Transactivation of the Epidermal Growth Factor Receptor by Heat Shock Protein 90 via Toll-like Receptor 4 Contributes to the Migration of Glioblastoma Cells*

Extracellular heat shock protein HSP90α was reported to participate in tumor cell growth, invasion, and metastasis formation through poorly understood signaling pathways. Herein, we show that extracellular HSP90α favors cell migration of glioblastoma U87 cells. More specifically, externally applied HSP90α rapidly induced endocytosis of EGFR. This response was accompanied by a transient increase in cytosolic Ca2+ appearing after 1–3 min of treatment. In the presence of EGF, U87 cells showed HSP90α-induced Ca2+ oscillations, which were reduced by the ATP/ADPase, apyrase, and inhibited by the purinergic P2 inhibitor, suramin, suggesting that ATP release is requested. Disruption of lipid rafts with methyl β-cyclodextrin impaired the Ca2+ rise induced by extracellular HSP90α combined with EGF. Specific inhibition of TLR4 expression by blocking antibodies suppressed extracellular HSP90α-induced Ca2+ signaling and the associated cell migration. HSPs are known to bind lipopolysaccharides (LPSs). Preincubating cells with Polymyxin B, a potent LPS inhibitor, partially abrogated the effects of HSP90α without affecting Ca2+ oscillations observed with EGF. Extracellular HSP90α induced EGFR phosphorylation at Tyr-1068, and this event was prevented by both the protein kinase Cδ inhibitor, rottlerin, and the c-Src inhibitor, PP2. Altogether, our results suggest that extracellular HSP90α transactivates EGFR/ErbB1 through TLR4 and a PKCδ/c-Src pathway, which induces ATP release and cytosolic Ca2+ increase and finally favors cell migration. This mechanism could account for the deleterious effects of HSPs on high grade glioma when released into the tumor cell microenvironment.

Glioma ranges from slowly growing low grade tumors to rapidly growing high grade tumors, including anaplastic astrocytoma and glioblastoma (1,2). High grade gliomas include anaplastic tumor cells, necrotic foci, and rich vascularity, due largely to the aberrant expression of angiogenic factors by tumor cells (3). Tumor cells are highly proliferative and inva-sive within the brain. Despite the development of various treatments, the life expectancy of patients remains poor (4).
One of the molecules that could contribute to invasion is the stress or heat shock protein 90 (HSP90). In several tumor types, cell surface expression of HSP90 correlates with metastatic potential (5), and its inhibition with antibodies (6,7) or with cell-impermeable inhibitors (8) reduces cell migration in vitro. Of the two HSP90 isoforms, only HSP90␣ has been described extracellularly, which argues against cell lysis as the source of extracellular HSP90 (8). The metalloprotease MMP-2 can be associated with extracellular HSP90␣ (8), which is favored by acetylation of the stress protein (9). Surface HSP90␣ also participates in extracellular matrix proteininduced c-Src/integrin association and reorganization of the actin cytoskeleton (10). The cell-impermeable inhibitor of HSP90, 17-dimethylaminoethylamino-17-demethoxygeldanamycin (DMAG)-N-oxide, displays anti-invasive and anti-metastatic activity in vitro and in vivo, respectively, probably through the inhibition of actin polymerization and focal adhesion formation (11). Actin polymerization can be activated by tyrosine kinase receptors of the EGFR 2 family (12). Cell surface HSP90␣ interacts specifically with the extracellular domain of HER-2 (also known as ErbB2), a ligand-less receptor, forming heterodimers with ligand-bound members ErbB1 (EGFR), ErbB3, and ErbB4. The HSP90␣/HER-2 interaction favors HER-2 heterodimerization with ErbB3 and signal transduction pathways via MAPK and PI3K-Akt, leading to actin rearrangement and cell motility (7). EGFR (ErbB1, or HER1) is overexpressed in 60% of multiform glioblastomas (13,14) and promotes invasion, proliferation, and metastasis (15)(16)(17) in response to several ligands known as EGF-related peptide growth factors (18). Binding of these ligands to the extracellular domain of EGFR leads to the formation of homo-and heterodimers, which triggers autophosphorylation of specific tyrosine residues within the cytoplasmic domain of EGFR, inducing signaling cascades. Ligand-independent EGFR activation, referred to as EGFR transactivation, can also be observed in response to various agents such as Toll-Like receptor (TLR) agonists and stress conditions (19,20). * This work was supported by grants from the CNRS, INSERM, and by Conseil Ré gional de Bourgogne (CRB, Faber Program) and by grants from are supported by the Ligue Nationale Contre le Cancer (to E. S. and C. G.). □ S The on-line version of this article (available at http://www.jbc.org) contains supplemental Figs. S1-S8. 1 To whom correspondence should be addressed. Tel.: 33-3-80-39-34-17; Fax: 33-3-80-39-34-34; E-mail: dominique.thuringer@u-bourgogne.fr. 2 The abbreviations used are: EGFR, epidermal growth factor receptor; TLR, Toll-like receptor; TSG, thapsigargin; M␤CD, methyl-␤-cyclodextrin; HB, heparin binding; MMP, matrix metalloproteinase; ER, endoplasmic reticulum; P, phospho; HC, heavy chain; ADAM, a disintegrin and metalloproteinase.
The aim of this study was to investigate the role of extracellular HSP90␣ in signaling events triggered by EGFR in the human astrocytoma cell line (U87). We show that exogenously applied HSP90␣ mediates a cross communication between TLR4 and EGFR and accelerates cell migration.
Specific Cell Treatments-EGF was used at a unique dose of 10 ng/ml. To rule out the possible implication of HER2 in EGF-induced cell responses, 5 M gefitinib was added. For cholesterol depletion, cells were incubated for 30 -60 min at 37°C in serum-free medium containing 0.1% bovine serum albumin (BSA) and 7.5 mM methyl-␤-cyclodextrin (M␤CD), a mixture known to reduce plasma membrane cholesterol levels with no cell death (21). Cells were preincubated with Polymyxin B, a potent LPS inhibitor (10 M; 30 -60 min), to avoid bacterial contamination of HSP90␣ before activation. An additional control was to boil HSP90␣ at 100°C for 1 h to suppress the specific effect of HSP90␣ because LPSs are resistant to heat inactivation but not HSPs (22)(23)(24).
Immunofluorescence Experiments-Cells were fixed in 4% paraformaldehyde and permeabilized with 0.1% Triton X-100. Cells were incubated with the primary antibody for 2 h and then secondary antibody (F(abЈ)2 fragment of IgG) conjugated with Alexa Fluor 488 nm or 568 nm (1:500) or with Alexa Fluor 488-conjugated phalloidin for 30 min. Substituting the primary antibody with preimmune serum demonstrates specificity of labeling. Counterstaining of nuclei was made using 40,6-diamidino-2-phenylindole (DAPI; 3 g/ml, 1-3 min). Images were collected on Axioplan 2 microscope using Axiovision Viewer 3 software (Zeiss). The intensity of labeled area (10 m 2 ) was measured with ImageJ analysis software. In some experiments, imaging was performed with a Leica SP2 RS confocal microscope, and Z-series of optical sections taken from 0.4-m intervals were collected using the scanning format 1024 ϫ 1024 pixels.
Cytosolic Free Ca 2ϩ Concentration-Changes in Fura-2 fluorescence were used to measure [Ca 2ϩ ] i . Briefly, confluent cells were incubated with 2 M fura-2-acetometoxymethyl ester in culture medium (40 min at 37°C) and then in HEPESbuffered saline solution containing in mM: 120 NaCl, 5.4 KCl, 0.8 MgCl 2 , 10 glucose, and 20 HEPES, pH 7.4. In nominally Ca 2ϩ -free conditions, no CaCl 2 addition was made. Dimethyl sulfoxide (DMSO) or ethanol, used to dissolve drug, was added in other solutions. Measurements were made on an inverted microscope (Nikon TS100) equipped with a 20ϫ objective (Nikon S Fluor, 0.75 NA) attached to a dual-excitation spectrofluorometer with excitation wavelengths alternating between 340 and 380 nm. Emission fluorescence (510 nm) was collected at a rate of 20/min. The 340/380 nm ratio was determined for each cell, and average values were reported as a function of time.
ATP Assay-Cells were exposed to HSP90␣ and/or EGF, and then the supernatants were collected to assess the amount of ATP release with luciferin-based ENLITEN ATP assay (Promega) at various time periods. Briefly, 100 l of luciferin-luciferase solution were added to supernatants, and light emission was recorded with a FLUOstar luminometer (BMG LABTECH). We calibrated light emission by using standard samples furnished by the manufacturer.
Internalization Assay-Cells were incubated with 200 g/ml anti-HSP90 for various time periods from 5 min to 2 h and then washed and fixed in paraformaldehyde. To detect possible internalization of anti-HSP90 antibody, permeabilized cells were incubated with Alexa Fluor 488-conjugated secondary antibody. For endocytosis analysis, cells were incubated 15 min in HEPES-buffered saline solution with CaCl 2 before experiment. The control solution containing 5 M BSA-Alexa Fluor 568 was applied 15 min to attest the absence of nonspecific endocytosis. EGF-BSA-Alexa Fluor 488 or BSA-Alexa Fluor 568 (1:20) were applied for 15 min (about 20 nM EGF) with or without HSP90␣. Cells were post-fixed and mounted on slides to be analyzed.
Immunodetection of Protein Phosphorylation-Cell monolayers were washed five times with ice-cold PBS and lysed (45 min on ice) using 1 ml of lysis buffer containing: 50 mM Tris-HCl, pH 7.4, 1% Nonidet P-40, 0.25% sodium deoxycholate, 150 mM NaCl, 1 mM EGTA, 1 mM NaF, 100 M sodium orthovanadate, 100 M phenylmethylsulfonyl fluoride, and protease inhibitor mixture (Roche Applied Science). Cells were scraped and centrifuged, and lysates were collected. Proteins were boiled for 5 min in 2 l of SDS buffer, fractionated using 8% SDS-PAGE, and transferred to nitrocellulose membranes (Bio-Rad). Membranes were blocked for 1 h with 5% BSA in Tris-buffered saline with Tween 20 (0.1%) (TBS-T) and incubated overnight with antibodies. For EGFR phosphorylation, proteins (20 -60 g/lane) were immunoblotted with anti-P-EGFR (1:50) and revealed with ECL reagents (Amersham Biosciences). Membrane was stripped and reprobed with antibody against EGFR to evaluate the total amount of EGFR.
Wound Healing Migration-A 1-mm-wide cell-free area was generated by gently scratching the cell monolayer with a sterile yellow Gilson pipette tip. The recording DMEM containing drugs or specific antibodies was applied (5% CO 2 in air at 37°C). Migration of astrocytes within the gap was monitored at given time intervals, using a Zeiss microscope equipped with a video camera. The extent of healing was defined as the ratio of denuded areas of the original wound and the wound after 11-22 h.
Bromodeoxyuridine Incorporation Assay-Serum-starved U87 cells cultured in the presence or absence of agents were exposed to bromodeoxyuridine at a concentration of 3 mM for 1 h. Cells were subsequently rinsed and fixed in 70% ethanol overnight, then washed and reincubated for another 10 min in 2 M HCl containing 0.1% Triton X-100 followed by a 20-min incubation in 0.1 M sodium borate. Cells were processed for immunofluorescence using an anti-bromodeoxyuridine antibody (Dako) and an appropriate Alexa Fluor 488-conjugated secondary antibody.
Statistical Analysis-One-way analysis of variance (Statview Software) was used to compare data groups. p values Ͻ 0.05 were significant.

Extracellular HSP90␣ Accelerates Glioblastoma Cell Migration in
Vitro-Immunofluorescence analyses detected the HSP90␣ isoform at the cell surface of glioblastoma U87 cells (Fig. 1A). HSP90␣ isoform was mainly expressed at the front edge of migration of cells. The extracellular localization of HSP90␣ was confirmed by flow cytometry. In contrast, no HSP70 was detected at the cell surface (Fig. 1A, right panel). To analyze the cellular basis for the invasive effect of extracellular HSP90␣, we used a wound healing assay. The addition of recombinant HSP90␣ protein to serum-free medium enhanced cell invasion in a dose-dependent manner (i.e. in the range from 60 ng/ml to 6 g/ml). This resulted in higher rates of gap closure observed as early as the first hour of cell incubation and confirmed in the following hours, 6 and 12 h (Fig.  1B). It is worthy to note that the effect of HSP90␣ is directed more toward cell migration than toward cell proliferation because a lower percentage of bromodeoxyuridine incorporation was observed (Fig. 1B, table). For instance, cell proliferation and invasion were respectively increased by 5 Ϯ 1 and 15 Ϯ 2% from control values following 6 h of cell exposure to 6 g/ml HSP90␣. Such an increase in cell migration was also observed with 10 ng/ml EGF and amplified by the combination of the two molecules (Fig. 1C). An additive effect of the activation of cell invasion was observed when both agents were added together into the cell bath medium. These results suggested that extracellular HSP90␣ exerted an additive effect on the ability of EGF to promote U87 cell migration.
Extracellular HSP90␣ Increases [Ca 2ϩ ] i in U87 Cells-To determine the role of Ca 2ϩ store in U87 response to HSP90␣, we measured [Ca 2ϩ ] i , in the presence and absence of extracellular calcium, in fura-2-loaded cells. Recombinant HSP90␣ induced a transient increase in [Ca 2ϩ ] i within 1-3 min ( Fig.  2A, left panel), whereas EGF induced a more sustained biphasic calcium response with a rapid peak followed by a plateau phase ( Fig. 2A, right panel). In a few experiments, the agonist was applied in the absence of external Ca 2ϩ to reveal release from internal stores. Then, Ca 2ϩ was restored extracellularly to reveal increase in plasma membrane Ca 2ϩ permeability (Fig. 2B). A subsequent application of 2 M TSG was made at the end of some recordings to estimate the Ca 2ϩ content of intracellular endoplasmic reticulum (ER) pools. In nominally Ca 2ϩ -free conditions, 6 g/ml HSP90␣ induced a transient Ca 2ϩ release from intracellular pools (Fig. 2B, left panel). The addition of Ca 2ϩ to the bath medium did not induce a significant increase in [Ca 2ϩ ] i , whereas TSG drastically increased [Ca 2ϩ ] i in the absence of external Ca 2ϩ , demonstrating that HSP90␣ did not deplete completely internal stores. In Ca 2ϩfree conditions, the amplitude of Ca 2ϩ peak induced by EGF remained unchanged, but the duration of the plateau phase was decreased, and Ca 2ϩ readdition to the bath medium resulted in a transient Ca 2ϩ influx through plasma membrane channels due to capacitative Ca 2ϩ entry (Fig. 2B, right panel). Upon stimulation with EGF, the cells showed calcium oscillations in response to HSP90␣ (i.e. with a frequency of three spikes per 5 min; 98%, n ϭ 50), which were rapidly abolished by the removal of HSP90␣ (Fig. 2C) or the addition of the EGFR/ErbB1 inhibitor Tyrphostin (AG1478), whereas the addition of ErbB2 inhibitor Gefitinib had no effect (supplemental Fig. S2). These results suggested that HSP90␣ could promote calcium oscillation through EGFR/ErbB1. ADP/ATP Contributes to HSP90␣-induced [Ca 2ϩ ] i Rise in Glioblastoma Cells-To determine whether the calcium oscillatory process elicited by HSP90␣ and EGF may involve the release of ATP/ADP and their interaction with purinergic (P2) receptors (25,26), we pretreated U87 cells with the P2 receptor antagonist suramin (10 M for 10 min) and the extracellular ATP scavenger apyrase (40 units/ml for 10 min). Both compounds drastically attenuated Ca 2ϩ oscillations generated by exogenously added EGF plus HSP90␣ (Fig. 2C, right  panel). To assess the effect of HSP90␣ and EGF on the subsequent ATP release, cells were exposed to these agents for various periods of time (from 15 min to 1, 6, and 12 h), and supernatants were collected to evaluate amounts of ATP by bioluminescence assay (Fig. 2D). Clearly no significant release of ATP could be detected in the absence or presence of HSP90␣. In contrast, HSP90␣ plus EGF released ATP from U87 cells to a higher extent than did EGF alone within the first 15-30 min of cell stimulation. Altogether, EGF triggers Ca 2ϩ oscillations that are amplified by extracellular HSP90␣ through release of ATP acting on purinergic P2 receptors.
Extracellular HSP90␣ Induces EGFR Internalization and Tyrosine Phosphorylation-Cell exposure to HSP90␣ for 15 min markedly reduced EGFR surface expression (Fig. 3A). U87 cells were incubated in the presence of BSA-Alexa Fluor 568, a membrane-impermeable macromolecule, before adding HSP90␣ for 15 min and then staining permeabilized cells.
Green EGFR was identified at the cell periphery and in the juxtanuclear region (Fig. 3B). In HSP90␣-stimulated cells, BSA co-localized with EGFR in the juxtanuclear region, suggesting a rapid internalization (Fig. 3B). Pretreatment with M␤CD to deplete cholesterol and disrupt lipid raft organization prevented the calcium response to HSP90␣ in the absence or presence of EGF, suggesting that lipid raft integrity may be essential for HSP90␣-induced Ca 2ϩ signaling (Fig.  3C). Cell exposure to HSP90␣ did not change significantly the level of ErbB1 versus total EGFR proteins (not shown). Both EGF and HSP90␣ induced EGFR phosphorylation at Tyr-1068 within 15 min, which was attenuated by cell pretreatment with the PKC␦ inhibitor, rottlerin (Fig. 4A). Because PKC   isozymes do not directly phosphorylate proteins at tyrosine residues, we examined the role of c-Src as an intermediate kinase between PKC and EGFR (supplemental Fig. S3). We found that the c-Src inhibitor PP2 inhibited HSP90␣-induced EGFR phosphorylation, whereas its inactive form (PP3) did not. Because this EGFR transactivation was attenuated by the matrix metalloproteinase (MMP) inhibitor (GM6001) and the heparin-binding (HB)-EGF inhibitor (CRM197; see supplemental Fig. S4), MMP activation and proHB-EGF shedding slightly contribute to this process. Thus, HSP90␣ interacts with PKC␦, which in turn phosphorylates Src to activate EGFR.
HSP90␣-induced EGFR Activation and TLR4 Signaling in U87 Cells-Extracellular HSP90␣ can interact with a variety of membrane receptors, including ErbB2 and TLR4 (7), and FIGURE 3. Extracellular HSP90␣ induced internalization of a pool of EGFR. A, flow cytometry analysis of EGFR expression in unstimulated (basal) and stimulated cells with 6 g/ml HSP90␣ alone or with the combination of HSP90␣ and EGF (10 ng/ml) for 15 min. Gray filled histograms are isotype IgG controls. B, HSP90␣-induced sequestration of BSA-Alexa Fluor 568 in EGFR endosomes. Cells were incubated for 15 min in HEPES-buffered saline solution containing HSP90␣ plus 5 M BSA-Alexa Fluor 568 (room temperature). The merged panel indicates co-localization of EGFR and BSA (yellow; n ϭ 3). C, raft-disrupting drug inhibits HSP90␣-induced Ca 2ϩ signaling. Cells pretreated with 7.5 mM M␤CD for 60 min were exposed to HSP90␣ and/or 10 ng/ml EGF. (2 M TSG; average of 50 cells; n ϭ 3). FEBRUARY (19). As observed for HSP90␣, the bacterial LPS (1 g/ml), a representative TLR4 ligand, also induced EGFR phosphorylation at Tyr-1068 within 15 min (Fig. 4A). Flow cytometry analysis indicated that TLR4 was expressed, whereas TLR2 could not be detected at the cell surface of U87 cells (Fig. 4B). No significant cell expression of TLR2 was shown (supplemental Fig. S5). Following external application of HSP90␣, TLR4 totally disappeared at the cell surface within 15 min. Cell surface expression of EGFR was also significantly reduced on U87 cells treated with LPS. Confocal microscopy observations revealed distinct patterns of cell expression, i.e. EGFR was mainly located at the cell periphery, whereas TLR4 was found throughout the cell (Fig.  4C). However, the two receptors co-localized on the plasma membrane and concentrated in pseudopodia (or filopodia). Upon cell stimulation with HSP90␣, EGFR was redistributed from the cell membrane to the juxtanuclear region where it co-localized with a small pool of internalized TLR4 (Fig. 4C). (Internalized EGFR by HSP90␣ is not targeted for degradation; see supplemental Fig. S7.) To determine the downstream signaling events of both TLR4 and EGFR activation, cells were incubated with 6 g/ml HSP90␣ for 0, 5, 10, 30, 60, and 320 min (Fig. 4D). Phosphorylated and total AKT, p44/42 MAPK, SAPK/JNK, NFB p65, and IRF-3 were examined by Western blot analysis. The AKT phosphorylation was induced within 5-10 min and then decreased progressively with time. The phosphorylation of p44/42 MAPK observed within 5 min reached a peak at 10 min and then returned to basal level by 6 h. The phosphorylation of SAPK/JNK also occurred rapidly within 5 min and followed a similar time course as AKT but to a lesser extent. The transcription factors, NFB p65 and IRF-3, related to TLR4 activation, were phosphorylated within the first 10 min of HSP90␣ application and maintained for at least 6 h. (Intracellular calcium is required for NFB p65 phosphorylation; see supplemental Fig. S8.) Because clathrin-mediated endocytosis is the major pathway of EGFR internalization, we also examined the co-localization of internalized receptor with clathrin-HC (heavy chain). As shown in Fig. 4E, P-EGFR (Tyr-1068) was internalized with FIGURE 4. Tyrosine phosphorylation of EGFR and partial co-localization of TLR4 and EGFR in glioblastoma cells. A, extracellular HSP90␣ induces EGFR tyrosine phosphorylation. U87 cells were unstimulated (control (Cont.)) or stimulated with EGF or HSP90␣ and 1 g/ml LPS for 15 min; EGFR phosphorylation at Tyr-1068 was analyzed by Western blot. In a few cases, U87 cells were pretreated for 60 min with the PKC␦ inhibitor, rottlerin (10 M). The membrane was stripped and reprobed for total ErbB1 subtype. On the bottom of the figure, values indicate the relative optical density (OD) of P-EGFR versus ErbB1 analyzed by densitometry (mean Ϯ S.D., n ϭ 2). B, flow cytometry analysis of cell surface expression of EGFR (ErbB1), TLR4, and TLR2 in unstimulated (basal) and stimulated U87 cells with HSP90␣ or LPS (with EGF) for 15 min. Gray filled histograms are isotype IgG controls (n ϭ 3). C, U87 cells were unstimulated (control) or stimulated with HSP90␣ for 15 min and double-stained for TLR4 and EGFR (ErbB1). In the merged panel, yellow labeling indicates co-localization of both receptors. Confocal optical sections of 0.4 m thickness are shown (n ϭ 3). D, phosphorylation of Akt, p44/p42 MAPK, and SAPK/JNK (left panel) and of the transcription factors, NFB p65 and IRF-3 (right panel). U87 cells were exposed to 6 g/ml HSP90␣ for the indicated periods, and whole cell lysates were subjected to immunoblot analysis using the indicated specific antibodies recognizing phosphorylated or total forms of proteins (n ϭ 4). E, co-localization of HSP90␣-induced P-EGFR and clathrin-HC. Serum-starved U87 cells, stimulated with 6 g/ml HSP90␣ for 30 min, were fixed and double-stained for the phosphorylated form of EGFR (Tyr-1068) (red) and the clathrin-Alexa Fluor 488 (green). In the merged panel, yellow labeling indicates co-localization of clathrin and P-EGFR. Confocal optical sections of 0.4 m thickness are shown (n ϭ 3). DiC, differential interference contrast.

External HSP90␣ Transactivates EGFR
clathrin after 30 min of cell stimulation with HSP90␣. No co-localization with P-EGFR and caveolin-1 was observed (data not shown). Furthermore, chlorpromazine, a well known inhibitor of clathrin-mediated endocytosis, significantly blocked the internalization of P-EGFR and the MAPK signaling induced by HSP90␣ (see supplemental Fig. S6). Thus, clathrin-mediated internalization is essen-tial for a sustained HSP90␣-mediated EGFR signaling. Because cholesterol depletion by M␤CD also blocks calcium signaling and endocytic processes, our findings support the view that HSP90␣-induced endocytosis recruits at least two kinds of membrane rafts: one including TLR4 and the second including EGFR. Altogether, our data suggest that signaling via the PKC␦/c-Src pathway is involved in FIGURE 5. shRNA-mediated depletion of TLR4 inhibits HSP90␣-induced Ca 2؉ signaling in U87 cells. A, protein lysates from both normal human astrocytes (NHA) and glioblastoma U87 cells were probed with anti-TLR4 (HSC70 as loading control). TLR4 expression in normal human astrocytes and U87 cells was shown, and cell stimulation with 1 g/ml LPS for 60 min was performed. Flow cytometry analysis of TLR4 expression in transfected U87 cells (white histogram) when compared with IgG control (gray filled histogram; n ϭ 3) as performed. B, effects of HSP90␣, LPS (1 g/ml), TSG (2 M), and HSP90␣ plus EGF on [Ca 2ϩ ] i in TLR4 shRNA (grey curves) and control shRNA (black curves) transfected U87 cells (average of 6 cells; n ϭ 3). C, consequences of shRNA-mediated depletion of TLR4 and EGFR (ErbB1) onto the cell invasion. The quantitative effect of HSP90␣ (6 g/ml) and/or EGF (10 ng/ml) on the closure of wound after a 12-h treatment is depicted in the table (mean Ϯ S.D., n ϭ 2; *, p valuesϽ0.05 versus control). The following cells were used as control in this specific experiment; cultures of U87 cells and normal human astrocytes (Lonza Walkersville, Inc.) were grown to confluency in DMEM plus 10% fetal bovine serum (Lonza). For lentiviral transduction particles and transfections, human TLR4 shRNA, EGFR (ErbB) shRNA and control lentiviral particles were used according to the manufacturer's recommendations (Santa Cruz Biotechnology). FEBRUARY (Fig. 5A) abolished the Ca 2ϩ response to HSP90␣ and LPS but did not affect the TSG-induced increase in [Ca 2ϩ ] i (Fig. 5B). This down-regulation induced a significant reduction in both the release of internal Ca 2ϩ stores and the secondary external Ca 2ϩ influx in response to EGF plus HSP90␣. Thus, TLR4 may be required for HSP90␣-induced [Ca 2ϩ ] i increase in U87 cells. The contribution of TLR4 and EGFR (ErB1) to HSP90␣-mediated cell invasion is shown in Fig. 5C. In shRNA-mediated knockdown of TLR4 or EGFR, the cell migration was severely impaired in response to HSP90␣. Thus, in agreement with our previous results, TLR4 is required for HSP90␣ effects. To rule out a role of contamination of recombinant HSP90␣ with LPS, cells were preincubated with Polymyxin B, an endotoxin-binding agent (10 M; 30 -60 min). HSP90␣ without contaminant LPS was always able to amplify the EGF-induced cell migration (Fig. 6A). This cell pretreatment reduced but did not inhibit HSP90␣-induced transient rise in [Ca 2ϩ ] i (Fig. 6B, left panels). Preincubation of U87 cells with anti-TLR4-blocking antibody (27) abolished the increase in [Ca 2ϩ ] i and the cell migration induced by HSP90␣ (Fig. 6, right panels) and decreased EGFinduced cell responses (Fig. 6, right panels). Altogether, HSP90␣, through a direct or indirect activation of TLR-4, upregulates EGF-induced cell responses.

DISCUSSION
Cell surface expression of HSP90␣ has been associated in various tumor types to their migration capabilities and metastatic potential (5)(6)(7)(8). Here, we show in glioblastoma U87 cell line that extracellularly applied HSP90␣ activates a signal transduction pathway involving TLR4 to amplify the effects of EGF on cell shape and cell migration. More specifically, extracellular HSP90␣ triggers EGFR phosphorylation at Tyr-1068 through a pathway that can be inhibited by rottlerin and PP2, suggesting a role of PKC␦ and c-Src (Fig. 7). Extracellular HSP90␣ also transiently increases cytosolic [Ca 2ϩ ] in U87 cells and converts calcium response induced by EGF from a transient to an oscillatory pattern. Ca 2ϩ mobilization induces ATP release that amplifies the cell response through a suramin-inhibited pathway. Altogether, these results identify the additive effect of extracellular HSP90␣ and EGF on glioma cell migration capabilities in human brain tumors.
TLR4-mediated transactivation of EGFR has been identified in epithelial cells (8,19). For example, this receptor crosstalk has been observed in response to Helicobacter pylori protein in gastric cells (19). Here, we show that extracellular HSP90␣ can induce TLR4-mediated EGFR transactivation and its downstream effects, i.e. cell migration and Ca 2ϩ signaling, in U87 cells. Furthermore, shRNA-mediated downregulation of TLR4 or its neutralization with a specific antibody prevents HSP90␣-induced activation of EGFR. Co-localization of EGFR, TLR4-ligand complexes, and extra-cellular HSP90␣ in specific membrane domains could favor an interaction between TLR4 and EGFR that would allow EGFR activation. Depletion of cholesterol by methyl-␤-cyclodextrin disrupted this co-localization and impaired HSP90␣induced Ca 2ϩ rise, even in the presence of EGF. M␤CD could prevent the activation of PKC␦, Src, and Lyn kinases, as demonstrated in other cell types (19,28,29). Interestingly, EGFR and TLR4 were observed to co-localize in pseudopodia or filopodia of non-stimulated glioblastoma cells whose polarization and stabilization determines cell motility (30). Upon cell exposure to HSP90␣, EGFR and TLR-4 were rapidly internalized with kinetics similar to that induced by their respective ligands, EGF and LPS.
HSP90␣ appears to trigger EGFR transactivation through a pathway that is sensitive to both PKC␦ and c-Src inhibitors (Fig. 7). Because PKC isozymes do not directly phosphorylate proteins at tyrosine residues, c-Src would be an intermediate kinase between PKC␦ and EGFR (35,38). The expression and activity of PKC isozymes are highly elevated in gliomas and glioma cell lines (31)(32)(33), and PKC inhibitors markedly reduce glioma cell proliferation (34). PKC␦ isoform was shown to phosphorylate several kinases including c-Src (35), ADAM, and Lyn (36). Cellular Src and EGFR act synergically to facilitate the progression of human brain tumors (35,37,38), whereas ADAM and Lyn kinases regulate activation of MMPs, proHB-EGF shedding, and EGFR tyrosine phosphorylation (20,36). Inhibition of extracellular HSP90␣ was shown to decrease MMP-2 activity, blocking invasiveness (8). The EGFR (ErbB1) kinase inhibitor, AG1478, only attenuated HSP90␣induced transactivation of EGFR at Tyr-1068 in U87 cells, at a dose known to completely abrogate EGF-induced phosphorylation of EGFR. This suggests that HSP90␣-induced EGFR Tyr-1068 phosphorylation through PKC␦ and c-Src is upstream of EGFR kinase activity. This result agrees with similar studies that used phorbol 12-myristate 13-acetate to transactivate the EGFR (35). Intracellular Ca 2ϩ modulates the proliferative effects of EGF (16,39,40), and complex Ca 2ϩ oscillations are associated with human astrocytoma cell migration (25,41,42). A major component of Ca 2ϩ homeostasis is the ER, which stores micromolar levels of Ca 2ϩ under control of Ca 2ϩ -ATPases known as sarcoplasmic reticulum Ca 2ϩ -ATPase (SERCAs), which maintain a low cytosolic [Ca 2ϩ ]. Specific inhibition of sarcoplasmic reticulum Ca 2ϩ ATPase activity with TSG results in ER Ca 2ϩ depletion and consequent opening of storeoperated calcium channels located in the plasma membrane. These channels elevate cytosolic [Ca 2ϩ ], and under physiological conditions, permit refilling of the ER Ca 2ϩ pool. The magnitude of the increase in [Ca 2ϩ ] i evoked by an agonist is thus dependent on the size of stores, leakiness of the ER membrane for Ca 2ϩ , and activity of store-operated calcium channels. Here, we show that extracellular HSP90␣ transiently increases cytosolic [Ca 2ϩ ] in glioblastoma U87 cells and converts calcium response induced by EGF from a transient to an oscillatory pattern. Similar calcium oscillations induced by EGF have been previously reported in astrocytes in response to glutamate (42), pro-inflammatory cytokines (41), or thimerosal, which directly activates the inositol 1,4,5- Our data are summarized in the box and linked to pathways (outside the box) previously proposed by other authors for cell migration. Specific inhibitors used in this work are written in grey. Upon HSP90␣ stimulation of TLR4, PKC␦ phosphorylates c-Src to activate EGFR in rafts. Assemblage of EGFR, c-Src, and phospholipase C␥ (PLC␥) might form a platform for the EGFR, resulting in the release of ATP from U87 cells. The released ATP, through its P2 receptors, triggers intracellular Ca 2ϩ waves, leading to the activation of ADAM protein(s), which cleaves pro-HB-EGF at the cell surface. HB-EGF in an autocrine and/or paracrine fashion binds EGFR, amplifying signals into the intracellular signaling network. The initial activation of PKC␦ could also involve Lyn, MMP2/9, and HB-EGF release. FEBRUARY 4, 2011 • VOLUME 286 • NUMBER 5 triphosphate receptor and induces calcium release from internal stores (43). These stimuli also induce release of ATP by astrocytes. Elevation of cytosolic Ca 2ϩ is a major regulator of nucleotide export in astrocytes, either by channel-mediated influx or through vesicle exocytosis. Rho activation and other changes in cytoskeletal organization have also been implicated in ATP release (44). Here, we show an elevated cytosolic Ca 2ϩ and an increase in F-actin content by stimulation of EGFR in U87 cells (supplemental Fig. S1). Stimulation of EGFR was reported to activate RhoA/Rho-associated protein kinase signaling, which could synergize with Ca 2ϩ mobilization to induce ATP release in response to HSP90␣ plus EGF.

External HSP90␣ Transactivates EGFR
Altogether, our data indicate that the ability of extracellular HSP90␣ to increase cytosolic [Ca 2ϩ ] in tumoral astrocytes and to induce complex Ca 2ϩ oscillations in the concomitant presence of EGF probably facilitates the migration of these cells. It remains to be demonstrated whether amplification of this pathway in tumor cells accounts for the relationship established between extracellular HSP90␣ and tumor cell invasion and metastasis formation.