eIF2α Confers Cellular Tolerance to S. aureus α-Toxin

We report on the role of conserved stress–response pathways for cellular tolerance to a pore forming toxin. First, we observed that small molecular weight inhibitors including of eIF2α-phosphatase, jun-N-terminal kinase (JNK), and PI3-kinase sensitized normal mouse embryonal fibroblasts (MEFs) to the small pore forming S. aureus α-toxin. Sensitization depended on expression of mADAM10, the murine ortholog of a proposed high-affinity receptor for α-toxin in human cells. Similarly, eIF2αS51A/S51A MEFs, which harbor an Ala knock-in mutation at the regulated Ser51 phosphorylation site of eukaryotic translation initiation factor 2α, were hyper-sensitive to α-toxin. Inhibition of translation with cycloheximide did not mimic the tolerogenic effect of eIF2α-phosphorylation. Notably, eIF2α-dependent tolerance of MEFs was toxin-selective, as wild-type MEFs and eIF2αS51A/S51A MEFs exhibited virtually equal sensitivity to Vibrio cholerae cytolysin. Binding of S. aureus α-toxin to eIF2αS51A/S51A MEFs and toxicity in these cells were enhanced as compared to wild-type cells. This led to the unexpected finding that the mutant cells carried more ADAM10. Because basal phosphorylation of eIF2α in MEFs required amino acid deprivation-activated eIF2α-kinase 4/GCN2, the data reveal that basal activity of this kinase mediates tolerance of MEFs to α-toxin. Further, they suggest that modulation of ADAM10 is involved. During infection, bacterial growth may cause nutrient shortage in tissues, which might activate this response. Tolerance to α-toxin was robust in macrophages and did not depend on GCN2. However, JNKs appeared to play a role, suggesting differential cell type and toxin selectivity of tolerogenic stress responses. Understanding their function or failure will be important to comprehend anti-bacterial immune responses.


Introduction
Membrane perforation by pore forming toxins (PFT) is an ancient mode of attack employed by many bacteria, which helps them to establish or sustain infection (1)(2)(3). PFT represent a large group of bacterial toxins, which can be divided into various structural families (2). Many PFT have been discovered based on their ability not only to lyse red blood cells but they also affect nucleated cells, with effects ranging from induction of cell death to proliferation, time scales of occurrence from seconds to days after attack (4). Of particular relevance in the present context, cell autonomous defenses are in place to limit or reverse damage of nucleated target cells of PFT (4)(5)(6)(7)(8)(9)(10)(11)(12)(13); they have been discussed as an integral part of the innate immune system, defending against bacteria (14). Work in C. elegans identified MAPK as master regulators of defense against PFT (7,15). Whereas the importance of p38 MAPK is well established (3,4,9,(16)(17)(18), data on the role of jun-N-terminal kinases (JNKs) are somewhat conflicting (19)(20)(21). Large scale analyses of perforated cells have identified multiple additional changes taking place in response to PFT (15,20,22,23), many of which appear to be triggered by the drop of cytosolic potassium (11,20,24,25). Although the contribution of the various pathways to cell autonomous defense against PFT remains to be established in most cases, a basic concept emerges according to which removal of membrane pores (3,10,13,(26)(27)(28)(29)(30)(31)(32) and metabolic homeostasis (10,12,13,20,33,34) are cornerstones of early cell autonomous defense against PFT. Importantly, mechanisms involved in pore removal depend on PFT and cell type (4,8,29). It appears that MAPK p38 and autophagy are required if the recovery process is prolonged as with S. aureus α-toxin and aerolysin (8,20).
Phosphorylation of eukaryotic translation initiation factor 2 α (eIF2α) is a conserved stress-response activated by various PFT (12,13,20,(33)(34)(35). How this pathway impacts survival of target cells remains incompletely understood. Eukaryotic translation initiation requires assembly of a 43S ternary pre-initiation complex, consisting of met-t-RNAi(Met), eIF2, and GTP. In mammalian cells, this step is controlled through phosphorylation of eIF2α at serine 51 by 1 of 4 eIF2α-kinases (GCN2, PERK, PKR, and HRI), which respond to different types of stress (36). GCN2 serves as nutritional sensor, which is activated by uncharged t-RNAs (37,38). Several lines of evidence indicate that membrane stress triggers this pathway: first, mutations that affect vesicular transport in yeast trigger phosphorylation of eIF2α (39). Second, in mammalian cells, plasma membrane perforation by bacterial PFT leads to activation of GCN2 (12,33), phosphorylation of eIF2α, transient attenuation of translation, and activation of autophagy (12,13,20). Also, membrane damage by chlorpromazine or detergent triggers GCN2 (40). S. aureus αtoxin inhibited uptake of leucine by cells, providing an explanation for activation of GCN2 in target cells of PFT (12). In human epithelial cells, eIF2α, eIF2α-kinases and the regulatory eIF2α-phosphatase subunit CReP/Ppp1r15B are all required for efficient recovery from α-toxin-dependent loss of ATP (13). Surprisingly, these proteins served to remove membrane pores, thus, linking control of translation initiation and membrane traffic (13).
Conspicuously, many rodent cell types are not affected even by comparably high concentrations (micromolar range) of αtoxin; but the cause is not known. Receptor density on murine cells might be low, or murine ADAM10 might be an inefficient receptor as compared to its human counterpart. Alternatively, murine cells might be particularly tolerant to the consequences of successful attack. At any rate, to better understand results of in vivo experiments with S. aureus or α-toxin in mice, it is important to comprehend the mechanisms underlying tolerance of murine cells.
In ecoimmunology, "tolerance" denotes the ability of an organism to cope with high-pathogen load and resulting damage (41)(42)(43). To elucidate mechanisms of tolerance, it will be important to investigate the phenomenon at the cellular level using defined noxious agents. Here, we focus on cellular tolerance to PFT. A PFT may fail to cause overt damage to a cell if it cannot bind to, or attack, membranes in the first place. Alternatively, target cells may be able to cope with membrane perforation. In both cases, we consider the target cell "tolerant to the PFT. " The term "susceptibility" shall denote responsiveness of a cell to a PFT as measured by loss of ATP or loss of potassium ions from the cytosol.
In the present work, we have investigated tolerance of mouse cells to S. aureus α-toxin. The results support a broader protective function of MAPK and document a cell type-and toxin-selective role of eIF2α.

Normal Mouse Embryonal Fibroblasts are Tolerant to S. aureus α α α-Toxin
Exposure of human keratinocytes (HaCaT) to nanomolar concentrations of α-toxin for 2 h leads to significant loss of ATP. In contrast, murine keratinocytes (PDV) or mouse embryonal fibroblasts (MEFs) appeared to be not susceptible to α-toxin ( Figure 1A). In contrast, MEFs were exquisitely susceptible to pVCC (Figure 1B), another PFT of the small β-barrel family (44).

Inhibitors of Various Signaling Pathways Sensitize MEFs for α α α-Toxin
One potential explanation for selective tolerance of MEF to αtoxin could be the lack of an appropriate receptor, or lower expression levels of receptor. Amino acid sequences of human and murine ADAM10, the proposed high-affinity receptor of αtoxin, are not identical. Western blots with an antibody against ADAM10 yielded weaker bands with the murine cells ( Figure 1C). Although it cannot be excluded that the antibody binds human and murine cells with different efficiency, it is therefore possible that qualitative or quantitative differences in receptor expression could explain lower susceptibility of murine vs. human cells to α-toxin. Alternatively, however, efficient ongoing repair of membrane damage and fitness to cope with metabolic stress could also play a role. Therefore, we tested a panel of small molecular weight compounds that we knew to inhibit recovery from α-toxindependent ATP-depletion in HaCaT cells. Several (combinations of) inhibitors sensitized MEFs for α-toxin (Figure 2A). The effect depended on the concentration of inhibitor, as exemplified for JNK3XIISR3576 ( Figure 2B). Although JNK3XIISR3576 is selective for JNK3 if applied at nanomolar concentrations, MAPK other than JNK3 seem to be involved here because MEFs do not express JNK3 (45). As shown for Sal/Dyn, inhibitors sensitized wild-type MEF for α-toxin, provided they expressed ADAM10 ( Figure 2C). Thus, inhibitors did not sensitize cells by increasing unspecific toxicity of α-toxin.
First, we compared ATP levels in these MEFs to assess metabolic perturbation after treatment with α-toxin. EeF2K −/− cells were not susceptible to 10 µg/ml of α-toxin and thus behaved like wild-type MEFs. In contrast, lack of GCN2 or Ppp1r15B both sensitized cells to α-toxin. The strongest effect was observed with eIF2α S51A/S51A MEFs, which lost up to~80% of ATP ( Figure 3A). Although α-toxin led to dose-dependent loss of ATP in eIF2α S51A/S51A cells, there was barely any effect on wild-type cells ( Figure 3B). Notably, pVCC-dependent loss of ATP was similar in wild-type or mutant MEFs ( Figure 3C). Next, we determined the frequency of sub-G1 events, a measure of DNA fragmentation, in cells treated with α-toxin for 48 h. No significant difference was observed between wild-type and eIF2α S51A/S51A cells in the presence of β-ME and non-essential amino acids (NEAA). However, without these supplements, the number of sub-G1 events was doubled in eIF2α S51A/S51A cells ( Figure 3D), although toxin-dependent loss of ATP and phosphorylation of eIF2α were equal in media with or without additives (data not shown). Additives thus appear to protect cells from secondary damage.

GCN2/EIF2AK4 Contains α α α-Toxin-Dependent Stress
Next, we assessed basal or toxin-dependent eIF2αphosphorylation in the various MEFs lines. In wild-type cells, some basal phosphorylation of eIF2α was noted, which was moderately increased by α-toxin; after normalization for eIF2α, the effect was, however, statistically insignificant. As expected, P-eIF2α was not detected in eIF2α S51A/S51A MEFs ( Figure 4A). No P-eIF2α was also discerned with samples of untreated GCN2 −/− cells; this indicated that basal phosphorylation of eIF2α in cultured MEFs depends on GCN2. Paradoxically, however, α-toxin-dependent phosphorylation of eIF2α was enhanced in cells lacking this eIF2α-kinase. This showed that increased susceptibility of GCN2 −/− MEFs to α-toxin ( Figure 3A) cannot be accounted for by diminished toxin-dependent eIF2αphosphorylation. Membranes were re-probed with antibodies against p38, which becomes phosphorylated in response to PFT. Strikingly, α-toxin caused robust phosphorylation of stress activated protein kinase p38 in GCN2 −/− MEFs, Ppp1r15B MEFs, and eIF2α S51A/S51A MEFs ( Figure 4A), but not in wild-type MEFs, supporting the notion that cells with imbalanced eIF2αphosphorylation experienced more severe toxin-dependent stress. p70S6K, substrate of mTORC1 became de-phosphorylated ( Figure 4A), indicating that α-toxin inhibits mTORC1, master regulator of translation and autophagy.

CHX Does Not Tolerize eIF2α α α S51A/S51A MEFs to α α α-Toxin
Results from Western blots ( Figure 4A) raised the question whether α-toxin impacts translation differentially in eIF2α S51A/S51A vs. wild-type MEFs. Paradoxically, treatment with α-toxin caused dose-dependent attenuation of translation in eIF2α S51A/S51A MEFs, but had no such effect on wildtype MEFs ( Figure 4B). This suggested that translation was inhibited in response to α-toxin through a mechanism that was independent of eIF2α-phosphorylation. Together with results from the foregoing ATP-assays, this also revealed that attenuation of translation per se is insufficient to maintain

Wild-Type eIF2α α α Modulates Binding and Action of α α α-Toxin
That inhibition of translation did not protect eIF2α S51A/S51A MEFs provoked the question how phosphorylation of eIF2α tolerizes MEF. So, we investigated a primary event underlying many of the rapid molecular changes induced by PFT, i.e., disturbance of natural ion gradients (4). Loss of potassium appears to be one major trigger (11,20,24,49). We measured loss of intracellular potassium by flame photometry. In line with results from ATP-assays, net loss of potassium was enhanced in GCN2 −/− or Ppp1r15B −/− cells and even more so in eIF2α S51A/S51A MEFs (Figure 5A), suggesting that membrane damage was more severe in cells with deficiencies in regulation of eIF2α-phosphorylation.
Pore formation by α-toxin depends on oligomerization and insertion into the plasma membrane. Because oligomers resist SDS, they can be detected by SDS-PAGE. Using cell surface labeling after incubation with internally radio-labeled α-toxin, we compared the amount of α-toxin on the surface of wild-type MEFs and eIF2α S51A/S51A MEFs. The fluorographic analysis shown in Figure 5B reveals that more α-toxin is present at the surface of eIF2α S51A/S51A MEFs as compared to wild-type cells, providing a straightforward explanation for enhanced loss of potassium and ATP from these cells.

Wild-Type eIF2α α α Modulates Expression of ADAM10
Increased amounts of toxin associated with eIF2α S51A/S51A MEFs could be due to higher abundance of α-toxin receptors. Therefore, we compared ADAM10 expression in eIF2α S51A/S51A MEFs and wild-type MEFs. Western-blot analysis revealed that ADAM10 is over-expressed in eIF2α S51A/S51A MEFs; and more ADAM10 is exposed on the cell surface of eIF2α S51A/S51A MEFs as compared to wild-type cells (Figure 5C). No significant differences between wild-type cells and eIF2α S51A/S51A MEFs were found in a lipidomics analysis (data not shown). Treatment with α-toxin led to down-regulation of ADAM10 at the cell surface of both wildtype and mutant cells. Because basal eIF2α-phosphorylation in MEFs depends on GCN2 (Figure 4A), the collective data indicate that nutrient stress or basal activity of GCN2 modulates levels of ADAM10 in MEFs.

Role of GCN2 for Tolerance to α α α-Toxin is Not Conserved in BMDM
Like MEFs, BMDM proved to be highly tolerant to α-toxin although they were susceptible to Vibrio cholera cytolysin (VCC) (Figure 6A). However, lack of GCN2-expression in BMDM did not significantly alter susceptibility to α-toxin (Figures 6A,D); α-toxin-dependent phosphorylation of eIF2α appeared slightly reduced ( Figure 6B). Further, infection of BMDM by S. aureus was equally efficient with both strains whether or not bacteria produced toxin or not ( Figure 6C). Therefore, GCN2 seemed to play no major role for resistance or tolerance to α-toxin of BMDMs.
Small MW inhibitors of JNK and dynamin, a cocktail, which efficiently breaks tolerance to α-toxin in MEFs, moderately sensitized BMDM to purified α-toxin ( Figure 6D). Inhibitors did not enhance ADAM10 expression ( Figure 6E). Thus, MAPK enhance natural tolerance of BMDM to α-toxin, but modulation of ADAM10 expression appears not to be involved.

Discussion
One conclusion of this work is that regulators of translation initiation (eIF2α, GCN2, and Ppp1r15B) render murine embryonic fibroblasts tolerant to S. aureus α-toxin. This is consistent with our previous finding in human epithelial cells that these proteins promote recovery from successful attack (13). Although it remains to be investigated whether a priori tolerance of MEFs to αtoxin is likewise based on efficient endocytic removal of α-toxin pores from the cell surface, the present results document that the tolerogenic effect of eIF2α, GCN2, and Ppp1r15B is conserved in mice and man, and that it is observable in fibroblasts. That both lack of an eIF2α-kinase and of constitutive eIF2α-phosphatase reduce tolerance to α-toxin could be interpreted to show that balanced phosphorylation of eIF2α, or cycling of eIF2α between phosphorylated and unphosphorylated state is required for the tolerogenic effect.
A role of translational regulation for various aspects of innate immunity, and a protective function against PFT has been discussed in the recent literature (34,50,51). One established function of eIF2α-phosphorylation and the integrated stress-response is to reprogram expression of genes, including of genes that regulate immunity. Translational attenuation in host cells might

FIGURE 4 | Continued
Cycloheximide does not sensitize wild-type MEFs (A) MEFs wt, MEFs eIF2α S51A/S51A , MEFs GCN2 −/− , or MEFs Ppp1r15b −/− were treated or not with 10 µg/ml α-toxin for 2 h. Cell lysates were analyzed by Western blot for P-eIF2α, eIF2α, P-p70S6 kinase, p70S6 kinase P-p38, and p-38. One of three similar blots is shown; lower panel summarizes densitometric data for (P)-eIF2α, mean ± SE; n = 3 (B) MEFs wt or MEFs eIF2α S51A/S51A cells were treated or not with indicated concentrations of α-toxin and incubated for 1 h at 37°C. Treatment with CHX served as positive control for translational arrest. Subsequently, the cells were incubated for 1 h at 37°C with 10 µg/ml puromycin, which incorporates during ongoing synthesis into nascent proteins. Eventually, cells were analyzed by Western blot for puromycin. (C) MEFs wt or MEFs eIF2α S51A/S51A cells were treated or not with 10 µg/ml α-toxin and incubated with or without CHX at 37°C. Cellular ATP levels were determined after 2 h; data are percent of untreated controls; mean values ±SE; n = 5.
also help them to conserve energy, as proposed by some authors (20). However, (P)-eIF2α-dependent tolerance of MEF could not be explained by toxin-dependent translational attenuation per se: actually, α-toxin caused inhibition of translation in eIF2α S51A/S51A MEFs, but not in wild-type MEFs, and CHX did neither protect nor hyper-sensitize wild-type cells from/for α-toxin. Yet, eIF2α S51A/S51A MEFs proved to be significantly less tolerant to αtoxin. Probably as a consequence, protein synthesis was halted through alternative pathways. This could occur, for instance, by deactivation of mTOR, as indeed suggested by α-toxin-dependent dephosphorylation of p70S6K, a target of mTOR. The data may reflect a hierarchy of stress responses: Activation of eIF2α via nutritional sensor GCN2 contains α-toxin-dependent damage and stress, which would otherwise lead to exaggerated hyperphosphorylation of eIF2α by (an)other eIF2α-kinase(s), possibly PERK and PKR. If eIF2α-phosphorylation fails, translation would be halted via robust deactivation of mTOR. Sustained inhibition of translation is obviously not tolerated by cells (35), but marked α-toxin-dependent inhibition of translation, observed in eIF2α S51A/S51A MEFs, does not explain the increased loss of ATP and potassium from these cells, because CHX did not affect these parameters in wild-type cells.
The fact that lack of phosphorylatable eIF2α was associated with increased toxin-dependent loss of potassium and ATP prompted us to measure the binding of α-toxin and expression of ADAM10, the proposed high-affinity receptor for α-toxin (52). This led to the unexpected finding that ADAM10 levels were significantly higher in eIF2α S51A/S51A cells. Therefore, tolerance to α-toxin in MEFs may be well due to GCN2/P-eIF2α-dependent modulation of its receptor, ADAM10. Whether or not the apparent link between basal nutrient stress and expression of ADAM10 was shaped by co-evolution of S. aureus and humans, it may also bear on functions of ADAM10 that are not related to infection.
How basal phosphorylation of eIF2α-levels modulates ADAM10-expression remains to be elucidated. Because P-eIF2α is required for starvation-dependent autophagy (53), a possible role of eIF2α in this context could be to maintain basal autophagic flux, which in turn could impact ADAM10 levels. Recently, Atg16L1, a protein essential for autophagy, has been shown to confer tolerance to α-toxin; the authors proposed that autophagy constitutively dampens ADAM10 levels in a cell-type selective manner (54). Similarly, eIF2α-dependent modulation of ADAM10 shown in the present work is constitutive and cell-type selective. Together this seems to suggest that basal eIF2α-phosphorylation functions upstream of autophagy to mediate tolerance. Alternatively, P-eIF2α, eIF2α-kinases, and -phosphatases could function through mechanisms acting in parallel to autophagy, for instance, by regulating endocytosis of membrane pores, as has been shown in epithelial cells (13). Whatever the effector mechanism(s) downstream of eIF2α are, our data reveal that GCN2-dependent basal phophorylation of eIF2α in MEF modulates ADAM10 levels as well as binding and action of α-toxin. Because GCN2 is activated by low levels of amino acids in cells, basal nutrient stress might be the driving force; the potential links discussed here are summarized in a model (Figure 7). Inhibitors of various pathways hyper-sensitized MEFs to S. aureus α-toxin, which supports the notion that multiple signaling pathways are required to confer cellular tolerance to α-toxin. This raises the possibility that drugs used in pharmaco-therapy may have potential tolerance-modulating effects, an issue that warrants further investigation. The present data support a tolerogenic role of JNKs in both MEFs and BMDM. More work is required to understand the underlying mechanisms, but they seem to be distinct from eIF2α-dependent cellular tolerance to α-toxin. Conserved stress responses may fail to protect against some PFT, as exemplified here by V. cholerae cytolysin. Whether cells of the immune system tolerate attack of a given PFT will co-determine the ability of an organism to mount an effective "classic" immune response to infection with corresponding bacteria.

Chemicals
RNase, propidium iodide, and puromycin were purchased from Sigma. NHS-fluorescein and EZ-Link Sulfo-NHS-LC-biotin were from Thermo Fisher Scientific and Streptavidin-Alexa 647 was obtained from Molecular Probes. Rapamycin was from Calbiochem.

Puromycin Incorporation
The assay has been described elsewhere (62). MEFs eIF2α S51A/S51A cells were seeded at a density of 2 × 10 5 cells/well into six-well plates. Cells were incubated as indicated with several concentrations of pVCC or α-toxin for 1 h at 37°C. Thereafter, 10 µg/ml puromycin was added and cells were incubated for an additional hour at 37°C. Subsequently, medium was removed; cells were washed with PBS and analyzed by Western blot using anti-Puromycin-antibody in 2.5% (w/v) skim milk in TBST.

ATP-Measurements
Measurements of intracellular ATP were performed, as described elsewhere (22).

Potassium Efflux Measurements
For measurements of intracellular potassium, cells were seeded at a density of 3 × 10 5 cells/well into six-well plates. Cells were incubated or not with 8 µM Rapamycin for 3 h before adding 10 µg/ml α-toxin. After 2 h incubation at 37°C, cells were washed three times on ice with cold Choline-chlorid-buffer and solubilized in 2 ml Choline-chlorid-buffer containing 0.5% Triton X-100 (63) for 30 min at room temperature under constant shaking. Potassium content of supernatants of centrifuged cell lysates was determined using Sherwood single channel flame photometer M401.

Quantification of Internalized Bacteria
The assay has been performed, as described elsewhere (64).

Sub-G1 Analysis
Cells were seeded at a density of 1 × 10 5 cells/well into six-well plates and treated or not with 10 µg/ml α-toxin permanently for 48 h at 37°C. Cells were harvested including the detached cells in the supernatant, washed in PBS and centrifuged. Cells were resuspended and fixed in cold 70% ethanol in PBS/EDTA for 2 h at 4°C. Subsequently, the cells were washed and treated with RNase for 10 min at 37°C. Then, propidium iodide was added and cells were incubated for 5 min at room temperature. Eventually, cells were analyzed using a FACScan flow cytometer (BD Biosciences) and CellQuest software.

Statistical Analysis
Two-sided, unpaired Student's t-test was employed to assess statistical significance of differences between mean values. Significance was assumed at p < 0.05.