The protective antigen (PA) of the anthrax toxin binds to a cell surface receptor and thereby allows lethal factor (LF) to be taken up and exert its toxic effect in the cytoplasm. Here, we report that clustering of the anthrax toxin receptor (ATR) with heptameric PA or with an antibody sandwich causes its association to specialized cholesterol and glycosphingolipid-rich microdomains of the plasma membrane (lipid rafts). We find that although endocytosis of ATR is slow, clustering it into rafts either via PA heptamerization or using an antibody sandwich is necessary and sufficient to trigger efficient internalization and allow delivery of LF to the cytoplasm. Importantly, altering raft integrity using drugs prevented LF delivery and cleavage of cytosolic MAPK kinases, suggesting that lipid rafts could be therapeutic targets for drugs against anthrax. Moreover, we show that internalization of PA is dynamin and Eps15 dependent, indicating that the clathrin-dependent pathway is the major route of anthrax toxin entry into the cell. The present work illustrates that although the physiological role of the ATR is unknown, its trafficking properties, i.e., slow endocytosis as a monomer and rapid clathrin-mediated uptake on clustering, make it an ideal anthrax toxin receptor.
Abbreviations used in this paper: ATR, anthrax toxin receptor; ATR-HA, COOH-terminal HA epitope-tagged version of ATR; β-MCD, β-methyl cyclodextrin; DRM, detergent-resistant membrane; EF, edema factor; LF, lethal factor; PA, protective antigen; PA63, COOH-terminal 63-kD moiety; PA83, protective antigen of 83-kD; PASNKE, PA mutated in the consensus furin cleavage site.
PA is an 83-kD protein (PA83) that binds to a widely expressed, 368 amino acid, type 1 membrane protein termed anthrax toxin receptor (ATR; Bradley et al., 2001). Receptor-bound PA is then cleaved by members of the furin family of proteases, causing release of an NH2-terminal 20-kD fragment and leaving the COOH-terminal 63-kD moiety (PA63) bound to ATR. It is important to stress that furin cleavage occurs at the cell surface, even though this enzyme is more abundant intracellularly and in particular, in the TGN (Chaudry et al., 2002; Mourez et al., 2002). Unlike PA83, PA63 can oligomerize to form ring-shaped heptamers (Petosa et al., 1997). Interaction of LF and EF with PA63 occurs at the cell surface after heptamerization has occurred (Singh et al., 1994; Mogridge et al., 2001; Cunningham et al., 2002). The complex of PA63 with LF and/or EF is then internalized and transported to endosomes where the low pH triggers membrane insertion of the PA63 heptamer and channel formation (Milne et al., 1994; Mourez et al., 2002). Delivery of EF and LF to the cytosol is concomitant with PA63 channel formation and may involve passage of these proteins through the channel. Once in the cytoplasm, LF and EF modify their respective targets.
A crucial step in the mode of action of anthrax toxin that has received surprisingly little attention is the initial entry. Interestingly, PA63 is endocytosed, whereas the precursor PA83 remains at the cell surface (Beauregard et al., 2000). Here, we have analyzed the mechanism that triggers the specific cellular uptake of PA63 and thereby of the enzymatic units, LF and EF.
Results And Discussion
We investigated whether the selective uptake of PA63, and not of PA83, was due to a change in surface distribution on conversion of PA83 to PA63. The similarity between the structure and the mode of action of PA and that of certain bacterial pore-forming toxins such as aerolysin (Abrami et al., 2000) prompted us to determine whether PA63 was associated with raftlike lipid microdomains of the plasma membrane. These domains are thought to form through lateral movement and assembly of cholesterol and glycosphingolipids. A specific subclass of rafts form flasklike invaginations at the plasma membrane and are then called caveolae (Simons and Ikonen, 1997; Brown and London, 1998). Rafts act as surface platforms in signal transduction, cholesterol homeostasis, and endocytosis (Brown and London, 1998; Simons and Toomre, 2000). Lipid rafts have also been implicated in various infectious processes (Fivaz et al., 1999), and in particular, were shown to favor heptamerization of the pore-forming toxin aerolysin (Abrami and van der Goot, 1999) via mechanisms that could well apply to PA. One biochemical characteristic of rafts is their resistance to nonionic detergents at 4°C, which allows their purification on density gradients. Native, full-size PA83 was associated with detergent-soluble domains of the plasma membrane (Fig. 1
A) in agreement with previous observations (Beauregard et al., 1999), as was a variant of PA mutated in the consensus furin cleavage site (PASNKE; Gordon et al., 1995). In contrast, PA63 (which can be obtained either in vitro by trypsin cleavage or in vivo by cell surface furin processing) was almost exclusively associated with detergent-resistant membranes (DRMs; Fig. 1 A). To confirm that the presence of PA63 in more buoyant, upper fractions of the gradient, is indeed due to its association with lipid rafts, we investigated whether the distribution of PA63 was altered by removal of plasma membrane cholesterol using β-methyl cyclodextrin (β-MCD) or sequestration of cholesterol using filipin (Simons and Toomre, 2000). As shown in Fig. 1 B, PA63 was partially (β-MCD) or completely (filipin) redistributed to the bottom of the gradient, in a manner similar to that of the normally raft-associated protein flotillin-1 (Bickel et al., 1997), but in contrast to the caveolar protein caveolin-1 (Kurzchalia and Parton, 1999). Because rafts are also rich in sphingomyelin, we analyzed whether reduction of sphingomyelin levels would affect the surface distribution of PA63. In sphingomyelin-deficient mutant CHO cells (Hanada et al., 1998), PA63 was no longer associated with DRMs (Fig. 1 B). Flotillin-1 was also partially redistributed to detergent-soluble fractions, in contrast to caveolin-1, which remained at the top of the gradient. Thus, PA63 is associated with noncaveolar cholesterol and sphingolipid-rich domains of the plasma membrane.
The fact that PA83, which is the form of the toxin that initially encounters the cell, was not raft-associated suggested that the PA receptor (ATR) is not itself a raft resident protein. Because antibodies against ATR are not available, we stably expressed a COOH-terminal HA epitope–tagged version of ATR (ATR-HA) in CHO PR230, a spontaneous ATR-deficient mutant derived from CHO-K1 cells (Liu and Leppla, 2003). As predicted, the receptor was solubilized by the detergent (Fig. 1 C), indicating that ATR has little, if any, affinity for rafts. Therefore, although aerolysin and other toxins owe their rafts association to the fact that their receptors reside in these microdomains (Abrami and van der Goot, 1999), PA appears to force the raft association of its receptor that is normally in the glycerophospholipidic, nonraft region of the membrane.
The fact that only the mature PA63 form can promote raft association suggests that it might be heptamerization of PA63 that leads to clustering of ATR, which in turn would increase raft partitioning of the receptor. If so, PA63 heptamers should be associated with cell surface rafts.
To probe for cell surface heptamers, we measured binding of LF to cells treated with PA63 at 4°C because LF interacts with heptameric, but not with monomeric PA63 (Mogridge et al., 2001; cell surface PA63 heptamers cannot be revealed by SDS-PAGE because they dissociate in SDS). As shown in Fig. 1 D, surface-bound LF was indeed associated with DRMs of PA63-treated cells (but was not detectable on cells incubated with PASNKE), indicating that heptameric PA63 and its ligand LF are predominantly present in lipid rafts.
Our hypothesis also predicts that clustering ATR by means other than PA heptamerization, for example by cross-linking ATR using a sandwich of primary and secondary antibodies (Harder et al., 1998), should lead to raft association. Not having anti-ATR antibodies, ATR was labeled with PA83, which cannot heptamerize unless processed, and thus clustering at 4°C was induced with antibodies against PA. Whether a monoclonal or a polyclonal antibody was used, the clustered PA83 was predominantly present in DRMs, in contrast to the nonclustered PA83 (Fig. 2
A). That antibody clustering modified the surface distribution of ATR was confirmed by fluorescence microscopy. As shown in Fig. 2 B, a very punctate and intense signal was observed on cross-linking, whereas on cells that had been fixed before the addition of antibodies, the staining was weaker and more diffuse (note that some punctate structures can be observed that could well be due to cross-linking after PFA fixation as previously observed for GPI-anchored proteins; Maxfield and Mayor, 1997). The antibody-induced DRM association was cholesterol-dependent, as shown by the sensitivity to filipin and β-MCD (Fig. 2 C), strengthening the notion that lipid rafts control this process. Importantly, clustering of PA83 also led to the redistribution of a population of ATR-HA to DRMs (Fig. 2 D ; the majority of ATR-HA remains in detergent soluble fractions because of the vast excess of receptor with respect to PA, and because intracellular receptors are unavailable for PA63 binding). We further confirmed the raft association of clustered PA83 by cellular localization studies. As a typical raft marker, we used antibody-clustered GPI-anchored proteins (Harder et al., 1998) that were labeled using an inactive mutant of aerolysin, a specific probe for this class of proteins (Fivaz et al., 2002). As shown in Fig. 2 E (arrows), we found strong colocalization with GPI-anchored proteins not only for PA63, as expected, but also for clustered wild-type PA83 and PASNKE, consistent with our biochemical analysis. However, we could not see any significant colocalization at the cell surface of clustered PA83 with caveolin-1 (Fig. 2 F), strengthening the notion that PA associates with noncaveolar rafts (Fig. 1). The above experiments show that clustering of ATR, either by PA63 heptamerization or by antibody cross-linking, promotes its partitioning into rafts (Fig. 2 G).
Next, we investigated whether ATR clustering is also a trigger for internalization. Again, we made use of the PA variant that cannot be processed by furin because such mutants remain at the cell surface (Beauregard et al., 2000), and investigated whether antibody clustering would promote internalization. Cells were incubated with PASNKE at 4°C, and a sandwich of primary and secondary antibodies was then added (Fig. 3
A, middle and bottom rows) or not (Fig. 3 A, top row) at 4°C. Cells were then shifted to 37°C to allow internalization (Fig. 3 A, top and middle rows) or kept on ice (Fig. 3 A, bottom row). Control cells (PASNKE with no antibodies; Fig. 3 A, top row) were shifted back to 4°C and were then treated with primary and secondary antibodies. Finally, cells were submitted or not to an acid wash in order to remove surface-bound antibodies and to reveal only the intracellular PASNKE. Intracellular PASNKE could only be detected when antibody cross-linking was performed before the internalization step (Fig. 3 A, middle row), demonstrating that clustering of PA83 is necessary and sufficient to promote its cellular uptake. When cells were treated with β-MCD before PA and antibody addition, intracellular accumulation was significantly reduced (Fig. 3 B).
To evaluate the physiological relevance of the role of rafts in the delivery of PA to endosomes, we measured the effect of cholesterol depletion on the appearance of the SDS-resistant PA63 heptamer because this form only appears in an acidic environment. As shown in Fig. 4
A, formation of the membrane-inserted SDS-resistant heptamer was strongly delayed on β-MCD treatment.
Finally, to emphasize the physiological relevance of the association of the PA63–LF complex with lipid rafts, we analyzed the effects of β-MCD on the kinetics of cleavage of MEK1, one of the intracellular MAPK kinases targeted by LF. Processing of MEK1 by LF can be easily followed by Western blotting because it is accompanied by the loss of NH2-terminal epitopes due to the removal of seven amino acids form the NH2 terminus (Duesbery et al., 1998). In control cells, NH2-terminal cleavage of MEK1 was observed with time, whereas total MEK1, detected with an anti-COOH-terminal antibody, remained constant (Fig. 4 B). In marked contrast, MEK1 remained intact within the time frame studied in β-MCD–treated cells (Fig. 4 B). We confirmed, as a control, that similar amounts of LF were associated with control and β-MCD–treated cells (Fig. 4 B). Thus, LF cannot reach its intracellular target if lipid rafts are disrupted with the drug.
The fact that internalization of the ATR was both ligand-triggered and raft-dependent raised the possibility that caveolin-dependent endocytosis might be involved (Pelkmans and Helenius, 2002). However, the observation that PA63 and caveolin-1 showed different behaviors on cells at 4°C (Fig. 1 B and Fig. 2 F) and the fact that caveosomes are neutral and do not appear to communicate with any acidic organelle (Pelkmans and Helenius, 2002; knowing that the PA63 heptamer requires an acidic pH for membrane insertion) argue against a caveolar mediated uptake. To clarify this issue, we investigated whether expression of a dominant-negative mutant of caveolin-1, namely caveolin-1 with the GFP fused to its NH2 terminus (GFP-Cav1; Pelkmans et al., 2001), would affect anthrax toxin entry. The kinetics of appearance of the SDS-resistant PA63 heptamer (Fig. 5
A) as well as the kinetics of MEK1 cleavage (unpublished data) were very similar in control and GFP-Cav1–expressing cells. Also, we found little colocalization between 30-min internalized antibody-clustered PA83 and caveolin-1-GFP, used here as a marker of caveolae and caveosomes (large green structures seen in Fig. 5 B, which were never positive for PA; note that fusion of GFP to the COOH terminus of caveolin-1 does not lead to a dominant-negative phenotype; Pelkmans et al., 2001). Finally, to further rule out a caveolin-mediated entry pathway and transport to caveosomes, we found that intracellular SDS-resistant PA63 heptamers also form in macrophages RAW264.7 and FRT cells; two cell types that do not express caveolin-1 (Zurzolo et al., 1994; Kiss et al., 2000; unpublished data), in agreement with the fact that macrophages are sensitive to LF (Tang and Leppla, 1999).
As shown in Fig. 5 C, significant internalization of antibody-clustered PASNKE occurred within 5 min, and a similarly short time was required to observe the SDS-resistant PA63 heptamer (Fig. 5 D). Subsequent degradation of PA63 heptamers was apparent within 30 min (Fig. 5 D). These rapid kinetics suggest that initial entry of PA might be clathrin-mediated. First, we investigated whether we could detect clustered PASNKE in clathrin-coated structures by electron microscopy, after warming the cells for 5 min at 37°C. As shown in Fig. 5 E, gold particles could be detected both in clathrin-coated pits and vesicles (Fig. 5 E, a–c) as well as sometimes in nonclathrin-coated invaginations of the plasma membrane (Fig. 5 E, d), but never in grape-like caveosomes (not depicted). Next, we investigated whether PASNKE was transported to endosomes. As shown in Fig. 5 F, colocalization was found between intracellular PASNKE and internalized transferrin (5 min), as well as internalized dextran (15 min), indicating early endosomal localization. Next, we investigated whether the GTPase dynamin was involved in anthrax toxin uptake. Dynamin has been implicated in both internalization of caveolae and in clathrin-mediated endocytosis (for review see McNiven et al., 2000). We made use of a well-described HeLa cell line that can be induced to express either wild-type dynamin-1 or a mutant defective in GTP binding and hydrolysis (DynK44A; Damke et al., 1994). As shown in Fig. 5 G, appearance of the SDS-resistant PA63 heptamer was severely delayed in DynK44A-expressing cells when compared with DynWT, as was the cleavage of MEK1 (unpublished data), showing that PA is preferentially internalized via a dynamin-dependent pathway. Because we have excluded the caveolar uptake pathway, these results suggest that clathrin is involved. To inhibit the clathrin-dependent pathway more specifically, we overexpressed a dominant-negative mutant of Eps15 (EΔ95/295), a protein required for clathrin-mediated endocytosis that binds directly to the plasma membrane adaptor AP2 (Benmerah et al., 1999). Transient transfection of HeLa cells was performed using either wild-type or dominant-negative Eps15 fused to GFP. Transfection rates of ∼75% were reached, and it was apparent when measuring uptake of rhodamine transferrin that very high expression levels of dominant-negative Eps15 (followed by GFP fluorescence) were required to inhibit clathrin-mediated uptake (unpublished data); levels that were reached in only few cells. Despite this, we have analyzed biochemically the kinetics of appearance of the SDS-resistant PA63 heptamer. As shown in Fig. 5 H, appearance of the SDS-resistant heptamer was significantly delayed, as was the subsequent degradation of the complex, even within this population of HeLa cells with mixed levels of inhibition of the clathrin pathway. Together, the inhibitory effects of dominant-negative mutants of dynamin and Eps15 show that the anthrax toxin preferentially enters the cell via a clathrin-dependent pathway.
The behavior of ATR is reminiscent of that of the B cell receptor, which undergoes ligand-dependent clustering and raft association (Cheng et al., 1999), and is subsequently internalized via a clathrin-dependent mechanism (Stoddart et al., 2002). Further studies will be required to determine whether ATR moves laterally out of rafts before endocytosis or, more likely, whether clathrin directly associates with ATR-containing rafts as for the B cell receptor.
The physiological role of ATR, as well as that of its splicing variant tumor endothelial marker 8 (St. Croix et al., 2000), is unknown. However, the present work highlights why anthrax toxin has highjacked this surface protein in particular. It indeed appears that independently of its function, ATR has a number of properties that make it the ideal anthrax toxin receptor. For a successful intoxication, PA requires a receptor that keeps it at the cell surface as long as it is monomeric, but drags it into the cell as soon as heptamerization, with concomitant binding of LF and/or EF, has occurred. Indeed, if on binding, PA83 were rapidly internalized by the cell, processing by furin could still efficiently occur in endosomes (Chapman and Munro, 1994), but binding of extracellular LF and/or EF could never take place, and the toxin would remain inactive. The initial surface localization of ATR and its low endocytosis rate ensure that cleavage by furin (which is also in the glycerophospholipidic, nonraft region of the membrane; Abrami and van der Goot, 1999) and heptamerization occur before uptake. The ability of ATR to redistribute to lipid rafts and to then be rapidly internalized via a clathrin-dependent mechanism on clustering guarantees the intracellular delivery of the PA heptamer–LF complex. This study clearly shows that subtle changes in the surface distribution of ATR are crucial for its function as an efficient anthrax toxin receptor, and might also be required for its physiological function. Lipid rafts appear to play an important role in regulating the assembly and the intracellular delivery of the anthrax toxin, and therefore constitute a potential therapeutic target.
Materials And Methods
Proteins produced as described previously (Leppla, 1988) include wild-type PA, PASNKE (a non-furin cleavage variant of PA in which the furin cleavage site was replaced by SNKE; Gordon et al., 1995), and LF. The aerolysin mutant (ASSP) was obtained as described previously (Fivaz et al., 2002). pAbs against PA (Liu et al., 2001) and LF were generated in the Leppla lab (Liu and Leppla, 2003). Anti-COOH-terminal MEK1 antibodies were obtained from Santa Cruz Biotechnology, Inc., anti-NH2-terminal MEK1 antibodies obtained from Upstate Biotechnology, anti-caveolin-1 and anti-flotillin-1 mAbs were obtained from Transduction laboratories, and FITC-conjugated secondary antibodies obtained from Molecular Probes, Inc. β-MCD and filipin were obtained from Sigma-Aldrich.
Sphingomyelin-deficient (CHO SPB) and recomplemented (CHO SPB(lcb1)) CHO cells were provided by K. Hanada (National Institute of Infectious Diseases, Tokyo, Japan) and maintained as described previously (Hanada et al., 1997). CHO PR230 is a previously described spontaneous PA receptor–deficient mutant (Liu and Leppla, 2003). Transfection of CHO PR230 with the ATR cDNA restored PA binding. Stable transfectants of CHO PR230 were used expressing human ATR having an HA-tag at the COOH terminus (Liu and Leppla, 2003). HeLa cell lines expressing wild-type and K44A dynamin mutant were maintained as described previously (Damke et al., 1994).
DRMs were prepared from CHO cells using OptiPrep gradients as described previously (Lafont et al., 2002). 7–8 fractions were collected from the top, the total protein content of each fraction was precipitated with 6% TCA in the presence of 375 μg sodium deoxylcholate as a carrier, and were analyzed by SDS-PAGE. For Western blot analysis, gels were transferred onto a nitrocellulose membrane either using a semi-dry system or, in order to detect the PA heptamer, by wet transfer for 5 h at 86 V.
Coverslip-grown CHO cells were incubated with 500 ng/ml PA. Cell surface–bound PA was submitted to clustering by an antibody sandwich. In brief, cells were put on ice and incubated for 20 min with polyclonal anti-PA antibodies followed by 20 min with secondary antibody FITC-anti–rabbit. To remove surface-bound antibodies, live cells were submitted to an acid wash, i.e., incubated for 5 min at 23°C with 140 mM NaCl, 100 mM citrate, pH 2.0. For all experiments, cells were fixed with 3% PFA and then mounted on microscopy slides.
Cells were incubated with 1 μg/ml PASNKE for 1 h at 4°C followed by a sandwich of polyclonal anti-PA, mouse anti–rabbit IgG and 10-nm gold–labeled protein A. Cells were then washed, shifted 5 min at 37°C, and fixed with 2% glutaraldehyde in culture medium buffered with Hepes for 10 min at 37°C and with 2% glutaraldehyde in phosphate buffer for 30 min at RT. After washes, the cells were pelleted in agar and incubated with 2% osmium for 1 h at 4°C and overnight with 0.25% uranylacetate before degradation and Epon embedding.
We would like to thank Kentaro Hanada for providing us with the sphingomyelin-deficient cells, S.L. Schmidt (Scripps Research Institute, La Jolla, CA) for the dynamin mutant cells, A. Dautry-Varsat (Pasteur Institute, Paris, France) for the Eps15 constructs, A. Helenius (ETH Zurich, Zurich, Switzerland) for the Cav1-GFP and GFP-Cav1 constructs, Dana Hsu (National Institutes of Health) for making toxins, and Dan Kolakovsky and Jean Gruenberg for critical reading of the manuscript.
This work was supported by grant 31-61 507.00 of the Swiss National Science Foundation (G. van der Goot).