Reactive oxygen species are essential for autophagy and specifically regulate the activity of Atg4

ROS serve as signaling molecules in a variety of cellular processes, including growth, differentiation, adhesion and PCD. To test whether ROS play a role in autophagy, we measured ROS production under starvation using a fluorescent probe, dihydroethidum (DHE). DHE reacts with peroxides to form a compound that upon binding to DNA fluoresces and can be visualized by confocal fluorescent microscopy (Vanden Hoek et al, 1997). Indeed, upon nutrient starvation, DHE staining was evident, indicating that cells accumulate peroxides under these conditions (Figure 1A). We then used another probe, 2′,7′‐dichlorofluorescin diacetate (DCF‐DA), which reacts mainly with H₂O₂ to form a fluorescent compound at the point of interaction (Cathcart et al, 1983; Vanden Hoek et al, 1997). DCF staining was already evident 15 min after the induction of starvation (Figure 1Bd and Supplementary Figure 1A, see text below). The pattern of DCF staining was punctated as well as cytosolic, and it was also apparent following overnight starvation (Figure 1Bh). These results indicate that H₂O₂ is most likely the oxidative agent accumulating in starved cells. To rule out the possibility that the oxidative signal associated with starvation is leading to PCD, we tested the viability of cells grown under nutrient starvation conditions. As shown in Supplementary Figure 1B, 5 h of nutrient starvation did not affect the viability of the cells, and the percent of cell death was equally low in control and starved cells. Taken together with the finding that the oxidative signal appears minutes after induction of starvation, these results suggest that H₂O₂ serves as a signaling molecule in the autophagic pathway, and not as part of an oxidative stress leading to PCD.

The pattern of DCF staining resembled mitochondrial staining, and therefore we double stained the cells with DCF and MitoTracker Red, a marker for functional mitochondria. We found significant overlap between the two markers (Figure 1Bf and j; enlargements in g and k, respectively). All the DCF puncta either colocalized with or were in close vicinity to MitoTracker‐stained structures, suggesting mitochondria as the source of ROS generated during starvation. This finding coincides with our observation that phagophores, identified by a double staining with Atg5 and Atg16 (Mizushima et al, 2003), tend to colocalize with MitoTracker Red‐stained mitochondria (Supplementary Figure 1C), suggesting a functional link between ROS generated in the mitochondria during starvation and autophagosome formation. This result is consistent with recently reported findings (Kissova et al, 2006).

In order to assess the rise in ROS in a more accurate manner, we measured the increase in DCF fluorescence using a fluorimeter set at 485 nm excitation and 535 nm emission in two different cell lines (CHO and HeLa). The cells were either deprived of serum, completely starved (deprived of serum and nutrients) or grown in control medium for 2 h, after which they were treated with DCF and the fluorescent signal of the whole population (3 × 10⁴ cells) was monitored. As shown in Figure 1C, complete starvation of the cells resulted in a dramatic increase in the fluorescent signal over time, as compared to the control, whereas serum starvation had only a mild effect on the fluorescent signal, indicating a slight increase in ROS production. These measurements were further processed and summarized as the increase in DCF fluorescence over 20 min of incubation with the reagent (Figure 1D).

In the attempt to understand where ROS production fits in the autophagic signaling pathway, we monitored the effect of two known autophagy inhibitors, wortmannin and 3‐methyladenine (3MA), on starvation‐dependent DCF staining (Petiot et al, 2000). Starvation of cells for a 2‐h period in the presence of either drug reduced the level of DCF fluoresence by 25–30% (Figure 2A, left panel). As expected, the autophagic activity in the treated cells, tested by the rate of degradation of long‐lived proteins (known to increase during autophagy, see Mizushima et al, 2001), was significantly reduced (Figure 2A, right panel). Wortmannin and 3MA are phosphatidyl inositol 3 kinase (PI3K) inhibitors. Class III PI3K plays a critical role in the early stages of autophagosome formation in mammals through formation of an essential complex with Beclin 1 and, hence, inhibition of its activity inhibits the autophagic process (Petiot et al, 2000; Tassa et al, 2003). This finding, therefore, indicates that ROS production is partially PI3K dependent. Indeed, partial silencing of Beclin 1 (previously shown to cause a partial autophagic defect; Reef et al, 2006) or the PI3K hVps34 resulted in 15 and 20% (respectively) reduction in starvation‐induced ROS production (Supplementary Figure 2A). Notably, partial silencing of hVps34 caused 30% reduction in the number of autophagosomes per cell (Supplementary Figure 2B, lower right panel).

Next, we assessed the level of DCF staining in cell lines deleted for the Atg5 gene, whose product is essential for autophagosome formation, acting downstream of the class III PI3K (Suzuki et al, 2001; Tanida et al, 2004b). DCF staining in the mutant cells did not decrease, as compared with control cells (Figure 2B), but rather increased significantly (40–50%), indicating that ROS production takes place upstream of the Atg5‐dependent step of autophagosome formation, and suggesting that in the absence of a functional autophagic pathway, cells accumulate ROS.

The results described above suggest that starvation induces ROS formation. But are ROS essential for autophagy? To address this question, we tested the effect of N‐acetyl‐l‐cysteine (NAC), a general antioxidant, and catalase, which specifically decomposes H₂O₂, on the formation of autophagosomes using GATE‐16 and LC3 as markers. LC3 and GATE‐16 were shown to label autophagosomes under starvation conditions (Kabeya et al, 2004; Supplementary Figure 3A), and as shown in Supplementary Figure 3B, the two proteins exhibit significant colocalization under these conditions. Addition of either NAC or catalase to the growth medium (Sakurai and Cederbaum, 1998; Preston et al, 2001; Xu et al, 2003) of the cells for 10 min before a 2‐h starvation period resulted in reduced lipidation of GATE‐16 and LC3 and abolished the formation of GATE‐16‐ and LC3‐labeled autophagosomes (Figure 3A and Supplementary Figure 3C). As expected, the same concentration of antioxidative reagents significantly reduced the level of ROS in the treated cells (Figure 3B). These results were further validated by testing the effect of antioxidants on starvation‐induced protein degradation (Figure 3C). Treatment with NAC caused 60% inhibition in starvation‐induced degradation, and catalase caused 25% inhibition in starvation‐induced degradation. Taken together, we conclude that starvation conditions induce formation of ROS, which are required for induction of autophagy. Notably, treatment of cells with H₂O₂ alone did not result in formation of autophagosomes (data not shown).

Next, we searched for a cellular target of the starvation‐induced oxidative signal. The amount of the lipidated form of Atg8s increases significantly during starvation (Kabeya et al, 2004). As shown in Figure 4A, this elevation, in mammalian Atg8s, does not result from increased transcription, as previously shown for ScAtg8 (Kirisako et al, 1999). Consistently, we could not observe changes in the protein levels of mammalian Atg8s under these conditions (Figure 4B and data not shown). We therefore deduce that the increase in lipidated mammalian Atg8 proteins in response to nutrient starvation is due to post‐translational regulation of the processing enzymes. To test this hypothesis, cells were starved for 3 or 13 h, after which they were lysed and the lysates were tested for their ability to cleave GATE‐16. In this assay, recombinant GATE‐16, tagged at both termini (His₆‐GATE‐16‐HA), is used so that cleaved GATE‐16 can be detected by its faster mobility in SDS–PAGE. As shown in Figure 4C, incubation of GATE‐16 in a lysate prepared from cells that were starved for a short period of 3 h resulted in a decrease of up to 40% in the cleavage activity as compared with control cells. Starvation of cells for longer periods of 13 h (during which the cells are still viable) resulted in further inhibition in cleavage activity. However, addition of 1 mM DTT to the lysates brought about recovery of the cleavage activity, suggesting that the inhibition of Atg4 was caused by oxidation of the protein. Consistently, treatment of cells with 1 mM H₂O₂ for 1 h before lysis resulted in strong inhibition of the cleavage activity of Atg4 as compared with control cells. Addition of DTT to the lysate restored most of the lost activity, confirming that the inhibition by oxidation is reversible. Notably, the protein levels of the GATE‐16‐specific protease, HsAtg4A, did not change under any of the tested conditions (Supplementary Figure 4A), supporting the idea that the regulation of Atg4 activity is post‐translational. Specific anti‐HsAtg4A antibodies were utilized to confirm that the cleavage activity shown in Figure 4C is mediated by HsAtg4A (for details see Supplementary Figure 4B), indicating that Atg4 is a target for redox regulation under starvation, and suggesting that accumulation of lipidated Atg8s under starvation might result from inhibition of Atg4 activity. Indeed, we found that treatment with 1 mM H₂O₂ resulted in accumulation of lipidated LC3 (Supplementary Figure 4C).

Atg4 acts both in the initial processing of Atg8s (priming) and in the delipidation step. The priming step was previously shown to occur immediately after translation under nutrient‐rich conditions (Kabeya et al, 2000). To test whether this step is also inhibited during starvation, we transfected cells with GATE‐16 tagged with Myc at its N‐terminus and HA at the C‐terminus, starved them for 30 min (a time period in which autophagosomes are expected to form; Fass et al, 2006) or 13 h and subjected them to a [³⁵S]methionine pulse–chase experiment as detailed in Materials and methods (Figure 4D). The control experiment shows that two bands at approximately 19 kDa are detected immediately after pulse labeling in the transfected cells (Figure 4D, left panel). A 1 h chase resulted in disappearance of the upper band, suggesting that this band corresponds to unprimed GATE‐16 (Myc‐GATE‐16‐HA), whereas the lower band corresponds to primed GATE‐16 (Myc‐GATE‐16). Short starvation does not induce accumulation of the unprimed form (Figure 4D, middle panel), and this form accumulates only in cells that were starved for a long period of 13 h.

These results indicate that HsAtg4A is redox regulated during starvation. As the priming activity is not affected by short starvation, we conclude that the delipidation activity is the main target of this regulation.

To characterize the redox regulation of HsAtg4, we expressed HsAtg4A as a recombinant protein and tested its activity in the cell‐free cleavage assay. As depicted in Figure 5A, the protease was gradually activated upon reduction of the reaction mixture and the activity saturated at 1 mM DTT. Next, we tested the inhibition of HsAtg4A activity by H₂O₂. The protein was preincubated in a low concentration of DTT for activation followed by addition of H₂O₂ for 5 min before incubation with GATE‐16 for the cleavage assay. The activity of HsAtg4A decreased as the H₂O₂ concentration increased above 100 μM (Figure 5B). In some experiments, a slight increase in the protease activity was detected upon addition of 30 μM H₂O₂ to the reaction mixture. Finally, we tested the reversibility of HsAtg4 inhibition by H₂O₂. In this experiment, the protein was preincubated in a low concentration of DTT (Figure 5C, lane 2) followed by addition of H₂O₂ (lane 3) and then addition of a high concentration of DTT (lane 4; each incubated for 5 min before addition of the next reagent) before incubation with GATE‐16. Under these conditions, HsAtg4A completely regained its activity. Taken together, these results show that reducing conditions activate HsAtg4A, whereas an oxidizing environment inhibits its activity. Moreover, these results indicate that H₂O₂ alone is sufficient to reversibly inhibit the activity of HsAtg4A in vitro, and therefore suggest that H₂O₂ might also act directly on the protease in vivo.

HsAtg4A contains 12 cysteines, seven of which are highly conserved among tetrapod homologues of Atg4A and Atg4B. The latter share an overall high conservation level with respect to the whole family (Figure 6). One of these conserved residues is the putative active residue of the protease Cys⁷⁷ (Kirisako et al, 2000; Tanida et al, 2004a), the only cysteine conserved also in the yeast S. cerevisiae (Figure 6A, large arrow, numbering is according to HsAtg4B). Another conserved residue (Cys⁸¹ in HsAtg4A) is located four sites downstream of the active residue. We first sought to determine whether Cys⁷⁷ is the catalytic residue, as suggested by homology to ScAtg4 and HsAtg4B. As shown in Figure 7A, recombinant HsAtg4A harboring a mutation of cysteine to alanine at position 77 (His₆‐HsHsAtg4A^C77A) did not cleave GATE‐16 in vitro, neither in the absence nor in the presence of DTT. Based on this result and the sequence homology described above, we conclude that Cys⁷⁷ is part of the active site of HsAtg4A.

To study the mechanism of redox regulation of HsAtg4A, we mutated the conserved cysteine residues 81 and 92 (Figure 6A, small arrows), which are closest to the active residue, to serine (His₆‐HsAtg4A^C81S and His₆‐HsAtg4A^C92S, respectively) and tested their activity in the cell‐free cleavage assay of GATE‐16 under different redox conditions. As shown in Figure 7A, mutation in Cys⁸¹ significantly reduced the redox sensitivity of HsAtg4A so that His₆‐HsAtg4A^C81S was partially active even in the absence of DTT. Mutation in Cys⁹², however, had no effect on the regulation and the protease remained under the same conditions as active as His₆‐HsAtg4A^WT (data not shown). Next, we tested the effect of oxidation by H₂O₂ on the activity of His₆‐HsAtg4A^C81S. We found that the mutant protein retained significant activity in the presence of H₂O₂ (Figure 7B) under conditions in which the WT was completely inhibited (Figure 5C). Similar to the WT, the partial inhibition of His₆‐HsAtg4A^C81S by H₂O₂ could be reversed by addition of DTT. From these results, we conclude that Cys⁸¹ is important for redox regulation of HsAtg4A in vitro.

To examine whether Cys⁸¹ plays a role in the redox regulation of HsAtg4A in vivo, we monitored the formation of GFP‐GATE‐16‐labeled autophagosomes in cells transfected with HsAtg4A mutants. Expression of HsAtg4A^C81S significantly impaired the formation of GATE‐16‐labeled autophagosomes in starved cells (Figure 8A), as compared with cells expressing HsAtg4A^WT (Figure 8A) or HsAtg4A^C92S (data not shown) at similar levels. The average number of autophagosomes formed in HsAtg4A^C81S‐expressing cells was 70% lower than in HsAtg4A^WT‐expressing cells (Figure 8A, lower panel). One could suggest that in the presence of HsAtg4A^C81S, autophagosomes are formed, but then rapidly degraded, and therefore cannot be seen. To exclude this possibility, we inhibited the degradation of autophagosomes in the lysosome using bafilomycin A (Mousavi et al, 2001). This treatment did not affect the mutant phenotype, indicating that autophagosomes do not form in cells transfected with HsAtg4A^C81S (data not shown). The finding that GATE‐16‐labeled autophagosomes do not form in HsAtg4A^C81S‐expressing cells was further supported by GATE‐16 lipidation measurements. As the HsAtg4A^C81S mutant is less sensitive to oxidation, we expect the delipidation activity to be higher in starved cells transfected with this mutant as compared with cells transfected with HsAtg4A^WT. Indeed, we found nearly 40% less of the lipidated form of GATE‐16 in cells transfected with HsAtg4A^C81S as compared with cells transfected with HsAtg4A^WT (Figure 8B). It should be mentioned that expression of HsAtg4A^C81S did not affect the pattern of LC3 staining in the cells (data not shown), as expected, as HsAtg4A is specific for GATE‐16 and does not process LC3. Taken together, our findings support the notion that Cys⁸¹ takes part in the redox regulation of HsAtg4A during autophagy.

In order to extend our study to other Atg4 family members, we decided to test the LC3‐specific protease, HsAtg4B. To that end, we transfected cells with GFP‐HsAtg4B^WT, GFP‐HsAtg4B^C78S or with GFP as control and tested the effect on lipidation of LC3 and formation of LC3‐labeled autophagosomes. We found that overexpression of both GFP‐HsAtg4B^WT and GFP‐HsAtg4B^C78S reduced autophagosome formation and lipidation of LC3 (data not shown). This phenomenon was previously reported for overexpression of HsAtg4B^WT (Tanida et al, 2004a). To overcome this problem, we added H₂O₂ to the starvation media. Under these conditions, HsAtg4B^WT was inhibited and a normal autophagic phenotype was observed (Figure 9). HsAtg4B^C78S, however, was resistant to H₂O₂ and the cells expressing this construct failed to form lipidated LC3 and LC3‐labeled autophagosomes. These results indicate that HsAtg4B, similar to HsAtg4A, is redox regulated.

Recombinant His₆‐HsAtg4A (WT and the mutants described in Supplementary data) and His₆‐GATE‐16‐HA were expressed from the pQE‐30 plasmid and purified as described previously (Scherz‐Shouval et al, 2003). Purification of His₆‐HsAtg4A (WT and mutants) was carried out in the presence of 10 mM βME to prevent formation of nonspecific disulfide bonds; however, before its use, the protein was diluted to at least 30‐fold or dialyzed in a buffer lacking βME.

All cell lines were grown in α‐MEM medium (Biological Industries, Israel) supplemented with 10% fetal calf serum and referred to as control medium. Atg5−/− MEFs and anti‐Atg16 polyclonal antibodies were a kind gift from the laboratory of Dr Mizushima (The Tokyo Metropolitan Institute of Medical Science). Stable clones of GFP‐GATE‐16‐, GFP‐LC3‐ or YFP‐Atg5‐transfected CHO cells were selected in 1 mg/ml geneticin (G418). Starvation was carried out in EBSS. Details regarding the use of different reagents for tissue culture treatments are described in Supplementary data. Transfections of CHO cells were preformed with lipofectamine (Invitrogen) according to the manufacturer's protocol. Transfections of HeLa and HEK 293 cells were performed by the standard calcium phosphate technique. For immunofluoresence, unless described otherwise, cells were fixed in methanol, permeabilized with acetone and then incubated with the indicated antibodies and visualized by a Nikon eclipse TE300 fluorescent microscope or by an Olympus IX‐70 confocal microscope.

H₂O₂ was measured using 50 μMDHE (Molecular Probes) and 30 μM DCFDA (Molecular Probes). See Supplementary data for details. The fluorescent signal was detected by an Olympus IX‐70 confocal microscope. Each experiment was repeated at least three times and representative images are shown. The DCFDA signal was also measured by a SPECTRAmax gimini fluorimeter, set to 485 excitation and 535 emission and kept at 37°C, for 40 min. Data collected from the fluorometric measurements were analyzed as detailed in Supplementary data. The results presented for all fluorimetric measurements are the means±s.d. of at least three experiments in duplicate or triplicate.

The assay was preformed on CHO cells according to the protocol described by Fass et al (2006). When required, 10 mM 3MA, 100 nM wortmannin, 1000 μ/ml catalase or 10 mM NAC were added. Notably, NAC was added for 10 min before starvation induction and catalase was added 14 h before starvation. [¹⁴C]valine release was calculated as a percentage of the radioactivity in the TCA‐soluble supernatant relative to the total cell radioactivity.

Cells were homogenized in Ripa buffer (0.1 M NaCl, 5 mM EDTA, 0.1 M Na₂HPO₄/NaH₂PO4, pH 7.5, 1% Triton, 0.5% DOC, 0.1% SDS and protease inhibitors) by vortexing on ice. The homogenates were centrifuged for 15 min at 20 000 g and the supernatant containing the lysate was collected.

RNA was isolated by Tri‐reagent (Molecular Research Center Inc.) as detailed in Supplementary data.

HeLa cells were transiently transfected with Myc‐GATE‐16‐HA or with an empty vector. Cells were then kept in full medium for 24 h, after which they were either incubated in methionine‐free medium for 10 min or starved in EBSS for 30 min (2 × 10 cm plates). For long starvation, the cells (three plates) were shifted to EBSS 10 h post‐transfection for 13 h, after which the medium was replaced with fresh EBSS. All cells were then pulse‐labeled with 0.2 mCi of [³⁵S]methionine for 10 min at 37°C, and either chased in full medium for 1 h at 37°C or lysed immediately on ice, in Ripa buffer supplemented with 0.5 mM N‐ethylmaleimide (sigma). Lysates were immunoprecipitated with anti‐Myc antibodies and analyzed on 15% SDS–PAGE.

Supplementary data are available at The EMBO Journal Online (http://www.embojournal.org).