Cell Surface-bound Heat Shock Protein 70 (Hsp70) Mediates Perforin-independent Apoptosis by Specific Binding and Uptake of Granzyme B*

Cell surface-bound heat shock protein 70 (Hsp70) renders tumor cells more sensitive to the cytolytic attack mediated by natural killer (NK) cells. A 14-amino acid Hsp70 sequence, termed TKD (TKDNNLLGRFELSG, aa450–463) could be identified as the extracellular localized recognition site for NK cells. Here, we show by affinity chromatography that both, full-length Hsp70-protein and Hsp70-peptide TKD, specifically bind a 32-kDa protein derived from NK cell lysates. The serine protease granzyme B was uncovered as the 32-kDa Hsp70-interacting protein using matrix-assisted laser desorption ionization time-of-flight mass peptide fingerprinting. Incubation of tumor cells with increasing concentrations of perforin-free, isolated granzyme B shows specific binding and uptake in a dose-dependent manner and results in initiation of apoptosis selectively in tumor cells presenting Hsp70 on the cell surface. Remarkably, Hsp70 cation channel activity was also determined selectively in purified phospholipid membranes of Hsp70 membrane-positive but not in membrane-negative tumor cells. The physiological role of our findings was demonstrated in primary NK cells showing elevated cytoplasmic granzyme B levels following contact with TKD. Furthermore, an increased lytic activity of Hsp70 membrane-positive tumor cells could be associated with granzyme B release by NK cells. Taken together we propose a novel perforin-independent, granzyme B-mediated apoptosis pathway for Hsp70 membrane-positive tumor cells.

Elevated cytoplasmic Hsp70 1 levels have been found to protect tumor cells against programmed cell death (1). However, evidence has accumulated indicating that the presence of Hsp70 on the plasma membrane or in the extracellular milieu is highly immunogenic and exposes target cells to immunological attack (2). Following receptor-mediated uptake (3) and re-presentation by antigen presenting cells, HSP-chaperoned peptides elicit a cytotoxic, CD8ϩ T cell response (4). Recently, several receptors, including CD91, Toll-like receptors 2/4 (TLR2/4), and CD40 (5) have been identified to mediate the interaction of HSP90-(gp96), HSP70-(Hsp70, Hsc70), and HSP60-peptide complexes with antigen presenting cells (6 -9). A peptide-independent "chaperokine effect" has been described for members of the HSP70 family. Binding of exogeneous HSP70 to monocytes via TLR2/4 induces receptor clustering in a CD14-dependent pathway (10) and the secretion of proinflammatory cytokines via the MyD88/IRAK/NF-B signal transduction pathway (11)(12)(13). We detected Hsp70, the majorstress inducible member of the HSP70 family, selectively in the plasma membrane of tumor cells, but not in normal cells by cell surface biotinylation and immunofluorescence (14). This finding was confirmed most recently by proteomic profiling of tumor cell membranes (15).
The amount of membrane-bound Hsp70 on tumor cells positively correlates with their sensitivity to lysis mediated by natural killer (NK) cells. Physical (i.e. heat, irradiation) and chemical (i.e. cytostatic drugs, alkyllysophospholipids) stress has been found to increase Hsp70 surface expression on tumor cells and thereby renders them better targets for NK cells (16 -18). Incubation of purified NK cells with recombinant Hsp70 increases the cytolytic activity against Hsp70 membrane-positive tumor cells (19). NK cells have been found to interact specifically with a 14-amino acid Hsp70 sequence, termed TKD (TKDNNLLGRFELSG, aa 450 -463 ), on the C-terminal of this protein. This region (TKD) is present in the ectoplasmic domain of viable tumor cells (20). Therefore, it was not surprising that similar to full-length Hsp70 (19), Hsp70peptide TKD exhibits a comparable immunostimulatory capacity to NK cells (20). Although the preceding observations indicate that Hsp70-peptide functions as a tumor-selective target recognition structure for NK cells (21), the mechanism by which NK cells lyse Hsp70 membrane-positive tumor target cells remained to be elucidated.

Cells
The NK cell line YT was cultured at low cell densities ranging between 0.1 and 0.5 ϫ 10 6 cells/ml in RPMI 1640 medium (Invitrogen) containing 10% fetal calf serum (Invitrogen) supplemented with 6 mM L-glutamine and antibiotics (100 IU/ml penicillin and 100 g/ml streptomycin; Invitrogen).
Columns were washed with 10 column volumes of 20 mM Tris buffer, bound proteins were eluted with 3 M sodium chloride in 20 mM Tris buffer, in 5 fractions (1 ml). Each fraction was subjected to a SDS-PAGE using a 10% polyacrylamide slab gel and transferred onto polyvinylidene difluoride membranes.

Western Blot Analysis
Blots were blocked with skim milk (0.1%) and incubation with monoclonal antibody directed against granzyme B (2C5, IgG2a, BD Biosciences, Heidelberg, Germany), for 5 h at 4°C. Blots were washed and incubated with a secondary mouse anti-IgG horseradish peroxidase antibody (Dianova, Hamburg, Germany), for 1 h at 4°C. Proteins were detected using the ECL kit (Amersham Biosciences) for 5 s.

Protein Identification by Peptide Mass Fingerprinting
A 32-kDa protein band was isolated by affinity chromatography on immobilized Hsp70-protein or Hsp70-peptide TKD, which was excised from Coomassie Blue-stained gels, digested with trypsin, and desalted using reversed phase ZIP tips (Millipore, Eschborn, Germany). The samples were embedded in 4-hydroxy-␣-cyanocinnamic acid and the peptide masses were determined with a Perseptive Voyager DePro matrix-assisted laser desorption ionization-time of flight (MALDI-TOF) mass spectrometer in reflective mode. A peaklist was compiled with the m/z software (Proteometrics) and used for peak selection; the resulting peptide mass fingerprint was used to search the non-redundant NCBI protein data base using the Profound search engine (Proteometrics). Granzyme B was identified with 100% probability and Ͼ95% confidence.

Membrane Preparation
Membrane purification was performed by Dounce homogenization (100 ϫ 10 6 tumor cells) in hypotonic, EDTA-free buffer containing the protease inhibitor phenylmethylsulfonyl fluoride followed by sequential centrifugation at 1,000 ϫ g for 5 min and at 100,000 ϫ g, at 4°C, for 60 min. The pellet containing membranes was resuspended in 2 ml of 0.3 M NaCl in 50 mM Tris buffer, 0.5% Nonidet P-40, pH 7.6.

Cation Channel Activity Measurements
Large unilamellar liposomes were prepared from plasma membranes derived of Hsp70 membrane-positive and membrane-negative tumors and applied to an orifice of about 100 -120 m in diameter with a Teflon film separating two compartments. The ionic solutions containing ei-ther symmetrical concentrations of KCl (200 cis /200 trans mM) or asymmetric concentrations (200 cis /50 trans mM) and 5 mM K-Hepes, 7.1 mM MgCl 2 , 0.5 mM CaCl 2 , and ATP (2 mM). The two ionic compartments were electrically connected via agar bridges and Ag/AgCl pellet electrodes to the input of a voltage clamp amplifier. Current was recorded using patch clamp amplifier (Axopatch-1D; Axon Instruments, CA) and data were stored on a PCM/VCR digital system (Toshiba) with a frequency response in the range from direct current to 25,000 Hz. Off-line analysis of the channel activity was carried out using the software package Pclamp 6 and Axosope (Axon Instruments).

Treatment
Stock solutions of camptothecin (4 mg/ml, Sigma, Munich, Germany) were diluted in dimethyl sulfoxide and stored at 4°C in the dark. Granzyme B (Hölzel Diagnostics, Cologne, Germany) solutions were freshly prepared directly before usage. Exponentially growing cells (0.5-1.5 ϫ 10 6 /ml) were incubated either with camptothecin at a final concentration of 4 g/ml or with purified, enzymatically active granzyme B (10 ng/ml, 1 g/ml, 2 g/ml, 4 g/ml) (22) for 10 min, and 30 min either at 4 or 37°C. After washing in RPMI 1640 medium binding and uptake was determined in non-permeabilized and permeabilized tumor cells by flow cytometry and fluorescence microscopy on a Axioscop 25 scanning microscope (Zeiss, Jena, Germany) equipped with a ϫ40 objective and standard filters. Images were treated by multiplicative shading correction using the software Axiovison (Zeiss Vison, Jena, Germany). Granzyme B was visualized in red by using the HC2-PE antibody.
Apoptotic cell death was measured after incubation of tumor cells with 10 ng/ml granzyme B for 4, 12, and 24 h by different apoptosis assays, as described below.

Apoptosis Assays
Annexin V-FITC Staining-Briefly, cells were washed twice in Hepes buffer containing 5 mM CaCl 2 and incubated with annexin V-FITC (Roche Diagnostics) for 10 min at room temperature. Annexin V-FITC positively stained cells were measured on a FACSCalibur flow cytometer (BD Biosciences).
DAPI Staining-Methanol/acetone fixed cells (0.1 ϫ 10 6 cells/100 l) were incubated with 0.5 g/l 4,6-diamino-2-phenylindole (DAPI) in PBS/glycerol (3:1) for 15 min in the dark. Following washing in PBS the cells were mounted with fluorescent mounting medium (Dako, Glostrup, Denmark) and analyzed for fluorescence using a Zeiss model Axioscop 2 scanning microscope (Zeiss, Jena, Germany) equipped with a ϫ40 objective and standard filters. Apoptosis was visualized with DAPI staining in 50 cells, each. Images were treated by multiplicative shading correction using the software Axiovision (Zeiss).
Cytochrome c-Cytochrome c was determined using a quantitative immunoassay (DCDCO, R&D Systems, Wiesbaden, Germany). Briefly, untreated, camptothecin (4 g/ml), or granzyme B (10 ng/ml)-treated cells (1.5 ϫ 10 6 /ml) were washed in PBS and treated with lysis buffer for 1 h at room temperature. Following centrifugation at 1,000 ϫ g for 15 min, supernatants were removed and 200 l of a 1:100, 1:250, and a 1:500 dilution was used for a sandwich enzyme-linked immunosorbent assay. Following incubation with substrate solution in the dark for 30 min the reaction was stopped. The optical density of each well was determined on an enzyme-linked immunosorbent assay reader at 450 nm. The amount of cytochrome c was determined according to a calibration curve.

Cr Release Assay and Inhibition Assay
NK cell-mediated cytotoxicity was measured using a 12-h 51 Cr radioisotope assay. As target cells the colon carcinoma sublines CXϩ and CXϪ were used. For blocking studies the monoclonal antibody C92F3B1 and an isotype matched control antibody (IgG1) were used at a final concentration of 5 g/1 ϫ 10 6 cells. Following incubation of CXϩ and CXϪ target cells with the antibodies for 30 min at 4°C, the cells were labeled with 51 Cr and the cytotoxicity assay was performed as described by MacDonald (23). The percentage of specific lysis was calculated as: [(experimental release Ϫ spontaneous release)/(maximal release Ϫ spontaneous release)] ϫ 100.

Granzyme B ELISPOT
Granzyme B release was compared in unstimulated NK cells (NK d0) and TKD-stimulated NK cells (NK d3) after a 4-h co-incubation period with tumor cell lines CXϩ/CXϪ and Coloϩ/ColoϪ, at different effector to target cell ratios (E:T) ranging from 20:1 to 2:1. For detection a granzyme B ELISPOT kit (number 552572, BD Biosciences) was used. Briefly, 96-well ELISPOT plates (MAIPN45, Millipore) were coated overnight at 4°C with capture antibody, blocked with RPMI 1640 culture medium containing 10% fetal calf serum and incubated with tumor and effector cells for 4 h at 37°C, as specified before. After washing in deionized water and wash buffers A and B, biotinylated anti-granzyme B antibody was added (2 g/ml) for 2 h. After another two washing steps granzyme B was visualized by the addition of freshly prepared avidin-horseradish peroxidase (1 h) and substrate solution (25 min incubation period). Spots were counted automatically using Immuno Spot Series I analyzer.

Granzyme B Is a Partner of Hsp70 and Hsp70-Peptide TKD-
Proteins that interact with Hsp70 were identified by affinity chromatography on immobilized Sepharose columns coupled either with human Hsp70-protein or a peptide derived from the C-terminal region of Hsp70, termed TKD (TKDNNLL-GRFELSG, aa 450 -463 ). TKD corresponds to 14-amino acids localized in the extracellular domain of Hsp70, present on Hsp70 membrane-positive tumor cells, which mediate recognition by NK cells (20). TKD was used in the affinity column to exclude the recognition of proteins by the substrate binding domain of Hsp70. Recently, we showed that Hsp70-protein and TKD, specifically bind to the NK cell line YT (24). Therefore, we assumed that YT cells provide an ideal tool for identifying Hsp70-interacting partners. YT cell lysates were fractionated on immobilized Hsp70 and TKD columns. The material bound to the columns was eluted with 3 M sodium chloride in five fractions. As controls, YT cell lysates were administered to carrier or BSA-conjugated columns. Moreover, lysate of a non-NK cell line (K562) was fractionated on TKD-conjugated affinity column. The eluted fractions were separated by SDS-PAGE and visualized by silver staining. A dominant protein band of apparent molecular weight of 32,000 was observed in fractions two (F2) and three (F3) of YT cell eluates derived from Hsp70-protein (Hsp70) or Hsp70-peptide (TKD) columns (Fig.  1A). This band was not detectable in eluates of carrier (data not shown) or BSA-conjugated columns nor in material eluted from the TKD affinity columns loaded with K562 cell lysates (Fig.  1A). Additionally, the 32-kDa protein band derived from F3 was excised from a SDS-PAGE stained with Coomassie Blue and digested with trypsin (Fig. 1B). The resulting peptides were analyzed by MALDI-TOF peptide mass fingerprinting. Sequences of the tryptic peptides exhibited 100% homology with granzyme B (Fig. 1B). The identity of the 32-kDa protein band as granzyme B was further confirmed by Western blot analysis using a specific antibody (2C5) against granzyme B. Fractionation of YT cell lysates on Hsp70-protein (Hsp70) and Hsp70-peptide (TKD) columns revealed the presence of granzyme B protein band by Western blotting (Fig. 1C). However, no granzyme B was detected in the eluted fraction of Hsp70 and TKD affinity columns loaded with K562 cell lysates (Fig.  1C). Also flow cytometry using a phycoerythrin-conjugated granzyme B antibody HC2-PE showed positive staining for cytoplasmic granzyme B in YT cells, but not in K562 cells (Fig. 1D). In summary, these data indicate that granzyme B interacts with full-length Hsp70-protein as well as Hsp70peptide TKD.
Specific Binding and Internalization of Granzyme B in Hsp70 Membrane-positive Tumor Cells-The preceding findings posed the question whether membrane-bound Hsp70 might enable specific binding and entry of granzyme B into the cytosol. Therefore, perforin-free, purified granzyme B was coincubated with tumor cell sublines CXϩ/CXϪ and Coloϩ/ ColoϪ that exhibit differential Hsp70 membrane expression. A light microscopical analysis of untreated CXϩ and Coloϩ cells (control) at 4 versus 37°C is shown in the upper row of each panel ( Fig. 2A). The corresponding immunofluorescence microscopy of the cells at 4 and 37°C is illustrated below (control). Initially, none of the cells showed any granzyme B staining, neither on the cell surface nor in the cytoplasm. However, after a 15-min incubation period of the cells with purified granzyme B (grB) at 4°C, a ring-shaped fluorescence, indicating a typical cell surface staining, was detected on Hsp70 membrane-positive CXϩ and Coloϩ tumor sublines ( Fig. 2A, left panel). A temperature shift from 4 to 37°C during the 30-min incubation period resulted in uptake of granzyme B, as determined by a cytoplasmic staining pattern in CXϩ and Coloϩ tumor sublines ( Fig. 2A, right panel). In contrast, the Hsp70 membrane-negative counterparts CXϪ and ColoϪ neither exhibited any granzyme B cell surface binding at 4°C nor uptake at 37°C (data not shown). Flow cytometry analysis of permeabilized cells revealed a faint shift of the granzyme B peak to the right selectively in Hsp70 membrane-positive CXϩ and ColoϪ, but not in CXϪ and ColoϪ tumor sublines, if the cells were coincubated with 1 g/ml granzyme B for 30 min at 37°C (Fig.  2B, upper graph). A dose-dependent increase in granzyme B uptake, in Hsp70 membrane-positive tumor cells (CXϩ/Coloϩ), was detected after co-incubation with 2 and 4 g/ml granzyme B (Fig. 2B, lower graph). However, even at the highest concentration of 4 g/ml, granzyme B was internalized much more pronounced by Hsp70 membrane-positive as compared with Hsp70-negative tumor cells (CXϪ/ColoϪ).
Potential ion channels formed by Hsp70 may play a role in the mechanism of selective granzyme B uptake in Hsp70 membrane-positive tumor cells. Indeed, a particular ion conductance pathway was observed after incorporation of vesicles derived from purified phospholipids of Hsp70 membrane-positive (CXϩ) tumor sublines. This was not seen in vesicles obtained from Hsp70 membrane-negative (CXϪ) tumor cells (data not shown). Based on these results one might speculate about an ion channel activity facilitating uptake of granzyme B selectively into Hsp70 membrane-positive tumor cells.
In Vitro Provided Granzyme B Induces Apoptosis Selectively in Hsp70 Membrane-positive Tumor Cells-Differences in the inducibility of apoptosis were studied by co-incubation of Hsp70 high-(CXϩ/Coloϩ) and low-(CXϪ/ColoϪ) expressing carcinoma cells (21) with 10 ng/ml enzymatically active granzyme B (22) for 4, 12, and 24 h. The topoisomerase inhibitor camptothecin at a final concentration of 4 g/ml served as a positive control for apoptosis. Programmed cell death was de-termined by using three different apoptosis assays including annexin V-FITC, DAPI staining, and mitochondrial cytochrome c release. After a 4-h incubation period, neither camptothecin nor granzyme B initiated apoptosis in any of the tested tumor cells (data not shown), indicating that our tumor carcinoma cell lines are more resistant to apoptotic cell death, as compared with the acute T cell leukemia cell line Jurkat. After a 12-and 24-h incubation period with camptothecin significant apoptosis was observed in all tumor sublines (Fig. 3A). It appeared that the colon carcinoma sublines CXϩ/CXϪ are better protected toward a camptothecin-mediated cell death as compared with the pancreas carcinoma sublines Coloϩ/ColoϪ. However, no significant differences in the inducibility of apoptosis by using camptothecin was observed between Hsp70 membrane-positive and Hsp70 membrane-negative tumor cells. Interestingly, this was not the case if the tumor sublines were incubated with granzyme B at a concentration that is found in the serum under physiological conditions (10 ng/ml): 12 h post-treatment the amount of annexin V-FITC positive cells was equally up-regulated in Hsp70 membrane-positive CXϩ/Coloϩ tumor cells (1.3-fold); after 24 h the increase in apoptotic cells was significant in Hsp70 membrane-positive CXϩ (1.8-fold) and in Coloϩ (2.4-fold) tumor cells (Fig. 3A). In line with these results the amount of Hsp70 membrane-positive leukemic K562 cells (25) was similarly up-regulated (1.8fold) following contact with granzyme B (data not shown). In contrast, the amount of apoptotic CXϪ and ColoϪ tumor cells with stably low Hsp70 membrane expression levels remained unaltered before and after a 12-and 24-h co-incubation period with granzyme B (Fig. 3A).
To exclude apoptosis initiated by anoikes light microscopical analysis of untreated (control), camptothecin-(cam), and granzyme B (grB)-treated CXϩ/CXϪ and Coloϩ/ColoϪ tumor cells were performed. As shown in Fig. 3B, 24 h post-treatment with granzyme B, neither Hsp70 membrane-positive nor -negative tumor cell lines exhibited any signs of loss in plastic adherence. Regarding these findings we ruled out the possibility that anoikes might be a possible mechanism for the induction of apoptotic cell death in Hsp70 membrane-positive tumor sublines. It is important to note that all apoptosis assays were determined within the adherent cell population following a short term (Ͻ1 min) trypsinization.
Consistent with the results derived by annexin V-FITC staining all cell types, CXϩ/CX-, Coloϩ/ColoϪ, exhibited a positive DAPI nuclear fragmentation staining, as a typical sign of apoptosis at a later stage, following treatment with camptothecin, as compared with untreated control cells (Fig. 3C). Again, DNA fragmentation was detected only in Hsp70 membrane-positive CXϩ and Coloϩ tumor cells, 24 h post-treatment with granzyme B (grB). In line with the annexin V-FITC staining results, no signs of DNA fragmentation were observed in Hsp70 membrane-negative CXϪ and ColoϪ cells (Fig. 3C). As an additional test for apoptotic cell death, cytochrome c release was measured following incubation of CXϩ and CXϪ cells with granzyme B and camptothecin. As summarized in Table I, following incubation with granzyme B for 24 h, cytochrome c concentration was elevated 1.8-fold (0.382 mg/ml versus 0.690 mg/ml) in CXϩ cells. However, no increase in cytoplasmic cytochrome c was observed in CXϪ cells following treatment with granzyme B (0.452 versus 0.425 mg/ml). An incubation with camptothecin (4 g/ml) for 24 h results in a comparable, 1.5-fold increase in cytochrome c concentrations in CXϩ and CXϪ tumor cells. In summary these results indicate that following binding and selective uptake, via membranebound Hsp70, granzyme B initiates apoptosis in a perforinindependent manner.

Granzyme B Released by TKD-activated NK Cells Mediates Apoptosis in Hsp70
Membrane-positive Tumor Cells-The physiological role of our findings was tested in functional assays using naive and Hsp70-peptide (TKD)-stimulated human NK cells. Previously, we have shown that incubation of NK cells with Hsp70-protein at concentrations between 10 and 50 g/ml or with equivalent TKD concentrations between 0.2 and 2.0 g/ml resulted in increased cytolytic activity against Hsp70 membrane-positive tumor target cells. Concomitantly, the expression of CD94, the killer cell inhibitory/activatory C-type lectin receptor, was up-regulated (24). Although it was known that Hsp70 acts as a tumor-selective recognition structure for NK cells (21,25), it remained unclear whether tumor cells die by necrosis or apoptosis. To elucidate this question, NK cells were incubated with Hsp70-peptide TKD (2 g/ml) for 3 days. A significant increase (1.4-fold) in cytoplasmic granzyme B levels was observed within 3 days of stimulation in CD3-negative NK cells. In contrast, granzyme B expression was not up-regulated in CD3-positive T cells following identical treatment conditions.
Killing of CXϩ/CXϪ and Coloϩ/ColoϪ tumor cells following contact with freshly isolated, unstimulated (NK d0), or TKDstimulated NK cells (NK d3) was compared in a standard 12-h 51 Cr release assay (Fig. 4). Because of experimental limitations, the co-incubation period of NK and tumor cells in the cytotoxicity assay could not be extended to 24 h, like in our in vitro apoptosis assays.
Concomitant with the increase in cytoplasmic granzyme B, the cytolytic activity of TKD-stimulated NK cells (NK d3) against CXϩ target cells was enhanced 1.5-fold and that of Coloϩ target cells 2.0-fold at effector to target (E:T) ratios ranging between 20:1 and 2:1. In contrast, the cytolytic activity against Hsp70 membrane-negative CXϪ and ColoϪ cells was not elevated. The increased lysis of Hsp70 membrane-positive tumor cells was decreased in both cell systems down to the degree of lysis of Hsp70 membrane-negative tumor cells, by Hsp70-specific monoclonal antibody, that is known to recognize membrane-bound Hsp70-peptide TKD on viable tumor cells (25). In contrast, the lower lysis of Hsp70 membrane-negative tumor cells remained unaffected after incubation with Hsp70 antibody. Technically, it is not possible to quantify the absolute amount of granzyme B that is transferred from NK cells into tumor cells by cell-to-cell contact. However, relative values of granzyme B release could be determined by ELISPOT analysis. Therefore, a comparison of the cytolytic response of freshly isolated, unstimulated (NK d0) and TKD-stimulated NK cells (NK d3) against CXϩ/CXϪ and Coloϩ/ColoϪ tumor cells was performed concomitantly with the definition of the granzyme B release. Irrespectively of the tumor cell line and the E:T cell ratio, co-incubation of tumor cells with unstimulated NK cells (NK d0) always results in very low granzyme B release; the number of spots was always less than 20. After a 3-day stimulation period with TKD (NK d3) followed by a 4-h co-incubation time with tumor cells, granzyme B release was significantly up-regulated. At an E:T ratio of 5:1, the number of granzyme B spots, as determined in three independent experiments, was as follows: CXϩ, 260 Ϯ 20; CXϪ, 165 Ϯ 6; Coloϩ, 137 Ϯ 55; ColoϪ, 66 Ϯ 8. Concomitantly, Hsp70 membraneof adherent growing CXϩ/CX and Coloϩ/ColoϪ tumor cell clusters either untreated (control) or following treatment with camptothecin (cam, 4 g/ml) or granzyme B (grB, 10 ng/ml), for 24 h. Scale bar indicates 100 m. C, in parallel, either untreated (control), camptothecin (cam), or granzyme B (grB)-treated CXϩ/CXϪ and Coloϩ/ColoϪ tumor cells (24 h) were stained with DAPI. Considerable nuclear DNA fragmentation was observed in all tumor sublines following incubation with camptothecin (middle panel). After incubation with granzyme B only CXϩ and Coloϩ cells exhibited nuclear DNA fragmentation (lower panel, left). No signs of apoptosis were observed in CX-and ColoϪ tumor cells following incubation with granzyme B (lower panel, right). Scale bar represents 10 m.  positive tumor target cells (CXϩ/Coloϩ) were lysed significantly better as compared with their negative counterparts (CXϪ/ColoϪ). These data strongly suggest that lysis of Hsp70 membrane-positive tumor cells by TKD-activated NK cells is associated with granzyme B release. Proteases, like granzyme B, initiate apoptosis by a intracellular mechanism (35). In Fig.  2, A and B, specific binding and a dose-dependent internalization of in vitro provided granzyme B was detected selectively in Hsp70 membrane-positive tumor cells. With respect to these findings, we assumed that differences in lysis of CXϩ/CXϪ and Coloϩ/ColoϪ tumor cells are because of a different capacity of the tumor cells to internalize granzyme B.

DISCUSSION
Previously we have demonstrated that tumor cells, but not normal cells, present Hsp70 on their plasma membrane (14). Antibody mapping revealed that part of the C-terminal substrate binding domain is exposed to the extracellular milieu (26). The epitope of this antibody corresponds to a 14-amino acid sequence, termed TKD (TKDNNLLGRFELS, aa 450 -463 ). Thus, exposure of TKD on the cell surface sensitizes tumor cells to the cytolytic activity of NK cells (20). Furthermore, Hsp70peptide TKD exhibits activatory properties on NK cells, which is comparable with the activity observed by equivalent amounts of full-length Hsp70 (19). Although it is obvious that Hsp70-protein and Hsp70-peptide TKD efficiently trigger NK cell activity, the mechanism how NK cells kill Hsp70 membrane-positive tumor target cells remained unclear.
To identify molecules that are involved in the interaction of NK cells with Hsp70-positive tumor cells, Hsp70 conjugated to Sepharose was used as affinity bait to isolate Hsp70-binding proteins from lysates of the human NK cell line YT (27). In the cytoplasm, Hsp70 binds hydrophobic residues of denatured polypeptides and co-chaperones via the substrate binding pocket, localized in the C-terminal region of the protein, in an ATP-dependent manner (28 -31). To characterize interacting partners specific for membrane-bound Hsp70, an affinity column with Hsp70-peptide TKD, was also used (20). A dominant 32-kDa protein band was eluted from Hsp70-protein and Hsp70-peptide columns with lysates of NK cells. This protein band was not detected with lysates of non-NK cells. The intensity of the 32-kDa protein band was more pronounced in eluates of TKD columns because at the molar level a 50 -100-fold excess Hsp70-peptide was coupled to the column, as compared with full-length Hsp70-protein. By MALDI-TOF and Western blot analysis the Hsp70-interacting partner could be identified as granzyme B (M r expected 32,000). This observation is consistent with a prior report indicating binding of Hsp70 and Hsp27 to granzyme-immobilized Sepharose columns (32).
One mechanism of NK cell-mediated killing involves the exocytosis of cytotoxic granules containing perforin and serine proteases (33). After internalization the protease granzyme B cleaves procaspases into their activated form, and thereby induces programmed cell death by promoting DNA fragmentation (34 -36). Also gene deletion studies demonstrated that granzyme B is critical for the induction of apoptosis (37). It is assumed that granzyme B enters target cells either in a perforindependent or perforin-independent way. Presently, the mannose 6-phosphate receptor is discussed to be involved in the process of endocytosis/pinocytosis of granzyme B (38 -40). Here we provide evidence for a perforin-independent induction of apoptosis mediated by the interaction of granzyme B with cell surface-bound TKD. Incubation of human colon and pancreas carcinoma sublines that differ in Hsp70 membrane expression, with isolated, enzymatic active granzyme B, at concentrations found in human serum under physiological conditions (41), induces apoptosis selectively in Hsp70 membrane-positive tu-mor cells, as determined by annexin V-FITC staining, nuclear DNA fragmentation, and cytochrome c release. Cell viability of Hsp70 membrane-negative tumor sublines was not affected by granzyme B. These differences in the induction of apoptosis were not observed with the topoisomerase inhibitor camptothecin. Therefore, we speculate that initiation of apoptosis by granzyme B and camptothecin involves different routes. Our results that binding of granzyme B to the extracellular exposed region of Hsp70, defined by the Hsp70-peptide TKD, may be critical for uptake and for the induction of apoptosis. This observation is supported by the finding that the enhanced cytolytic activity of NK cells against Hsp70 membrane-positive tumor cells could be blocked by an Hsp70-specific monoclonal antibody, which binds to TKD. It is conceivable that cell surface-bound Hsp70 antibody prevents binding and internalization of granzyme B to tumor cells and thus protects them from apoptosis.
Earlier observations have shown that only an Hsp70 antibody directed against the C-terminal part TKD was able to recognize cell surface-localized Hsp70 on viable tumor cells (26). This suggests that the protein may be incorporated into the lipid bilayer. This option has been supported by recent evidence indicating interaction of HSP70s with phospholipids of the plasma membrane (42). Moreover, an ion conductance pathway was observed in artificial lipid bilayers after incorporation of Hsc70 (43). More recently, these proteins were found to interact with lipids in a liposome aggregation assay (44). Therefore, it is possible that Hsp70 is also embedded in the plasma membrane of tumor cells. Here we report on a selective cation channel activity in plasma membrane of Hsp70 positive tumor cells. Based on these results, one might speculate about channel formation, which facilitate binding and uptake of granzyme B into Hsp70 membrane-positive tumor cells. In the absence of Hsp70 in the plasma membrane, these channels cannot be created and thus uptake of granzyme B is prohibited. In the cytosol Hsp70 frequently cooperates with other proteins. Therefore, we were interested to identify molecules that may be tightly associated with Hsp70 in the plasma membrane. Unpublished data 2 of our group suggest that the silencer of death domain, also termed Bag-4, is co-localized with Hsp70 on the cell surface. Ongoing studies investigate the role of Hsp70 in concert with Bag-4 in the formation of ion channels.
Keeping in mind that tumor sublines with differential Hsp70 membrane expression did not differ with respect to their intracellular Hsp70 levels, neither under physiological conditions nor following stress (21), differences in the sensitivity to granzyme B-mediated apoptosis cannot be explained by differential cleavage of Hsp70 inside the cell. Also modulations in the expression of major histocompatibility complex class I molecules, known to differentially induce killer cell inhibitory/activating receptors on NK cells, could be ruled out, because the major histocompatibility complex class I cell surface pattern was identical in Hsp70 membrane-positive and -negative tumor sublines (21).
Apart from the observation that Hsp70 serves as an entry port for granzyme B into Hsp70 membrane-positive tumor target cells, Hsp70-protein and TKD stimulate the production and release of granzyme B in primary NK cells. This finding can explain why high Hsp70 membrane expression not only predisposes tumor target cells to apoptotic cell death but also favors NK cell activation. In contrast to in vitro applied granzyme B, activated NK cells also kill Hsp70 membrane-negative tumor cells to a certain extent. However, this weak lysis was not blockable with Hsp70 antibody. Together with the result that lysis of Hsp70 membrane-positive tumor cells was reduced to the degree of lysis observed with Hsp70 membrane-negative tumor cells by Hsp70 antibody, we assumed that Hsp70-mediated granzyme B internalization and apoptosis induction is only one mode how activated NK cells kill tumor cells. Furthermore, in comparison to in vitro provided granzyme B, killing mediated by NK cells engage a physical cell-to-cell contact that is likely to result in high local granzyme B concentrations at the tumor cells that cannot be quantified experimentally. Therefore, our in vitro apoptosis assays were conducted with granzyme B concentrations similar to those found in human serum (41). These concentrations are undoubtedly different to that secreted by NK cells and hence it would be rather surprising if they had the same efficacy. Although the absolute amount of granzyme B transferred from NK cells into tumor cells could not be defined, the cytotoxic response of activated NK cells against Hsp70 membrane-positive tumor cells correlated with the relative release of granzyme B into the medium, as determined by ELISPOT analysis. Kinetical studies indicating that increasing amounts of in vitro provided granzyme B are still selectively taken up by Hsp70 membrane-positive tumor cells further support these findings.
In summary our data suggest a dual role for Hsp70-protein and Hsp70-peptide TKD in the immune response against cancer. On the one hand it stimulates the production and delivery of granzyme B by NK cells, on the other hand it facilitates uptake of granzyme B selectively into Hsp70 membrane-positive tumor target cells. We propose the Hsp70 epitope TKD is a naturally occurring interacting partner for granzyme B that elucidates one part of the puzzling role of membrane-bound Hsp70 in the natural defense mechanisms against tumor cells.