Cell Surface-expressed Phosphatidylserine and Annexin A5 Open a Novel Portal of Cell Entry*

Expression of phosphatidylserine (PtdSer) at the cell surface is part of the membrane dynamics of apoptosis. Expressed phosphatidylserine functions as an “eat me” flag toward phagocytes. Here, we report that the expressed phosphatidylserine forms part of a hitherto un-described pinocytic pathway. Annexin A5, a phosphati-dylserine-binding protein, binds to and polymerizes through protein-protein interactions on membrane patches expressing phosphatidylserine. The two-dimensional protein network of annexin A5 at the surface prevents apoptotic body formation without interfering with the progression of apoptosis as demonstrated by activation of caspase-3, PtdSer exposure

Programmed cell death eliminates cells from tissues in a silent and non-provocative manner. It is essential for development and tissue homeostasis of the multicellular organism. Apoptosis is the most abundant form of programmed cell death (1). Its well ordered biochemistry produces the hydrolysis of the vital macromolecular structures of the cell (2) and the appearance of a plethora of "eat me" signals at the cell surface (3). The latter causes tethering to and engulfment by phagocytes. Phos-phatidylserine (PtdSer) 1 is the most well explored "eat me" signal. It is expressed at the cell surface very early after the onset of execution of apoptosis, regardless of the cell type and the apoptosis-inducing trigger (4). The binding of expressed PtdSer on the apoptotic cell to the recently described PtdSerreceptor (5) evokes a macropinocytic response in the phagocyte (6). In addition to its involvement in phagocytosis, surfaceexpressed PtdSer constitutes a target for the imaging of apoptosis using annexin A5 in vitro (7,8) and in vivo in animal models (9 -11) and in patients (12,13).
Annexin A5 is a member of an evolutionary conserved multigene family of Ca 2ϩ and phospholipid-binding proteins (14). It binds reversibly to PtdSer-expressing membranes with high affinity in a calcium-dependent manner (15,16) such that ambient extracellular calcium conditions in vivo promote binding to apoptotic cells (17). Binding to PtdSer is accompanied by the formation of a two-dimensional network of annexin A5 molecules through protein-protein interactions (18,19). This phenomenon was explored using model membranes but is believed to occur on the apoptotic cell surface also (20).
Recently, we observed the internalization of annexin A5 by apoptotic cells in vivo. An exploration of this phenomenon led to the discovery of the existence of a novel endocytic pathway that can be activated during apoptosis. This study describes the elucidation of the mechanism of this PtdSer-dependent novel portal of cell entry. We furthermore show that this endocytic pathway can also be activated in living tumor cells. This knowledge opens novel avenues for delivery and cell entry of therapeutic compounds.

EXPERIMENTAL PROCEDURES
Chemicals-Human apo transferrin was purchased from Sigma, and Tris was purchased from Acros Organics.
Cloning, Expression, Purification, and Labeling of Annexin A5, M23, M1234, and Annexin A1-cDNA encoding for the annexin A5 mutant M1234 (E72D, D144N, E228A, and D303N) having four defective calcium binding sites was kindly provided by Prof. Dr. F. Russo-Marie (BIONEXIS-Pharmaceuticals). The M1234 cDNA was recloned to remove the 5Ј-base pairs responsible for the three extra N-terminal amino acids after expression (21). The annexin A5 mutant M23 was generated by subcloning the mutations D144N and E228A into the annexin A5 cDNA. Annexin A5, M23, and M1234 cDNAs were cloned into the bacterial expression vector pET-5a (Novagen). Proteins were expressed in Escherichia coli and purified to homogeneity as assessed by silverstained SDS-PAGE and Western blotting. Annexin A5, M23, and M1234 were labeled with fluorescein-isothiocyanate and Alexa568-succinimidylester according to the manufacturer's protocol (Molecular Probes). The labeled proteins with 1:1 stoichiometry were purified from the mixtures by MonoQ chromatography with Ä cta Explorer (Amersham Biosciences). Annexin A1 was kindly provided by Dr. E. Solito (London, UK). Annexin A1 was labeled with Ä lexa568-succinimidylester according to the manufacturer's protocol. The PtdSer-binding capacity of annexin A1-Alexa568 was verified by ellipsometry.
Analysis of Apoptotic Body Formation-Jurkat cells were washed and resuspended at 10 6 cells ml Ϫ1 in medium 199 (Invitrogen) supplied with 0.5 mM CaCl 2 . Apoptosis was induced with anti-Fas (200 ng ml Ϫ1 ) in the absence or presence of annexin A5 and M1234. The course of apoptosis was determined by flow cytometry using the annexin A5-FITC staining protocol (Nexins). Apoptotic body formation was quantified with flow cytometry and subsequent off-line calculation of the percentage of events with reduced forward and sideward scatter. Offline analysis of the flow cytometer files was performed with WinMDI 2.8 (free share software designed by Joseph Trotter).
PtdSer exposure of anti-Fas-stimulated Jurkat cells was measured by flow cytometry using annexin A5-FITC. If apoptosis was executed in the presence of annexin A5, a mixture of 7.5 g ml Ϫ1 annexin A5 and 2.5 g ml Ϫ1 annexin A5-FITC was added to the cells prior to the addition of anti-Fas.
ROCK-1 activation was determined by lysing Jurkat cells in Tris buffer containing 50 mM Tris, 50 mM EDTA, 0.1% Triton X-100, and 0.5% protease inhibitor mixture (Sigma). Lysates of 10 6 cells were loaded in each well and submitted to SDS-PAGE and Western blotting using anti-ROCK-1 antibody (Santa Cruz).
DNA fragmentation was analyzed by resuspending 5 ϫ 10 6 cells in 600 l of ice-cold lysis buffer (10 mM Tris, 10 mM EDTA, 0.5% Triton X-100, pH 8.0). The mixture was incubated for 10 min on ice. The lysates were cleared by centrifugation (140,000 ϫ g/min, 4°C). DNA was extracted from the supernatant with an equal volume of phenol: chloroform:isoamyl alcohol (16:16:1). The water phase was adjusted to 300 mM NaCl and diluted with 2.5 volumes of ice-cold 100% ethanol. Following overnight precipitation at 4°C, the DNA was pelleted and washed once with ice-cold 70% ethanol. The DNA was dried in a speed VAC and resuspended in 10 l of Tris-EDTA containing RNaseA. The DNA fragments were separated on a 1.2% agarose gel.
HeLa cells were seeded on glass coverslips and left to adhere in culture medium. After 24 h one or a combination of the fluorescentlabeled proteins was added in a concentration of 100 nM. HeLa cells were incubated at 37°C and 5% CO 2 for 3 h. The HeLa cells were washed sequentially with phosphate-buffered saline (PBS), EDTA buffer, and PBS. The cells were then fixed for 15 min in Ca 2ϩ buffer containing 4% paraformaldehyde, pH 8.0, and washed three times for 20 min in PBS.
Plasmid and Transfection-The plasmid caveolin-1-green fluorescent protein was a kind gift of Dr. D. Mundy (University of Texas, Southwestern Medical Center, Dallas, TX). The plasmid was transfected in HeLa cells with the transfection agent FuGENE 6 (Roche Applied Science).
Immunoelectron Microscopy-2 ϫ 10 6 Jurkat cells were stimulated with 200 ng ml Ϫ1 anti-Fas in the presence of 20 g ml Ϫ1 annexin A5. The cells were pelleted by centrifugation (15,000 ϫ g/min) and resuspended in Ca 2ϩ buffer containing 2% paraformaldehyde, 0.2% glutaraldehyde. After 10 min the cells were transferred to Ca 2ϩ buffer containing 1% paraformaldehyde. The fixed cells were washed with PBS, resuspended in 20 l of PBS containing 10% gelatin, and solidified by cooling down. The gelatin drops were stored overnight at 4°C in PBS containing 2.3 M sucrose and then vitrified in liquid nitrogen for subsequent cryosectioning (Ϯ 90 nm) on a Leica EM FCS Cryo-Microtoom. The ultrathin sections were stained with anti-annexin A5 antibody (Hyphen Biomed) and protein A-conjugated with 10 nm gold (a kind gift of Dr. G. Postuma, University Medical Center Utrecht, Department of Cell Biology), washed with PBS containing 0.1% w/v bovine serum albumin, postfixed in 1% glutaraldehyde, rinsed in ice-cold 1.8% methylcellulose, 0.4% uranylacetate, and stained in ice-cold 1.8% methylcellulose, 0.4% uranylacetate for 5-10 min. The grids were air-dried and examined in a Philips CM 10 microscope at 80 keV.
Transmission Cryoelectron Microscopy-Extruded phospholipid vesicles of dioleoylphosphatidylcholine, dioleoylphosphatidylserine, and cholesterol (molar ratio 63.3:3.3:33.3) were prepared by passing a mixture of 10 mM hand-shaken liposomes of the above composition at least three times through a 100-nm filter (Anotop 10, Whatman). The extruded vesicles were submitted to a mild hypotonic shock to obtain 100-nm diameter vesicles with a perfect spherical shape.
Annexin A5 or M23 was added to the vesicles at a weight ratio of 1:4 in the absence of Ca 2ϩ ions. At this ratio maximally 50% of the phospholipid surface can be covered with annexin A5 upon binding (15). Annexin A5 binding to the vesicles was achieved by adding Ca 2ϩ ions to the mixture such that the final Ca 2ϩ ion concentration was 2.5 mM, and the solution became hypertonic relative to the lumen of the vesicles. The vesicles were prepared for cryoelectron microscopy analysis on a Philips CM12 microscope (Philips) using the Vitrobot TM (36). Atomic Force Microscopy-Atomic force microscopy studies were performed following an adaptation of a described procedure (23). Supported lipid bilayers of egg-phosphatidylcholine, dioleoylphosphatidyl-serine, and cholesterol (46:20:34) were made on mica with the vesicle fusion method (24). The presence of a bilayer could be verified by the detection of pinholes in the bilayer. The bilayers were incubated for up to 3 h with the wild type annexin A5 or the M23 (final protein concentrations 15 g ml Ϫ1 ). Samples were imaged with a Nanoscope III using NP probes (both Digital Instruments) with an estimated spring constant of 0.06 N m Ϫ1 . Imaging was done in contact mode in fluid exerting low force (ϳ100 pN). All samples were made and imaged in Hepes buffer (10 mM, 150 mM NaCl, 2 mM CaCl 2 , pH 7.4). Images were low pass-filtered and/or flattened using the Nanoscope software. Ser can be measured with fluorescently labeled annexin A5 and starts rapidly after the onset of apoptosis (4). Short incubations of anti-Fas-stimulated Jurkat cells with annexin A5-Oregon Green result in plasma membrane staining of early apoptotic cells, massively blebbing apoptotic cells, and apoptotic bodies (Fig. 1a). We observed that co-incubation of Jurkat cells with anti-Fas and annexin A5 changed the apoptotic membrane dynamics such that apoptotic body formation was inhibited. This inhibition depends on the annexin A5 dose and its ability to bind to PtdSer, because its non-binding mutant M1234 was without effect (Fig. 1b). Co-incubation of Jurkat cells with anti-Fas and annexin A5 did not affect other parts of the apoptotic program such as caspase-3 activation (Fig. 1c), Ptd-Ser expression (Fig. 1d), DNA fragmentation (Fig. 1e), and ROCK-1 activation (Fig. 1f). The latter is involved in membrane blebbing, which precedes apoptotic body formation (25). Inspection of the co-incubated Jurkat cells with CSLM and transmission electron microscopy revealed apoptotic cells that had internalized annexin A5 in translucent vesicles of various sizes (Fig. 2, a and b). We demonstrated that these vesicular structures are endocytic vesicles and not part of an open cannicular system by (i) washing the cells with EDTA and (ii) incubating them shortly with annexin A5-Oregon Green prior to CSLM analysis. This procedure neither removed annexin A5-Alexa568 from the vesicular structure nor resulted in the co-localization of annexin A5-Alexa568 and annexin A5-Oregon Green. The latter bound to the plasma membrane only. We concluded that annexin A5 is taken up in endocytic vesicles via a PtdSer-dependent mechanism. On the basis of our results so far, we reasoned that annexin A5 first binds to PtdSer expressing membrane patches prone to become blebs and second induces their invagination and budding.

A Novel Pinocytic Pathway Internalizes Annexin A5 into
Cells Executing Apoptosis-The size distribution of the annexin A5 endocytic vesicles suggests that the uptake is a macropinocytic process. This process is preceded by membrane ruffling and dependent on the polymerization of actin (26). We never observed membrane ruffling when Jurkat cells were co-incubated with anti-Fas and annexin A5. The staining of F-actin with phalloidin-TRITC revealed the presence of an actin coat around the annexin A5-containing vesicles (Fig. 3, a-c), suggesting the involvement of the actin cytoskeleton. Latrunculin B (Sigma), an inhibitor of actin polymerization, did not prevent the uptake of annexin A5 in endocytic vesicles (Fig. 3d). These vesicles remained localized underneath the plasma membrane indicating that actin polymerization is involved in intracellular trafficking but is not required for invagination and budding. The actomyosin contractile apparatus is involved in bleb formation during apoptosis via ROCK-1 activation (25). Y-27632 (Tocris), an inhibitor of ROCK-1, did not inhibit the internalization of annexin A5-Alexa568 (Fig. 3e). Instead, ROCK-1 inhibition led to the formation of large annexin A5-containing vesicles suggesting that the actomyosin counteracts annexin A5-induced internalization thereby limiting vesicle size. Coincubation of Jurkat cells having colchicines (Sigma) disrupted microtubules with anti-Fas, and annexin A5-Alexa568 resulted in annexin A5-containing vesicles that remained attached to the plasma membrane (Fig. 3f), indicating that intracellular trafficking depends on microtubules. Taken together our data showed that annexin A5 is internalized through a process that differs mechanistically from macropinocytosis.
Two-dimensional Crystallization of Annexin A5 Is the Driving Force for Internalization-On basis of our results hitherto we reasoned that annexin A5 reverses membrane fission from blebbing into invagination. Because the annexin A5-PtdSer complex has no transmembrane orientation, direct intracellular signaling by this complex could be ruled out to be the cause for this reversion. Therefore, we started to look for an explanation on a nanomechanical basis. The tertiary structure of annexin A5 shows that its phospholipid binding side has a convex shape (27). When bound to the membrane annexin A5 forms trimers of which each monomer retains the convex shape at its phospholipid binding side (19,28). We hypothesized that the trimers bend the membrane and provide the driving force for the reversion of membrane movement. To test the hypothesis we generated the annexin A5 mutant M23 (21), which by prediction from the available structural data (19,27) binds to PtdSer but lacks the ability to form trimers. Using a novel FRET assay we showed that annexin A5 but not M23 develops FRET when bound to a phospholipid surface (Fig. 4a). Atomic force microscopic analysis of phospholipid-bound annexin A5 and M23 revealed the ordered array formation of the former and the disordered organization of the latter (Fig. 4b).
The convex shape of phospholipid-bound annexin A5 trimers was demonstrated to induce invagination of part of the membrane using large unilamellar vesicles. Structures with typically invaginated cups were obtained (see "Experimental Procedures") (Fig. 4c). The thickness of both the cup membrane with a negative curvature (Fig. 4d, nc) and the outer membrane with a positive curvature (pc) were 4.96 Ϯ 0.75 nm. Incubating these structures with annexin A5 at subsaturating conditions resulted in thicknesses of 9.57 Ϯ 1.46 and 5.26 Ϯ 0.76 nm for negative and positive curvature, respectively. This showed that the annexin A5 network bound to the membrane is bent with the convex shape of its phospholipid binding side. Incubation of the vesicles with M23 yielded thicknesses of 5.91 Ϯ 0.96 and 5.66 Ϯ 0.75 nm for negative and positive curvature, respectively. All together, it is evident that annexin A5 and not M23 organizes at the phospholipid surface in a bent network. Coincubation of Jurkat cells with anti-Fas and M23 resulted in the generation of large surface blebs with M23 at the outside and F-actin at the inside (Fig. 4e). No intracellular vesicles containing M23 were observed. Summarizing these results, we demonstrated that annexin A5 opens a PtdSer-dependent novel portal of cell entry by bending the membrane inward into the cell through two-dimensional crystallization.
Living Cells Bear a PtdSer-dependent Portal of Entry-The execution of apoptosis results in the surface expression of Ptd-Ser, which functions as a determinant for the entry of annexin A5. To determine whether apoptosis is an essential element for internalization, we incubated HeLa cells that were not subjected to apoptotic stimuli with annexin A5-Alexa568. After 3 h of incubation more than 50% of the HeLa cells had internalized annexin A5 in vesicles of varying sizes, similar to those observed in apoptotic Jurkat cells (Fig. 5a). The internalization was not coupled to the execution of the apoptotic program because (i) Z-VAD-fmk (IDN 1529, Idun Pharmaceuticals) did not inhibit annexin A5 internalization, and (ii) the cells with internalized annexin A5 were viable (data not shown). Apart from this discrepancy viable HeLa cells share the modus operandi with Jurkat cells executing apoptosis concerning the entry of annexin A5 into the cell. Annexin A5 internalization is dependent on the PtdSer binding and two-dimensional crystallization properties of annexin A5, because HeLa cells did not internalize fluorescently labeled M1234 or M23. Furthermore, Latrunculin B did not inhibit the internalization of annexin A5. Both Latrunculin B and colchicine prevented the intracellular trafficking of the annexin A5 containing vesicles. Further examination of the pathway of uptake demonstrates that it differs from known portals of cell entry (26). Annexin A5 uptake occurs neither through fluid phase internalization (Fig. 5b) nor through clathrin-and caveolin-mediated entry (29) (Fig. 5, c  and d). Taken together our data showed that both cells executing apoptosis and living tumor cells have the same PtdSer-dependent portals of cell entry.
The Specificity of the Novel Portal of Cell Entry-To investigate further how mandatory the two-dimensional crystallization is for endocytosis, we used annexin A1. This annexin binds to PtdSer but does not form a two-dimensional protein network on the membrane surface (30). Co-incubation of Jurkat cells with anti-Fas and annexin A1-Alexa568 results only in plasma membrane staining of the apoptotic Jurkat cells but not in the formation of endocytic vesicles typically seen with annexin A5 (Fig. 6). Similar results were obtained with living HeLa cells. Interestingly, annexin A5 shuttles annexin A1 into the cell, demonstrating that this novel pathway leads to the internalization of proteins bound to or in the vicinity of surface-expressed PtdSer. The latter has been demonstrated for tissue factor, which is internalized in PtdSer-expressing cells by annexin A5. 2

DISCUSSION
Plasma membrane dynamics during apoptosis include the expression of PtdSer at the cell surface and membrane blebbing. Both phenomena occur during apoptosis downstream of caspase-3 activation (4) but are regulated independently. Ptd-Ser expression occurs under conditions wherein membrane blebbing is inhibited (25). With fluorescently labeled annexin A5 it was shown that membrane blebs have surface-expressed PtdSer, indicating that the PtdSer-expressing membrane patches are flexible and bendable. Recently, it was reported that annexin A5 inhibits apoptotic body formation by applying a physical constraint on the plasma membrane and inhibiting the apoptotic program (20). We confirmed that annexin A5 inhibits apoptotic body formation, but it does so in our system without interfering with the key steps of the apoptotic program such as caspase-3 activation, PtdSer expression, ROCK-1 cleavage, and DNA fragmentation. Because the apoptotic body formation is downstream of membrane blebbing (25), the inhibitory mechanism may operate on membrane blebs. Our investigations into this phenomenon revealed the existence of an endocytic pathway that is activated by annexin A5 and surface- FIG. 7. Model for the pinocytic pathway mediated by surface-expressed PtdSer and annexin A5. Cell surface expression of PtdSer results in disassembly of the cortical actin network underlying the PtdSer-exposing membrane patch. Annexin A5 binds to PtdSer, crystallizes, and bends the membrane patch toward invagination. The invaginated membrane cup is closed, encapsulated with F-actin, and transported into the cytosol in a microtubule-dependent manner. expressed PtdSer. We demonstrated that annexin A5 reverses the movement of the PtdSer-expressing membrane patch from blebbing into invagination. This results in vesicle formation and the intracellular trafficking of the endocytic vesicle (Fig. 7). From the experiments with M23 and annexin A1 we concluded that the process of invagination is driven nanomechanically by the formation of an annexin A5 two-dimensional crystal that bends the membrane, which is likely due to the bent shape of the trimer building blocks (19). The mechanism of membrane bending differs from that operating in clathrin-coated pits. Here, the insertion of proteins into the lipid layer pushes lipid head groups of one leaflet apart thereby inducing curvature of the bilayer (31). If annexin A5 would operate comparably it is expected that annexin A5 facilitates blebbing instead of inducing invagination. Therefore, it is concluded that annexin A5 binds extrinsically and molds the membrane patch according to the convex shape of its phospholipid binding side. The energy required for this action is likely released by the two-dimensional crystallization process.
The size distribution of the annexin A5 containing endocytic vesicles suggests macropinocytosis rather than clathrin-or caveolin-mediated endocytosis and clathrin-and caveolin-independent endocytosis (26). Despite a morphological resemblance the two endocytic pathways are mechanistically distinct. Annexin A5-induced endocytosis is neither actin-driven nor preceded by membrane ruffling. Hence, we concluded that the PtdSer-expressing membrane patch forms a novel portal of cell entry, which is opened by annexin A5.
The relevance of the annexin A5 endocytic pathway during apoptosis is debatable. Why should a cell determined to die be capable of activating endocytic pathways? An inspection of the time course revealed that the endocytic pathway can be activated early after the onset of apoptosis, likely as soon as the first PtdSer molecules appear at the surface. Because cell surface expression of PtdSer has been reported for viable cells (32)(33)(34)(35), 3 the annexin A5 endocytic pathway in apoptotic cells could be a remnant of the pinocytic capabilities of the living cell. In this paper we demonstrated the likelihood of this possibility by showing that the annexin A5 endocytic pathway exists in viable HeLa cells. The pathway in HeLa cells is dependent on PtdSer expression and annexin A5 crystallization and independent of membrane ruffling and actin polymerization. Furthermore, the pathway is distinct from clathrinand caveolae-dependent endocytosis and macropinocytosis. In addition, we have strong indications that the pathway also exists in vivo in cardiomyocytes that express PtdSer reversibly on their surface because of a mild ischemia/reperfusion stress. 3 The relevance of the annexin A5 endocytic pathway in living cells remains to be established. Its activation leads to the down-regulation of surface-expressed proteins such as the tissue factor. 2 It may therefore play a role in the regulation of processes such as blood coagulation 2 and phagocytosis. 4 The annexin A5 endocytic pathway is apparently not operative ubiquitously in healthy non-disturbed tissues (17). The reported cases of reversible PtdSer expression point toward situations of mild stress such as hypoxia in tumors and the transient ischemic heart that causes reversible PtdSer expression.
Although its physiological relevance needs to be investigated, the annexin A5 endocytic pathway already offers an attractive mechanism for targeted delivery and cell entry of drugs designed to kill (tumor cells) or to rescue (ischemic/ reperfused cardiomyocytes).