Podosomes in osteoclast-like cells: structural analysis and cooperative roles of paxillin, proline-rich tyrosine kinase 2 (Pyk2) and integrin αVβ3

Cultured cells develop discrete sites of contact with the extracellular matrix, in which clustered transmembrane proteins, particularly of the integrin family of adhesion receptors, link extracellular matrix proteins to the actin cytoskeleton within the cell (Hynes, 1992EF17). Focal adhesion plaques represent the most ubiquitous and well known of these contacts. They are characterized by their connection to easily detectable bundles of actin filaments, called stress fibers (Burridge and Chrzanowska-Wodnicka, 1996EF6; Jockusch et al., 1995EF18). Most cells additionally develop smaller peripheral contact sites (focal complexes) often associated with lamellipodia, which can develop into bona fide focal adhesions (Rottner et al., 1999EF37). By contrast, only very few cell types form different, much less understood adhesion sites that consist of dot-shaped, F-actin-rich close contacts, which have been initially described in cells transformed by Rous sarcoma virus (RSV; David-Pfeuty and Singer, 1980EF13) and later in monocyte-derived cells including macrophages (Marchisio et al., 1987EF27) and osteoclasts (Marchisio et al., 1984EF26). These cytoskeletal structures have been named `rosettes', owing to their appearance as small black rings in interference reflection microscopy (IRM; David-Pfeuty and Singer, 1980EF13). However the name `podosomes', which denotes their foot-like appearance in electron-microscopy is now most commonly accepted (Tarone et al., 1985EF49).

It has been recognized that podosomes share major structural components with focal adhesions, e.g. α-actinin, vinculin and talin (David-Pfeuty and Singer, 1980EF13; Marchisio et al., 1984EF26; Marchisio et al., 1987EF27). Furthermore, both structures are sites of increased protein tyrosine phosphorylation (Tarone et al., 1985EF49; Burridge and Chrzanowska-Wodnicka, 1996EF6). On the other hand, proteins, like gelsolin (Yin, 1987EF58) and fimbrin (Tplastin; Bretscher, 1981EF4), which regulate actin filament organization, are enriched only in podosomes (Wang et al., 1984EF55; Carley et al., 1986EF8; Marchisio et al., 1984EF26; Marchisio et al., 1987EF27). Recent genetic evidence underlines the particular importance of two such proteins for podosome function in vivo, of gelsolin (Chellaiah et al., 2000EF9) and WASP (Linder et al., 1999EF25). These data suggest that the mechanisms linking cell adhesive contacts to the F-actin cytoskeleton are similar in focal adhesions and podosomes, but that both structures differ in the mechanisms that regulate the microarchitecture of actin filaments.

Using time-lapse video and photobleaching techniques in RSV-transformed rat kidney cells microinjected with fluorescent α-actinin, Stickel and Wang (Stickel and Wang, 1987EF46) observed that podosomes underwent rapid formation, movement and breakdown. Fluorescing α-actinin molecules were replaced in podosomes with a halftime of fluorescence recovery of 4 seconds, more than 10 times faster than in focal adhesions of non-transformed cells. Thus, podosomes are highly dynamic structures, that rapidly move and change shape and size. Chen and co-workers have noted that podosomes in RSV-transformed cells, but not focal adhesions in the non-transformed cells, colocalize with proteolytic activities that degrade extracellular matrix proteins (Chen et al., 1984EF11; Chen, 1989EF10). In transmission electron microscopic images, they associated these proteolytic activities with membrane protrusions, called invadopodia, formed adjacent to the rosette contact sites. Two other studies (Nitsch et al., 1989EF31; Ochoa et al., 2000EF32) have described invaginations of the plasma membrane reaching into the center of podosomes. These are associated with membrane transport phenomena, as indicated by the presence of the GTPase dynamin 2 (Ochoa et al., 2000EF32), and might represent sites of local protease release. Moreover, podosomes co-distribute with matrix metalloproteinases in osteoclasts (Blavier and Delaissé, 1995EF2; Sato et al., 1997EF39). Therefore, podosomes appear to be structures that uniquely combine adhesive functions with proteolytic degradation of the extracellular matrix, characteristics that are especially important for highly invasive cells.

The adhesive mechanisms establishing the podosome contact with the extracellular matrix are not well understood. Experimental evidence that demonstrates a direct implication of integrins in podosome function is scarce compared with their well-established role in the formation of focal adhesion plaques (Burridge and Chrzanowska-Wodnicka, 1996EF6). Yet integrins have been localized to podosomes that co-distribute either with F-actin (Marchisio et al., 1988EF28; Nakamura et al., 1999EF30) or with the plasma membrane region directly adjacent to it (Marchisio et al., 1988EF28; Helfrich et al., 1996EF16; Zambonin-Zallone et al., 1989EF61). One study (Johansson et al., 1994EF19) has localized a subpopulation of β1 integrin subunits that contain a phosphorylated tyrosine in its cytoplasmic domain in podosomes of RSV-transformed cells. In osteoclasts, podosomes are putative precursors of the sealing zone, a tight adhesion structure that seals the cells onto bone during resorptive activities (Väänänen and Horton, 1995EF53). Osteoclasts express integrin αVβ3 as major integrin adhesion receptor, which is required for the formation of the sealing zone ex vivo and for bone resorbing activity in vivo (McHugh et al., 2000EF29). However, it is highly questioned whether this integrin provides the adhesive link between the sealing zone and bone, because it could not be localized in this adhesive structure (Lakkakorpi et al., 1991EF21; Helfrich et al., 1996EF16; Väänänen et al., 2000EF54). By contrast, several reports demonstrate an association of this integrin with podosomes in non-resorbing osteoclasts or osteoclast-like cells (Nakamura et al., 1999EF30; Helfrich et al., 1996EF16; Zambonin-Zallone et al., 1989EF61).

We have decided to study podosomes in cells of the monocytic lineage, which represent the sole untransformed cell types forming podosomes. We observed the organization and the molecular architecture of these structures during the differentiation of human and chicken monocyte-derived precursor cells to bone-resorbing osteoclasts. This differentiation was initiated by the cytokine RANKL-ODF (ligand of receptor activator of NFκB-osteoclast differentiation factor; Yasuda et al., 1998EF57; Anderson et al., 1997EF1), which induced high expression levels of integrin αVβ3, reinforced the distribution of podosomes at the cell periphery, and initiated their partial fusion to larger F-actin-containing structures. We also observed changes in the distribution of tyrosine-phosphorylated proteins in the podosome zone, together with the terminal steps of differentiation. Moreover, we demonstrate the colocalization of integrin αVβ3 in the juxtamembrane region adjacent to podosomes with the non-receptor tyrosine kinase Pyk2 and the cytoskeletal adapter protein paxillin. Both of these latter proteins, especially paxillin, incorporated high amounts of phosphate into tyrosine residues during cell adhesion. Finally, we provide evidence from in vitro binding experiments pointing to a novel, direct interaction of both paxillin and Pyk2 with the C-terminal region of the integrin β3 tail. This interaction might be a crucial element in podosome function in osteoclast-like cells.

We used the following antibodies: mAb 23C6, directed against integrin αVβ3, was kindly donated by Dr M. Horton, London; mAbs clone 5 (anti-tensin), 21 (anti-zyxin), 11 (anti-Pyk2), 21 (anti-p130cas), 349 (anti-paxillin), 30 (anti-cortactin) were purchased from Transduction Laboratories; mAb BM-75.2 (anti-α-actinin) was from Sigma; mAb 4F11 (anti-cortactin) from Upstate Biotechnology; anti-phosphotyrosine mAbs were 4G10 (produced in our laboratory) and PY99 from Santa Cruz Biotech; mAb 327 (anti-pp60src) was donated by Dr J. Brugge, Boston; and mAb 3F3A5 (anti-WASP) was donated by Dr David Nelson (Bethesda, MD). The following antibodies were purchased from the Developmental Studies Hybridoma Bank maintained by the John Hopkins University, Iowa City: mAb 8e6 (anti-talin) developed by C. Otey and K. Burridge; and mAb VN3-24 (anti-vinculin) developed by S. Saga. Fluorescein- and peroxydase-labeled secondary antibodies were from Jackson Laboratories and rhodamine-phalloidin was from Molecular Probes. Human recombinant soluble RANKL-ODF was provided by Immunex (Seattle, WA) and human recombinant M-CSF was purchased from R&D SYSTEMS. David Schlaepfer (La Jolla, CA) kindly provided a pCDNA3.1 expression vector for Myc-tagged human full-length Pyk2.

Chicken monocytes were isolated from Ficoll-separated peripheral blood cells and cultured as described previously (Solari et al., 1995EF45; Boissy et al., 1998EF3). After 2 days of initial culture, adherent macrophages were trypsinized and replated onto coverslips or plastic dishes. For RANKL-ODF treatment, cells were cultured in αMEM containing 10% fetal calf serum (HyClone Laboratories) and 30 ng/ml RANKL-ODF for 2 to 5 days. Adherent human peripheral blood macrophages, similarly obtained from Ficoll-separated peripheral blood leukocytes, were cultured in αMEM, 10% fetal calf serum (HyClone Laboratories) and treated for 6-9 days with 50 ng/ml RANKL-ODF and 20 ng/ml M-CSF. Murine RAW 264.7 cells were obtained from ATCC (Rockville, Maryland) and cultured in DMEM, 10% fetal calf serum (HyClone Laboratories).

Cells grown on glass coverslips were for a short period immersed in phosphate-buffered saline (PBS) and immediately fixed in 2.5% paraformaldehyde, 5% sucrose in PBS for 10 minutes at room temperature. Cells were permeabilized in 0.1% Triton X-100, 1 mM sodium orthovanadate in PBS for 5 minutes. After washing, coverslips were blocked with 1% bovine serum albumin, 1 mM sodium orthovanadate in PBS and incubated with first and secondary antibodies and rhodamine phalloidin in this same buffer. PBS-washed samples were mounted in Fluorsave^TM (Calbiochem) containing 0.05% p-phenylenediamine and observed the next day with an epifluorescence microscope (Axioplan 2, Zeiss) equipped for confocal immunofluorescence analysis (LSM510, Zeiss).

Peptides were synthesized as described (Boissy et al., 1998EF3). They consisted of N-terminally biotinylated penetratin peptide (RQIKIWFQNRRMKWKK), which was directly followed by integrin β cytoplasmic domain sequences (represented in Fig. 6A). Peptide purity was controlled by HPLC and their correct mass confirmed by electrospray mass spectrometry. Recombinant structural mimics of human integrin β1 and β3 tails were provided by David Calderwood and Mark Ginsberg (La Jolla, CA). These proteins consist of the full-length integrin cytoplasmic domains (Fig. 8) linked at their N terminus to a heptad-repeat and a His-Tag sequence, as described (Calderwood et al., 1999EF7; Pfaff et al., 1998EF35).

A pGEX-2TK^TM vector for bacterial expression of glutathione-S-transferase-(GST)-linked human paxillin was kindly provided by Dr Mark Ginsberg (La Jolla, CA) with kind permission of Dr Ravi Salgia (Boston, MA). Recombinant GST-paxillin was produced in Escherichia coli strain DH5α (GIBCO Life Technologies) and purified on glutathione crosslinked agarose (Sigma) from bacterial lysates. Lysis and purification were performed in 20 mM Tris/HCl pH 8.0, 0.1% Triton X-100, 100 mM sodium chloride, 1mM EDTA, 1 mM DTT. Bound paxillin was eluted from glutathione beads with 20 mM reduced glutathione, 0.75 M Tris/HCl pH 9.6 and added to 1/2 volume of 1.5 M Tris/HCl pH 6.8.

A pEGB mammalian GST-fusion expression vector to produce GST-linked human Pyk2 N-terminal domain (amino acids 1-407) (GST-Pyk2-NT) was provided by David Schlaepfer (La Jolla, CA; Sieg et al., 2000EF44). GST-Pyk2-NT was purified on glutathione agarose from lysates of transiently transfected 293T cells.

After two short washes with PBS, cells were lysed in buffer A (50 mM Tris/HCl pH 7.4, 75 mM sodium chloride, 50 mM sodium fluoride, 40 mM sodium pyrophosphate, 1 mM sodium orthovanadate, 1 mM EDTA and protease inhibitors (Cømplete^TM, Boehringer Mannheim)) containing 1% Triton X-100, 0.5% sodium deoxycholate and 0.1% SDS. After 30 minutes on ice, lysates were cleared by centrifugation and agitated with Protein G Sepharose 4B (Sigma) for 1 hour at 4°C. Precleared lysates were incubated overnight with antibodies and Protein G Sepharose at 4°C. Immunoprecipitates were washed five times with buffer A containing 1% Triton X-100, 0.5% sodium deoxycholate and 0.1% SDS, and processed for SDS polyacrylamide electrophoresis followed by electrotransfer onto nitrocellulose (Amersham). Transfer efficiency was verified by Ponceau S staining. Nitrocellulose membranes were blocked overnight with 100 mM Tris/HCl pH 7.4, 150 mM sodium chloride containing either 5% fat-free milk powder (Régilait, France) or 3% bovine serum albumin, 0.1% Tween 20 (for incubations with antiphosphotyrosine mAbs). After incubations with first and peroxydasecoupled secondary antibodies, bound antibodies were detected by chemiluminescence (ECL, Amersham).

For peptide binding experiments, cell lysates were prepared in buffer A containing 1% Triton X-100, 0.2% sodium deoxycholate. Cleared lysates were diluted with one volume buffer A and incubated with biotinylated peptides prebound to crosslinked streptavidin agarose (Sigma) overnight at 4°C. After five to six washes with buffer A, containing 0.5% Triton X-100 and 0.1% sodium deoxycholate, bound proteins were eluted by boiling in reducing sample buffer for SDS gel electrophoresis. The presence of equal amounts of the different peptides in each experiment was verified on Coomassie-stained 15% SDS polyacrylamide gels and bound proteins were identified by specific antibodies on 8-10% SDS polyacrylamide gels transferred to nitrocellulose. Binding to recombinant GST, GST-paxillin and GST-Pyk2-NT was performed under identical conditions, except that 0.5 mM DTT and 2 mg/ml bovine serum albumin were added to the binding buffer. Recombinant structural mimics of integrin β tails were used prebound to a Ni²⁺-resin (Novagen) as described (Pfaff et al., 1998). To maintain their interaction with the resin, EDTA in buffer A had to be replaced with 1 mM CaCl₂ and 2 mM MgCl₂ throughout the experiment.

Confocal images, scanned immunofluorescence photographs, as well as autoradiograms of western blots and scans of Coomassie Blue stained SDS-PAGE gels were digitally processed using the program Adobe Photoshop for the layout of the figures.

Chicken peripheral blood monocytes were seeded onto plastic dishes and cultured in the absence or presence of the osteoclast differentiation factor RANKL-ODF. Within the first days of culture, monocytes adhered and spread out developing podosomes as major dot-like structures labeled by phalloidin (Fig. 1). At this early stage, podosomes were distributed throughout the basal cell body (Fig. 1, left panel). During continued culture cell sizes increased by cell fusion and spreading, and podosomes became more and more limited to peripheral cell regions. In the absence of RANKL-ODF the sizes of the moderately multinucleated cells remained within 20-80 μm and podosomes appeared in irregular and interrupted peripheral rings or loops, possibly representing local extensions of the plasma membrane (Fig. 1, middle panel). RANKL-ODF treatment for 2-4 days led to larger, more multinucleated cells (up to several hundred μms) with a smooth periphery lined by a band of tightly packed podosomes (Fig. 1, right panel). These podosomes showed an increasingly dense phalloidin-staining and, occasionally, individual actin spots fused to larger F-actin containing structures (Fig. 3B,C). Similar changes in the distribution of podosomes were observed during RANKL-ODF/M-CSF-induced differentiation of human peripheral blood monocytes (Figs 2B, 3C, 5). Human osteoclast precursors generally contained much more podosomes, but only a reduced number of osteoclasts were generated in vitro compared with the chicken cells (Fig. 2B). RANKL-ODF addition increased the expression of integrin α Vβ3 from hardly detectable to moderate or extremely high levels in chicken (Fig. 2A), as well as in human osteoclast precursors (Fig. 2B; see also Quinn et al., 1998EF36). The highly expressed integrin was distributed throughout the osteoclast plasma membrane, including in the peripheral podosome zone, where it appeared especially enriched (Figs 2, 5). In experiments performed with RANKL-ODF-differentiated chicken osteoclasts cultured for 5-6 days on dentine slices, we revealed their capacity to resorb bone (P. Boissy, O. Destaing, M. P. and P. J., unpublished). Thus, the presence of podosomes is characteristic for all differentiation stages during the macrophage-osteoclast transition, but their increased density at the cell periphery associated with high expression of the integrin αVβ3 reflects differentiation towards a bone-resorbing osteoclastic cell phenotype.

To gain insight into the mechanisms that regulate podosome architecture and function, we have analyzed the association of tyrosine-phosphorylated proteins with these cytoskeletal structures during in vitro osteoclast differentiation. The major structures labeled by monoclonal anti-phosphotyrosine antibodies at all stages of differentiation were indeed podosomes (Fig. 3). Podosome-associated phosphotyrosine predominantly localized to the extracellular substrate-oriented podosome tips in early macrophages and in non-RANKL-ODF-treated cells (Fig. 3). This was especially evident in confocal images scanned in the z-axis, that sectioned individual podosomes in their centers (Fig. 3A,C, insets). However, the particularly F-actin-rich podosomes in RANKL-ODF treated osteoclast-like cells showed a reduced phospho-tyrosine staining (Fig. 3B,C, arrows in right-hand panels). Only the podosomes located at the side of the podosome ring facing the cell-centers were strongly stained at the tips (Fig. 3B,C, arrowheads in righthand panels). Reduced phospho-tyrosine staining of particularly large, F-actin-rich podosomes was also occasionally observed in non-RANKL-ODF-treated cells (see arrow in Fig. 3A, right-hand panel). These variations in the association of podosomes with tyrosine-phosphorylated proteins indicate a functional specialization of the podosome ring during RANKL-ODF-induced differentiation, which remains to be further elucidated.

To identify podosome proteins phosphorylated on tyrosine in a cell adhesion-dependent manner, we biochemically analyzed cellular lysates in western blotting with anti-phosphotyrosine antibodies. A distinct profile of tyrosine-phosphorylated proteins with major bands at 60, 70 and 85, and between 110 to 130 kDa was detected in lysates of adherent osteoclast precursors (Fig. 4). This profile was essentially unaffected during RANKL-ODF-induced differentiation (Fig. 4A). Tyrosine phosphorylation was strongly reduced in lysates of cells kept in suspension for 2 hours, but reappeared during subsequent cell adhesion in a time course reflecting the extent of cell spreading (Fig. 4B). The major tyrosine phosphorylated protein migrating at 70 kDa in SDS-PAGE gels was identified as paxillin (Fig. 4C). Because immunoprecipitation with antiphosphotyrosine antibodies did not markedly reduce paxillin levels in the lysates, we conclude that only a small portion of paxillin was actually phosphorylated on tyrosine (Fig. 4D). In similar experiments, we detected adhesion-dependent tyrosine phosphorylation of the 115 kDa protein kinase Pyk2 (Fig. 4E) and of p 130cas (data not shown). We also noted that the band at 60 kDa contained tyrosine-phosphorylated pp60c-src (data not shown), which remained markedly tyrosine phosphorylated in the suspended cells (Fig. 4B,E). So far, we were unable to identify the prominent tyrosine-phosphorylated protein migrating at 85 kDa. In conclusion, cell adhesion and spreading of osteoclasts and osteoclast precursors result in high-level tyrosine phosphorylation of a characteristic set of proteins, most notably, however, of a minor subpopulation of paxillin molecules.

We next analyzed the precise cellular location of podosome proteins by immunofluorescence (Fig. 5). Both paxillin and Pyk2 accumulated in the zone containing podosomes, but they were essentially confined to regions between the individual F-actin-containing core structures of podosomes (Fig. 5). Confocal images scanned in the z-axis narrowed down their location as being close to the extracellular substrate-oriented plasma membrane between the podosomes (Fig. 5, insets in left panels). A broader analysis showed that most of the proteins known to associate with focal adhesion plaques in other cells, like talin, vinculin, α-actinin, zyxin, tensin, p130cas and the integrin αVβ3, share a very similar subcellular location with paxillin and Pyk2 in the podosome zone (Fig. 5, and data not shown). A different distribution was observed for the src-substrate cortactin, which colocalized with the F-actin core of podosomes in horizontal confocal sections (resulting in yellow podosomes in images with merged F-actin and cortactin channels, Fig. 5, bottom panels). Vertical confocal sectioning (Fig. 5, inset in bottom left panel) revealed its preferential location in the basal tips of podosomes. A similar distribution was observed for the protein WASP (data not shown). Thus, the subcellular location of the proteins previously identified as major substrates of adhesion-dependent tyrosine phosphorylation (Fig. 4) was to some extent different from the subcellular structures stained during immunofluorescence with anti-phosphotyrosine antibodies (Fig. 3). Most notably, the prominent phosphotyrosine signal detected on the basal tips of podosomes (Fig. 3) corresponded best to the distribution of cortactin, but much less to that observed for paxillin, Pyk2 and integrin αVβ3 (Fig. 5). However, we could not detect adhesion-dependent phosphorylation on tyrosine in immunoprecipitates of cortactin. Moreover, although cortactin migrates as a 85 kDa band in SDS-PAGE gels, its immunodepletion from the cell lysates did not remove the major tyrosine-phosphorylated 85 kDa band (see Fig. 4; data not shown). Hence, an as yet unidentified protein might be responsible for the phosphotyrosine signal detected in situ on the podosome tips or, alternatively, the small subpopulation of paxillin molecules, which is actually phosphorylated on tyrosine (Fig. 4D), might be located close to the basal podosome tips.

To gain insight into molecular interactions that link adhesion receptors to the particular actin cytoskeleton in macrophages and osteoclast-like cells, we performed affinity precipitation experiments with peptides containing the C-terminal third of integrin β cytoplasmic domains (Fig. 6A). In a previous study, we have shown that such a peptide containing the 17 C-terminal amino acids of the chicken integrin β3 tail, which is rendered membrane permeable by its coupling to penetratin (Derossi et al., 1994EF14), blocked the spreading of cultured macrophages expressing high levels of αVβ3 (Boissy et al., 1998EF3). When incubated with a lysate of macrophages or osteoclast-like cells, this peptide precipitated high amounts of two tyrosine-phosphorylated proteins, which were identified as paxillin and Pyk2 (Fig. 6B, D). Two other cytoskeletal proteins, talin and vinculin, did not bind to this peptide (Fig. 6B and data not shown). Peptides containing single amino-acid substitutions (S-752-P; Y-759-A), previously reported to compromise β3 integrin function (Chen et al., 1992EF12; O'Toole et al., 1995EF33; Ylänne et al., 1995EF60; Schaffner-Reckinger et al., 1998EF40) or a peptide containing the homologous region of the integrin β1 tail bound these proteins only very weakly or not at all (Fig. 6B). Experiments with lysates of suspended and attached cells indicated that the interactions with both paxillin and Pyk2 did not correlate with the extent of adhesion-triggered tyrosine phosphorylation (Fig. 6C). Moreover, Pyk2 binding to the β3 cytoplasmic tail was unaffected by immunodepleting cell lysates for paxillin (Fig. 6D). This indicates that it was independent of the known interaction between Pyk2 and paxillin (Schaller and Sasaki, 1997EF41). Finally, binding of both latter proteins to the β3 tail was observed in additional experiments using the chicken β3-tail peptide (shown in Fig. 6A) with lysates of a murine monocyte-derived cell line RAW 264.7 and of human 293T cells transfected with human Pyk2 (data not shown).

Purified glutathione-S-transferase fusion proteins of full-length human paxillin and of the N-terminal part of Pyk2 (amino acids 1-407) also strongly bound to the unmodified integrin β3 tail peptide, whereas glutathion-S-transferase alone showed negligible binding (Fig. 7). Therefore, both interactions appear to be direct.

To further confirm these results, we used recombinant structural mimics of integrin β tails containing full-length human cytoplasmic domains in similar experiments (Fig. 8; Pfaff et al., 1998EF35). These studies revealed again significant binding of both paxillin and Pyk2 in lysates of chicken osteoclast precursors (Fig. 8) and in mammalian cell lines (data not shown) to the integrin β3 cytoplasmic domain. We also noted somewhat reduced, but still significant binding of Pyk2 and paxillin to the full-length β1 tail (Fig. 8). As expected from other studies, both full-length β1 and β3 tails bound to talin, and talin binding was disrupted by tyrosine-alanine mutations in their membrane-proximal NPXY motifs (Pfaff et al., 1998EF35; Calderwood et al., 1999EF7; Fig. 8). Interestingly, this same mutation also abolished paxillin and Pyk2 binding to the integrin β3 tail, but not to the integrin β1 tail (Fig. 8), suggesting differing binding requirements for β1 and β3 tails. Therefore, these data provide evidence for novel and direct interactions of the integrin β3 tail C-terminal region with paxillin and Pyk2.

We have analyzed the podosome organization in peripheral blood monocyte-derived cells during their in vitro differentiation to osteoclasts. We present the following observations.

1. Osteoclast differentiation induced by RANKL-ODF is accompanied by profound changes in the distribution, density and tyrosine phosphorylation of podosomes. These changes could correlate with the acquisition of novel podosome functions related to their role as precursors of the sealing zone in bone-resorbing osteoclasts.

2. The protein paxillin is the major acceptor of cell adhesion-induced tyrosine phosphorylation in early and late osteoclast precursors. It colocalizes with the integrin αVβ3 and the protein kinase Pyk2 in the juxtamembrane region directly adjacent to podosomes of human and chicken osteoclast-like cells.

3. Both paxillin and Pyk2 bind in vitro to the C-terminal 17 amino acids of the integrin β3 tail. Both of these interactions are direct and not affected by adhesion-induced tyrosine phosphorylation of paxillin and Pyk2. Moreover, they can be observed in lysates of chicken as well as of mammalian cells.

Our results suggest key roles of these three proteins in the regulation of cell adhesion-triggered cytoskeletal organization and signal transduction in podosome-containing cells derived from the monocytic lineage.

During RANKL-ODF-induced osteoclast differentiation of peripheral blood monocytes, podosomes formed early in macrophage-like precursor cells. At this stage, podosomes were found throughout the basal cell body (Fig. 1). During subsequent culture, the cells increasingly spread and began to fuse; concomitantly, their podosomes redistributed towards the cell periphery. This change in podosome distribution might reflect a change in the mechanism of podosome-mediated cell movement: we have observed that mononuclear osteoclast precursors migrate as `whole cells', whereas the movement of large multinucleated osteoclast precursors rather consists in extensions and retractions of peripheral cell parts (S. Ory and P. J., unpublished). In these osteoclast precursors, the podosome tips facing the extracellular substrate are strongly enriched in tyrosine-phosphorylated proteins. RANKL-ODF-treatment reinforced the peripheral podosome distribution and led to more tightly packed podosomes, which occasionally fused to larger F-actin containing aggregates. This possibly reflects initial differentiation of the podosome ring to a functional sealing zone, which normally forms only in bone-resorbing osteoclasts (Väänänen and Horton, 1995EF53). The densely packed podosomes in RANKL-ODF-treated cells contained reduced amounts of phosphotyrosine, which did not appear to be topologically associated with the prominent F-actin containing structures. Hence, the osteoclastogenic factor RANKL-ODF induced discrete alterations in the microarchitecture of podosomes and in their association with tyrosine-phosphorylated proteins.

However, a very similar set of proteins in cell lysates was phosphorylated on tyrosine in a cell-adhesion-dependent manner in both RANKL-ODF-treated and untreated cells (Fig. 4A). Thus, changes in the cellular distribution (Fig. 3) rather than in the tyrosine phosphorylation of these proteins predominate during osteoclast differentiation. Most of the phosphotyrosine generated in adherent and spread macrophages or osteoclasts was detected in the LIM protein paxillin (Turner, 1994a), a scaffolding protein with binding sites for many proteins, including the cytoplasmic protein tyrosine kinases FAK, Pyk2 and c-src (Schaller and Sasaki, 1997EF41; Weng et al., 1993EF56; Sabe et al., 1994EF38), the protein-tyrosine phosphatase PTP-Pest (Shen et al., 1998EF43), and the cytoskeletal protein vinculin (Turner and Miller, 1994EF51). Paxillin also provides a link between regulators of p21 GTPases and adhesion complexes by its binding to a protein complex containing the p21 GTPase-activated kinase (PAK) and the guanine nucleotide exchange factor PIX (Turner et al., 1999EF52). Thus, paxillin is in a key position to regulate adhesion-dependent cytoskeletal organization and signal transduction. So far, its role in osteoclast adhesion and in podosome function has not been thoroughly addressed, although several recent studies noticed its colocalization with F-actin, vinculin, Pyk2 and p130cas in the sealing zone of mouse osteoclasts (Lakkakorpi et al., 1999EF22; Duong et al., 1998EF15). In addition, a 70 kDa protein, probably identical to paxillin, has been identified as predominant tyrosine-phosphorylated adhesion substrate in human monocytes (Lin et al., 1994EF24). We now identify paxillin as the major tyrosine-phosphorylated protein in differentiating osteoclasts. It co-distributes with the integrin αVβ3, the non-receptor tyrosine kinase Pyk2 and with many other proteins, which are typically found in focal adhesion complexes, in the membrane-proximal region immediately adjacent to podosomes (Fig. 5; M. P. and P. J., unpublished; see also Lakkakorpi et al., 1999EF22; Duong et al., 1998EF15; Nakamura et al., 1999EF30; Helfrich et al., 1996EF16; Zambonin-Zallone et al., 1989EF61; David-Pfeuty and Singer, 1980EF13; Marchisio et al., 1984EF26; Marchisio et al., 1987EF27).

Cell adhesion phenomena associated with podosomes are not very well characterized. It remains to be demonstrated in molecular detail, how podosomes contact the extracellular substrate. IRM techniques have revealed that podosomes contact the substrate within a small ring (a rosette) co-distributing with the protein vinculin around the central actin core (David-Pfeuty and Singer, 1980EF13; Marchisio et al., 1984EF26). This topology of the contact zone of a podosome is consistent with the subcellular location of paxillin, Pyk2 and the major osteoclast integrin, αVβ3, revealed in this study. It implies that strong integrin-dependent adhesion and cytoskeletal linkage occur predominantly in this delimited region, where adhesion receptors most closely approach the central actin-rich podosome core. Moreover, these locally restricted cell adhesive contacts could also trigger the strong protein tyrosine phosphorylation that we observed at the podosome tips. This phosphotyrosine staining observed in situ could correspond to the predominant tyrosine phosphorylation of paxillin detected in lysates of adherent osteoclast precursors, assuming that only a small subpopulation of paxillin molecules located near the podosome contact zone is involved.

We report for the first time that the 17 C-terminal residues of the integrin β3 chain strongly and directly bind to both paxillin and Pyk2 (Figs 6,7,8). An interaction between paxillin and integrin β cytoplasmic domains has been suggested in earlier studies (Schaller et al., 1995EF42; Tanaka et al., 1996EF47). Schaller et al. reported paxillin binding to the membrane-proximal regions of integrin β1, β2 and β3 tails, but they did not provide evidence that the interaction was direct (Schaller et al., 1995EF42). Tanaka et al. demonstrated direct paxillin binding to the integrin β1 cytoplasmic domain (Tanaka et al., 1996EF47). Consistent with the former studies, we detected paxillin binding to full-length β1 tails, which was not abolished by a point mutation adjacent to and not overlapping the proposed binding motif (Schaller et al., 1995EF42; Fig. 8). In contrast, our results show that paxillin interacts in a clearly different way with the integrin β3 tail. A peptide consisting of its 17 C-terminal amino acids alone contains strong paxillin binding activity and mutations at three critical positions in its C-terminal region abrogate this interaction. The serine(752)-proline mutation has been identified in a patient with Glanzmann's thrombasthenia and resulted, like the tyrosine-alanine mutations at positions 747 and 759, in compromised capacities of the integrin to become competent for ligand binding, to localize to focal adhesion plaques and to promote cell spreading (Chen et al., 1992EF12; O'Toole et al., 1995EF33; Ylänne et al., 1995EF60; Schaffner-Reckinger et al., 1998EF40). Hence, our results provide a potential molecular explanation for at least some of the defects observed with these mutants, notably those that interfere with cell spreading and focal adhesion localization.

We also observed strong binding of the protein kinase Pyk2 to the integrin β3 tail. Although the binding requirements in the β3 cytoplasmic domain were very similar for Pyk2 and for paxillin, we excluded the possibility that Pyk2 binds via paxillin to the β3 tail (Fig. 6D). By using GST-fusion proteins, we obtained additional evidence indicating that both interactions are indeed independent and direct (Fig. 7). However, the requirements for β3 tail binding to paxillin and Pyk2 differed strongly from its binding to talin. This was expected, as talin binding requires more membrane-proximal regions of integrin β tails, which are absent in our synthetic peptides (Tapley et al., 1989EF48; Patil et al., 1999EF34).

We further show that neither paxillin nor Pyk2 binding to peptides containing the C-terminal 17 amino acids of the integrin β3 cytoplasmic domain are altered by their cell adhesion-triggered tyrosine phosphorylation (Fig. 6C). Therefore, tyrosine phosphorylation of paxillin and Pyk2 is unlikely to provide a direct regulatory cue for their binding to the β3 integrin cytoplasmic domain. This binding has then to be regulated by other means, for example by the availability of free β3 tails. Previous studies indicate indeed that the availability of integrin β tails for cytoskeletal interactions is constrained in unoccupied integrins. This constraint involves integrin α tails and it is released during binding of the extracellular ligand (LaFlamme et al., 1992EF20; Briesewitz et al., 1993EF5; Ylänne et al., 1993EF59). Pyk2- and paxillin-binding to the integrin β3 tail could thus trigger their recruitment to β3 integrin-dependent cell contacts. A major current issue will now be to study the role of this novel interaction observed primarily in a cell-free system in the context of intact cells.

On the basis of these observations, podosomes emerge as dynamic cytoskeletal structures with many molecular and functional homologies to focal adhesions. Both adhesion structures use a similar set of adhesion receptors and of adaptor proteins that link the extracellular contact to the actin cytoskeleton as well as to signal transduction pathways. The tyrosine kinase Pyk2, which is normally absent in focal adhesions (Schaller and Sasaki, 1997EF41), could functionally replace in podosomes the focal adhesion kinase FAK, which is only weakly expressed in osteoclasts and in cells of the monocytic lineage (Duong et al., 1998EF15; Lin et al., 1994EF24; Li et al., 1998EF23). The integrin αVβ3, which is highly expressed during later stages of osteoclast differentiation, might be important to adapt podosomes to their function in osteoclast migration on bone and in the formation of the sealing zone during bone resorption (McHugh et al., 2000EF29). Finally, the molecular complex that contains paxillin and Pyk2 bound to the C terminus of the integrin β3 chain could represent a molecular core structure that governs the distribution of regulatory cues linking integrin-dependent cell adhesion to podosome functions in osteoclasts.

We kindly acknowledge the expert help of Damien Ficheux (Institut de Biologie et Chimie des Protéines, Lyon) during the synthesis and characterization of peptides, and of Wilfried Condemine during the preparation of recombinant GST-paxillin. We are grateful to Mark Ginsberg, David Calderwood, Shouchun Liu and David Schlaepfer (The Scripps Research Institute, La Jolla, CA), Ravi Salgia (Dana-Farber Cancer Institute, Boston, MA), Michael Horton (London, UK), and to Immunex (Seattle, WA) for providing reagents.