Endoplasmic reticulum: reduced and oxidized glutathione revisited

The endoplasmic reticulum (ER) constitutes the starting point of the secretory pathway where secretory and membrane proteins are synthesized. Correct folding of these ER client proteins is required for their proper function, which critically depends on different ER-resident determinants such as chaperones, folding catalysts, ATP and Ca²⁺ levels (Braakman and Bulleid, 2011). Imbalances between the burden of protein synthesis in the ER and the capacity of its folding machinery activate an adaptive cellular program called the unfolded protein response (UPR), which is conserved from yeast to man (Hetz, 2012). The signaling pathways of the UPR, which include the phosphorylation of protein kinase RNA-like ER kinase (PERK) and eukaryotic initiation factor 2α (eIF2α) as well as the splicing of X-box binding protein 1 (XBP1) mRNA, mediate the transduction of information on the load of unfolded proteins (so-called ‘ER stress’) from the ER lumen to cytosol and nucleus.

Protein homeostasis in the ER is strongly connected to the formation of native disulfide bonds during client protein folding, which also requires reduction of non-native bonds (Braakman and Bulleid, 2011). Important catalysts of these thiol–disulfide exchange reactions are the members of the protein disulfide isomerase (PDI) family. Upstream of PDIs, endoplasmic oxidoreductin 1 (Ero1) enzymes, among other pathways (Ruddock, 2012), generate disulfide bonds by reducing molecular oxygen (Ramming and Appenzeller-Herzog, 2012). This oxidative mechanism is antagonized by the low molecular weight thiol compound glutathione (GSH), which maintains a reduced fraction of PDIs (Chakravarthi et al., 2006). GSH-mediated reduction results in the formation of its dimeric oxidized form glutathione disulfide (GSSG). The presence of oxidizing and reducing components in the ER allows a dynamic control of the redox state (Appenzeller-Herzog, 2011).

The status of the glutathione redox couple (GSH–GSSG) is an accepted indicator of intracellular redox conditions (Schafer and Buettner, 2001). For instance, compared to the cytosol, the GSH∶GSSG ratio in the ER is significantly lower (Hwang et al., 1992; Appenzeller-Herzog, 2011). In keeping with this, glutathione is mainly transported through the ER membrane in its reduced form, whereas GSSG is retained and accumulated in the ER lumen (Bánhegyi et al., 1999). Unfortunately, there is wide disagreement in the literature as to the actual oxidative capacity, i.e. the electrochemical reduction potential of GSH–GSSG (E_GSH) in the ER (Hwang et al., 1992; Dixon et al., 2008; van Lith et al., 2011; Delic et al., 2012; Kolossov et al., 2012), which is given by the Nernst equation (see below). One reason for this is that it is not possible to deduce E_GSH from the GSH∶GSSG ratio alone without knowledge of the absolute glutathione concentration (Appenzeller-Herzog, 2011).

Green fluorescent protein (GFP) is a jellyfish protein of 26.9 kDa emitting bright green light when excited with light around 480 nm (Shimomura et al., 1962). The chromophore of wild type GFP (wtGFP) is formed by the three amino acids Ser⁶⁵, Tyr⁶⁶ and Gly⁶⁷. It exhibits two excitation peaks at 395 nm and 475 nm, the molecular reason of which is the reversible protonation of Tyr⁶⁶ (Brejc et al., 1997). Critical improvements in expression and brightness of GFP in mammalian cells were achieved by substitution of Ser⁶⁵ with threonine and by systematic codon optimization (Yang et al., 1996). These alterations led to a new GFP variant called enhanced GFP (EGFP).

In order to obtain two variants of redox-sensitive GFP (designated roGFP1 and 2), Ser¹⁴⁷ and Gln²⁰⁴ were replaced by cysteines, which allowed formation of a disulfide bond under oxidizing conditions (Dooley et al., 2004; Hanson et al., 2004). Formation and dissolution of this disulfide influences the protonation of Tyr⁶⁶, thus rendering roGFP a ratiometric sensor, which exhibits redox-dependent changes in emission intensity following excitation at two different wavelengths (Meyer and Dick, 2010). RoGFP-based probes have extensively been used in mitochondria and cytosol (Meyer and Dick, 2010) where they specifically sense E_GSH. This specificity, which has first been described for the related rxYFP sensor (Østergaard et al., 2004), depends on the expression of endogenous glutaredoxins, which catalyze thiol–disulfide exchange between the GSH–GSSG and roGFP redox pairs (Meyer et al., 2007). Based on this principle, cytosolic and mitochondrial E_GSH sensors of high responsiveness and specificity have been created by intramolecular fusion of human glutaredoxin-1 (Grx1) to roGFP2 (Gutscher et al., 2008). Redox measurements using roGFPs targeted to the ER have also been reported. In contrast to roGFP2 (Merksamer et al., 2008), modified, more oxidizing variants based on roGFP1 (Lohman and Remington, 2008) were found not to be quantitatively oxidized at steady state in the ER (Zito et al., 2010; van Lith et al., 2011; Delic et al., 2012). However, since no endogenous glutaredoxins are expressed in the ER of human cells (Izquierdo et al., 2008), the specificity of any roGFP variant in this compartment is unclear.

On these grounds, we expressed ER-targeted and codon-optimized fusion proteins of Grx1 and different roGFP variants with the aim to determine E_GSH in the ER [E_GSH(ER)]. Indeed, the steady-state redox distribution of Grx1-coupled roGFP was significantly more oxidized than that of their uncoupled counterparts, indicating that the latter were not in equilibrium with GSH–GSSG. Deduced values for E_GSH(ER) prompt the reassessment of existing models.

RoGFP sensors have been found to monitor E_GSH in the cytosol, and their specific response depends on catalysis of their reaction with GSH–GSSG by endogenous glutaredoxins (Meyer et al., 2007). However, there is no evidence for expression of any glutaredoxins in the non-fungal ER (Izquierdo et al., 2008), and the switch disulfide in conventional roGFPs is so stable [E°′ <−270 mV (Meyer and Dick, 2010)] that the reduced dithiol form is virtually absent under steady-state conditions in the oxidizing environment of the ER (Merksamer et al., 2008) (and see below). We therefore used roGFP variants carrying amino acid insertions adjacent to Cys¹⁴⁷ (termed roGFP-iX where X denotes the inserted amino acid in single letter code), which display a substantially lowered thermodynamic stability of the switch disulfide, resulting in midpoint potentials that range between −229 and −246 mV (Lohman and Remington, 2008). Furthermore, to enable equilibration of the probes with GSH–GSSG specifically we also expressed roGFP-iX probes as Grx1 fusion proteins in the ER (Fig. 1A). Expression of Grx1 in the ER led to artifactual N-glycosylation of a consensus site at position 52 (supplementary material Fig. S1), which is located in a solvent exposed loop ∼16 Å away from the active site (Sun et al., 1998). However, since comparison with a glycosylation-site mutant revealed that neither the folding of the sensor nor the catalysis of thiol-disulfide exchange between roGFP and GSH–GSSG was inhibited by glycosylation (see below, supplementary material Fig. S4B; Fig. S7A), we proceeded with the glycosylated variants.

Initially, we fused HA-tagged roGFP1-iE, derived from the original jellyfish wtGFP cDNA (Lohman and Remington, 2008), to an N-terminal signal peptide and a C-terminal ER retrieval motif (roGFP1-iE_ER) (Fig. 1A). Although similar constructs have previously been published (Zito et al., 2010; van Lith et al., 2011), this probe was found to be poorly expressed (Fig. 1B) and difficult to detect by fluorescence readouts (supplementary material Fig. S2). Comparable results were obtained with wtGFP-derived Grx1-roGFP-iE_ER (Fig. 1B; supplementary material Fig. S2).

We therefore equipped EGFP-based roGFP2 and Grx1-roGFP2 constructs (Gutscher et al., 2008) with ER-targeting sequences and HA-epitope and subsequently introduced the iE or iL insertion. In addition, to precisely match the roGFP1-iX amino acid sequence (Lohman and Remington, 2008), we replaced Thr⁶⁵ with serine and introduced the ‘folding mutations’ F99S, M153T, V163A and I167T (Fig. 1A). Of note, Ser versus Thr in position 65 defines the principal difference between wtGFP/roGFP1 and EGFP/roGFP2. The roGFP2 constructs, which are composed of high-usage (‘humanized’) codons for expression in mammalian cells, were thus converted into roGFP1-iX and Grx1-roGFP1-iX constructs with proper codon composition. These readily showed a reticular fluorescence pattern that colocalized with PDI, a bona fide ER marker (supplementary material Fig. S3). Consistent with the literature (Haas et al., 1996; Yang et al., 1996), comparison of the humanized roGFP1-iE_ER constructs with the wtGFP-based constructs by western blot (WB) documented enhanced expression of the former (Fig. 1B). As expected for two sensors with the same amino acid sequence, the steady-state degree of probe oxidation in the ER was not influenced by the codon composition of the construct (Fig. 1B).

Formation of the switch disulfide in roGFP, which bridges two neighboring strands in the native GFP β-barrel (Meyer and Dick, 2010), likely reflects the correct conformation of the protein and can be easily monitored by non-reducing SDS-PAGE (Fig. 1B). Hence, we monitored the folding rate of different roGFP_ER variants by pulse-chase analysis. While roGFP2_ER and Grx1-roGFP2_ER (Fig. 1A) were readily folded, acquisition of the correct conformation was delayed for all roGFP1-iX_ER constructs (supplementary material Fig. S4A). This delay was most prominent for roGFP1-iL_ER.

Evidence for hampered folding of iX variants in the ER was also obtained, when analyzing the sensors by WB. Here, the proportion of reduced sensor was significantly higher in cell lysates than in corresponding immunoprecipitates (supplementary material Fig. S4C). This higher proportion of the oxidized form corresponded better to the results obtained by fluorescence measurements (see below). This suggested that the reduced fraction in cell lysates contained misfolded (i.e. non-oxidizable) roGFP-iX, which was removed by the centrifugation step of the immunoprecipitation (IP) procedure. Also for roGFP2_ER and Grx1-roGFP2_ER, a minor reduced fraction was visible in whole-cell lysates (supplementary material Fig. S4D).

Does the low folding efficiency of roGFP1-iX_ER variants affect their performance as real-time redox sensors, i.e. their responsiveness to changes in redox? In a previous study (van Lith et al., 2011) on cells expressing roGFP1-iL in the ER, recovery of the reduced fraction of the sensor occurred only very slowly upon washout of the oxidant 2,2′-dithiodipyridine (DPS). To recapitulate this, we subjected cells transfected with roGFP1-iX_ER sensors to a similar washout procedure using either diamide or DPS as the oxidant and analyzed the redox state of the sensors at different time points after washout by IP and WB. As shown in Fig. 2A and supplementary material Fig. S5, both Grx1-fused and unfused roGFP1-iX_ER readily recovered from diamide/DPS-induced hyperoxidation at a rate comparable to the recovery rates of endogenous PDIs (Appenzeller-Herzog et al., 2010). Likewise, oxidative recovery of roGFP1-iX_ER sensors following washout of the reductant dithiothreitol (DTT) occurred promptly (van Lith et al., 2011) (see below). To probe for their ability to report modest redox changes, we also incubated (Grx1-)roGFP1-iE_ER-expressing cells in a range of concentrations of DTT or diamide and determined their redox state by IP and WB. Consistent with what has been observed with roGFP-iL (van Lith et al., 2011), the sensors readily responded to submillimolar concentrations of reductant/oxidant (Fig. 2B). Taken together, although initial folding of roGFP1-iX_ER variants appears to be relatively inefficient, they are, once properly folded, capable of reporting redox changes in the ER.

We next quantified the steady-state redox state of the four ER-targeted roGFP1-iX sensors as well as of Grx1-roGFP2_ER in transiently transfected HeLa cells using three independent techniques: in situ fluorescence spectrum analysis, IP/WB, and ratiometric live-cell fluorescence microscopy. For fluorescence spectrum analysis in a plate reader, transfected cells were trypsinized and seeded into 96 well plates. To obtain oxidized or reduced control spectra, cells in some wells were treated with DPS or DTT. As expected, Grx1-roGFP2_ER was fully oxidized in the ER (99±0.5%) (Fig. 3A,B). Conversely, roGFP1-iX_ER sensors produced excitation spectra, in which the curve recorded from untreated cells clearly ran intermediate to the curves recorded from oxidized and reduced cells (Fig. 3A). In general, the fluorescence signals from iL sensors were weaker compared to the ones from iE sensors. This was particularly evident for roGFP1-iL_ER, which yielded a low-quality spectrum (Fig. 3A) with limited reproducibility (Fig. 3B). Interestingly, at steady-state Grx1-roGFP1-iE_ER was more oxidized (91±3%) than roGFP1-iE_ER (82±7%).

The redox states of all sensors were also examined by IP and WB. Again, all roGFP1-iX_ER sensors displayed a characteristic distribution of reduced and oxidized species, whereas Grx1-roGFP2_ER was fully oxidized (Fig. 4A). As in the excitation spectrum analyses (Fig. 3), roGFP1-iE_ER was less oxidized (54±1%) than Grx1-roGFP1-iE_ER (70±5%) (Fig. 4B). Probably owing to incomplete removal of the misfolded subfraction (see supplementary material Fig. S4), the WB readout yielded slightly more reduced redox distributions than fluorescence scanning. Compared to iE variants, iL variants consistently formed more high-molecular weight complexes of unknown identity. Accordingly, the reduced and oxidized monomeric forms of iL variants were less prominently detected compared to iE (Fig. 4A), which correlated with their weaker fluorescence (Fig. 3). Altogether, the data so far indicated roGFP1-iE_ER and Grx1-roGFP1-iE_ER to be the most reliable ER redox sensors.

To complement the above results with a data set, which also reports cell-to-cell differences, we next turned to live imaging of single cells. Due to poor fluorescence of roGFP1-iL_ER variants, these analyses were focused on the variants carrying the iE insertion. Transfected cells were subjected to laser scanning microscopy using two different laser lines for excitation. To obtain the excitation ratio corresponding to unperturbed, fully oxidized, and fully reduced sensor, the cells were first imaged at steady state and then treated with 0.5 mM diamide followed by addition of 20 mM DTT. Calculation of the steady-state redox distributions yielded numbers that were comparable to the results obtained by fluorescence spectrum analysis. As shown in Fig. 5, Grx1-roGFP2_ER was completely oxidized. By contrast, the oxidized fractions of roGFP1-iE_ER and Grx1-roGFP1-iE_ER were 82±13% and 93±2%, respectively (Fig. 5B). For unknown reasons, variations in the excitation ratio between individual cells were only evident at steady state and under oxidizing conditions, but disappeared upon reduction of the sample (Fig. 5A).

Taken together, the results in Figs 3–f04,5 confirmed the previously described effects caused by amino acid insertions on the stability of the disulfide bridge in roGFP1 (Lohman and Remington, 2008). Moreover, N-terminal fusion of Grx1 led to increased oxidation of roGFP1-iE in the ER at steady state.

Since Grx is normally absent from the ER, the finding that fusion of Grx1 to roGFP1-iE_ER increases its steady-state oxidation could potentially be explained by Grx1-mediated changes in overall ER redox conditions, which would be expected to elicit the UPR (Hansen et al., 2012). We therefore set out to test these possibilities.

First, the ability of the newly developed sensors to report oxidative changes in the ER was examined. For this purpose, we generated a cell line overexpressing a hyperactive form of Ero1β (Ero1β-C100/130A) (Wang et al., 2011) under a doxycyline-inducible promoter. Compared to wild-type Ero1β, induction of hyperactive Ero1β for 24 h caused increased oxidation of the ER-resident protein ERp57 (Fig. 6A). ER hyperoxidation by Ero1β-C100/130A was also evident when analyzing the redox states of roGFP1-iE_ER or Grx1-roGFP1-iL_ER, the sensors with the least oxidized steady-state distribution in the ER (Fig. 2), by IP/WB (Fig. 6B) (Grx1-roGFP1-iE_ER was omitted because of its already high steady-state oxidation and roGFP1-iL_ER because of its poor folding; see above). Unfortunately, fluorescent readouts in this setup were not possible owing to the fluorescence of doxycycline.

We next checked whether expression of Grx1 in the ER showed effects comparable to those of Ero1β-C100/130A, which could explain the observed differences between roGFP1-iE_ER and Grx1-roGFP1-iE_ER. For this purpose, we transfected cells with the different roGFP_ER sensors, with ER-localized Grx1 (Grx1_ER; supplementary material Fig. S6), or with Ero1β-C100/130A and determined the redox state of endogenous ERp57. In contrast to Ero1β-C100/130A, neither roGFPs nor Grx1_ER affected the redox state of ERp57 (Fig. 6C). Likewise, Grx1_ER did not influence the redox state of HyPer_ER (Fig. 6D), a ratiometric probe, which reacts with H₂O₂ (Belousov et al., 2006; Enyedi et al., 2010) and, probably, with PDIs (Ruddock, 2012). To monitor possible signs of Grx1-dependent ER stress, we analyzed markers of the UPR and found that neither Grx1-fused roGFP_ER sensors nor Grx1_ER by itself elicited UPR signaling (Fig. 6E). We concluded that expression of Grx1 in the ER does not broadly affect ER redox homeostasis. The formal possibility that Grx1 catalyzes deglutathionylation of ER proteins, which would increase GSSG levels in the ER, can at present not be excluded. However, we consider this possibility unlikely, because (i) the abundance of glutathionylated ER proteins is low (Hansen et al., 2009) and (ii) unnaturally increased GSSG levels would elicit the UPR (Hansen et al., 2012).

We next examined whether the increased oxidation of Grx1-roGFP1-iE in the ER could be the result of altered redox properties of this sensor compared to roGFP1-iE. Thus, the midpoint reduction potentials E°′ of Grx1-fused and unfused sensors were determined using purified proteins. Consistent with published data (Lohman and Remington, 2008), E°′_roGFP1-iE was −233±5 mV (Fig. 6F). However, N-terminal fusion of Grx1 resulted in a slightly more reducing E°′ of −239±3 mV (Fig. 6G), which may partially explain the observed differences in ER sensor oxidation.

Reaction of Grx1-roGFP1-iE with glutathione in vitro was markedly accelerated compared to roGFP1-iE (Fig. 7A), thus confirming previous findings with the N-terminal fusion of Grx1 to roGFP2 (Gutscher et al., 2008). To examine whether such catalysis also occurs in the ER, we exploited the fact that ER GSSG levels readily rise after DTT treatment and rapid washout (Appenzeller-Herzog et al., 2010). Thus, DTT recovery assays using cells transfected with roGFP_ER variants with or without fused Grx1 were performed and analyzed by IP/WB. Indeed, recovery of roGFP2_ER was accelerated by the addition of Grx1 (Fig. 7B). The overall rapidity of roGFP2_ER re-oxidation is presumably determined by the strong sensitivity of this sensor to trace amounts of GSSG (Gutscher et al., 2008). Likewise, the re-oxidation of Grx1-roGFP1-iE_ER, which occurred in the time scale of minutes, was faster than that of roGFP1-iE_ER (Fig. 7C). Grx1-mediated acceleration was not influenced by mutation of its glycosylation site, but depended on active-site cysteines (supplementary material Fig. S7). Mutation of the active site in Grx1 also significantly lowered the steady-state oxidation of Grx1-roGFP1-iE_ER (Fig. 7D,E). These results demonstrated that the catalytic activity of Grx1 specifically enhanced the equilibration of Grx1-roGFP1-iE_ER with GSH–GSSG.

Based on these findings and assuming that Grx1-roGFP1-iE_ER is in (quasi-)redox equilibrium with GSH–GSSG, we then calculated E_GSH(ER) from E°′_{Grx1-roGFP1-iE} = −239 mV and the steady-state redox state of Grx1-roGFP1-iE_ER by using the Nernst equation. As we figured the fluorescent fraction of the sensor to be correctly folded, we considered the numbers obtained by fluorescence spectrum analysis (Fig. 3) and laser scanning microscopy (Fig. 5) the most reliable, which yielded values for E_GSH(ER) at pH 7.0 of −209±4 mV and −205±3 mV, respectively.

Having established two in situ sensor molecules, one of which specifically reports E_GSH(ER) (Grx1-roGFP1-iE_ER) and the other the redox state of (an) as yet ill-defined ER redox pair(s) (roGFP1-iE_ER; see Discussion), we set out to examine the relationship between induction of ER stress and redox state in real-time experiments. Since DTT elicits ER stress via manipulation of ER redox and the membrane trafficking inhibitor brefeldin A was found to impede accurate time-lapse analyses, we concentrated our investigations on the well-established ER stress inducers thapsigargin (TG, an ER Ca²⁺ pump inhibitor) and tunicamycin (TM, an inhibitor of glycosylation).

Recent data by Enyedi et al. (Enyedi et al., 2010) using TG and the HyPer_ER probe demonstrated that lowered ER Ca²⁺ content promptly diminishes oxidation of the HyPer probe in the ER. Since the underlying mechanism is obscure and TG-mediated hypo-oxidation may be restricted to the H₂O–H₂O₂ redox couple (as potentially measured by HyPer_ER), we were interested whether inhibition of the Ca²⁺ pump would also have an influence on GSH–GSSG. As published (Enyedi et al., 2010), real-time monitoring indicated a prominent reduction of the HyPer_ER redox state upon addition of TG, which occurred within 5 minutes (Fig. 8A). RoGFP1-iE_ER as well as Grx1-roGFP1-iE_ER showed a very similar behavior, suggesting that TG-induced Ca²⁺ depletion from the ER broadly influences thiol–disulfide equilibria in the ER. As expected for the highly reducing roGFP2 sensor (Meyer and Dick, 2010), addition of TG to cells expressing Grx1-roGFP2_ER did not elicit any change in fluorescence.

Since treatment of yeast cells with TM causes pronounced under-oxidation of ER-targeted roGFP2 after a lag time of 1 h (Merksamer et al., 2008), we tested whether this was also the case in mammalian cells using roGFP1-iE_ER. However, no reduction of the sensor was observed in HeLa cells even 6 h after the application of TM (Fig. 8B). Conversely, a subset of cells manifested a slight hyperoxidation of roGFP1-iE_ER, which started roughly 100 min after TM addition. We concluded that the relationship between ER stress and ER redox is likely indirect and depends on the context.

Using a glutathione-specific ER-targeted redox sensor, our work provides quantitative data to resolve the long-standing question of how oxidizing the ER is. Based on data from two independent fluorescence readouts, we determined E_GSH(ER) in HeLa cells to be −208±4 mV (at pH 7.0), which, as expected, is more oxidizing than E_GSH(cytosol) of ∼−300 mV (Gutscher et al., 2008). Given the necessity for both oxidative and reductive processes in the ER (Braakman and Bulleid, 2011) and the central position of this organelle in many physiological and pathological processes (Rutkowski and Hegde, 2010; Hetz, 2012), this knowledge is of broad relevance.

Initial evidence for a relatively oxidizing E_GSH in the ER/secretory pathway was obtained twenty years ago in hybridoma cells (Hwang et al., 1992). Using a membrane-permeable tetrapeptide, which was glycosylated and trapped in the secretory pathway and which reacted in situ with GSH–GSSG, the authors elegantly demonstrated a more oxidized GSH∶GSSG ratio in secretory compartments compared to the overall ratio in the whole cell. By estimating cell volume and glutathione concentration, the deduced GSH∶GSSG ratios between 1∶1 and 3∶1 led to an estimate of E_GSH between −172 and −188 mV (Hwang et al., 1992). The accuracy of these numbers, however, depends on the validity of several bold assumptions, in particular (i) that the tetrapeptide probe specifically and fully equilibrated with GSH–GSSG and (ii) that the glutathione concentration in the secretory pathway equaled the overall cellular concentration. Many years later, GSH was derivatized in rat liver microsomes, and the GSH∶GSSG ratio determined with comparable results to the Hwang study (Bass et al., 2004). However, as microsomes are permeable to GSH (Bánhegyi et al., 1999), the significance of these results is unclear. Indeed, it was later shown that perfusion of livers with the sulfhydryl-modifying agent iodoacetic acid (IAA) prior to preparation of microsomes resulted in intra-microsomal GSH∶GSSG ratios of up to 6∶1 (Dixon et al., 2008). Still, whether in this setup, leakage of GSH was fully prevented by IAA was not shown, and a further confounding factor is the poor membrane permeability of IAA (Hansen and Winther, 2009).

More recent estimates of E_GSH(ER) were obtained using GFP-based sensors in the ER. By monitoring roGFP1-iE in the ER of the yeast Pichia pastoris, E_GSH(ER) was suggested to be −242 mV (Delic et al., 2012). A similar value (−231 mV) was obtained in human HT1080 cells stably transfected with ER-targeted roGFP1-iL (van Lith et al., 2011). Conversely, the most oxidizing value in the spectrum of suggested reduction potentials of the ER (−118 mV) was obtained using an engineered, redox-sensitive Förster resonance energy transfer sensor protein (Kolossov et al., 2012). These inconsistent outcomes apparently reinforce the concept that multiple ER redox couples, which are kinetically separated from each other, exist in parallel (Appenzeller-Herzog, 2012). Considering the high concentration of glutathione ([GSx]_ER, see below) though, E_GSH is probably the most accurate measure for the degree of ER oxidation, as has also been suggested by others (Schafer and Buettner, 2001).

In this study, we report that appending Grx1 to roGFP1-iE_ER conferred specificity towards GSH–GSSG (Fig. 7), which was exploited to measure E_GSH as −208 mV. {In terms of absolute reduction potential, it is important to consider that pH(ER) is probably slightly above 7.0. Thus, since pKa_{Grx1-roGFP-iE} is > 8 (our unpublished data), E°′_{Grx1-roGFP1-iE} (and therefore also E_GSH) decreases by 6 mV/0.1 pH unit (Schafer and Buettner, 2001).} Although roGFP1-iE_ER by itself may also to some extent interact with GSH–GSSG, its redox state was highly variable in our hands, as indicated by large standard deviations (Figs 3, 5). Moreover, its theoretical redox state assuming quasi-equilibration with GSH–GSSG (as calculated with E_GSH and E°′_roGFP1-iE) is 87% oxidized, which clearly diverges from the measured average value of 82% (Figs 3, 5). We therefore believe that roGFP1-iE_ER does not preferentially react with the glutathione redox couple but also with other redox partners that influence its steady-state redox distribution. It is possible that this undefined reactivity is the reason for the increased trapping of mixed-disulfide partners with roGFP1-iX_ER compared to Grx1-roGFP1-iX_ER (Fig. 4). Likewise, as per Ero1β-C100/130A, we noticed detectable levels of spliced XBP1 mRNA upon expression of unfused roGFPs (Fig. 6E).

Given E_GSH(ER) of −208±4 mV and E°′_GSH of −240 mV (Rost and Rapoport, 1964), the Nernst equation returns a value for [GSH]²∶[GSSG]. Accordingly, tentative calculation with a molar GSH∶GSSG ratio of 6∶1 (Dixon et al., 2008) yields a [GSx]_ER of 19±6 mM. Notably, this concentration would be significantly higher than the concentration measured in rat liver microsomes [4.5 mM (Dixon et al., 2008)] and in whole HeLa cells [∼7 mM (van der Schans et al., 1986)]. In the light of the ability of GSH to diffuse through microsomal membranes (Bánhegyi et al., 1999), the former discrepancy may not be surprising. As for a presumably higher [GSx] in the ER than in the rest of the cell, there are two non-mutually exclusive explanations: (i) Cellular glutathione is trapped in the ER, and/or (ii) the GSH∶GSSG ratio in the ER of live cells is greater than 6∶1. Specifically measuring this ratio in situ is an important future challenge.

PDIs rapidly react with GSH–GSSG in vitro (Ruddock, 2012). It was therefore unexpected that the obtained E_GSH(ER) was apparently not oxidizing enough to maintain the known steady-state redox states of PDIs. For instance, the a′ domain in ERp57 is typically 15–40% oxidized in cells (Appenzeller-Herzog et al., 2008)(see also Fig. 6C). However, calculation of the oxidized fraction of ERp57a′ using its E°′ [−156 mV (Frickel et al., 2004)] and E_GSH(ER) (i.e. assuming quasi-equilibrium with GSH–GSSG) yields a value of 2%. This indicates that oxidation of ERp57a′ is specifically catalyzed and that, at variance with previous models (Appenzeller-Herzog, 2011), GSSG-driven oxidation of reduced PDIs is probably a minor pathway (at least in HeLa cells). Whether the oxidized fraction of ERp57a′ [and of other family members such as TMX3 (Haugstetter et al., 2005)] is maintained by Ero1 (Ramming and Appenzeller-Herzog, 2012) or by other oxidative pathways (Ruddock, 2012) is presently unclear. Another interesting implication following the determination of E_GSH(ER) relates to the question, how a catalytically active (i.e. reduced) fraction of reducing ER oxidoreductases such as Sep15 [E°′ = −225 mV (Ferguson et al., 2006)] can be maintained. Here, equilibration with GSH–GSSG would be expected to result in a reduced fraction of 21%.

Our results also demonstrated differences between roGFP1-iE_ER and roGFP1-iL_ER, with the latter being less stable, less efficiently folded, less fluorescent and more prone to form mixed-disulfide complexes (Figs 3, 4; supplementary material Fig. S4A). Together with the implementation of codon-optimized cDNAs, this likely means that the expression of roGFP1-iE_ER sensors used herein is preferable to that of the previously used roGFP1-iL-KDEL (van Lith et al., 2011). It is conceivable that weak expression of the latter sensor and the concomitant necessity to use strong illumination produced the reported photo-oxidation phenomenon (van Lith et al., 2011), which we never observed in our setup.

ER re-oxidation following the washout of DTT is a rapid process, which takes less than five minutes and depends on Ero1α (Appenzeller-Herzog et al., 2010; van Lith et al., 2011; Rutkevich and Williams, 2012). Although PDI is the principal substrate of Ero1α (Inaba et al., 2010; Ramming and Appenzeller-Herzog, 2012), evidence exists that, unlike at steady state (see above), other family members such as ERp57 are also oxidized via Ero1α during early time points of recovery after DTT (when Ero1α is maximally active) (Appenzeller-Herzog et al., 2010). In agreement with this, ERp57 reacquires its steady-state redox distribution within seconds in an Ero1α-dependent manner (Appenzeller-Herzog et al., 2010). By contrast, Grx1-roGFP1-iE_ER was re-oxidized within around two minutes (Fig. 7B; supplementary material Fig. S7), which likely reflected its inability to accept disulfide bonds directly from Ero1α. As expected for a glutathione-specific probe, the kinetics of sensor re-oxidation rather corresponded to the re-oxidation kinetics of GSH–GSSG (Appenzeller-Herzog et al., 2010). In case of roGFP1-iE_ER or Grx1SS-roGFP1-iE_ER, oxidative recovery was slower compared to Grx1-roGFP1-iE_ER. Indeed, as recently modeled (Ruddock, 2012), ‘generic’ protein disulfides such as these are expected to form at a slower rate upon DTT washout than GSSG.

It has been suggested that UPR signaling following ER stress favors a more reduced ER to promote a more ‘cautious’ state where release of misfolded proteins is prevented at the expense of an overall diminished secretion rate (Delic et al., 2012). This model was based on yeast strains, which expressed ER-targeted roGFP and were modified to signal an ER stress response (Merksamer et al., 2008; Delic et al., 2012). In keeping with published HyPer_ER experiments in HeLa cells (Enyedi et al., 2010), we also observed a swift drop in probe oxidation following inhibition of the ER Ca²⁺ pump (Fig. 8A), which is known to trigger the UPR. While evidence has been presented that ER redox conditions impact on Ca²⁺ handling at the ER membrane (Li and Camacho, 2004; Higo et al., 2005; Anelli et al., 2012), the mechanism underlying this opposite finding is not yet known. Whether or not it depends on the UPR, remains to be determined. However, in agreement with recent findings in pancreatic β-cells (Schuiki et al., 2012), we did not observe ER hypo-oxidation following treatment of cells with another ER stress inducer, TM (Fig. 8B). By contrast, roGFP2 expressed in the ER of yeast cells was massively reduced in response to TM (Merksamer et al., 2008), which might potentially be explained by ER-stress-induced relocation of this sensor to the cytosol (Rubio et al., 2011).

Several human diseases are caused (or at least accompanied) by manifestations of ER stress and oxidative injury (Kim et al., 2008). In some cases, PDI has been identified as a cytoprotective factor, indicating that modulation of the ER thiol–disulfide system including E_GSH(ER) confers therapeutic advantage (Tanaka et al., 2000; Severino et al., 2007; Walker et al., 2010; Wang et al., 2012). By developing a glutathione-responsive ER sensor, which allows an authentic estimate of E_GSH(ER) in living cells, this study now offers an important tool to study these processes in vivo. It will be interesting to test, whether there is a direct relationship between changes in E_GSH(ER) and the survival rate of cells experiencing pathological ER stress.

Cloning of Grx1-roGFP2_ER was initiated by amplifying the coding sequence from pLPCX/Grx1-roGFP2 (Gutscher et al., 2008) and extending it to encode KDEL* at the C-terminus using 5′-GAA GGT ACC ATG GCT CAA GAG TTT GTG AAC TGC-3′ and 5′-GGT GAA GCT TTT ACA GTT CGT CCT TGT ACA GCT CGT CCA TGC-3′ (Restriction sites underlined). The PCR product was ligated via Asp718/HindIII into a plasmid harboring the ER signal sequence of ERp44 and an HA epitope upstream of an Asp718 site (Otsu et al., 2006). To obtain Grx1-roGFP1-iX_ER constructs, Grx1-roGFP2_ER was mutated by the QuikChange protocol (Stratagene) as follows (only sense strand primers, mutations underlined, amino acid numbers derived from roGFP): iE insertion [C147CE+H148S; (Lohman and Remington, 2008)]: 5′-GGA GTA CAA CTA CAA CTG CGA GAG CAA CGT CTA TAT CAT GG-3′; iL insertion (C147CL+H148S): 5′-GGA GTA CAA CTA CAA CTG CCT GAG CAA CGT CTA TAT CAT GG-3′; T65S: 5′-CGT GAC CAC CCT GAG CTA CGG CGT GC-3′; F99S: 5′-GGA GCG CAC CAT CAG CTT CAA GGA CGA CGG C-3′; M153T: 5′-GCA ACG TCT ATA TCA CCG CCG ACA AGC AGA AGA ACG-3′, V163A+I167T: 5′-GAA GAA CGG CAT CAA GGC TAA CTT CAA GAC CCG C-3′.To obtain roGFP1-iX_ER, the roGFP moiety was amplified from Grx1-roGFP1-iX_ER using 5′-AGA GGT ACC GTG AGC AAG GGC GAG GAG-3′ and 5′-GGT GAA GCT TTT ACA GTT CGT CCT TGT ACA GCT CGT CCA TGC-3′ and cloned via Asp718/HindIII as described above.

WtGFP(jellyfish)-derived roGFP1-iE_ER constructs were generated by amplifying the coding sequence from pQ30/roGFP1-iE [obtained by QuikChange (5′-TTG GAA TAC AAC TAT AAC TGC GAG TCC AAT GTA TAC ATC ACG GCA-3′) using pQ30/roGFP1-iL (Lohman and Remington, 2008) as template] using 5′-GAA GGT ACC AGT AAA GGA GAA GAA CTT TTC ACT GG-3′ and 5′-AGA CTT AAG TTA CAG TTC GTC TTT GTA TAG TTC ATC CAT GCC ATG T-3′. The resulting fragment (featuring a KDEL* motif) was then ligated into Asp718/AflII-digested pcDNA3.1-ERp57-HA (see above). N-terminal fusion with Grx1 was achieved by excising roGFP1-iE_ER with EcoRI/AflII and exchanging it with the roGFP1-iL sequence in Grx1-roGFP1-iL_ER. For the cloning of pQE-30/Grx1-roGFP1-iE, BamHI sites at the 5′ end of Grx1 and the 3′ end of the glycine–serine linker were introduced into Grx1-roGFP1-iE_ER by QuikChange (sense strand primers 5′-CCT GAC TAC GCA GAA GGA TCC ATG GCT CAA GAG TTT G-3′ and 5′-GAT CAG GAG GAG GAT CCG TGA GCA AGG GCG-3′), and the resulting cDNA digested with BamHI to generate Grx1-GGSGG₆, which was ligated into BamHI-digested pQE-30/roGFP1-iE.

Grx1_ER was amplified from Grx1-roGFP2_ER using 5′-GAA GGT ACC ATG GCT CAA GAG TTT GTG AAC TGC-3′ and 5′-GCA AGC TTT TAC AGT TCG TCC TTC TGC AGA GCT CCA ATC TGC TTT A-3′ and ligated into Asp718/HindIII-digested pcDNA3.1-ERp57-HA (see above). The N52Q and C23S+C26S mutants of Grx1-roGFP1-iE_ER were generated by QuikChange (sense strand primers: 5′-CGA TAT CAC AGC CAC CCA GCA CAC TAA CGA GAT TCA AG-3′ and 5′-CAT CAA GCC CAC CTC TCC GTA CTC CAG GAG GGC CCA AG-3′). PcDNA5/FRT/TO/Ero1β-C100/130A was obtained by introducing C100A and C130A (sense strand primers: 5′-GAC TGT CAT GTG GAG CCC GCA CCA GAG AGT AAA ATT CCG-3′ and 5′-GGC AAA CAA TAC CAA AGA ATT AGA AGA TGC TGA GCA AGC TAA TAA ACT GGG-3′) into pcDNA5/Ero1βmyc6his (Appenzeller-Herzog et al., 2008) by QuikChange.

HeLa cells were cultivated in Dulbecco's Modified Eagle's Medium (DMEM) containing 4.5 g/l glucose supplemented with 10% fetal bovine serum, 100 U/ml penicillin, 100 µg/ml streptomycin at 37°C in 5% CO₂. HeLa cells were transfected with Turbofect (Fermentas) or, for microscopy and fluorescence scanning, with Fugene HD (Promega) according to the manufacturers' protocols. Flp-in TRex 293 cells stably transfected with empty vector (Mock) or wild-type Ero1β have been described previously (Appenzeller-Herzog et al., 2008). A corresponding cell line inducibly expressing Ero1β-C100/130A was generated as suggested by Invitrogen. Flp-in TRex 293-derived cell lines were transfected with Metafectene PRO (Biontex).

Transfected cells were starved for 15 min in DMEM without methionine and cysteine (Sigma). Subsequently the cells were labeled with 100 µCi/ml [³⁵S]protein labeling mix (PerkinElmer) and chased in normal growth medium containing 10 mM L-methionine for indicated times. Cells were washed once with PBS containing 20 mM N-ethylmaleimide (NEM), incubated in the same buffer for 20 min on ice, and then subjected to denaturing anti-HA immunoprecipitation as described elsewhere (Appenzeller-Herzog et al., 2005). Bound proteins were analyzed by non-reducing SDS-PAGE and autoradiography.

To obtain completely reduced or oxidized control samples, cells were pretreated with either 10 mM DTT or 5 mM diamide in normal growth medium for 5 min at 37°C. Cells were then washed with phosphate buffered saline (PBS) containing 20 mM NEM and incubated in the same buffer for 20 min on ice. The cells were lysed on ice for 1 h in 100 mM NaPO₄, pH 8, containing 1% Triton-X-100 and 200 µM phenylmethylsulfonylfluoride. Insoluble cell debris were pelleted for 1 h at 4°C and 16,000 g, and the supernatant was immunoprecipitated using NHS-activated agarose (Thermo Scientific) decorated with anti-GFP (Kojer et al., 2012) (generously provided by Jan Riemer, University of Kaiserslautern, Germany). Primary antibodies used for WB were anti-HA (a kind gift from Hans-Peter Hauri University of Basel, Switzerland), anti-ERp57 (a kind gift from Ari Helenius, ETH Zürich, Switzerland), anti-PERK, anti-eIF2α and anti-P-eIF2α (Cell Signaling Technology). DTT and diamide washout experiments were performed as described previously (Appenzeller-Herzog et al., 2010). Where indicated, diamide was replaced by 1 mM DPS. AMS modification of originally disulfide-bound cysteines is delineated elsewhere (Appenzeller-Herzog and Ellgaard, 2008).

For roGFP measurements, HeLa cells were trypsinized 48 h post transfection and washed twice in HEPES buffer (20 mM HEPES pH 7.4, 130 mM NaCl, 5 mM KCl, 1 mM CaCl₂, 1 mM MgCl₂, 5.5 mM D-glucose). A total of 3×10⁵ cells per well were seeded in HEPES buffer onto a flat-bottom 96-well plate. Plates were briefly centrifuged at 180×g and the cells were left untreated or treated with 0.5 mM DPS or 10 mM DTT (three wells per condition). Fluorescence intensities at 530 nm emission were measured from the bottom on a Spectramax Gemini EM (Molecular Device) within a 350–500 nm excitation range. HyPer_ER (Enyedi et al., 2010) (kindly provided by Miklos Geiszt, Semmelweis University, Hungary) measurements were performed within a 410–510 nm excitation range at 535 nm emission, using 100 µM H₂O₂ instead of DPS, and omitting D-glucose from the buffer.

Cells were grown on glass bottom dishes (Ibidi). For determination of the extent of sensor oxidation, roGFP-expressing cells were washed twice with HEPES buffer and the dishes transferred to an Olympus Fluoview 1000 laser scanning confocal microscope equipped with a 60×oil immersion objective (NA 1.40) and a 405 nm laser diode, a 440 nm laser diode, and a 488 nm argon laser. Pictures (1024×1024 pixels) were taken with an open pinhole within an emission window of 505–605 nm at steady state, 1 min after addition of 5 mM diamide, and 5 min after addition of 20 mM DTT. One region of interest (ROI) per cell was chosen, which remained immobile for the duration of image acquisition, and 405/440 or 405/488 nm ratios were determined from average intensities in background-subtracted ROIs. False color images were generated using the Fluoview 1000 software. For time-lapse imaging, cells were washed twice with DMEM without phenol red and subjected to ratiometric laser scanning microscopy using a stage incubator (37°C), CO₂ control, a 20/80 excitation and a 50/50 emission beam splitter. Fluorographs were analyzed with the ImageJ software.

Total RNA was extracted from HeLa cells using TRI Reagent (Sigma), and reverse transcribed with Superscript III (Invitrogen). The resulting cDNA was subjected to PCR using 5′-AAA CAG AGT AGC AGC TCA GAC TGC-3′ and 5′-TCC TTC TGG GTA GAC CTC TGG GAG-3′ to generate XBP1 amplicons, which were separated on a 2% agarose gel.

Transiently transfected HeLa cells were grown for 48 h on glass coverslips, fixed with 4% paraformaldehyde for 20 min at room temperature, quenched with 50 mM NH₄Cl and permeabilized with 0.1% Triton X-100. The cells were blocked with 1% BSA in PBS and incubated in the same buffer with anti-PDI (RL90, Abcam; 1∶100) or anti-HA (1∶1000) for 1 h at room temperature. The coverslips were washed with PBS and incubated with a goat-anti-mouse Hilyte 555 conjugated antibody (AnaSpec). After several washes with PBS, the cells were mounted in Mowiol 4–88 (Hoechst). Staining was analyzed on an Olympus Fluoview 1000 laser scanning confocal microscope (pinhole: 80 µm).

N-terminally hexahistidine-tagged Grx1-roGFP1-iE and roGFP1-iE were purified as reported elsewhere (Meyer et al., 2007). To get their standard (midpoint) potentials, the sensors (1 µM final concentration) were equilibrated for 2–3 h with defined mixtures of oxidized/reduced lipoic acid (2.5 mM) in degassed 100 mM HEPES pH 7.0, 300 mM NaCl, 1 mM EDTA. From the fluorescence intensities at 520 nm for excitation at 390 and 480 nm, respectively, as determined on a fluorescence plate reader (POLARstar Omega; BMG, http://www.bmglabtech.com) with filter-based excitation, OxD_roGFP values were derived as described above. 1-OxD (RedD) was plotted against the potentials of the respective lipoic acid redox buffers. Single data points were fitted to a sigmoidal dose-response curve using GraphPadPrism5 (GraphPad Software, San Diego California USA) to calculate E°′ of both sensors using E°′_{lipoic acid} of −290 mV (Lees and Whitesides, 1993).

Ratiometric time-course measurements with isolated protein were performed as described previously (Meyer et al., 2007). Briefly, glutathione solutions were automatically injected to a final concentration of 2 mM into phosphate-buffered solutions of roGFP (1 µM), 0.1 µM recombinant Glutathione Reductase (Sigma) and 100 µM NADPH. The sensors were pre-oxidized for 30 min with 10 mM H₂O₂, which was then removed by Zeba™ Spin Desalting Columns. H₂O₂ and DTT to a final concentration of 10 mM were separately used to define maximum oxidation and maximum reduction of the sensors.

Data sets were analyzed for statistical significance using Students t-test. To evaluate the relative changes in the roGFP redox ratios upon induction of Ero1β-C100/130A, the data from different experimental batches were logarithmically transformed and fitted to a linear model using a batch specific offset value. Linear regression, 95% confidence intervals, and P values were calculated using Microsoft Excel.

We thank Lars Ellgaard for help with Ero1 experiments, Anne Spang for critical reading of the manuscript, and Jim Remington, Miklos Geiszt, Jan Riemer, Ari Helenius, and Hans-Peter Hauri for reagents.