MCL‐1 is a stress sensor that regulates autophagy in a developmentally regulated manner

The results presented above suggest that MCL‐1 has a second, apoptosis‐independent role in the regulation of cellular stress. A major response activated by nutrient deprivation under conditions that trigger MCL‐1 degradation is autophagy. To address the physiological significance of MCL‐1 degradation and establish a link between MCL‐1 degradation and the induction of autophagy, we analysed the temporal relationship between MCL‐1 loss and autophagosome formation. Conversion of a GFP‐tagged version of the autophagy marker LC3 (GFP–LC3) from GFP–LC3‐I to GFP–LC3‐II, its translocation to punctate structures in the cytosol, as well as changes in levels of the LC3‐associated protein p62 (which is degraded in the lysosomes under these conditions) have extensively been used to follow autophagy induction in cells. Incubation of H1299 cells stably expressing GFP–LC3 (H1299‐LC3) (Chang et al, 2010) for up to 8 h in glucose‐free media or in EBSS did not cause BAX activation or cytochrome c release (Figure 2A and B), but induced loss of p62 protein, processing of GFP–LC3‐I into GFP–LC3‐II (Figure 2C) and appearance of LC3‐positive structures in the cytosol (Figure 2B), indicative of autophagy induction. Under these conditions, loss of MCL‐1 protein was an early event, as it occurred before LC3 processing and p62 degradation (Figure 2D).

The observation that MCL‐1 degradation is associated with activation of autophagy, but occurs before major changes in autophagy markers, suggests that MCL‐1 degradation regulates autophagy initiation upstream of autophagosome formation. We took several approaches to test this possibility. First, to assess whether MCL‐1 degradation lies upstream of autophagosome formation, MCL‐1 levels were measured in the presence of 3‐methyladenine (3MA), a class III PI3 kinase inhibitor that blocks autophagosome formation. However, while 3MA caused an increase in p62 levels under both basal and autophagic conditions, reflecting the inhibition of autophagy, it had no effect on MCL‐1 degradation (Figure 2E). Similarly, the lysosomal inhibitor bafilomycin blocked p62 degradation, but had no effect on MCL‐1 degradation (Figure 2F). Altogether, these results suggest that MCL‐1 degradation occurs upstream of autophagosome formation, consistent with a possible regulatory function of MCL‐1 in the initiation of autophagy.

We then addressed the function of MCL‐1 degradation during autophagy. To preserve MCL‐1 levels following induction of autophagy, MCL‐1 was overexpressed in H1299‐LC3 cells using an adenovirus. Overexpression of MCL‐1 decreased the number of GFP–LC3 foci (autophagosomes) (Figure 3A). The total surface area occupied by GFP–LC3 vesicles was also quantified, showing a significant decrease in vesicle formation in MCL‐1‐overexpressing cells (Figure 3B). Cells overexpressing MCL‐1 also had increased basal levels of p62, while p62 degradation was reduced in Ad MCL‐1‐starved cells compared with control infection (Ad LacZ; Figure 3E). These results indicate that increased MCL‐1 levels block autophagy.

As these experiments were consistent with an inhibitory role of MCL‐1 on the initiation of autophagy, we tested the effect of MCL‐1 loss on the autophagic response. Compared with control siRNA, MCL‐1 levels were decreased by ∼70% in cells transfected with MCL‐1 siRNA (Figure 3F). Downregulation of MCL‐1 greatly increased the formation of GFP–LC3‐positive vesicles (Figure 3A and B), as well as the lipidation of endogenous LC3 (Figure 3C and D). In addition, MCL‐1 knockdown resulted in a 50% decrease in p62 levels in cells grown in complete media and a further decrease when cells were grown in glucose‐free media for 3 h (Figure 3F). Similarly, MCL‐1 knockdown increased processing of GFP–LC3I to GFP–LC3II following starvation, and also caused the accumulation of cleaved GFP (generated through lysosomal degradation of GFP–LC3; Klionsky et al, 2008) even in nutrient‐replete cells (Figure 3F). Of note, none of these manipulations caused cytochrome c release or BAX activation for up to 4 h (Figures 2A and 3A), although some activation of apoptosis was observed at later time points (<10% of the cells at 8h; Figure 2A) when MCL‐1 was acutely depleted before starvation rather than as a normal consequence of starvation. Altogether, these results support the conclusion that loss of MCL‐1 is not sufficient to induce apoptosis even under starvation conditions.

The role of MCL‐1 in the regulation of autophagy was also studied in primary cortical neurons from MCL‐1^flox/flox animals, in which MCL‐1 was removed in vitro using a lentiviral vector expressing the Cre recombinase. Consistent with the siRNA results, ablation of MCL‐1 in primary neurons caused a marked decrease in p62 levels (Figure 3G). Similarly, an increase in endogenous LC3‐II was observed in SV40‐transformed MCL‐1‐null MEFs (Figure 3H). Taken together, these results indicate that MCL‐1 negatively regulates autophagy, and that its degradation is required for autophagy to proceed.

One potential explanation for the increase in autophagosomes observed following loss of MCL‐1 is the failure of autophagosome fusion with lysosomes, thus resulting in their accumulation. To test this possibility, we further studied the effect of MCL‐1 on the fusion between GFP–LC3 vesicles and lysosomes. A fraction of GFP–LC3 vesicles colocalised with the lysosomal marker LAMP1 in EBSS‐treated control cells and the percentage of GFP–LC3 vesicles that colocalised with LAMP1 was not affected by overexpression or downregulation of MCL‐1 (Figure 3J; quantified in Figure 3I), indicating that MCL‐1 does not affect the fusion between autophagosomes and lysosomes. Accordingly, inhibition of lysosomal function with bafilomycin increased endogenous LC3‐II levels in both control and MCL‐1 knockdown cells (Figure 3C and D). This is consistent with the accumulation of cleaved GFP observed following MCL‐1 downregulation. These findings, together with the observation that MCL‐1 degradation is upstream of autophagosome formation, suggest that MCL‐1 serves as an early regulator of autophagy in cells exposed to nutrient deprivation, its effects being proportional to the strength of the autophagic stimulus.

BCL‐2 and BCL‐X_L inhibit autophagy by blocking the function of Beclin‐1 (Pattingre et al, 2005; Maiuri et al, 2007a), a key upstream regulator of autophagy (He and Klionsky, 2009). MCL‐1 was also reported to interact with GST‐tagged Beclin‐1 (Erlich et al, 2007), an observation that we have extended to the endogenous proteins (Figure 4A). This, along with the observation that MCL‐1 inhibits autophagy caused by overexpression of Beclin‐1 (Maiuri et al, 2007a), suggests that, similar to BCL‐2 and BCL‐X_L, MCL‐1 regulates autophagy at least in part by controlling the activity of Beclin‐1. Interestingly, while the functional interaction between BCL‐2 and Beclin‐1 occurs at the endoplasmic reticulum, where both proteins partially localise (Pattingre et al, 2005) (Figure 4B), MCL‐1 is absent from this organelle (Figure 4B). Rather, MCL‐1 is a mitochondrial protein (Chou et al, 2006; Germain and Duronio, 2007) in which Beclin‐1 also partially localises (Figure 4B) (Pattingre et al, 2005; Maiuri et al, 2007a), suggesting that MCL‐1 interacts with Beclin‐1 on mitochondria. To directly test this possibility, cells were fractionated into HM containing mitochondria and LM containing ER, and MCL‐1 was immunoprecipitated. As shown in Figure 4C, Beclin‐1 co‐precipitated with MCL‐1 only in the fraction containing mitochondria, indicating that the two proteins interact at the mitochondria and not the ER.

At the level of the organism, previous studies have indicated that autophagy has an important role in the homeostasis of the central nervous system (CNS). For example, deletion of key autophagy regulators (ATG5, ATG7) in the CNS leads to progressive accumulation of ubiquitin‐containing aggregates and neurodegeneration (Hara et al, 2006; Komatsu et al, 2006), whereas deletion of Ambra1, a component of the Beclin‐1/Vps34 complex, causes major neural tube defects (Fimia et al, 2007). To determine whether MCL‐1 regulates autophagy in the CNS, we examined the consequences of MCL‐1 deletion in post‐mitotic cortical neurons. Mice homozygous for a floxed MCL‐1 allele (MCL‐1^flox/flox) were crossed with animals carrying a CamKIIα Cre transgene, resulting in the expression of the Cre recombinase in neurons of the cortex, hippocampus and, to a lesser extent, striatum (Casanova et al, 2001). MCL‐1^Δ/Δ mice were born at the expected ratio, but were smaller in size (Figure 5A) and had a median lifespan of 49 days, with a few animals surviving to 3 months (see Figure 8E). Recombination of the floxed alleles was confirmed by western blot of tissue from the cortex of P14 animals (Figure 5B). No change in protein expression was observed for BAX and BCL‐X_L (Figure 5B), suggesting that there were little compensatory changes in these other BCL‐2 homologues highly expressed in the brain. Cresyl violet staining of brain sections revealed progressive cellular loss in cortical layers adjacent to the corpus callosum in MCL‐1^Δ/Δ animals, with extensive loss evident by P14 (Figure 5C). As we hypothesised that MCL‐1 loss would result in increased autophagy, we measured several autophagic markers in MCL‐1^Δ/Δ mice. Brain sections from control and MCL‐1^Δ/Δ mice were stained for endogenous LC3, which appears as punctate staining in the cytosol of autophagic cells. As shown in Figure 5E, LC3‐positive cells lined the cortical region where cell loss was observed in P14 MCL‐1^Δ/Δ animals, while minimal staining was present in control mice. Higher magnification images of the affected region confirmed punctate LC3 staining in cells of mutant, but not control animals (Figure 5G; quantification in Figure 5D). Co‐staining with the neuronal marker NeuN confirmed that neurons, but not glia, were affected (Figure 5H). To further substantiate these results, the cortex of P14 control and MCL‐1^Δ/Δ mice was examined by electron microscopy. Compared with control, multiple vesicles containing portions of cytoplasm, including mitochondria, were present in the cortex of MCL‐1^Δ/Δ mice (Figure 5F and I; asterisks denote autophagosomal structures; arrows denote mitochondria). These structures were delineated by a double membrane, indicative of autophagosomes (Figure 5I, arrowheads point to the double membrane). As accumulation of autophagosomes could reflect a decrease in their clearance, we also determined p62 levels in MCL‐1^Δ/Δ brains as a measure of lysosomal degradation of autophagosomes. As shown in Figure 6A, p62 levels were decreased in MCL‐1^Δ/Δ cortical extracts, suggesting that loss of MCL‐1 does not prevent lysosomal degradation of the autophagosomes. Altogether, these results indicate that the deletion of MCL‐1 in cortical neurons results in the deregulated activation of autophagy.

In addition to its role in the regulation of autophagy, MCL‐1 has a well‐characterised antiapoptotic role during development, including in the CNS (Opferman et al, 2003, 2005; Arbour et al, 2008). To determine whether MCL‐1 has a similarly important antiapoptotic role in post‐mitotic neurons, we analysed MCL‐1^Δ/Δ mice for the presence of apoptotic markers. The activated form of BAX (6A7‐reactive) and caspase‐3 were first detected at P7 in MCL‐1^Δ/Δ mice and were increased by P14 (Figure 6B), suggesting that apoptosis is activated to some extent in these animals. To determine the relationship between autophagy and apoptosis activation, we further analysed BAX activation in relation to LC3 staining. In P7 MCL‐1^Δ/Δ mice, virtually all (97±3%) NeuN‐positive neurons were also positive for LC3 (Figure 5H), while only 18±5% of the cells were positive for active BAX (Figure 6B). These BAX‐positive cells are neurons with activated autophagy, as they co‐stained with LC3 (Figure 6C). In addition, while there are no new cortical neurons generated after birth, the number of LC3‐positive neurons decreased at P14 compared with P7 with a concomitant increase in both active BAX and active caspase‐3‐positive cells that paralleled cellular loss. This suggests that a subset of the neurons that initially activate autophagy in response to MCL‐1 deletion undergo apoptosis over time, although we cannot exclude the possibility that apoptosis and autophagy are concomitantly activated in this subset of neurons. Nevertheless, the idea that autophagic MCL‐1 KO neurons eventually activate apoptosis is supported by the observation that while all BAX‐positive cells were also positive for LC3 at P7, a portion of these cells (8%) was LC3‐negative by P14 (Figure 6C; quantification in Figure 8D), in accordance with the notion that apoptosis can inhibit autophagy (Luo and Rubinsztein, 2010).

To further examine the effect of MCL‐1 loss on apoptosis in post‐mitotic neurons in a system where dying cells are more readily tractable, we deleted MCL‐1 from MCL‐1^flox/flox neurons in culture, which activates an autophagic response (Figure 3G), and analysed the cells for the presence of apoptotic markers. There was no increase in apoptosis in MCL‐1 knockout neurons compared with control neurons for up to 10 days in culture, as determined by caspase‐3‐dependent cleavage of p130^CAS into a 31 kDa fragment (Kook et al, 2000) (Figure 6D) and cytochrome c release (Figure 6E). Altogether, these results suggest that the primary cellular response to MCL‐1 deletion in neurons is the induction of autophagy, consistent with our results in cell lines (Figures 1, 2, 3).

We previously reported an increase in apoptosis when MCL‐1 was deleted in neuronal progenitors (using Foxg1 and Nestin Cre promoters), both in vivo and in neural progenitor cultures (Arbour et al, 2008). This suggests that the cellular context dictates the final outcome following degradation of MCL‐1. To test this possibility, we first examined whether Foxg1‐MCL‐1^Δ/Δ embryos (deletion of MCL‐1 in neural progenitors) also showed increased autophagy. In contrast to CamKIIα‐MCL‐1^Δ/Δ animals (Figure 6A), active caspase‐3 was readily detected by western blot in E12.5 Foxg1‐MCL‐1^Δ/Δ brains (Figure 7A), consistent with our previous immunohistochemistry results (Arbour et al, 2008). However, LC3 lipidation (LC3‐II) and p62 levels were similar between heterozygous and knockout mice (Figure 7A). To further characterise the autophagic response of Foxg1‐MCL‐1^Δ/Δ mice, E15.5 embryos were stained for endogenous LC3 and analysed by fluorescence microscopy. As shown in Figure 7B, no major differences in LC3 staining were observed in the ventricular zone (VZ) between control and MCL‐1(Foxg1)^Δ/Δ embryos (Figure 7B; top panels, VZ). However, there was a striking increase in LC3 staining in the cortical plate (where post‐mitotic neurons reside) of control embryos (Figure 7B; middle left panel). A similar increase in LC3 staining was seen in MCL‐1(Foxg1)^Δ/Δ embryos, although, as previously reported (Arbour et al, 2008), the cortex was much smaller (Figure 7B; middle right panel). As the percentage of LC3 vesicles that colocalised with the lysosomal marker LAMP1 did not differ between the cortex and the VZ (Figure 7C), these results suggest that the autophagic machinery is upregulated as neurons differentiate, consistent with the autophagic response observed in MCL‐1^Δ/Δ neurons.

To determine whether post‐mitotic neurons and neural progenitors express a different complement of apoptotic and autophagic genes, cortical neurons and neurospheres (neural progenitors) were cultured from E14.5 embryos. There was a marked difference in the expression of proapoptotic and autophagy‐related proteins between progenitors and post‐mitotic neurons (Figure 7D). For example, expression of the BH3‐only proteins BIM and Puma was higher in neurospheres. Conversely, ATG7 and the ATG5–ATG12 conjugate, two important autophagy regulatory proteins, were expressed at higher levels in neurons (Figure 7D). This suggests that autophagy and apoptosis are differentially regulated as neurons differentiate, consistent with the pattern of LC3 staining observed in E15.5 embryos. As a control, the purity of the cultures was assessed by blotting for the neuronal marker Tuj‐1 and the progenitor marker Nestin (Figure 7D). As BH3‐only proteins are required for BAX activation, a key step in determining cell fate downstream of MCL‐1 degradation (Figure 1), these differences suggest that neural progenitors have a higher propensity than neurons to undergo apoptosis under low‐nutrient conditions. This was tested by measuring apoptosis and autophagy in EBSS‐treated neurospheres and neurons. We first verified that starvation caused MCL‐1 degradation in neurospheres (Figure 7E). While both neurons and neurospheres had increased LC3‐II levels following induction of autophagy, only neurospheres showed an increase in apoptosis, as evaluated by p130^CAS cleavage and cytochrome c release (Figure 7F and G). This indicates that neural progenitors are more susceptible than neurons to apoptosis following nutrient deprivation.

Altogether, these results indicate that neurons, as long‐lived post‐mitotic cells, preferentially respond to stress by activating autophagy rather than apoptosis, in an attempt to restore homeostasis. Thus, inhibiting autophagy in neurons should drive these cells towards an apoptotic response. To test this hypothesis, MCL‐1^Δ/Δ mice were crossed with mice deficient for Beclin‐1, as it is an essential autophagy protein regulated by MCL‐1 (Figure 4; Maiuri et al, 2007a). As homozygous deletion of Beclin‐1 is embryonically lethal, we examined Beclin‐1^+/− animals, which have a 30–50% reduction in autophagy (Qu et al, 2003; Yue et al, 2003). MCL‐1^Δ/Δ Beclin‐1^+/− mice were born at the expected ratio, but started to die around 2 weeks of age (Figure 8E). MCL‐1^Δ/Δ Beclin‐1^+/− mice had a reduced lifespan compared with their MCL‐1^Δ/Δ Beclin‐1^+/+ littermates (median lifespan of 21 days for the former, compared with 49 days the latter). Nevertheless, as with MCL‐1^Δ/Δ mice, some MCL‐1^Δ/Δ Beclin^+/− animals lived to 3 months of age. However, cresyl violet staining of brain sections did not reveal major differences in morphology between MCL‐1^Δ/Δ and MCL‐1^Δ/Δ Beclin^+/− brains (Figure 8A).

To determine the extent of apoptosis, brain sections from each genotype were stained for active BAX (6A7) and active caspase‐3. The number of cells positive for active BAX and active caspase‐3 was significantly increased in MCL‐1^Δ/Δ Beclin‐1^+/− sections compared with MCL‐1^Δ/Δ Beclin‐1^+/+ (Figure 8B and C). Moreover, the number of 6A7‐positive, LC3‐negative cells was doubled in MCL‐1^Δ/Δ Beclin‐1^+/− mice compared with MCL‐1^Δ/Δ (Figure 8D). This suggests that a greater proportion of neurons activate apoptosis when Beclin‐1 levels are reduced, consistent with previous reports on the relationship between apoptosis and autophagy in stressed cells in culture (Boya et al, 2005; Luo and Rubinsztein, 2010). Altogether, our results suggest that MCL‐1 functions as an upstream regulator of a stress–response pathway controlling both apoptosis and autophagy in a developmentally regulated manner.

H1299 cells stably expressing GFP–LC3 (gift from Dr Gordon Shore, McGill University), HeLa cells and MEFs were cultured in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum. For the induction of autophagy, cells were washed in PBS and incubated for the indicated time in either EBSS(Sigma) or DMEM, containing no glucose or pyruvate supplemented with 10% fetal bovine serum. Where indicated, cells were pre‐treated 1 h with 3MA (5 mM) or Bafilomycin (200 nM) before inducing autophagy. Infections with an adenoviral vector coding for myc‐tagged MCL‐1 or HA‐tBID (gift from Gordon Shore) were carried out as described (Germain et al, 2005). For the siRNA experiments, cells were transfected with 50 nM Ctrl or MCL‐1 siRNA (Santa Cruz Biotechnology) using SilentFect (Bio‐Rad Laboratories) and analysed after 24 h. Subcelular fractionation was carried as described (Germain et al, 2002).

Neurospheres, cortical neurons and CGNs were cultured as described previously (Cregan et al, 1999; Fortin et al, 2001; Vanderluit et al, 2004). For immunofluorescence, neurosphere were grown on coverslips for 2 days in neurobasal containing 0.5 mM l‐glutamine, 1% N2 supplement, 2% B27 supplement, 10 ng/ml bFGF and 20 ng/ml EGF. Where indicated, neurons were infected at the time of plating with a lentiviral vector expressing the Cre recombinase using an MOI of 2. Hypoxic cell death was induced using a humidified environmental chamber (Coy Laboratory, Grass Lake, MI) set at 37°C. Oxygen concentration was maintained at 1% with a calibrated gas mixture of 95% nitrogen and 5% carbon dioxide. CGN cultures were treated with 10 μM cytosine–arabinoside after 2 days in vitro to inhibit glial growth, and incubated in the chamber after 7–8 days in vitro for 16–18 h at 1% oxygen in the presence of the NMDA receptor blocker, MK801 (10 μM). Cultures were then allowed to reoxygenate by incubating under normoxic conditions at 37°C. Cells were collected at the indicated times following reoxygenation.

The following antibodies were used: mouse anti‐mtHSP70 (ABR Bioreagents); mouse anti‐HSC70 (Abcam); mouse anti‐Tuj‐1 (Gift from Dr Dave Brown); mouse anti‐Nestin (Research Diagnostics Inc.); guinea‐pig anti p62 (Progen Biotechnik); rabbit anti‐mouse MCL‐1 (Rockland Immunochemicals); rat anti‐APAF‐1 and mouse anti‐NeuN (Chemicon); rabbit anti‐human MCL‐1, mouse anti‐human p62, rabbit anti‐mouse p62, rabbit anti‐Beclin‐1, rabbit anti‐ATG‐7, mouse anti‐BCL‐X, mouse anti‐human LAMP1 and rabbit anti‐BAX (Santa Cruz Biotechnology); mouse anti‐GFP (Clonetech), mouse anti‐actin (Sigma); rabbit anti‐LC3 and rabbit anti‐ATG5 (Novus Biologicals); mouse anti‐mouse p62, mouse anti‐cytochrome c, rabbit anti‐BIM, mouse anti‐BCL‐2, mouse anti‐active BAX (6A7), mouse anti‐calreticulin (BD Pharmingen); rabbit anti‐BAK (Upstate); rabbit anti‐BCL‐W, rabbit anti‐HA, rabbit anti‐PUMA and rabbit anti‐active caspase‐3 (Cell Signaling Technology). Cells were lysed in 10 mM Tris–HCl, pH 7.9, 150 mM NaCl, 1 mM EDTA, 1% Triton X‐100 supplemented with a protease inhibitor mixture (Sigma‐Aldrich). For immunoprecipitation, lysates were incubated for 2 h with the antibody and precipitated with protein G‐Sepharose. For immunoblot analysis, proteins were subjected to SDS–PAGE, transferred to nitrocellulose membranes, and blotted with specific antibodies. Blots were incubated with horseradish peroxidase‐conjugated secondary antibodies and visualised by enhanced chemiluminescence (Amersham Biosciences).

For immunofluorescence, cells were grown on coverslips and treated as indicated in the figure legends. Cells were then fixed with 4% paraformaldehyde (PFA) and analysed by immunofluorescence using AlexaFluor secondary antibodies (Molecular Probes). All images were taken using a Zeiss 510 meta confocal microscope. Quantification of GFP–LC3 staining was performed by calculating the surface area of the GFP–LC3 fluorescence above a set threshold (that was kept constant throughout the experiments) over the total surface area of the cell, using the Image J software.

All experiments were approved by the University of Ottawa's Animal Care Ethics Committee adhering to the Guidelines of the Canadian Council on Animal Care. To generate forebrain‐specific MCL‐1 KO mice, the previously described floxed MCL‐1 mice were crossed with CamKIIα Cre mice (Casanova et al, 2001). MCL‐1 mice were also crossed with Beclin‐1^+/− mice (gift from Dr Beth Levine). All mice were kept on a C57Bl/6 background. Animals were genotyped according to standard protocols with previously published primers for MCL‐1 Beclin‐1 and Cre.

Mice were euthanised with a lethal injection of sodium pentobarbitol. For immunohistochemistry, mice were perfused with 1 × PBS followed by cold 4% PFA. Brains were then removed, post‐fixed overnight in 4% PFA, cryoprotected in 20% sucrose in 1 × PBS, and frozen. Sections were collected as 14 μm coronal cryosections on Superfrost Plus^® slides (Fisher Scientific). For western blot analysis, brains were removed, cortices dissected and flash‐frozen in liquid nitrogen. For EM, mice were perfused with a combination of 2% PFA and 2% glutaraldehyde. Pieces of cortex not >1 mm² were taken and processed as described (Cheung et al, 2006).

Statistical differences were determined using Student's t‐test or one‐way ANOVA. P<0.05 was considered statistically significant.