (Anelli et al., 2012; Kiviluoto et al., 2013; Li et al., 2009). Furthermore, the degradation of IP₃Rs and RyRs is inhibited during ER stress (Belal et al., 2012) and the ER membrane protein σ-1 receptor dissociates from BiP to chaperone and stabilize type 3 IP₃R at MAM (Hayashi and Su, 2007). Moreover, ER–mitochondrion contact zones are tightened in response to chronic ER stress, stimulating Ca²⁺-dependent mitochondrial apoptosis (Csordás et al., 2006). It is possible that such tightening is facilitated by PERK, which is required for integrity and functionality of MAM (Verfaillie et al., 2012). Finally, although not formally proven, it is to be expected that UPR-dependent inhibition of anti-apoptotic Bcl-2 family proteins (by BH3-only proteins, see above) modulates IP₃-induced Ca²⁺ release (Parys, 2014). Taken together, these data demonstrate that the stimulated directed Ca²⁺ transport to mitochondria in response to ER stress can trigger MPTP opening and MOMP.

As exemplified by ROS-mediated amplification of ER Ca²⁺ release downstream of MOMP (Raturi et al., 2014), apoptotic ER–mitochondrial crosstalk entails various positive-feedback loops, including allosteric activation of IP₃R by cytochrome c (Boehning et al., 2003), caspase-3- or calpain-catalyzed activating cleavage of IP₃R (Alzayady et al., 2013) and inhibition of SERCA in response to cellular ATP depletion (Moore et al., 2011) (Fig. 3). In fact, it is probable that the propagation of MOMP through the mitochondrial network that precedes the ‘point of no return’ relies on these amplification loops, the generation of which is under the control of several signaling nodes.

The UPR is interconnected with other redox-relevant pathways in addition to those highlighted above, i.e. NOX proteins and MOMP. One of these is initiated by ASK1 and leads to the activation of the mitogen-activated protein kinases (MAPKs) JNK or p38 (Matsuzawa and Ichijo, 2008). ASK1 responds to various apoptotic stimuli, including ROS. Mechanistically, inhibition of ASK1 by the major cytosolic antioxidant protein thioredoxin (TRX; also known as TXN) (Saitoh et al., 1998) or by glutathione S-transferase pi (GSTp) (Adler et al., 1999) is abolished upon ROS stimulus, or reactive cysteine residues in ASK1 itself are oxidized in response to ROS (Jarvis et al., 2012). In addition, ASK1 is activated by ER stress through IRE1 (Urano et al., 2000). Sustained activation of ASK1, and thus JNK, following ER stress crucially depends on JNK-dependent inhibition of mitochondrial respiration and resulting ROS production (Win et al., 2014) (Fig. 3). JNK also stimulates BIM and represses BCL2 through phosphorylation, which provides a further link to mitochondrial dysfunction (Lei and Davis, 2003; Putcha et al., 2003; Yamamoto et al., 1999).

Another redox-relevant pathway involves thioredoxin-interacting protein (TXNIP; also known as TBP-2 or VDUP1), a cytosolic ER stress effector that is upregulated both transcriptionally and by a de-repression mechanism that involves IRE1-catalyzed degradation of microRNA 17 (Lerner et al., 2012; Oslowski et al., 2012). TXNIP fulfills many functions, including negative regulation of mammalian target of rapamycin (mTOR) complex 1 (Jin et al., 2011), activation of the NACHT, LRR and PYD domains-containing protein 3 (NLRP3) inflammasome (Zhou et al., 2010) and inhibition of the antioxidant enzyme TRX (Nishiyama et al., 1999). The latter function is associated with higher susceptibility to oxidative stress (Liyanage et al., 2007), which again links UPR signaling with oxidative challenge.

However, in contrast to TXNIP-dependent TRX inhibition, UPR signals can also stimulate programs that have antioxidant effects to counterbalance oxidative injury (Fig. 2). These programs prominently include the antioxidant response that is orchestrated by nuclear factor erythroid 2-related factor 2 (NRF2; also known as NFE2L2) (Box 3). NRF2 can be directly phosphorylated and stabilized by PERK downstream of ER stress (Cullinan and Diehl, 2004). Conversely, in the context of ER stress during liver cirrhosis, NRF2 is degraded by an IRE1-regulated ubiquitin ligase (Wu et al., 2014). Extensive transcriptional crosstalk between UPR and the antioxidant response has been documented in C. elegans (Glover-Cutter et al., 2013), which establishes the basis for future investigations in mammalian models.

Together with CHOP, ATF4 is responsible for the upregulation of many genes that are involved in protein synthesis (Han et al., 2013; Harding et al., 2003), including genes encoding amino acid transporters that are essential for intracellular GSH synthesis. Consequently, ATF4-knockout cells display low GSH levels and are hypersensitive to oxidative challenge or amino acid withdrawal, highlighting an antioxidant function of the PERK–ATF4 pathway (Harding et al., 2003). Recently, stimulation of the hexosamine biosynthetic pathway, which provides the building blocks for stress-induced O-linked N-acetylglucosamine (O-GlcNAc) modifications that are involved in ROS defense, has been demonstrated downstream of IRE1 and XBP1 activation (Vincenz and Hartl, 2014). Taken together, both PERK and IRE1 effectors can contribute to ER-stress-regulated oxidant and antioxidant levels, whereas their relative specificities are likely to be context dependent.

There is overwhelming evidence that in many cases ER-stress-related pathologies are accompanied or even caused by intracellular redox disturbances. Importantly, these disturbances can be upstream and downstream of ER stress or UPR and therefore have the potential to foster vicious circles that further amplify the initial disturbance (Figs 2 and 3). Depending on the context, the self-amplifying pathological redox imbalance can be oxidizing, as exemplified by ROS, or reducing, as outlined in the recent hypothesis that ER hypo-oxidation contributes to the etiology of diseases as diverse as type 2 diabetes, Alzheimer's disease and cancer (Watson, 2014). However, it should be noted that connections between the redox state and UPR, so far, have been mostly derived from cell culture-based observations, and studies showing their in vivo relevance are scarce. Thus, examinations using suitable mouse models and genome-wide association and exome studies in humans are needed to fill this gap.

Does ‘redox’ provide a platform for therapeutic intervention? Although preclinical data suggest that treatment with free radical quenchers can improve ER-stress-dependent atrophy (Back et al., 2009; Malhotra et al., 2008), the translation of such therapies into the clinic is challenging (see, for example, Persson et al., 2014). One of the associated problems is illustrated by the dual role that NRF2 has in cancer; by increasing the endogenous antioxidant capacities, NRF2-inducing treatments lower the sensitivity to carcinogens but, at the same time, increase the risk of resistance against anti-cancer chemotherapy (Ma, 2013). More targeted interventions, such as BCL2 mimetics (Rong et al., 2009) or ASK1 inhibitors (Kim et al., 2009), might prove more useful to specifically interrupt pro-apoptotic redox amplification loops. On a different note, an emerging therapeutic concept is to kill solid tumors by redox attack, which is most prominently illustrated by the preclinical success of photodynamic therapies (Garg and Agostinis, 2014), as well as by the use of PDI inhibitors (Higa et al., 2014; Xu et al., 2014). Finally, defective insulin secretion and ER stress that is caused by folding-impaired genetic variants of proinsulin can be rescued by expression of hyperactive Ero1 (Wright et al., 2013), although increasing Ero1 activity in healthy β-cells predisposes them to developing ER stress and diabetes (Awazawa et al., 2014). Thus, targeted ER hyper-oxidation might be a viable therapeutic strategy in a subset of diabetic syndromes.

Apart from these translational challenges, the redox–UPR interplay will no doubt continue to elicit academic interest. Among the important questions to be addressed, we would like to highlight the nature and role of catalyzed thiol–disulfide exchange reactions within and between UPR sensor molecules (as tentatively proposed in Fig. 1). Moreover, it will be interesting to scrutinize the reactions of ROS with biological membranes during oxidative challenge and the potential function of lipid peroxides as second messengers, e.g. at the ER–mitochondrion interface (van Vliet et al., 2014). The final goal of a comprehensive description of the impact of redox on UPR and vice versa is still far ahead.